Entitled to Work: Urban Property Rights and Labor Supply in Peru

Transcription

1 Entited to Work: rban Property Rights and Labor Suppy in Peru Erica Fied Harvard niversity This version: Juy 003 Abstract: Over the past decade, the Peruvian government issued property tites to over 1. miion urban househods, the argest government titing program targeted to urban squatters in the deveoping word. This paper examines the abor market effects of increases in tenure security resuting from the program. In particuar, I study the direct impact of securing a property tite on hours of work, ocation of entrepreneuria activity and chid abor force participation. To isoate the causa roe of ownership security I make use of differences across regions induced by the timing of the program and differences across target popuations in the eve of pre-program tenure security. My estimates suggest that titing resuts in a substantia increase in abor hours, a shift in abor suppy away from work at home to work in the outside market and substitution of adut for chid abor. For the average squatter famiy, granting of a property tite is associated with a 17% increase in tota househod work hours, a 47% decrease in the probabiity of working inside the home, and a 8% reduction in the probabiity of chid abor. Keywords: Property rights, and titing, deveopment poicy, urban economics, time aocation and abor suppy, empoyment determination and creation JEL Categories: P14, Q15, J0, J, R0, O18, O54 ACKNOWLEDGEMENTS: I am indebted to Hank Farber and Anne Case for generous support throughout this project and to Danie Andauz in the COFOPRI office for providing the survey data. I aso thank Attia Ambrus, David Autor, Meissa Cark, Javier Escoba, Eszter Hargittai, Chang-Tai Hsieh, Jeff King, Lewis Kornhauser, David Linsenmeier, Kristin Mammem, Aex Mas, Ted Migue, Ceci Rouse, Máximo Torero, Diane Whitmore, IRS abor unch, RPDS workshop, GRADE seminar and NY cooquium participants for numerous usefu comments.

2 1 Introduction Strengthening economic institutions is widey argued to foster investment in physica and human capita, boster growth performance, reduce macroeconomic voatiity and encourage an equitabe and efficient distribution of economic opportunity (Acemogu et a., 00; North,1981). As one of the basic roes of institutions and fundamenta to a economic transactions, codifying and protecting property rights is seen in many academic discussions as requisite for economic deveopment and poverty reduction. 1 Among poicy-makers as we, property titing is increasingy considered one of the most effective forms of government intervention for targeting the poor and encouraging economic growth (Baharogu, 00; Binswanger et a, 1995). Despite the consensus on the importance of institutiona factors for economic performance, there is a shortage of reiabe estimates of the infuence of property reforms on a range of market outcomes. This paper studies the impact of property rights on abor markets in deveoping countries by anayzing househod abor suppy responses to exogenous changes in forma ownership status. In particuar, I assess the vaue to a squatter househod of increases in tenure security associated with obtaining a property tite in terms of hours of abor suppy gained and improved efficiency of abor aocation between home and market work and between chid and adut abor. An obstace to measuring the infuence of tenure security is the potentia endogeneity of ownership rights. I circumvent the probem by using data from a dramatic natura experiment in Peru, in which a nationwide program issued forma property tites over a five-year period to more than 1. miion urban househods. This approach in arge measure breaks the ink between tenure security and income and heps isoate the causa effect of property titing on market outcomes. Athough no pane data are avaiabe on program participants, extensive cross-sectiona data were coected on past and future tite recipients midway through the program, generating a natura set of comparison groups composed of treated and yet-tobe-treated househods. The Peruvian titing program constitutes the first arge-scae urban property rights reform that has occurred in the deveoping word, and its impact has impications for many deveoping countries in which urban squatting is a widespread phenomenon. An important contribution of this paper is the specific focus on non-agricutura househods and the vaue to urban residents of increased ownership security. In deveoping countries, arge proportions of urban and rura residents aike ack tenure security. Yet, presumaby because of historic interests in agricutura investment and reated poitics of and reform, the majority of both academic and poicy 1 See, generay, Demsetz (1967), Achian and Demsetz (1973) and Sheifer et a. (001). Direct evidence of this is provided by Micei et a. (001), who anayze the extent of endogeneity of forma agricutura property rights in Kenya. 1

3 attention to property rights reform has centered on rura househods tenure insecurity. Nevertheess, in most of the deveoping word, the popuation and particuary the impoverished popuation is increasingy urban. 3 Though advocates of urban property reform cite many of the same benefits to and titing for non-agricutura as for farm househods, the reationship between tenure security and economic efficiency is ikey to be distinct in the urban setting. In particuar, as wi be addressed in this paper, there is cause to beieve that urban empoyment eves are particuary sensitive to the degree of residentia formaization. In this manner, the paper aso contributes to the iterature by examining a unique aspect of the wefare gains to property titing: the effect of improvements in tenure security on abor suppy and abor aocation decisions within the househod. The fundamenta consequence of successfu residentia formaization is a reduction in the househod s ikeihood of forced eviction by the government or expropriation by other residents. As ong as untited househods expend their own human resources in an effort to soidify informa caims to and, the acquisition of a property tite has direct vaue in terms of freeing up hours of work previousy devoted to maintaining tenure security through informa means and securing forma rights. As the foowing quote iustrates, there is ampe anecdota evidence that urban squatters are commony constrained by the need to keep a famiy member at or cose to home to protect against residentia property invasion: I go to work, and my mother ooks after the house, says Aejandrina Matos Franco, who ses cassettes on the street in Lima and who worries that peope coud seize her house when she is away. (Conger, 1999) In addition, the ega process of acquiring forma property tites traditionay invoved substantia monetary and time costs. 4 Both factors ceary raise untited househods abor needs for production of home security and in turn the opportunity cost of empoyment outside the home. As a resut, untited househods make constrained decisions in aocations of eisure, home production, and the amount of chid reative to adut abor. To study these reationships, I impement a quasi-experimenta empirica strategy using crosssection micro-data from a survey of past and future beneficiaries of the Peruvian titing program. Two 3 In Latin America and the Caribbean, for instance, the popuation shifted between 1950 and 000 from 41% to 75% urban (nited Nations, Word rbanization Prospects: The 1999 Revision, 000). 4 According to one report, In Peru, the process of getting a deed from the bureaucracy invoved 07 steps divided among 48 government offices, took an average of 48 months to compete, and was too expensive for sma property owners. (Economist, 1995)

4 sources of variation in program infuence are used to isoate the effect of titing: neighborhood program timing and program impact based on prior househod ownership status. In particuar, staggered regiona program timing enabes a comparison of househods in neighborhoods aready reached by the program with househods in neighborhoods not yet reached. Meanwhie, variation in pre-program tenure security aows residents not subject to changes in security to serve as a quasi-contro group for residents who experience reativey arge changes as a resut of the program. The fact that the program targeted neary a untited househods regardess of househod demand for forma property rights aso enabes a broader exporation of heterogeneity in response to the program. Heterogeneity in the demand for property tites has been shown to depend heaviy on factors which contribute to the cost of maintaining informa rights. 5 For this purpose, both residentia tenure a proxy for informa tenure security and househod size are used as indicators of the reative vaue of a property tite for a given househod. Given that overa de facto property rights are observed to increase with residentia tenure, the vaue of a property tite and therefore the program impact shoud be ower for househods with onger residentia tenure (De Soto, 1986). Likewise, since (for a given property size) househods with more aduts have greater capacity to provide home security, the tenure security vaue of a forma tite shoud be ower for arger famiies. Severa interesting findings emerge. My estimates of eary program impact suggest that househods with no ega caim to property spend an average of 16. hours per week maintaining informa tenure security, refecting a 17% reduction in tota househod work hours for the average squatter famiy. Aso, househods are 47% more ikey to work inside of their home. Thus, the net effect of property titing is a combination of an increase in tota abor force hours and a reaocation of work hours from inside the home to the outside abor market. My estimates further support the predictions that informa property rights and househod size infuence the home security demands facing an untited househod. For a abor suppy measures, the effect of obtaining a property tite is decreasing in residentia tenure and in the number of working-age househod members. Finay, for househods with chidren, urban and titing is associated with a 8% ower probabiity of chid abor force participation. The resuts are particuary convincing in ight of a number of possibe downward biases. The next section of the paper reviews the theoretica and empirica iterature on and rights in deveoping countries. The third section describes the titing program in greater detai. The fourth section presents a mode of househod abor suppy in which, under very genera conditions, tota abor suppied 5 In fact, heterogeneity in the demand for property tites is modeed expicity in Micei et a. (001). 3

5 to the outside market unambiguousy rises with an increase in forma property rights, and both abor hours in home production and chid abor unambiguousy fa. The fifth section describes the empirica mode and discusses the identification strategy for program effect. The sixth section presents resuts and robustness checks. The seventh section discusses ong-run predictions and the eighth section concudes. Reated iterature There exists a wide body of iterature demonstrating the positive infuence of property institutions on market outcomes. Severa macroeconomic anayses have shown a reationship between economic deveopment and cross-country variation in institutiona strength, which encompasses property institutions (Knack et a., 1995; Mauro,1995; Ha et a.,1999; Rodrik,1999). In the microeconomic iterature, the ink between property rights and wefare enhancement has generay been confined to three channes estabished in a semina paper by Besey (1995) that expores the benefits of ownership rights for agricutura househods. These are: increased tenure security and greater investment incentives, ower transactions costs and gains from trade in and, and greater coatera vaue of and and improved credit access. The reationship between and rights and abor markets has been mentioned ony in the context of residentia mobiity and abor market adjustment, a coroary impication of higher transaction costs in rea estate (Yao, 1996; Word Deveopment Report, 000; Moene, 199). Empirica estimates of the vaue of property tites in agricutura settings corroborate these predictions. Studies such as Aston et a. (1996), Lopez (1997) and Carter and Ointo (1997) ink and tites with improved credit access, whie many authors incuding Feder (1998), Besey (1995), Banerjee et a. (00) and Aston et a. (1996) provide evidence that ack of property tite indeed affects agricutura investment demand. 6 In urban settings, the vaue of property tites has been measured far ess often and empirica work has focused primariy on rea estate prices. A major contribution is a paper by Jimenez (1984), invoving an equiibrium mode of urban squatting in which it is shown that the difference in unit housing prices between the non-squatting (forma) sector of a city and its squatting (informa) sector refects the premium associated with tenure security. The accompanying empirica anaysis of rea estate markets in the Phiippines finds equiibrium price differentias between forma and informa sector unit dweing prices in the range of 58%, and greater for ower income groups and arger househods. 6 Other work, such as Migot-Adhoa et a. (1998) and Kimuyu (1994) detect itte impact of and titing on investment. The mixed resuts are commony attributed to the difficuty of addressing the endogeneity of tite status. 4

6 Consistent with the agricutura investment iterature, Hoy and Jimenez (1996) find that and tites are aso associated with greater oca pubic goods provision in squatter communities in Indonesia. A separate ine of research on property institutions reates to the roe of informa or de facto property rights. A number of authors such as Carter (1994, 1996) and Gaa and Razzaz (001) note that, in many settings, informa institutions arise to compensate for the absence of forma property protection. Thus, ega enforcement constraints are binding ony insofar as they correspond to rea tenure insecurity. Lanjouw and Levy (00) find that eves of informa property rights vary greaty in urban communities in Ecuador, and de facto tenure security varies systematicay with observabe househod characteristics such as sex of househod head and ength of residence. In addition, their paper demonstrates that the vaue of a forma tite can be overestimated by rea estate price differentias when non-transferabe informa rights are ignored. In my paper, the concept of informa rights is further extended to comprise not ony exogenous househod characteristics, but aso security investment choices made by individua househods. Incuding resource aocation decisions in the definition of informa protection mechanisms has the additiona effect of narrowing the bias in rea estate price differentias between tited and untited properties in the Lanjouw and Levy mode. 3 Project Background This paper examines the effects of the Peruvian government s recent series of ega, administrative and reguatory reforms aimed at promoting a forma property market in urban squatter settements. Peru's informa urban settements grew out of the massive urban-rura migration that occurred over the ast hafcentury as a resut of the coapse of the rura economy (due in part to a faied and reform program) and the growth of terrorism. The existence of extensive barren and owned by the state on the perimeters of major cities aong with an impicit housing poicy during the 1980s that aowed squatter settements on unused government ands ed to an extended era of urban migration, often in the form of organized invasions by squatters from the same area of emigration (Oórtegui, 001). 7 It is estimated that in 1997, a quarter of Peru s urban popuation ived in margina squatter settements in peri-urban areas and many more untited residents occupied inner-city neighborhoods (Word Bank, 1997b). 8 7 Invasion of privatey-owned property was aowed by aw if the and had been unused for a period of four years. The aw has since changed (in 1990) so that invasions of private property are not aowed under any circumstances. 8 See Appendix A for a country map of the untited popuation and properties targeted for formaization. 5

7 Prior to the reforms, obtaining a property tite for a Peruvian househod was neary impossibe due to heavy bureaucratic procedures and prohibitive fees. As described in the initia project report: Peru s traditiona system of titing and registration is compex, inefficient, expensive prohibitivey so for poor peope and prone to rent-seeking. Fourteen different agencies are invoved in the generation of each tite, the courts have rarey been abe to vaidate these tites as the aw requires (Word Bank, 1998a). 9 Due to acute housing shortages and ack of ega transparency, tenants strugged not ony with the government but aso among themseves to secure residentia properties. The common faiure of the government to defend or even recognize informa tenure rights in individua disputes gave rise to rentseeking behavior in the form of invasions of untited and (Oórtegui, 001). In 1991, a Peruvian non-governmenta organization embarked on an innovative property titing project in the capita city of Lima whose goa was the rapid conversion of informa property into securey deineated and hodings by the issuing and registering of property tites (Word Bank, 1998b). Between 199 and 1995, roughy 00,000 tites were issued at an extremey ow cost, convincing the government and a growing internationa audience of the potentia for efficiency gains from urban property formaization (Word Bank, 1998a). In 1996, under the auspices of the pubic agency COFOPRI (Committee for the Formaization of Private Property) and Decree 44: Law for the Formaization of Informa Properties, the Peruvian government estabished a nationa property registry based on the eary mode to formaize the remaining properties in Lima and extend the program to seven other cities. 10 Just as in the piot project, impementation of the nationa program invoved area-wide titing by neighborhood, which was presumed to foster, through community participation and education, a demand for formaization, reduce the unit cost of formaization, and rapidy generate a minimum critica mass of beneficiaries (Word Bank, 1997c). Whie the od process of acquiring a property tite was prohibitivey sow and expensive, the new process was free and extremey rapid. Once a oca property registration system was set up, oca program officias were trained, and the city s target areas were propery identified and mapped, severa project teams simutaneousy entered neighborhoods starting from 9 In his groundbreaking study of the underground economy, economist Hernando de Soto documented the same phenomenon: In The Other Path, de Soto and aids concuded that to get tite to a house in an informa settement whose permanence the government had aready acknowedged took 78 steps from one agency aone, and ten other agencies aso required approva (Rosenberg, 000). 10 According to the Word Bank Project Appraisa Document (1998), target cities were chosen according to a formua based on city size, density of informa settements, and distance from commercia centers, measures indicating the ikey ease and cost of formaization and the expected poverty impact. 6

8 different points in the city. 11 To be eigibe for program participation, tite caimants were required to verify residency predating 1995, and had to ive on eigibe pubic properties. 1 As a resut of the reforms, by December 001 neary 1. miion of the country s previousy unregistered residents became nationay registered property owners, affecting approximatey 6.3 miion of the roughy 10 miion untited residents iving in the range from just above to beow the poverty ine. 13 In the ream of iterature on the economic benefits of tenure security, the Peruvian experience provides a unique research opportunity for many reasons. Briefy, the nationa formaization pan constitutes a one-of-a-kind natura experiment wordwide in terms of providing neary cost-free improvements in ownership security on such a arge scae. Furthermore, unike many arge-scae government programs, the titing efforts took pace at an extremey rapid pace, which faciitates program evauation by eiminating much of the need to consider time trends that coud obscure the independent effects of program participation. At the same time, in the absence of pane data on participating househods, the fact that program timing was staggered proves to be an asset for evauation purposes. A survey of 750 urban househods was conducted in March 000 midway through program impementation. Because the sampe was drawn from the universe of a target popuations for eventua program intervention, the data contain a number of househods in neighborhoods in which the program has not yet entered. 4 Conceptua framework 4.1 Tota Househod Labor Suppy This section presents a simpe variation of the basic agricutura househod mode to formaize the 11 In campaigns of two months each, project teams entered 50 to 70 neighborhoods encompassing roughy 30,000 to 35,000 pots. Within a neighborhood, teams spent five to seven weeks estabishing residentia caims and deineating properties before conferring state-registered property tites onto a eigibe residents. The registration process for these tites took an additiona period of one to six months. 1 Ineigibe properties incuded archeoogica sites and food panes, among other exceptions see page 15 for description. In the COFOPRI data, 9.4% of samped househods are ineigibe according to reported ength of residence, and an additiona 10% remain untited after severa years of program operation. 13 Though the grant period is not yet over unti December 00, thus far, 1.64 miion ots have aready been formaized and 1.1 miion tites granted, the vast majority of which took pace between 1998 and 000. Whie no residents who previousy possessed registered municipa tites are incuded in this figure, it is uncertain what fraction of this number had ocay registered saes documents before the nationa reforms as these househods were incuded in the government s definition of untited, though in reaity the program simpy transferred such tites to the nationa registry. In my paper, the term squatter refers ony to househods with no saes or judicia tites prior to the reforms, which is estimated to be 37% of the target popuation. 7

9 intuition that, in a setting of incompete property rights, the standard abor-eisure choice wi be infuenced by househod demand for security of property. The main innovation is the incorporation of a tenure security function, s( ), into the utiity function, such that both eisure and home production enter househod utiity through two separate channes: through their respective consumption and production vaues and through their effect on home security. 14 Furthermore, the security vaue of time at home is sensiby modeed as a househod pubic good, such that individua utiity depends on the eisure and home production hours of a other members via s ( ). In this framework, utiity, given a set of househod characteristics ψ and resource endowment E, is an increasing function of per capita eisure, consumption, and home security, and home security is determined by the foowing three parameters: tota hours of househod time at home (time spent by famiy members protecting property), an exogenous parameter, θ, which refects the househod s eve of forma property rights, and a summary measure, τ, which refects the degree of informa or de facto rights the househod has acquired. For tractabiity, I make the foowing set of assumptions. First, the househod is assumed to maximize per capita eisure and not the eisure of individua members. Given that this mode is concerned with the effect of θ on tota househod abor, ignoring the second stage of the househod decision probem in which eisure is aocated across individua members is inconsequentia to the centra resuts. Second, there is no outside abor market for the provision of home security. Assuming a missing abor market for property protection is easiy justified by an incompete contracts argument (there is risk invoved in empoying non-members to guard property), athough a more compicated mode woud have this market depend on θ. 15 Furthermore, whie the mode does not expicity incude hired security, there is room to incorporate the existence of a back market for property protection into τ. Fourth, as opposed to modes of joint production such as Graham and Green (1984), in this mode eisure and home production hours are assumed to be perfect substitutes for the hours an individua spends on property protection. 16 Finay, this is a unitary househod mode, and it is assumed that a househod members face a common wage, w. 14 As opposed to modes of joint production in the vein of Gronau (1977), I assume incompete substitution between market goods and home security due to the absence of an outside market for home security protection. 15 Additionay, extension of this mode to a more compicated setting in which there is an imperfect (as opposed to nonexistent) market for the provision of home security is inconsequentia under the uniform wage assumption. 16 Whie this assumption might seem unreasonabe in ight of the fact that eisure time which contributes to home security is constrained reative to eisure which can be spent inside or outside of the home, incorporating a jointness function which measures the psychic vaue of home reative to market production does not change the comparative statics of the mode. 8

10 Let N be the number of househod members, and i be eisure, x i consumption, h fi abor hours in home production, and h oi outside abor hours of househod member i, and L = N i i = 1 N, H f = h fi, H o = h oi, X = x i, i= 1 N i= 1 N X x =, i= 1 N L =, Z = H f + L. N Labor hours of househod members are divided between work at home ( H f ) and work in the outside market ( H o ). Time spent at home (Z) is divided between work at home ( H f ) and eisure (L). The vaue of abor at home is given by the production function q( H f the home is the market wage w. 17 Househod utiity is then given by: ), whie the vaue of work outside ( x,, s; ψ, E), where s = s ( Z, θ, τ ). Here ( ) and s( ) are twice continuousy differentiabe, concave, and increasing in each argument. 18 Whie the tenure security function impies that the production of home security is determined purey by exogenousy given and rights (θ and τ ) and the amount of time spent in the home, s ( ) coud easiy be extended to incude other househod inputs such as secure ocks and doors. The parameter θ can be thought of either as a binary indicator of a egay registered property tite, or ese a more nuanced parameter which refects the eve of forma ega recognition of a househod s tenure status (eve of efficiency of court systems, eves of poice cooperation, etc.). The choice variabes for the househod are: H, H, X, L and s. The constraints to the maximization probem are: f o s = s( H f + L, θ, τ ) px = wh + q T = L + H o + H f o ( H f L, H, H, X 0 o f ) 17 Incorporating a market for hired abor in home production does not affect the mode s predictions. Inseparabiity in this mode comes from the ack of substitutabiity of househod members in the production of security, not q( ). 18 I assume that security inputs ( Z,θ and τ ) are substitutes in production, and make corresponding assumptions on the cross-partia derivatives of s( ). 9

11 where q( ) satisfies decreasing margina productivity ( q > 0, q < 0 ). Then, normaizing prices to one, the househod s optimization probem can be written: 19 1 max ( ( w* H Ho, H f N o + q( H f 1 )), ( T H N o H f ), s( T H o, θ, τ )) This yieds the foowing necessary first-order conditions for an interior soution ( H > 0; H > 0; H + H T ) : o f o f < 0 w 1 * x = * + S * sh N N o (1) q f * = H x () Equation 1 estabishes that, at the optimum, househods equate the margina vaue of an additiona hour of outside abor with the margina utiity of eisure. Equation states that they aso equate the margina utiity of eisure with the margina vaue of an additiona hour of work at home. For each househod invoved in both home and market work, the soution to this set of equations impicity defines demand functions for abor hours in the outside market and in home production which depend on θ, w, and τ : * f = * f * o = * o H H ( w, θ, τ ), H H ( w, θ, τ ) Assume that 0, 0, 0. 1 Then tota differentiation yieds the foowing inequaities xs for vaues of w, θ, and τ corresponding to inner optima: x s H f θ < 0 H o and > 0 θ 19 For the remainder of the anaysis, househod characteristics and resource endowment are assumed to be fixed and omitted from the arguments of the utiity function. 0 The boundary conditions and guarantee that (H f + H o )<T and that at east one of H f 0 x x 0 and H o is stricty positive. It is shown on the foowing page that the corner soutions H f =0 and H o =0 do not affect the aggregate predictions of the mode. 1 Note that this incudes the additivey separabe case. 10

12 For househods invoved in both types of abor, an increase in forma tenure security decreases work hours at home and increases work hours in the outside market. At the corner soution H o =0, 0 and θ H o H 0, and at the corner soution H f =0, f H = 0 and o > 0. Thus, in aggregate, strengthening θ θ θ forma property rights decreases work hours at home and increases hours outside the home. Detais of the comparative statics are provided in Appendix B. Intuitivey, this refects the fact that an exogenous increase in the eve of forma property rights corresponds to a decrease in the househod s need to spend time on home security, thereby owering the opportunity cost of outside abor force hours. H f In the empirica anaysis, data imitations prevent me from separating empoyment hours inside and outside of the home. With respect to the net effect of a property tite on tota empoyment hours, my mode predicts that househods with zero home production hours ex ante (H f = 0) wi increase tota househod abor hours by some positive amount in response to stronger forma property rights. For househods with any amount of abor hours devoted to a home business, the net effect on tota hours is ambiguous. Whie the eve of outside work hours wi unambiguousy rise for househods invoved in both types of production, the resuting change in average houry earnings arising from the difference between wages earned in the externa abor market and the margina productivity of abor in home production wi generate both income and substitution effects. The net change in tota abor hours, ( H + Ho), wi depend on the reative sizes of these effects. In the empirica section, due to the fact f that ony 5% of househods are invoved in home production, the program effect on househods working outside the home is presumed to dominate the possibe negative effect on househods with home businesses. Thus, I predict ex-ante that a titing program wi be associated with an increase in tota empoyment hours. At the same time, I wi expore the effect on househods with home businesses by studying the probabiity that a househod uses their residence as a source of economic activity. Since work hours inside the home are predicted to fa unambiguousy, so shoud the percentage of househods that spend any time working at home. Two auxiiary impications foow from this mode. First, the effect of a change in forma property rights on abor suppy is decreasing in the househod s eve of informa property rights, τ : It is important to note at this point that I have ignored the consumption vaue of home security via its infuence on the market price of tradabe assets, which has a potentia income effect on abor suppy that coud counteract the impication stated in equation (). This is justified by two considerations: first, rea estate markets are often nonexistent in these settings; second, for the purposes of estimating a abor suppy effect, the possibe income effect of increases in home security which is being ignored biases downwards the effect on abor suppy. Hence, any finding of an effect is a ower bound on the impact of the program. 11

13 * H f > 0 θ τ * H o θ τ and < 0 Second, given average consumption eve x, the effects are decreasing in the number of working-age househod members, N. 3 * H f θ N > 0 * H o θ N < and 0 The intuition behind the famiy size effect is that, the more famiy members iving in a househod, the more ikey it is that someone chooses to stay at home independent of security considerations, thus arge househods are ess distorted by the need to keep watch over the residence. These predictions wi motivate me to test empiricay whether the effect of acquiring a forma property tite on abor suppy differentiay impacts househods of different sizes and with different engths of residentia tenure. 4. Labor Suppy of Chidren An extension of the mode, aso detaied in Appendix B, incorporates differences in the househod suppy of adut and chid abor when ony aduts contribute to home security provision. This extension formaizes the intuitive idea that, if aduts have a comparative advantage in the provision of home security, in the absence of secure property rights, chidren wi substitute for aduts in the abor market. In this case, whie tota househod abor hours rise with an increase in forma rights, chid abor hours wi actuay fa. For simpicity, in the foowing description I ignore the roe of home production, though the resuts hod under very genera conditions when production is incuded. Here, N A and N C are the number of adut and chid househod members, respectivey, A and C are per capita adut and chid eisure, and L are tota adut and chid eisure and T and T are tota adut and chid time endowments. In this C setting, the househod s maximization probem is: A C L A max(x,, A C, x A, C, s( L A, θ, τ )) such that wa ( TA LA) + wc ( TC LC ) = X 3 Given that members of extended famiies often divide their time between househods, some authors treat N as continuous peope hours instead of a discrete number of peope. The same resut can be proven for discrete N. 1

14 The first-order conditions corresponding to each empoyed adut member i and chid member j are: a i cj wa = x + + s s AL A N w = N A C C x + C = 0 = 0 From these conditions it can be shown that, for a interior optima, C > 0 and A < 0. In θ θ househods in which chidren are abor force participants, chid abor hours wi fa and adut abor hours wi rise with an increase in tenure security. For a other househods, adut abor hours wi aso rise and chid abor hours wi remain at zero. Thus, given a positive amount of ex-ante chid abor, the aggregate number of chid abor hours wi unambiguousy fa, whie the number of adut abor hours rises with an increase in forma property rights. 4 Whie the theoretica mode deas with changes in abor suppy at a fixed wage rate, the empirica mode wi capture changes in actua empoyment eves, which are functions of both suppy and demand. Given the size of the program, it is reasonabe to anticipate genera equiibrium effects on the wage rate. However, because increased abor suppy wi decrease the market wage, as ong as eisure is a norma good such effects woud ony bias downward the estimated program effect. Thus, the actua abor suppy response to titing is presumaby higher than what can be measured with changes in working hours. 5 Data and Estimation Methods 5.1 Data Set My empirica anaysis of househod abor suppy responses to changes in forma property rights rey on the COFOPRI baseine survey data. The sampe universe for the survey was a residences in nonincorporated urban and peri-urban settements identified in the 1993 census of the eight cities targeted by 4 Athough this mode focuses on optima abor aocation, the income effects that foow from reaxing the househod s time constraint provide a pausibe aternative expanation for a decrease in chid abor with an increase in forma rights, and one that has been proposed by other authors. In particuar, a decrease in chid abor woud foow from the uxury and substitution axioms of the Basu and Van (1998) mode of chid abor suppy, in which chidren can substitute for aduts in the abor market and a famiy wi send chidren to the abor market ony if the famiy s income from non-chid abor sources fas beow some threshod amount. 13

15 the titing program. The data consist of 750 househods distributed across a eight program cities. The survey was stratified on city, with custer units of ten househods randomy samped at the neighborhood eve within cities. The number of custers drawn from each city was based on the city s share of eigibe residents. The survey instrument cosey mirrors the Word Bank Living Standards Measurement Survey (LSMS) in content, and therefore contains a wide variety of information on househod and individua characteristics. In addition, there are five modues designed to provide information on the range of economic and socia benefits associated with property formaization. 5. Identification Strategy To study the impact of receiving a property tite on househod abor suppy, I expoit variation in the year in which the COFOPRI program entered a neighborhood to compare househods in program neighborhoods that have aready been reached by the survey date to househods in ate program neighborhoods. The first step in cassifying program timing was to identify whether or not a neighborhood had been reached by the time of the survey. The survey data do not directy identify program neighborhoods, nor can this variabe currenty be constructed by matching geographic identifiers to COFOPRI office data. Instead, a observations within a survey custer are assigned a program entry vaue of one if more than one househod in the custer reports owning a COFOPRI tite. 5 Custers in which no househod or ony one househod have a COFOPRI tite are assumed to be those in which the program has not entered, athough it is generay impossibe to separate the neighborhoods in which the program wi never enter from those which wi be treated eventuay. Nonetheess, such neighborhoods share the key feature of no expected program effect. 6 A breakdown of program and non-program neighborhoods by region is provided in Appendix C. Not every squatter househod that the program reaches is granted a COFOPRI tite by the time of the survey. Reasons that househods may be excuded incude: the househod cannot prove residence prior to 5 There is ceary some measurement error in this method of identifying treated neighborhoods. In particuar, it is possibe that neary a residences in the custer were not given tites athough the program did in fact enter the neighborhood. To address this, I aso estimate the mode excuding seven custers in which a samped househods had registered municipa property tites prior to the program, making it impossibe to observe whether or not the program entered. In none of my anaysis does excuding these 69 househods affect the estimate of program effect. 6 Incuding custer units with ony one reported COFOPRI recipient as non-program neighborhoods does not affect the resuts. Since it is extremey unikey that ony one househod is tited in a program neighborhood severa months into the program, such neighborhoods are ikey to refect either misreported tite data or recent program entry. If ony one househod has actuay received treatment, effectivey the neighborhood is at this stage untreated and neighborhood effects shoud not be observed. 14

16 1995; the househod beongs to a cooperative association; the residence ies on an archeoogica site, food pane, mining site or private property; and ambiguous or disputed ownership caims. nfortunatey, none of the above information is coected in the survey. 7 Since the househods in the treated neighborhoods may or may not actuay have received a government tite, this is an intent-to-treat (ITT) anaysis. The second step in cassifying variation in program timing was to identify the year in which the program entered. The effect of the program is presumed to increase over time in a fashion anaogous to a dose response measure from the experimenta design iterature for three reasons: First, titing an entire neighborhood can be a engthy procedure, such that the percentage of tited househods within a treated neighborhood increases (at a decreasing rate) over time. Secondy, househod abor suppy takes time to adjust. Finay, it is pausibe that confidence in the vaue of a COFOPRI tite is increasing over time. For purposes of exporing the program effect over time, year of program entry was defined as the eariest reported COFOPRI tite year within the custer. 8 Dynamic response was restricted to be inear in four time periods: January 1999 June 000, January 1997 December 1998, January 1995 December 1996, and January 199-December This division corresponds to three major waves of program expansion: From 199 to 1995, 00,000 tites were granted by the Institute of Liberty and Democracy as part of a piot project prior to COFOPRI; the first wave of COFOPRI tites was initiated in 1995 in Lima and Arequipa; and beginning in 1997 the program expanded into six other cities. 9 Furthermore, these intervas were consistent with the observed reationship between subjective statements on tenure security and years since program entry, as is reported for squatters in the city of Lima in Appendix D. 30 Athough target areas for wide scae economic deveopment programs are never randomy seected, these data have the advantage that a sampe members ive in areas that wi eventuay be targeted for program intervention, increasing confidence in the comparabiity of treated and untreated househods. Furthermore, the universa nature of the treatment and the participation rues of the program generay 7 According to anecdota evidence from program administrators, disputed caims within famiies or between neighbors are the most common reason that tite distribution is deayed for an untited househod in a treated neighborhood (Caros Gandofo, persona interview, Lima, August 9, 000). 8 Due to the fact that not a househods were given property tites right away and because of measurement error in tite year reporting, househods in the same custer who had received a COFOPRI tite did not necessariy report the same tite year. When the minimum reported tite year fe beow the first regiona tite year according to program data, the second owest tite year was assigned to the custer. 9 This region-specific pattern of intervention makes it important to incude city dummies in regression estimates of program effect. 30 The tabe in Appendix D reveas a tota change in average reported tenure security for residents of Lima of roughy 0.6 points on a four-point scae. The tabe aso iustrates that, whie newer househods have consistenty ower perceived tenure security than more estabished famiies, the change in perceived tenure security foows the same approximate trajectory over time since titing program for both groups. 15

17 rue out concern over individua seection bias that coud arise even if program pacement were random. Nonetheess, there is sti potentia for program timing bias, in which areas seected for eary program participation are different from the rest. If program timing is not randomy assigned to neighborhoods conditiona on observabes, a comparison of pre- and post-program neighborhoods wi produce a biased estimate of program effect. The infuence of non-random city timing is easiy resoved by incuding city fixed effects in the regression estimates. 31 A more compicated source of program timing bias concerns the order in which project teams entered neighborhoods within cities. Empirica evidence that this is not a reevant compication is provided from a comparison of eary and ate neighborhood characteristics prior to the program. Tabe 1 reports district eve poverty indicators from the Peruvian Ministry of Economics and Finance based on 1993 census data. The ast row reports the genera poverty indicator constructed from a weighted mean of eight district-eve measures, reported in the rows above: rates of chronic manutrition, iiteracy, fraction of schoo-aged chidren not in schoo, residentia crowding, adequacy of roofing, and the proportion of the popuation without access to water, sewerage, and eectricity. 3 Not ony is the genera poverty index simiar across program and non-program neighborhoods in 1993, but the differences in a eight base indicators reported in the rows above are sma and insignificant, and vary in sign across indicators. The observed simiarity between program and non-program neighborhoods in a range of poverty measures is strong evidence against a obvious sources of endogenous neighborhood program timing within cities. Further evidence that program timing was independent of neighborhood economic deveopment comes from a visua inspection of the entry patterns of the titing program in Lima, the ony program city in which a four waves of program expansion are represented. Figure 1 pots the basic progression of and titing through districts in Lima as reported in my sampe. In genera, program activity begins in the city center (during the ILD period), then moves to the perimeter of the city and graduay spreads back into the city center. The spatia pattern of poverty in Lima according to 1993 poverty indicators appears entirey unreated to program timing patterns. According to the corresponding poverty map in Figure 1, Wave 3 31 The ony information on the ordering of cities comes from a vague statement in the Word Bank Project Report (#18359), which specifies that the order was designated in advance according to ease of entry. As far as neighborhood program timing, there appears to have been no specific agorithm in the program guideines. The COFOPRI office caim ony that order was subject to geographica situation, feasibiity to become reguarized, dweers requests, existing ega and technica documents, and inkages with other institutions invoved in the existing obstaces (Yi Yang, 1999). 3 Higher vaues of the index refect higher poverty. For a detaied description of how the FONCODES indicator was constructed, see Schady (00). 16

18 ( ) and Wave 4 (1999+) program activity takes pace in districts that span the entire range of poverty eves (1-4). Wave 1 ( ) activity, which took pace in the center of the city, covers districts spanning poverty eves -4, whie Wave ( ) takes pace in districts ranging in poverty eve from 1-3. Worth noting is the fact that when the government took over the titing program during Wave, program activity in Lima was initiated simutaneousy for poitica reasons in each of the three regions of peri-urban settements, shown by the white squares on the map. Thus, in waves and 3, program activity is spread across districts from the Southern, Northern, and Eastern Cones of Lima. Whie the avaiabe information on program timing suggests that is was argey exogenous to the economic environment of neighborhoods, without precise knowedge of the formua for neighborhood timing I cannot safey assume random assignment to treatment nor accuratey specify a seection on observabes mode. Hence, cautious quasi-experimenta anaysis cas for an estimation strategy that is robust to potentia seection on unobservabes. To reduce the roe of endogenous program timing, my identification strategy makes use of a comparison group of non-beneficiary househods. In a framework anaogous to difference-in-difference (DID) estimation, I compare the difference in abor suppy of potentia program beneficiary and nonbeneficiary househods in neighborhoods that the program has reached to the difference in neighborhoods that have not been reached. The simpe idea underying this distinction is that the tenure security effect of titing disproportionatey (or soey) benefits househods with weak ex ante property caims, for whom the demand for tenure security is high. 33 To capture this, I make use of detaied survey data on past and present property tites to construct a binary indicator of whether or not a househod had a tite at the start of the titing program. Those who do not are abeed squatters, whie the term non-squatter refers to househods with pre-program tites. 34 Whie the abor suppy of squatters may systematicay differ from that of non-squatters due to any number of unobservabe factors, identification of program effect wi be robust as ong as this behavior is constant across program and non-program regions. To address the possibiity that it is not, I take two additiona steps. First, I contro for a arge set of observabe househod and neighborhood characteristics 33 There were severa ways a househod might have obtained a property tite in the era before the recent titing effort. First, there was aways the engthy and costy option of foowing the officia bureaucratic process for obtaining and registering a municipa property tite. Second, there were a handfu of past isoated attempts at property reform in which interim titing agencies were set up by municipa governments in an effort to incorporate some proportion of informa residents (De Soto, 1986). Finay, on a number of occasions, mayora and presidentia candidates were known to distribute property tites in an effort to win voter support prior to an eection (Yi Yang, 1999). 34 Throughout this paper, squatter wi refer to househods acking property tites prior to the program. 17

19 in an effort to capture exogenous differences in househod types between program versus non-program areas. Nonetheess, the conditiona independence assumption wi sti be vioated if there exist patterns across program and non-program neighborhoods in a reevant unobserved characteristic that affects the economic environment of squatters differenty than non-squatters. As a further step, I expoit two sources of predicted variation in the impact of the treatment on different househods types. As impied by the mode of Section 4, I expect the impact of receiving a tite to be decreasing in both the number of working age members and the eve of informa property rights. This aows me to additionay estimate modes that test for predicted heterogeneity in response to the program according to househod size and residentia tenure.. Residentia tenure is used as a summary measure of a househod s eve of informa property rights. This stems from the assumption that househods with onger community membership can rey more heaviy on community enforcement, documented in studies on informa property protection such as Lanjouw and Levy (00) and De Soto (1986). Furthermore, aside from refecting community ties, ength of residence coud enter positivey into home security by owering the househod s uncertainty about eviction ikeihood. Because both househod size and residentia tenure are highy correated with poverty but in opposite directions, the dua restriction that program effect be increasing in househod size and decreasing in residentia tenure heaviy reduces concerns over program timing bias by eiminating the possibe confounding roe of any unobservabe trends that are correated with househod poverty. 35 In order for a regiona trend in some unobservabe determinant of abor suppy to be mistakeny attributed to the program, its infuence woud have to be decreasing in both residentia tenure and househod size, and hence no such factor coud be correated with poverty in either direction. Tabe provides descriptive statistics on the sampe popuation, aowing an informa check for random assignment of program timing. As the means in the tabe indicate, there is variation in some demographic characteristics across program and non-program regions. Namey, sampe househods in program areas on average have smaer dweings (fewer rooms), are more ikey to have eectricity, and have higher nativity rates (percentage of members born in province). However, whie statisticay significant differences exist across program and non-program areas, no statisticay significant differences in differences are observed between squatters and non-squatters in program and non-program areas (coumn III). This finding supports the use of non-squatters as a comparison group. 35 Correations between a 3-eve poverty index and househod size and ength of residence verify these patterns in the COFOPRI baseine survey data. 18

20 5.3 Regression Mode The basic estimate of program effect is obtained from the foowing OLS regression: L i = β 0 + β 1 (N) + β (N) + β 3 (squatter) + β 4 (program) + β 5 (program*squatter) + α X i +e i, (1) where L i refers to some measure of househod abor suppy; N is number of househod members; squatter refers to a househod with no pre-program property tite; program indicates whether the househod ives in a neighborhood that has been reached by the program; and X i is a vector of demographic contros. The coefficient on the interaction between program and squatter, β 5, is the estimated program effect, which provides a measure of the conditiona (on X i ) average difference in time worked by ex-squatters in program areas versus non-program areas. The incusion of contros for squatter and program fixed effects corresponds to a standard DID empirica specification. The second estimate incorporates a gradient of the program effect over time. L i = + β 6 (program periods) + β 7 (program periods*squatter) () Here, the variabes of interest are the interactions between the dummy variabes for squatter househod and program entry, β 5, and between the squatter dummy and the number of periods since the titing program entered, β 7. Together, these pick up any differentia patterns in abor suppy of squatters reative to non-squatters that are consistent with the neighborhood s years of program experience. The combination of these interactions, β 5 + β 7 (mean # program periods), is the estimated average program effect. This can be interpreted as the margina change in the amount of abor suppied by the average squatter househod in a program neighborhood for each additiona period with a property tite. 36 Additiona variation in program response by residentia tenure and househod size is captured by the foowing modes: 36 The vaidity of the inear constraint on the program effect across periods of program entry is tested by running unconstrained versions of the regressions for a outcome measures, presented in Appendix E. In these modes, instead of the interaction term squatter*(program period), four dummy variabes are incuded corresponding to each period of program entry such that the sope of the program effect is not constrained to be inear over time. The coefficient estimates revea a strikingy consistent trend of increasing program effect over number of periods since the titing program began, supporting the use of a inear restriction. For a outcomes, adjusted Wad tests fai to reject the hypothesis that the differences between program periods are equa (and therefore that the sope of the program effect is inear). Furthermore, the estimates in Appendix E revea the necessity of aowing for a eve effect of the program that is arger than the period-to-period program effect for a outcomes except in-home work. 19

21 L i = + β 8 (tenure) + β 9 (tenure*squatter) + β 10(tenure*program) + β 11 (tenure*program*squatter) (3) L i = + β 1 (N*squatter)+β 13 (N*program) + β 14 (N*squatter) + β 15 (N*program) + (4) β 16 (N*program*squatter) + β 17 (N*program*squatter) The variabe tenure in equations (3) and (4) refers to the number of years a househod has ived in a residence, which is used as a summary measure of househod informa rights and corresponds to τ in the theoretica mode. In equation (3), the average program effect is captured by [β 5 + β 7 (mean # program periods) + β 11 (mean residentia tenure)], whie in equation (4) the estimated average program effect is [β 5 + β 7 (mean # program periods) + β 11 (mean residentia tenure) + β 16 (mean househod size) + β 17 (mean HH size) ]. A estimates are adjusted to account for the sampe custers and strata, the standard errors derived from the Huber-White robust estimator for the variance-covariance matrix. 37 The set of regressors contained in X i is common to a regressions in the empirica section, and incudes contros for the number of working-aged househod members, city fixed effects, ot size and residentia tenure, as we as a constant. In addition, X i incudes the foowing demographic contros: sex, age, education and degree eve of househod head; number of househod members, number of schoo-age chidren, number of babies (ages -4), fraction of aduts that are mae, fraction of aduts that are immigrants (born outside of province), and number of members age 70 and oder; size of property, househod residentia tenure, whether indoor pumbing, whether the property was acquired by invasion, and whether the property was inherited; whether dweing ies within waking distance of nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market, and this indicator interacted with waking time to each ocae; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and whether each of these existed two years ago, and whether neighborhood has government schoo, chid, food or genera socia assistance program. 38 A regressions aso incude a set of dummy interactions between cities and program entry, and between cities and pre-program tite status. The incusion of these interactions absorbs potentia regiona variation in program impementation and regiona differences in informa property institutions that coud be driving reative differences in program impact between tited and untited residents. It is arguabe that the incusion of such a wide set of demographic contros amounts to over-controing. However, as detaied in Appendix F, a of the proceeding resuts are robust to the excusion and incusion of a wide 37 For a description of the technique used to estimate standard errors, see Chapter. of Deaton (1998), The Econometrics of Custered Sampes. 38 A variabes used in the anaysis are defined in Appendix G. 0

22 variety of right-hand-side variabes. For a outcomes in Section 5, coefficient estimates from regressions with no demographic contros are presented aongside the saturated modes. 5.4 Endogeneity Concerns With respect to the choice of right-hand-side variabes, whie an effort was made to incude principay time-invariant househod characteristics, there remain many sources of potentia endogeneity in the set of regressors. Most notaby, endogenous migration of househod members, fertiity and housing investment are a behaviors arguaby correated with tenure security. The robustness of regression estimates to a wide range of specifications provides genera evidence against the roe of endogeneity bias (see Appendix D). With respect to investment, increased credit opportunities among post-program squatters shoud ony bias downward the estimated program effect, given that greater abiity to smooth income has the potentia to ower the margina utiity of wage income, thereby reducing the opportunity cost of eisure. Furthermore, credit has the potentia to increase educationa investment, an additiona pu factor reducing empoyment hours in post-program areas. Nonetheess, in order to minimize endogeneity concerns, ony ot size and underground residentia infrastructure are incuded among the characteristics of the residence, both of which are reasonaby beieved to be reativey time-invariant. 39 The potentia endogeneity of credit access generates one notabe compication in interpreting the home business outcome ony. Namey, it is possibe that the untited are sufficienty credit constrained to be unabe to cover the fixed cost of moving a business from inside to outside the home (this woud appy to non-sef-empoyed as we if abor force participation invoved a high enough fixed cost of participation). However, this is inconsistent with corresponding sampe data on business oans, as we as evidence from four separate studies of credit effects of COFOPRI, in which property tites were found to have no significant effect on residents access to business credit (Fied and Torero, 00; Cockburn, 000; Kagawa, 001; Torero, 000). Individua sampe seection arising from househod migration is unikey to be a reevant compication in this anaysis due to the fact that is was widey known that new residents were ineigibe for a property tite. Migration of individua househod members, however, coud compicate the anaysis 39 A 000 study of a sampe of COFOPRI participants by Kagawa reveaed that residentia eves of sub terra infrastructure, and in particuar the pubic water connection system, does not systematicay vary with neighborhood reguarization (Kagawa, 000). 1

23 if non-random migration rates differentiay atered famiy composition of treatment and contro groups. The principa evidence that this is not the case comes from direct comparisons of treatment and contro group data on residency of househod members, recent migration of past members, number of workingage members, and age and sex of househod head, none of which revea significant differences in famiy composition. As fertiity is potentiay infuenced by changes in tenure security, chidren under age two are excuded from right-hand-side measures of famiy size. A fina source of potentia endogeneity bias arises in a experimenta and quasi-experimenta settings in which participants are aware of treatment. In particuar, program timing woud not identify the treatment effect of obtaining a tite if the contro group adjusted their behavior in anticipation of treatment. Anecdota evidence from COFOPRI office personne suggests that there was much uncertainty as to the timing and choice of program ocations, making it is unikey that househods woud fee confident in advance that the program woud eventuay enter their vicinity. 40 More importanty, this behavior woud ony bias downward the estimated program effect in my mode. The ony possibiity for upward biases is an Ashenfeter dip response of future program participants, in which squatters spend disproportionate time safeguarding property when the program is about to enter. Whie possibe, there is no intuitive nor anecdota reason to expect demand for invasions to rise in anticipation of the program. 6 Empirica Resuts 6.1 Program Effect on Tenure Security The theory of Section 4 posits that obtaining a property tite affects househod abor suppy by increasing tenure security. Naturay, if becoming a tited property owner does not change househods perceived probabiity of eviction, there wi be no expected program effect. Survey data on househod perceptions of eviction ikeihood are therefore informative for verifying the presumed reationship between tite acquisition and tenure security before continuing with the anaysis. The foowing indicators are expored: whether the househod reported experiencing a change in tenure security with the acquisition of a property tite, whether eviction is considered very ikey and whether eviction is considered very unikey. Indeed, according to the simpe DID estimates in Tabes 3a 3c, the data provide evidence of a basic program effect that is consistent with the variations in program entry and groups of beneficiaries described above. Squatters in program neighborhoods report significanty higher 40 Interview with Caros Gandofo, COFOPRI Office, Lima, Peru, August 000.

24 current eves of home security (3a, 3b) and changes in tenure security associated with property tites (3c). Thus, it is reasonabe to concude that the program indeed ed to significant increases in tenure security. 6. Reduced-form Estimates of Effect on Labor Suppy Strong evidence of a corresponding program effect on househod abor suppy comes from a visua comparison of pre-program squatter and pre-program tited househods in program and non-program neighborhoods. Figures pots the distribution of annua abor force days per househod worker by these four sub-sampes. 41 The density marked by squares, which corresponds to squatters in neighborhoods not yet reached by the program, is visiby distinct from the densities corresponding to the two groups of residents in program areas and aso from that of the tited residents in non-program areas. Two important patterns are worth noting: First, among non-squatters, the empoyment hours distribution of residents across program regions is very simiar, whereas among squatters the distributions depend heaviy on whether or not the program has entered. 4 Second, not ony are the work patterns of the comparison group reativey constant across program and non-program areas, but they are aso simiar to the work patterns of pre-program squatters after the program has entered. These reguarities end confidence to the use of non-squatters as a comparison group. The program effect interpretation of such a picture is that the titing program eads squatter househods to shift outward their distribution of work hours to reach that of titehoders, as woud occur if ack of tenure security were responsibe for the empoyment hours differentia. To further expore this pattern, a inear regression framework is needed to contro for househod, neighborhood and regiona determinants of abor suppy which, if unbaanced, coud confound measures of program impact. Tabes 4 6 present the coefficient estimates of interest from modes (1) (4) of Section 5.3. Coumn 1 reports resuts from the sparsest regression, which constrains the program effect to be constant across househod type and time since titing, whie coumns, 3 and 4 aow the program effect to vary by time since program entry, ength of residence and famiy size, cumuativey. The outcomes of interest are tota househod weeky hours of work, tota househod annua months of work, and fraction of househod members in the abor force. 43 Weeky hours of work refer to ast week s 41 Whie my empirica estimates wi focus on weeky and not annua hours worked, the patterns refected in Figures are usefu in providing the cearest iustration of my identification strategy. The kerne density of annua hours per member is presented in Figure 3a. 4 In fact, the hours distribution of squatters in program areas stochasticay dominates that of squatters in nonprogram areas. See Figures 3b and 3c. 43 In tota, 99 househods are dropped from the anaysis due to missing abor suppy information (a househod is considered to have missing weeky hours data if it has one or more members who both report having worked ast 3

25 empoyment, and is constructed from survey questions on the number of days and mean hours per day worked ast week asked of a househod members who report having worked during the past week. Working-age members who are not in the abor force and those who are in the abor force but report not having worked ast week are assigned empoyment hours vaues of 0. Annua months of work is constructed from survey questions on the number of months worked of the ast tweve, asked of a househod members who report having worked during the past year (which incudes a those who worked ast week). 44 Labor force participation is measured as the fraction of working-age househod members who report either having worked, had a temporary absence from the abor force or searched for a job during the past week. In coumn 1 of Tabe 4, the margina effect impied by the estimated coefficient on the interaction term between squatter and program is roughy 13.4 hours per week. In coumn, which aows the program effect to increase with time since the program began, the margina effect impied by the estimated coefficient on the interaction term between squatter and program periods is roughy 14.5 hours per week, whie the fixed effect is -1.7 hours but insignificant. This impies a tota program effect of roughy 16. hours per week for the median squatter househod with two periods of property rights. For the average househod without a property tite, this impies a 17% increase in tota househod abor suppy per week or around two days of fu-time work. The ong-run, or steady state effect of the program, refected in the estimated effect on househods with the maximum number of program periods, is an average increase of 45 hours of empoyment per week across the entire target popuation of squatters roughy the same as one fu-time worker being added to the abor force. For new househods and househods with few working-age members, the program effect is even arger. The estimates in coumn 3, in which the program effect is aowed to vary by residentia tenure, provide evidence that newer residents increase abor hours more in response to an increase in tenure week and have positive reported vaues of either hours worked per day or days worked per week and missing vaues of the other variabe), 31 househods have missing data on property size and/or oca eementary schoo faciities, 0 househods are excuded in two custers in which program entry does not match institutiona data on regiona program timing, and 8 househods are excuded because a members are reported as over the age of 80, eaving a tota of 59 househods. Due to the survey design, information on daiy and houry work time was incompete (but not missing) for 69 individuas who reported not working in the ast week but working over the ast tweve months. For such individuas, ony the number of months out of the year worked was asked, and not days a week or hours a day worked. For the weeky hours variabe, these individuas are assigned vaues of 0 for days worked ast week and hours worked ast week. For the annua hours estimates in Figures, predicted vaues of hours and days a week were assigned to these observations based on a vector of househod and individuas characteristics. No predicted vaues, however, were used in the regression or probit estimates. 44 nfortunatey it is impossibe to combine months and hours responses to create a summary measure of annua abor suppy without using predicted vaues of weeky hours for peope who worked ast year but not ast week. 4

26 security. In the regressions that account for differences according to househod years of residence, the estimated program effect rises to.6 hours per week for the average squatter famiy with 15 years of residentia tenure a 3% increase in househod abor suppy. Furthermore, aowing the program effect to vary by residentia tenure accounts for the negative coefficient on the main effect of the program in coumn. When the program effect is aowed to vary by famiy size, we observe even stronger evidence that the impact of the titing program on abor suppy is concentrated among househods with few potentia workers. In coumn 4, when both sources of variation in treatment response are taken into account, both the eve effect and the dose effect of the program become significant. Athough the estimated effect on the average squatter househod fas to 1.3 hours and becomes insignificant, the estimates indicate that the size of response depends heaviy on househod type. Thus, sma famiies and famiies with few years of residence account for the majority of program effect captured in coumns 1 3. As mentioned in Section 4, the coumn 4 patterns of heterogeneity in program response according to residentia tenure and famiy size provide additiona evidence that unobservabe factors are not biasing the resuts. Whie poverty and program effect shoud decrease with residentia tenure, poverty and program effect move in opposite directions with respect to househod size. Thus, any unobserved heterogeneity between eary and ate program neighborhoods that is correated with poverty eve coud not be responsibe for both patterns of variation in program infuence. To expore in more detai how the program response varies by househods size and residentia tenure, Tabe 4a presents the estimated program effect for a range of househod types. At east two things are worth noting from this chart. First of a, the program effect does not appear to kick in unti more than a year after the first tite is distributed. This coud be driven by the fact that titing within a neighborhood takes an estimated eight months to compete, such that a disproportionate number of househods in the most recent program regions are sti untited by the time of the survey. 45 Aternativey, this coud refect an adjustment ag necessary for househods to either re-optimize abor suppy or to ascertain the increase in tenure security associated with their newy acquired and tite. Secondy, the program effect fas with famiy size ony for househods with ess than five workers. As shown above, the quadratic function estimating the program effect according to famiy size reaches a minimum at five working-age members, at which point the estimated program effect remains we above zero. This is inconsistent with the mode of Section 4, in which, as ong as it is significant, the program effect fas with househod size, in which case the minimum of the quadratic function shoud not be 45 Time estimate reported in a mimeo on the program procedure distributed by the COFOPRI office in Lima. 5

27 bounded away from zero additiona hours. Instead, the resuts above suggest a mode in which either desired eisure time per capita fas (equivaenty, desired consumption per capita increases) with househod size, or ese the demand for tenure security increases with househod size (as opposed to the mode s assumption that security demands are independent of number of members, controing for ot size, residentia tenure, and forma rights). Most ikey, the second association is responsibe on account of unobserved heterogeneity in househod type correated with househod size. As reported in Tabe 5, the effect of the titing program on tota househod months of work tes a simiar story to the estimates on weeky hours. The measured program effect on househod annua empoyment months is approximatey.9 months, sighty ess than the month effect impied by the weeky hours estimates. Differences between the sizes of the program effects reported in Tabe 4 and Tabe 5 argey refect the extent to which reductions in abor suppy driven by tenure insecurity are due to shorter average work weeks versus extended periods of unempoyment or non-participation. Thus, the combined estimates suggest at east some increase in the number of abor force participants. Indeed, Tabe 6 reveas that added workers account for a significant portion of the change in famiy abor suppy resuting from the titing program. When the same regressions are run on househod abor force participation rates, we observe an impied 6 7 percentage point increase in the number of workingage househod members who are empoyed or searching for work (coumns 1 and ). With an average 49% abor force participation rate among squatter househods with four working-age members, an effect of this size woud be accompished if one in every four househods that obtains a property tite adds a worker to the abor force (5%/4 = 6.5%). Even if every such added worker worked fu time (48 hours per week), additiona abor force participants coud not account for the entire impied program effect on hours. This suggests that average hours of the empoyed are aso higher among program participants. As evidenced in Figures, a rough comparison of average hours per worker reveas a difference in the average number of empoyment hours of workers in program areas and non-program areas of around 5 hours per week. In the average two-worker (four-member) famiy, this accounts for approximatey twothirds of the program increase in hours. Tabe 7 decomposes by gender the program effect on hours to study separatey the impact of titing on work hours of adut men and women. The regressions in coumns 1 and 4 of Tabe 7 are identica to the coumn 1 and 4 regressions in Tabe 4 except that further contros for famiy composition are incuded (number of adut men, adut women, boys and girs aged 1 16, and chidren aged 5 11). Furthermore, to reduce the dimensionaity of the program effect for anaytica purposes, in a Tabe 7 regressions the 6

28 program effect is constrained to be constant over time. 46 The estimates reported in coumns 1 3 indicate that changes in mae empoyment account for the majority of the program effect on hours. In coumn, we see that higher mae hours account for 10.3 out of the impied tota program effect of 1.9 hours. Meanwhie, the difference in femae hours (coumn 3) is sma and insignificant. However, not surprisingy, femae hours are much more eastic than mae hours. Athough the mean effect of acquiring a property tite on hours worked by women is cose to zero and insignificant for the average famiy, when the program effect is aowed to vary with famiy size and residentia tenure, we observe that the effect on femae hours depends heaviy on househod type. For instance, in famiies with ony two working-age members and ten years of residence, the impied program effect on femae abor is 18. additiona hours per week and statisticay significant. This is equivaent to one in three women joining the abor force fu time. In contrast, as observed in coumn 5, the average program effect on mae hours does not depend on either famiy size or ength of residence. 6.3 Effect on Chid Labor Force Participation As motivated by the mode of in Section 4, an increase in forma property rights is predicted to generate a decrease in the amount of chid empoyment if chidren have a comparative advantage in market work reative to home security. The next set of estimates ooks for an effect of property titing on chid abor force participation. In the sampe, ony 8.% of a househods report reguar abor force participation (excuding unpaid domestic work) by chidren between the ages of five and 16. This fraction coud easiy underreport the actua eve of work hours by chidren, as househods might be reuctant to admit to chidren working or not consider irreguar empoyment of chidren when answering survey questions. Yet, whie this number is ow, it is not ceary underreported. According to Internationa Labor Office estimates, 4.1% of a Peruvian chidren aged 6-14 were economicay active in Though the rate shoud be higher for the reativey poor househods in my study, it is aso true that urban househods have ower rates of chid abor than do rura househods in Peru (Ray, 000). To study the effect of urban property titing on chid abor, I estimate a probit mode where the dependent variabe is a dummy indicator of whether or not any househod members under age 16 are reported as working more than five hours per week. I estimate a binary mode rather than modeing the 46 When working hours of men and women are regressed separatey on the eve effect and the does response, (program periods)*squatter, it appears that femae hours change initiay but do not rise over time, whie mae hours change ess initiay but graduay increase with additiona years post-program. The discrete change in femae hours suggests that femae workers are ikey to be new abor force entrants, whereas men are more ikey to be od abor force participants increasing hours of work over time. 7

29 margina effect on chid abor hours due to the fact that the majority of famiies report no chid abor hours, necessitating a imited dependent variabe mode with more stringent functiona form assumptions. Tabe 8 reports the coefficients and margina effects from the probit estimates with a fu set of contros. 47 Coumn 1 estimates the program fixed effect on chid abor, where the coefficient on the interaction term is anaogous to the DID strategy in a inear framework. Coumns and 3 decompose the program effect across househods of different sizes, first aowing the program effect to change ineary with househod size, then by measuring the program effect on ony the smaest 85% of househods. Whie the first coumn shows no average program effect on the probabiity of chidren working, when the effect is aowed to vary by famiy size, we observe a significant effect of property titing on househods with fewer than four working-age members. As reported in Coumn, for househods with three working-age members, the impied margina effect of property titing is arge (.4 percentage points, where the mean is 7.8%) and significant. For arger famiies, the effect is cose to zero and insignificant. This is consistent with the theoretica predictions and with the estimates of Tabe 4: if famiies with more than four working-age members are unconstrained by the need to keep famiy members at home, neither shoud they have incentive to send chidren to work in pace of aduts. To estimate the average program effect on constrained househods and aso test for potentia noninearities in the famiy size effect, I aso run the same mode excuding the argest 13% of househods in the sampe, those with more than six working-age members. Coefficients from this mode are presented in Coumn 3. When famiies with many potentia workers are excuded, we observe that obtaining a property tite reduces the average ikeihood of chidren entering the abor market by. percentage points. According to this estimate, the impied program effect on chid abor force participation among famiies with 1 6 working-age members amounts to a reduced ikeihood of roughy 8%. Whie the estimated impact of property titing on the probabiity of chidren working is compeing, the mechanism by which property rights reduce chid abor is ambiguous. If chid eisure is a norma good, the prediction woud aso foow from the income effect of an increase in adut wage earnings due to added work hours. Both expanations are consistent with past research on the determinants of chid abor force participation in Peru, in which it was found that chid empoyment eves are responsive to changes in the adut mae wage (Ray, 000). Whie property titing does not necessariy generate an 47 In an effort to avoid mistaking young domestic workers for chidren, I excude singe mae-headed househods. Incuding these househods owers the point estimate of program effect sighty but the estimate remains significant. 8

30 increase in the adut wage, an anaogous resut shoud arise from a change in the opportunity cost of adut eisure, which in this mode is the wage minus the security vaue of eisure Effect on Rate of In-home Work The fina question addressed in this paper is whether or not members of a househod participate in market work at home. In the sampe, 4.3% of househods report running a business from home. 49 Whie a genera cass of modes of househod production treat abor suppy decisions as separabe from production decisions, in my mode, in-home work has the additiona feature of increasing tenure security and thereby reducing the househod demand for eisure. Thus, in the absence of a property tite, the mode impies that the decision to run a business from home is determined jointy with decisions about the tota number of hours worked by househod members. According to the predictions of Section 4.1, the margina vaue of in-home work fas when forma property rights are secured and there is no onger a security incentive to stay at home. As a resut, newy unconstrained decision-makers wi have incentive to more efficienty aocate resources by moving production outside of the home or finding work with an outside empoyer. The nature of this reationship between business investment and and titing is a surprising departure from the rura context, in which and tites are hypothesized to promote investment in home production (Besey, 1995). Given the amount of attention paid to increasing credit access via and titing programs, it is interesting to note that investment demand in the urban case may actuay fa with increases in tenure security if increased worker mobiity causes the rate of sef-empoyment to fa. The probit estimates presented in Tabe 9 support the theoretica prediction. In coumn 1, the margina effect impied by the coefficient on the interaction term between squatter and program periods is a 7.6 percentage point reduction in the ikeihood of owning a home business for the average squatter househod, though the estimate is not significant. However, when the program effect is aowed to increase over time, the impied program effect rises and becomes significant. In coumn, the impied margina change in the ikeihood of working inside the home fas by 11.6 percentage points for the average squatter famiy with two program periods impying a reduction in the rate of home business activity of approximatey 47%. Interestingy, as shown by the coefficient estimates in coumn 3, the 48 I observe no significant effect of titing on the probabiity of chid schooing. This is consistent with evidence from past studies on chid schooing and empoyment in Peru, which found schooing eves to be unresponsive to chid abor due to the country s high percentage of working chidren who are aso enroed in schoo (Ray, 1999). 49 The exact survey question is: Do you participate in some economic activity within your home or use part of your property as a source of economic activity? 9

31 program effect on in-home work does not appear to depend heaviy on famiy size or residentia tenure, a possibe indication of omitted variabes bias or other specification error. As an additiona test of variation in program response in which the covariates are not assumed to be constant across househod types, I run the probit estimate separatey for househods iving on properties acquired by invasion of first resident (3% of sampe) versus non-invaded properties (purchased, inherited, or acquired by some other transfer). 50 Househods on invaded properties generay suffer from acute tenure insecurity, and are therefore presumed to have higher demand for a property tite. Coefficients from this estimate are reported in the ast coumn of Tabe 9. As expected, the effect of obtaining a property tite on the decision to operate a home business is much more severe for the sampe of invaded properties and insignificant for a other residents. In fact, the estimated coefficient for reativey insecure househods is more than six times the size of the coefficient for a other residents. The sub-sampe of invaded residences aso exhibits the famiiar pattern of program response by residentia tenure and househod size. In contrast to the differentia effects on in-home work, the program effect on abor hours does not differ substantiay according to whether the househod was acquired by invasion. 51 This combination of abor suppy and business ocation responses suggests a more compicated mode of househod abor suppy and tenure security. The differentia impact of the home business resut for famiies on invaded properties is consistent with a story in which very insecure househods whose security needs require a arger amount of time spent at home and/or very poor househods with itte disposabe income fee particuary constrained by the amount of foregone earnings home security provision entais. Given the aternative to work inside the home, such famiies choose to reduce tota work hours ony up to a point after which it is more beneficia for househod members to shift production inside the home rather than substitute eisure for outside work hours. This woud expain why the home business effect is ony observed among very insecure househods, whereas the abor hours effect is universa The expected trends were aso observed in comparisons between other sub sampes, incuding mae versus femae househod heads and househods with and without chidren. 51 Nonetheess, the tota program effect on invaded househods is substantiay arger than it is for non-invaded househods, since this popuation experiences both an increase in hours as we as a shift from production inside the home to production outside of the home. 5 Further evidence of the home business effect is provided by a comparison of the average ages of home businesses before and after the titing program. If the impications of Tabe 10 on in-home work fit the proposed mode, not ony shoud the frequency of home businesses be ower but aso the average age of home businesses shoud be higher among squatters after the titing program. Given that househods on invaded properties appear to account for the vast majority of the estimated program effect on rates of in-home work, the estimates are run separatey on the sub sampes of invaded and non-invaded househods. Indeed, we observe that home businesses ocated on invaded properties are an average of 6.3 years oder in program regions than in non-program regions. For home businesses ocated on non-invaded properties, the age difference is sma, positive and insignificant. 30

32 6.5 Robustness Checks To end support to the previous set of estimates, I use propensity score matching based on the probabiity of residing in a program neighborhood to construct a comparison group of untited residents of non-program areas. Propensity score matching reduces bias created when the inear mode underying regression adjustment is incorrect. For comparabiity with the OLS estimates, the same covariates are used to derive the predicted probabiity of a neighborhood being reached by the program in a probit estimate. As reported in Tabe 10, average treatment effects based on kerne matching on the predicted z- score repicate the pattern of program effects found in the OLS estimates in both magnitude and pattern of program impact according to househod size. 53 The estimated abor suppy response to obtaining a property tite is 1.3 additiona hours of work, a 5.1 percentage point increase in the fraction of workingage househod members in the abor force, and a 9.1 percentage point decrease in the ikeihood of running a business from home. When the abor hours resut is broken down by househod size, the estimated average effect of a property tite is 14. weeky hours among househod with ess than four members, 7. hours among househods with 4-5 members, and insignificant among househods with more than 5 potentia workers. As an additiona robustness check, I run identica estimates on the sub sampe of househods that are ineigibe for receiving a tite on account of having moved into their current residence post Ceary, if property tites are responsibe for the observed change in abor suppy, we shoud observe no program effect among ineigibe residents. 54 Indeed, there is no measurabe program effect on ineigibe househods (in fact, the estimated program effect is negative, though insignificant), which is particuary compeing given that newer househods tend to have very ow tenure security. Finay, the previous resuts are robust to severa aternative definitions of squatter. For instance, atering the definition of squatter to incude househods with unregistered municipa tites actuay increases sighty the predicted effect. In addition, excuding Lima from the anaysis produces the same pattern of coefficients, but with much arger standard errors. The mean ITT effect is smaer, which is accounted for by the ower rate of titing in newer program areas. 53 Nearest neighbor and stratified matching produce a simiar pattern of outcomes. 54 Whie ineigibe residents coud serve as a contro group, there are too few (9.4%) identifiabe in the data. 31

33 7 Long-Run Predictions Given the size of short-run effects of the COFOPRI program, it is interesting to consider the scope of impact of the nationwide titing effort. Based on the previous estimates, what is the change in abor suppy that a program neighborhood woud experience once a eigibe squatters have been tited and the tota increase in perceived tenure security and adjustment ags have occurred? Two chaenges arise in predicting the average treatment effect on tited househods after severa periods with a tite from the current set of estimates. In particuar, the resuts presented in Tabes 4 6 underestimate the ong-term impact on abor suppy because both the rate of titing and the impact of titing within a program neighborhood presumaby increase over time. The first compication arises from the ITT nature of the identification strategy, which is anaogous to non-compiance in experimenta data. The fact that it is impossibe to observe in pre-program neighborhoods whom among the eigibe woud prompty receive a tite upon program entry makes it necessary to incude a eigibe recipients in an ITT anaysis. A disadvantage of this strategy is that it fais to isoate the program effect on the househods that actuay received a tite through the government program and therefore underestimates the ong-term impact of residentia formaization. Whie 74.% of squatters in tited neighborhoods did in fact receive a registered government property tite by the time of the survey, the incusion of the remaining 5.8% of untited program participants biases downward the estimate of program effect. The second compication is that the previous resuts average the short-term effect on newy tited househods with the ong-term effect on househods tited many years ago. If the security gains from receiving a tite increase after a tite is granted, the program effect wi aso increase in the ong-run among the 74.% who are aready tited. To generate ong-run predictions, it is necessary to isoate the treatment effect on the treated and to isoate ong-term from short-term gains. One method of isoating the average treatment effect on the treated is to assume that the program infuence is concentrated excusivey among tite recipients. This amounts to using the program as an instrument for whether or not the househod acquires a tite. Because IV attributes the measured ITT effect to ony those who actuay received treatment, it is equivaent to scaing the ITT estimates of program effect by the rate of titing that occurs through the program, a standard method of obtaining a correction factor for non-compiance in experimenta data. 55 These estimates are presented in coumns 1 55 For consistency with the previous estimates, IV is appied within the difference-in-difference framework, such that receiving a tite is instrumented with the interaction term between squatter and program neighborhood and the rate of titing among non-squatter househods is controed for by incuding a fixed effect for squatter househods among the right-hand side variabes. See Newhouse and McCean (1997) for a detaied description of IV in the context of difference-in-difference anayses. 3

34 and of Tabe 11. Whereas the ITT program effect associated an increase of 13.4 hours per week with program intervention, rescaing by the number of tited impies an average treatment effect of 17.8 empoyment hours per week and an 8.4 point increase in the fraction of abor force participants among househods that are actuay granted a tite. 56 However, ony if the benefits of titing are reaized immediatey wi these numbers accuratey approximate the ong-range impact of neighborhood titing among the remaining eigibe househods. This estimate wi sti be biased downwards if the infuence of receiving a tite does not kick in immediatey and therefore some tite recipients are not affected by the program by the time of the survey. For this reason, an arguabe improvement over assuming the program effect is concentrated among tited househods is to assume the program effect is concentrated ony among those tited househods that aso report experiencing a change in tenure security. The mode in Section 4 assumes that property tites encourage peope to work by increasing perceived tenure security and thereby decreasing the margina security vaue of eisure. If this is truy a necessary condition for the titing program to affect abor suppy, then a more pausibe excusion restriction is that the program ony operates through changes in security among tited househods. 57 In support of this assumption, the data provide direct evidence of a strong first stage (tabes 3a 3c demonstrate arge concomitant increases in perceived tenure security). Among the 74% of eigibe squatters that were tited, 81% report a change in tenure security associated with the program tite (that is, 81% more than the program/non-program difference reported among non-squatters). As a resut, the IV estimates in coumns 3 and 4 predict that titing efforts that are successfu in making peope fee more secure wi ead to an average abor suppy gain of 4 weeky hours per househod and 11. point increase in the fraction of househod abor force participants. Assuming that a tite recipients eventuay fee more secure, this provides an estimate of the ong-run effect on eigibe househods. However, this cacuation probaby sti underestimates ong-term gains since it assumes that the program effect is imited to a one-time improvement in tenure security (athough it does not necessariy happen right away). In fact, there is reason to beieve that perceptions of tenure security increase 56 Extrapoating these gains to future tite recipients aso requires that eigibe untited househods are simiar in type to the tited (permitting ignorabe non-compiance). nfortunatey, the compiers and non-compiers are ikey to be inherenty different with respect to abor suppy outcome, inducing non-ignorabe non-compiance, making IV best interpreted as average treatment effects on compiers. 57 Ceary, there are other potentia expanations for the observed positive correation between property tites and abor suppy. For instance, the reative vaue of eisure versus empoyment coud be higher for untited househods due to fewer work opportunities or more incentive to participate in community organizations. In fact, data coected on community organization participation reveas that househod days spent participating in community organizations increases with the acquisition of a tite, further evidence that househod members were ex-ante constrained to stay inside the home. With respect to empoyment opportunities, there is no anecdota evidence that home ownership directy affects empoyment offers. 33

35 graduay over time since tite is granted. In other words, the group of tite recipients who report having aready experienced a change in tenure security are ikey to be a mix of househods that have experienced sma improvements in a short time with househods that have experienced arge improvements over many years. One straight-forward method of isoating the ong-term program effect is to imit the sampe to househods in eary program areas. In coumns 5 and 6, the same IV estimates are run excuding the subpopuation of recent program neighborhoods, or those in which the program entered within the ast 16 months. 58 Consistent with the notion of a agged impact of titing, these estimates are consideraby arger rising from 4 to 38 hours per week, whie abor force participation rises to 1 percentage points. 59 Here, the same excusion restriction appies as in coumns 3 and 4 the program ony infuences tited househods that report changes in tenure security associated with the tite. The benefit of the estimates in coumns 5 and 6 is that it is more pausibe that after at east 16 months with a tite, househods have had sufficient time to become convinced of its security vaue and to adjust their behavior. Excuding the newest program areas generates a more convincing estimate of the average treatment effect on compiers aso because there is ikey to be an anticipatory positive effect of the program on those who are waiting in ine for a tite in new program areas, whereas the househods which have sti not received a tite in ate neighborhoods can be assumed to be ineigibe for a tite due to unobservabe factors. If the program has some degree of positive impact on non-tited squatter househods in program neighborhoods, the scaed ITT estimate wi be biased downwards and underestimate the effect of titing. 60 If a non-tited househods were ineigibe, this woud not be an issue, however, as evidenced by the growing rate of titing in program neighborhoods over time, this is not the case. In fact, in a survey question in which untited househods were asked whether or not they expected to receive a tite, haf of squatters in program neighborhoods said that they expected a tite in the next tweve months. Finay, using the eary neighborhoods to predict ong-run responses reduces the potentia roe of nonignorabe non-compiance. Whie househods ate to receive a tite may have systematicay different abor suppy responses than those tited eary, under the restriction that eary program neighborhoods have had time to reach a of the eigibe, the eary program estimates incorporate this potentia heterogeneity into the estimated average treatment effect on compiers. 58 Sampe size restrictions prevent me from isoating ony very eary program neighborhoods. For instance, a pre neighborhoods are concentrated in Lima, reducing the externa vaidity of corresponding predictions. 59 Coincidentay, the OLS estimates excuding ate treatment neighborhoods are identica to the IV estimate of the average treatment effect on a tited househods, or 3.8 hours. 60 Aternativey, if the program had a negative abor suppy effect on non-treated househods in treatment neighborhoods, the IV estimate woud overestimate average treatment effects on compiers. There is, however, no reason to beieve that untited househods in treatment neighborhoods fee ess secure as a resut of the program. 34

36 For these reasons, the eary neighborhood program response of coumns 5 and 6 arguaby constitutes a reasonabe ower-bound estimate of the ong-term impact of titing efforts on future neighborhoods. The estimates predict that, once a eigibe squatter househods have been tited for at east 16 months, the average increase in abor suppy attributabe to the program wi be in the order of 38 hours per week. This is consistent with a scenario in which untited househods commony keep one working-age member at home fu-time to protect property. This estimate is aso in the same range as the predictions of the ITT effect on Period 1 househods: According to coumn of Tabe 4, neighborhoods treated in the first program wave experience an increase of around 45 hours per week as a resut of the program, or approximatey 40% more work hours per househod. 7.1 Cost-Benefit Anaysis In my sampe, 37% of eigibe government tite recipients in non-program areas are squatters by the strict definition used in my paper. By the end of the program 1. miion tites were granted by the COFOPRI program, suggesting that the above ong-run predictions appy to approximatey 447,000 househods in Peruvian cities, or around ten percent of the country s popuation. This is equivaent to reaxing the time constraint tenure insecurity paced on neary haf a miion workers. In contrast, the cost to the government of nation-wide titing amounts to an estimated $66 per tite, around 0% of which is recovered from user fees and property taxes. 61 The additiona cost to the government of maintaining a nationa property registry in terms of abor hours is margina empoyment figures from pubic registry offices have actuay faen since the consoidation of the oca registries so it is reasonabe to assume that the majority of the program cost comes from the initia mapping and titing process. 6 Thus, it is safe to say that the ong-term benefit fows per househod in wages far exceed the net cost of government titing per househod, which is roughy haf the monthy minimum wage. 63 From a socia accounting perspective, the difference in abor hours expended by househods reative to governments to soidify property caims amounts to societa dead weight oss, and attests to the efficiency of pubic institutions in providing tenure security services. In a compete cost-benefit anaysis, this wefare gain shoud be considered in addition to capita gains resuting from the change in the vaue 61 Project costs reported in the cost-benefit anaysis section of the Project Appraisa Document (Word Bank, 1998). 6 There is no indication that enforcement costs have risen, as evidenced by the number of court cases and poice expenditures. 63 Given the possibiity of genera equiibrium effects on the wage, a ower bound estimate of the ong term wage gains per househod wi equa the minimum wage mutipied by the additiona time spent protecting property in the absence of a tite discounted over time by the time it takes to increase de facto rights (which is we over a month) 35

37 of property, the ony benefit fow typicay considered in project vaue assessments. This is reassuring from the perspective of project appraisa given that capita gains projections based on rea estate price differentias wi overstate increases in househod wefare in the presence of non-transferabe de facto tenure rights (Lanjouw and Levy, 001). Furthermore, the wefare enhancements from capita gains wi not be reaized by the househod unti the residence is sod or mortgaged. My estimates, on the other hand, demonstrate that the benefit fow to squatter househods from a nation-wide titing program in terms of the vaue of hours gained aone we surpasses the costs to the government of project impementation amost immediatey. 8 Concusions This paper has presented new evidence on the vaue of forma property rights in urban squatter communities in deveoping countries. By studying the reationship between the exogenous acquisition of a property tite and househod abor suppy, I have provided empirica support for the anecdota evidence that untited squatters commony attain informa rights by taking time off work to participate in such activities as guarding their property, participating in community groups and fiing administrative caims for formaization. My resuts indicate that the cost of maintaining informa rights via remova from the abor force and distortions in optima househod abor suppy decisions is substantia. There are three major findings. First, unike empoyment responses to most wefare programs, which tend to invove an income effect that potentiay removes peope from the abor force, government property titing programs appear to have the opposite impact on empoyment eves. Second, urban property titing is associated with a significant decine in the fraction of househods that use their residence as a source of economic activity. This finding, which inks property rights to ower rates of business investment, aso departs from the property rights iterature in other settings. Furthermore, property tites appear to reduce the househod demand for chid abor in the majority of househods by amost one-third. Whie eary program effects are noteworthy, the ong-run impications of the titing program are particuary striking. In the survey data, many of the treated househods are sti awaiting ega documents. The ITT estimates of program impact on househods tited very eary on suggest that over time, as a househods are actuay reached by the titing program and receive ega ownership rights, newy tited househods wi increase weeky abor force hours by an average of 45 hours per week or an increase in average weeky househod hours equivaent to one fu-time worker. This prediction is supported by IV 36

38 estimates of the effect of the program on those househods who have actuay been tited for at east 16 months, which predict a 40% gain in abor force hours. Addressing this gap in the iterature is important at this juncture for severa reasons. In recent years, a handfu of poicy initiatives have arisen to address tenure insecurity among untited urban residents of deveoping countries. 64 Whie cost-benefit anayses universay suggest that governments are more efficient suppiers of property rights, these caims tend to ignore actua quantifications of the immediate cost to househods of individua property protection, which appears to be substantia (Barber, 1970; Word Bank, 1998). As the resuts of this study indicate, accuratey measuring the return to property formaization requires adequate attention to the cost of informaity. In addition, understanding empoyment responses to property formaization may be critica to understanding and anticipating other market responses to area-wide titing programs. For instance, higher empoyment coud be an important channe for increasing access to credit, whie the income effect of increases in earnings coud simutaneousy ower demand for credit. Simiary, greater abor mobiity from increased tenure security coud encourage the deveopment of rea estate markets (as opposed to the other way around). Finay, given the evidence on the roe of institutiona causes underying bad macroeconomic performance, these resuts have potentia impications for genera understanding of abor market frictions in deveoping countries. In particuar, in settings characterized by a arge amount of residentia informaity, distortions resuting from informa urban property protection may constitute an important obstace to abor market adjustment and economic growth. 64 In particuar, the Word Bank has sponsored a number of projects aimed at promoting forma property institutions in urban sums wordwide. For an overview, see Land, Security, Property Rights and the rban Poor: Twenty Five Years of Word Bank Experience. Word Bank Briefing Note 8,

39 9 References [1] Acemogu, Daron, Simon Johnson and James A. Robinson (001). Coonia Origins of Comparative Deveopment: An Empirica Investigation. American Economic Review, 91, pp [] Acemogu, Daron, Simon Johnson and James A. Robinson (00). Reversa of Fortune: Geography and Institutions in the Making of the Modern Word Income Distribution. Forthcoming Quartery Journa of Economics. [3] A Matter of Tite, Economist, December 9, 1995, pp [4] Achian, A.A. and Demsetz, H. (1973). The property rights paradigm. Journa of Economic History, 33, pp [5] Aston, Lee J., G. D. Libecap and B. Mueer. Tites and Land se: The Deveopment of Property Rights on the Braziian Amazon. niversity of Michigan Press, [6] Angrist, J.D., Imbens, G.W., and Rubin, D.B. (1996). Identification of Causa Effects sing Instrumenta Variabes. Journa of the American Statistica Association 91:434, pp [7] Baharogu, Deniz (00). Word Bank Experience in Land Management and the Debate on tenure Security. Word Bank Housing Research Background Land Management Paper, Juy 00. [8] Banerjee, Abhijit, Pau Gerter and Maitresh Ghatak (00). Empowerment and Efficiency: Tenancy Reform in West Benga. Journa of Poitica Economy, Vo. 110 (), pp [9] Basu, Kaushik, and P. H. Van. (1998). The Economics of Chid Labor. American Economic Review 88(3), pp [10] Besey, Tim (1995). Property Rights and Investment Incentives: Theory and Evidence from Ghana. Journa of Poitica Economy, 103(5), pp [11] Binswanger, Hans P. and Kaus Deninger, (1999). The Evoution of the Word Bank s Land Poicy: Principes, Experience, and Future Chaenges. Word Bank Research Observer 14(), pp [1] Binswanger, Hans, Kaus Deninger, and Gershon Feder (1995). Power, Distortions, Revot, and Reform in Agricutura Land Reations, Handbook of Deveopment Economics, Voume IIIB, Amsterdam: Esevier. [13] Byamugisha, Frank F. K. (000). The Effects of Land Registration on Financia Deveopment and Economic Growth: A Theoretica and Conceptua Framework. Word Bank Working Paper. [14] Caderon Cockburn, J., (1998). Reguarisation of rban Land in Peru. Land Lines, May 1998, Lincon Institute of Land Poicy, Cambridge, Massachusetts,.S.A. [15] Carter, Michae, and Keith Wiebe (1994). Tenure Security for Whom? An Econometric Anaysis of the Differentia Impact of Land Titing Programs in Kenya. In S. Migot-Adhoa and J. Bruce (eds.) Land Tenure Reform in Sub-Saharan Africa, Kenda/Hunt Press. [16] Carter, Michae R. and Pedro Ointo (1996). Getting Institutions Right for Whom? The Weath Differentiated Impact of Property Rights Reform on Investment and Income in Rura Paraguay. Working Paper. [17] Carter, Michae R. and Eduardo Zegarra (000). Land Markets and the Persistence of Rura Poverty in Latin America: Conceptua Issues, Evidence and Poicies in the Post-Liberaization Era. Chapter 8 in A. Vades and R. Lopez (eds.) Rura Poverty in Latin America, MacMian Press. [18] COFOPRI Office, mimeo, June 000. [19] Coe, Wiiam E. (1996). Labor Migration and rban Empoyment in Deveoping Countries: The Impact of Popuation Growth and Property Institutions. In John Adams and Anthony Scaperanda, eds. The institutiona economics of the internationa economy. Boston; Dordrecht and London: Kuwer Academic, pages [0] Conger, Lucy (1999). Entited to Prosperity, rban Age Magazine, The Word Bank Group, Fa 1999 Issue. [1] Deaton, Angus (1998). The Anaysis of Househod Surveys: A Microeconomic Approach to Deveopment Poicy. The Johns Hopkins niversity Press. 38

40 [] Demsetz, (1967) "Toward a Theory of Property Rights," American Economic Review, 57, pp [3] De Soto, Hernando, (1990) The Other Path, Harper and Row Pubishers, Inc., New York, New York. [4] Dufo, Esther. (1999). Schooing and Labor Market Consequences of Schoo Construction in Indonesia: Evidence from an nusua Poicy Experiment. American Economics Review. September (4), pp [5] Eastery, Wiiam (001). The Lost Decades: The Deveoping Countries Stagnation in Spite of Poicy Reform. Journa of Economic Growth, 9, pp [6] Eickson, Robert C. (1991). Order Without Law: How Neighbors Sette Disputes, Cambridge: Harvard niversity Press. [7] Feder, Gershon, Tongroj Onchan, Yongyuth Chaamwong and Chira Hongadarom. (1988). Land Poicies and Farm Productivity in Thaiand, Batimore: Johns Hopkins niversity Press, for the Word Bank. [8] Fied, Erica M. and Maximo Torero (00). Do Property Tites Increase Credit Access among the rban Poor? Evidence from Peru. Mimeo, Princeton niversity. [9] Gaa, A. and O. Razzaz (001). Reforming Land Markets. Poicy Research Working Paper 616, Word Bank. [30] Graham, J., and C. Green (1984). Estimating Parameters of a Househod Production Function with Joint Production. The Review of Economics and Statistics 66, no., pp [31] Gronau, Rubin. (1977). Leisure, Home Production, and Work The Theory of Aocation of Time Revisited. Journa of Poitica Economy 85, 1977, pp [3] Gruber, Jonathan and Brigitte Madrian (1997). Empoyment Separation and Heath Insurance Coverage. Journa of Pubic Economics, 66(3), pp [33] Jacoby, H. (1993). Shadow Prices and Peasant Famiy Labour Suppy: An Econometric Appication to the Peruvian Sierra. Review of Economic Studies, 60(4), pp [34] Johnson, Simon, John McMian, and Christopher Woodru. (00). Property Rights and Finance. American Economic Review, forthcoming. [35] Jones, Eric L. (1981) The European Mirace: Environments, Economies and Geopoitics in the History of Europe and Asia, Cambridge niversity Press, Cambridge K. [36] Hoy, M. and Jimenez, E. (1991). Squatters' Rights and rban Deveopment: An Economic Perspective. Economica 58, pp [37] Jaan, Jyotsna and Martin Ravaion (1998). Are There Dynamic Gains from a Poor-Area Deveopment Program? Journa of Pubic Economics, Vo.67, No.1, pp [38] Jimenez, E. (1984). Tenure security and urban squatting. Review of Economics and Statistics, 66, pp [39] Jimenez, E. (1985). rban squatting and community organization in deveoping countries. Journa of Pubic Economics, 7, pp [40] Kagawa, Ayako (000). Poicy Effects and Tenure Security Perceptions of Peruvian rban Land Tenure Reguarisation Poicy in the 1990s. Workshop Paper, ESF/N-AERS Internationa Workshop, Leuven and Brusses, Begium, 3-6 May 001. [41] Knack, Stephen and Phiip Keefer (1995). Institutions and Economic Performance: Cross-Country Tests using aternative measures. Economics and Poitics, 7, pp [4] Lanjouw, J. O. and Phiip Levy (00). ntited: A Study of Forma and Informa Property Rights in rban Ecuador, The Economic Journa. 11(48), pp [43] Lopez, R. & Romano, C. (1997). Rura poverty in Honduras: asset distribution and iquidity constraints. Working Paper. [44] McCean, Mark and Joseph P. Newhouse (1997). The margina cost-effectiveness of medica technoogy: A pane instrumenta variabes approach. Journa of Econometrics, 77, pp [45] Micei, Thomas J., C.F. Sirmans, and Joseph Kieyah (001). The Demand for Land Tite Registration: Theory with Evidence from Kenya. American Law and Economics Review, Voume 3 Number, pp

41 [46] Moene, Kar Ove (199). Poverty and Landownership. American Economic Review, Voume 8, Issue 1, pp [47] North, Dougas C. (1981). Structure and Change in Economic History, W.W. Norton & Co., New York. [48] North, Dougass C. and Richard P. Thomas (1973). The Rise of the Western Word: A New Economic History, Cambridge niversity Press, Cambridge K. [49] Oórtegui, Ingrid G. Informa Setters in Lima, ESF/N-AERS Internationa Workshop, Leuven and Brusses, Begium, 3-6 May 001 [50] Ray, Ranjan, (000a). Anaysis of Chid abor in Peru and Pakistan. Journa of Popuation Economics, 13, pp [51], (000b). Chid Labor, Chid Schooing and their Interaction with Adut Labor. The Word Bank Economic Review, 14(), pp [5] Rosenberg, Tina. Looking at Poverty, Seeing ntapped Riches, Editoria Observer, The New York Times, October 1, 000. [53] Rosenbaum, Robert, and Donad Rubin (1983). The Centra Roe of the Propensity Score in Observationa Studies for Causa Effects. Biometrika 70:1, pp [54] Schady, Norbert (00) Picking the Poor: Indicators for Geographic Targeting in Peru. Word Bank Working Paper, Number 477. [55] Singh I., Squire L., and Strauss J. (1986) Agricutura Househod Modes: Extensions, Appications and Poicy, Word Bank, Washington, D.C. [56] Torrero, Maximo (1999). Estudio de a Oferta,, Demanda y Fuentes de Credito Informa, Documento Supementario de Estudio: Perfi de a Demanda y Oferta de Credito Forma y Informa, paper prepared for the COFOPRI office, Grupo de Anaisis para e Desaroo, November [57].S Department of Labor (1998). By the Sweat and Toi of Chidren: Voume V: Efforts to Eiminate Chid Labor Bureau of Internationa Labor Affairs, Washington DC, Avaiabe onine at: [58] Wasmer, E. and Zenou, Y. (1999). Does Space Determine Search Frictions? A Theory of rban nempoyment, CEPR discussion paper 157. [59] Wasmer, E. and Zenou, Y. (00). Does City Structure Affect Job Search and Wefare? Forthcoming, Journa of rban Economics [60] Word Bank Deveopment New Archives, Peru s rban Poor Gain Access to Property Markets, February, 000. [61] Word Bank, (1997a). The Lega and Institutiona Framework, ANNEX A1, rban Property Rights Project, Word Bank Interna Paper, Washington D.C.,.S.A. [6], (1997b). Socia Context, ANNEX A1, rban Property Rights Project, Word Bank Interna Paper, Washington D.C.,.S.A. [63], (1997c). Socio-Economic Assessment, ANNEX A3, rban Property Rights Project, Word Bank Interna Paper, Washington D.C.,.S.A. [64], (1997d). Impementation of the Nationa Formaisation Pan, ANNEX A5, rban Property Rights Project, Word Bank Interna Paper, Washington D.C.,.S.A. [65], (1998a). Project Appraisa Document, Report No.1845PE, Peru - rban Property Rights Project, Washington D.C., SA. [66], (1998b). Project Information Document, No. PID653. Peru - rban Property Rights Project, Washington D.C., SA. [67] Yao, Yang (1996). Three Essays on the Impications of Imperfect Markets in Rura China, Ph.D. dissertation, niversity of Wisconsin. [68] Yi Yang, Zoia Z. (1999) COFOPRI, an Experience of Land Tenure Reguarization in Informa Settements in Perú: Reguarisation process case study at the Sau Cantora Settement. Paper prepared for the Advanced Internationa Training Programme, Housing and Deveopment, Lund Institute of Technoogy Schoo of Architecture. 40

42 Tabe 1: FONCODES Poverty indicators, 1993 No program Program t water roofing eectricity sewerage % in schoo iiteracy residentia crowding manutrition overa poverty Note: Means weighted by city fraction of entered neighborhoods. Source: Peruvian Ministry of Economics and Finance, FONCODES office,

43 Tabe. Sampe Means* Pre-program Squatter HHs Pre-program tited HHs (N=668) (N=08) Ia Ib Ic IIa IIb IIc III Program No Program t Program No Program t t Femae head of HH Mean age of HH member Age of HH head HH size (# members) # of rooms in dweing Lot size (m) Highest grade head Residence acquired by invasion Age of dweing HH adut iteracy rate Pumbing Light Municipa service (water) HH monthy expend. (S/) Whether HH saves Number of members that have moved/eft HH Number of members that were born in province * Coumns Ic and IIc report the t-statistics of the difference between coumns Ia and Ib and IIa and IIb. Coumn III reports the t-statistic of the difference in difference (coumn Ic - coumn IIc). 4

44 Tabes 3a-3c: Evidence of Program Effect on Perceived Tenure Security Tabe 3a: Large change in tenure security with ast tite Difference-indifferences No Program Program Difference (not yet entered) (entered) (no covariates) Difference-indifferences No Program Program Difference (not yet entered) (entered) (no covariates) Not squatter (N=191) (0.01) (0.019) (0.03) Squatter ** (N=559) (0.000) (0.09) (0.037) (0.045) Tabe 3b: Do you consider dweing currenty at risk of eviction/invasion? Not squatter (N=191) (0.011) (0.013) (0.017) Squatter ** (N=559) (0.03) (0.019) (0.030) (0.035) Do you consider dweing currenty very secure from eviction/invasion? Difference-indifferences No Program Program Difference (not yet entered) (entered) (no covariates) Not squatter (N=191) (0.01) (0.00) (0.04) Squatter ** (N=559) (0.06) (0.030) (0.040) (0.046) Notes: Standard errors in parentheses. Ony eigibe HHs (according to residentia tenure) incuded. * Significant at the 0.05% eve. ** Significant at the 0.01% eve. 43

45 Tabe 4: Tota Househod Weeky Hours in Labor Force (N=379) (1) () (3) (a regressions incude demographic characteristics, city*program years, and city*initia rights) (4) (5) no demog. contros Number working-age members (3.373)** (3.366)** (3.36)** (6.451) (4.934)** Squatter*program (6.488)* (1.118) (16.685) (6.035)* (7.194)* Squatter*program periods (5.83)* (5.719)** (5.369)** (6.018)** Squatter*program* tenure (0.571)* (0.559)* (0.617) Squatter*program* working-age members (Squatter*program* working-age members) (11.657)* (11.895)* (1.309)* (1.356)* Impied program effect: N=4, T=15 Impied program effect: N=3, T=15 Impied program effect: N=3, T= (6.49)* (6.55)** (7.03)** (7.98) (8.65) (6.47)** (7.04)* (8.48)** (7.97)** (8.5)** Notes: OLS regression, dependent variabe is HH tota weeky hours of empoyment. Standard errors are in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns a and b incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing hours or days vaues for working members are excuded. Impied program effects evauated at median number of program periods (). Demographic contros incude: sex, age, iteracy and degree eve of HH head; # HH members, # of schoo-age chidren, # of babies, fraction mae (of working-age members), fraction immigrants, and # members 70 and oder; whether indoor pumbing, whether property acquired by invasion, and whether inherited ot; whether dweing ies within waking distance and this indicator interacted with waking time to nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and for the ast two years, and whether neighborhood has schoo, chid, food or genera socia assistance program. 44

46 Tabe 4a: Variation in Program Response according to Househod Type (Outcome: househod weeky empoyment hours) Residentia tenure=10 years HH size (# working-age) workers 3 workers 4 workers 5 workers 1 program period program periods 3 program periods 4 program periods (9.81) (8.71) (10.64) (14.43) (9.41) (7.97) (9.80) (13.64) (10.59) (9.3) (10.9) (14.47) (11.61) (10.69) (1.31) (15.69) HH size (# working-age) workers 3 workers 4 workers 5 workers 1 program period Residentia tenure=15 years program periods 3 program periods 4 program periods (8.69) (7.45) (9.65) (13.73) (8.17) (6.47) (8.63) (1.84) (9.41) (7.98) (9.81) (13.66) (10.45) (9.43) (11.4) (14.88) 45

47 Tabe 5: Tota Househod Annua Months in Labor Force (N=379) (1) () (3) (4) (a regressions incude demographic characteristics, city*program years, and city*initia rights) (5) no demog. contros Number working-age members Squatter*program Squatter*program periods Squatter*program* tenure Squatter*program* workingage members (Squatter*program* workingage members) (0.594)** (0.595)** (0.596)** (1.4) (0.944)** (1.93) (.591) (3.409) (5.78) (5.438) (1.175) (1.159)* (1.10)* (1.148)* (0.10) (0.10) (0.19) (.63) (.99) (0.55) (0.59) Impied program effect: N=4, T=15 Impied program effect: N=3, T=15 Impied program effect: N=3, T= (1.9) (1.7)* (1.31)** (1.51) (1.64) (1.0) (1.3) (1.59)** (1.50) (1.53) Notes: OLS regression, dependent variabe is HH tota annua months of empoyment. Standard errors are in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns a and b incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing hours or days vaues for working members are excuded. Impied program effects evauated at median number of program periods (). Demographic contros incude: sex, age, iteracy and degree eve of HH head; # HH members, # of schoo-age chidren, # of babies, fraction mae (of working-age members), fraction immigrants, and # members 70 and oder; whether indoor pumbing, whether property acquired by invasion, and whether inherited ot; whether dweing ies within waking distance and this indicator interacted with waking time to nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and for the ast two years, and whether neighborhood has schoo, chid, food or genera socia assistance program. 46

48 Tabe 6: Fraction of Househod Members in Labor Force (N=379) (1) () (3) (4) (a regressions incude demographic characteristics, city*program years, and city*initia rights) (5) no demog. contros Number working-age members Squatter*program Squatter*program periods Squatter*program* tenure Squatter*program* working-age members (Squatter*program* working-age members) (0.011)** (0.011)** (0.011)** (0.018)** (0.017)* (0.08)* (0.050) (0.065) (0.156) (0.173) (0.01)* (0.01)** (0.00)** (0.0)** (0.003) (0.003) (0.003) (0.060) (0.065) (0.005)* (0.006) Impied program effect: N=4, T=15 Impied program effect: N=3, T=15 Impied program effect: N=3, T= (0.08)* (0.07)* (0.07)** (0.09) (0.031) (0.034)* (0.038)* (0.035)** (0.038)* (0.041)* Notes: OLS regression, dependent variabe is percentage of working-age HH members who are either empoyed or searching for a job. Standard errors are in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns a and b incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing hours or days vaues for working members are excuded. Impied program effects evauated at median number of program periods (). Demographic contros incude: sex, age, iteracy and degree eve of HH head; # HH members, # of schoo-age chidren, # of babies, fraction mae (of working-age members), fraction immigrants, and # members 70 and oder; whether indoor pumbing, whether property acquired by invasion, and whether inherited ot; whether dweing ies within waking distance and this indicator interacted with waking time to nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and for the ast two years, and whether neighborhood has schoo, chid, food or genera socia assistance program. 47

49 Tabe 7: Gender Distribution of Househod Weeky Hours (1) () (3) (4) (5) (6) (N=379) Tota Hours Men Women Tota Hours Men Women Number working-age members Number adut men Number adut women Squatter*program Squatter*program* tenure Squatter*program* working-age members (Squatter*program* working-age members) (3.441) (.196) (1.93) (6.130) (4.010) (3.09) (5.795)** (4.010)** (5.781)** (4.046)** (4.737)** (.735)** (4.690)* (.693)** (6.168)* (4.59)* (4.118) (4.736)** (17.506) (15.408)** (0.543) (0.381) (0.348)** (11.499)* (8.460) (7.311)** (1.69)* (0.955) (0.800)** Mean Program Effect SE (6.17)* (4.6)** (4.1) (7.88) (5.38) (5.35) 48

50 Tabe 8: Whether any Househod Member Age 5-16 Works (1) () A househods with chidren 5-16 N=1557 (3) Househods with <6 members N=150 Number boys age 1-16 Number girs age 1-16 Number chidren age (0.157) (0.157)* (0.15)** (0.160) (0.160) (0.03) (0.14) (0.14) (0.155) Squatter*program (0.76) (0.619)* (0.300)* Squatter*program* working-age members 0.80 (0.10)* Mean program effect on HH w/ 3 potentia workers SE (0.7) (0.34)* (0.30)* Margina effect Notes: Binomia probit estimation, dependent variabe is a dummy indicator of whether HH members ages 5-16 report working more than 5 hours/week. Standard errors are in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns a and b incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing hours or days worked vaues for working members are excuded. Demographic contros incude: sex, age, iteracy and degree eve of HH head; # HH members, # of schoo-age chidren, # of babies, fraction mae (of working-age members), fraction immigrants, and # members 70 and oder; whether indoor pumbing, whether property acquired by invasion, and whether inherited ot; whether dweing ies within waking distance and this indicator interacted with waking time to nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and for the ast two years, and whether neighborhood has schoo, chid, food or genera socia assistance program. 49

51 Tabe 9: Whether Residence Source of Economic Activity (1) () (3) (4) (5) (6) (7) (N=97) (demographic characteristics, city*program years, and city*initia rights) no demog. contros ninvaded Invaded Squatter*program (0.178) Squatter*program periods Squatter*program* tenure Squatter*program* workingage members (Squatter*program* workingage members) Impied program effect: N=4, T= (0.091)* (0.137) (0.164) (0.090) (0.105) (0.178)* (0.01) (0.016) (0.186) 0.01 (0.00) (0.178) (0.18)* (0.183)* (0.1)* (0.181) Impied margina change Notes: Binomia probit estimate, dependent variabe is whether or not HH runs a business in residence. Standard errors in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns a and b incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing vaues for working members are Impied program effects evauated at median number of program periods (). Demographic contros incude: sex and age of HH head; # HH members, # of schoo-age chidren, # of babies, percentage mae (of working-age members) and percentage immigrants; whether indoor pumbing, whether property acquired by invasion and whether inherited ot; and whether neighborhood has municipa services, eectrica infrastructure, whether oca bus stop/market/commissary/primary and secondary schoo two years ago, whether neighborhood has schoo assistance program, custer average waking distance to oca primary schoo, and custer average waking distance to bus stop. 50

52 Tabe 10: Propensity Score Estimates: Kerne Matching Estimator (N=536) Mean of matched treated Mean of matched contros Average treatment effect HH Weeky Hours in Labor Force ** (4.350) -3 working-age members * (6.77) 4-5 working-age members * (3.35) 6-7 working-age members (0.0) Fraction of HH in Labor Force Whether Home Business ** (0.00) * (0.044) Notes: Propensity score estimated as probit mode, where dependent variabe is whether or not program enters neighborhood. Gaussian kerne, bandwidth 0.06, bootstrapped standard errors. 51

53 Tabe 11: Instrumenta Variabes Estimates (N=346) Number working-age members Registered property tite (Instrument=program ) HH Weeky Hours Fraction of HH in Labor Force (1) () (3) (4) (5) (6) 1.111** ** 11.89** ** * ** (3.365) (0.011) (3.460) (0.011) (4.190) (0.013) * * (8.617) (0.037) HH Weeky Hours Fraction of HH in Labor Force Late program neighborhoods excuded HH Weeky Hours Fraction of HH in Labor Force Whether change in tenure security (Instrument=program) 3.948* -0.11* 37.83* -0.10* (11.877) (0.051) (15.044) (0.06) Weighted compier average treatment effect * * 3.948* -0.11* 37.83* -0.10* (8.617) (0.037) (11.877) (0.051) (15.044) (0.06) Notes: Property tite and change in tenure security with tite instrumented with interaction between program area and squatter. Change in tenure security indicator comes from survey question: "Did the ast property document you obtained affect the security of your residence?", asked ony of househods with property documents. Set of regressors in a coumns corresponds to OLS regressions from Tabes 4 and 5. Robust standard errors account for sampe custering and stratification. Ony eigibe HHs (residentia tenure pre-1995) incuded. * Significant at the 0.05% eve. ** Significant at the 0.01% eve. 5

54 Figure 1: Timing of program intervention and poverty across districts in Lima 53

55 54

56 55

57 Figure. No prog, tite Prog, tite No prog, squat Prog, squat Kerne density estimates of annua abor force hours per HH worker 56

58 Figures 3(a) 3(c): Househod Annua Hours per Member 57

59 Appendix A: Maps of Program Areas and ntited Popuation Map No.1 Map No. NMBER OF NTITLED PROPERTIES (NATIONAL LEVEL) PROGRAM NIVERSE OF INFORMAL PROPERTY AND LEVEL OF FORMALIZATION VIA COFOPRI # IQITO S PIRA # CHICLAYO # TRJILLO CHIMBOTE # # # HARAZ LIMA # T i t e s Source: ENAHO TRIMESTER II, 1998 NMBER OF TITLES niverse of Informa Property COFOPRI Tites AREQIPA # 58

60 Appendix B: Section 4 Comparative Statics B.1 Tota Househod Labor Suppy H f w From the Section 4.1 first-order conditions, H = * * = 0 o x S s and H o N q H f = C L N * = 0, totay differentiate each expression and sove for H f : θ H f = θ wq H XX ( q LL ( s s )( wq + H θ SS H XX LL < H XX LL+ wxx LL+ qhh XLL+ qhh XXXw + sh ssqh XX+ sh ss LL+ sh ssqhh X + shh sqh XX+ shh s XX+ shh s XqHH) ) 0 Given my assumptions on and s, a individua terms in the numerator are negative, so the vaue of the numerator is positive. In the denominator, a terms are positive, making the vaue of the denominator negative excuding the first term which precedes the negative sign. However, since wq H ( w + q ), so is wq ( q + w + q H XX LL H XX LL XX LL HH X LL ). Hence, the second and third terms of H the denominator cance out the first term and the denominator is unambiguousy negative. Thus, f <0. θ H Since the denominators in the expressions for o H and f are identica, the same appies to Ho : θ θ H o = θ wq H XX LL ( q + H XX LL wxx LL + q HH X LL + q HH ( s s X H θ SS XX )( q H + HH L w + s q LL ss H XX + q + s H ss ) H XX LL + s q H ss HH + s X q + s HH s H XX HH s XX + s q HH s X HH > 0 ) In this expression, however, the numerator is aso negative, so that H o >0. θ B. Second-order resuts For tractabiity I ignore home production and derive the second-order resuts of the mode for househods maximizing over abor and eisure ony (those with no home business). 65 The househod s optimization probem is written as: 65 Deriving the resut with home businesses is straight forward but comparative statics are cumbersome. Contact the author for a detaied proof. 59

61 max(x,, s( L, θ, τ )) s.t. x, w ( H L) = X max( x, w N ( H L), L N, s( L, θ, τ )) Because of the assumptions = and =, the optima soution wi be interior. The X X =0 first-order condition for an interior optimum is: L L=0 L = w N x 1 + N + s L s = 0 Taking the tota derivative I obtain the foowing expression for L = θ w ( ) N Xs L : θ w w XX XL Xs sl + Ls sl + ss ( sl ) + 1 N w N s θ N 1 N Ls s θ N ss s θ s L s s Lθ N LL + s s LL Because of the assumptions on, XL, and, the numerator is positive and the denominator is L negative. Therefore, < 0. θ Xs Ls Continuing with the assumption of additive separabiity, I derive the second-order resuts of the mode by taking the second derivatives with respect to the impicit functions. Consider the case of τ. The expression for ( L = θ θ τ L L L is found by appying the chain rue to the above expressions for and : θ τ θ τ f L L f f f + ) + ( τ θ τ L L f ( ) L L f f + ) τ θ τ θ Assuming that a third derivatives of and s are weaky positive (as in the case of CARA or CRRA or quadratic utiity functions), the numerator of the above expression is positive and the denominator is negative, so that L <0. An anaogous proof foows for N. θ τ 60

62 B.3 Labor Suppy of Chidren Here I expand the mode to incorporate differences in the househod suppy of adut and chid abor, and show that under the simpifying assumption that ony adut eisure contributes to home security, in househods in which chidren are abor force participants, chid abor hours wi fa with an increase in tenure security whie adut abor hours rise. To see this, consider a househod with one chid and one adut, where A and C are adut and chid eisure, respectivey, and h A and h C are adut and chid time endowments. As before, househod utiity is a function of individuas eisure, per capita consumption, and home security. The same functiona form assumptions as before appy to s(.) and (.). The ony difference in this case is that ony adut eisure enters the security function, and 0 = A C. The househod s maximization probem is then: )),, (, max,, τ θ A C A x s, (x, C A s.t. X h w h w C C C A A A = + ) ( ) ( The first order conditions for a utiity maximum are: 0 1 = + + = s L x A A s w A 0 1 = + = C x C C w If is additivey separabe in its arguments, by taking the tota derivatives of the first order conditions, we can obtain the foowing expressions for θ C and θ A : 0 ] [ ) ( > = LL s L ss XX A LL s XX C L ss XX C XX C XX C A L S SS L C s s w s w s w w w w s s s C C C C A A C C A A θ θ θ 0 ] [ ) )( ( < = LL s L ss XX A LL s XX C L ss XX C XX C XX C L S SS L A s s w s w s w w w s s s C C C C A A C C A A C C θ θ θ Chid eisure wi rise with an increase in forma property rights, and adut eisure wi fa. 61

63 Appendix C: Distribution of househods in sampe City: No program Program Tota Lima Arequipa Trujio Chicayo Piura Chimbote Huancayo Iquitos Tota Note: Cities isted in order of timing of program entry 6

64 Appendix D: Tenure Security Leves according to Program Tming (Region=Lima ) Rank tenure security, high to ow (1-4) Time since program entry New househods Od househods (residence < µ*) (residence? µ) Frequency 0 (has not entered) (entered ) (entered ) (entered ) (entered ) Note: µ is sampe average residentia tenure. 63

65 Appendix E: Test of Linearity in Program Periods I II III (N=394) Weeky hours LFP Home Bus Number working-age members Squatter*program period 1 Squatter*program period Squatter*program period 3 Squatter*program period 4 Squatter*program* tenure Squatter*program* working-age members (Squatter*program* working-age members) (6.457) (0.018)** (0.03) (5.487)** (0.153) (0.061) (5.779)** (0.161)* (0.063) (7.018)** (0.157)* (0.108) (30.395)** (0.160)** (0.075)** (0.554)* (0.003) (11.75)* (0.061) (1.313)* (0.005)* Adjusted Wad Test: (PP4-PP3)=(PP3-PP)=(PP-PP1) F(, 58) = Prob > F =

66 Appendix F: Changes in Set of Covariates (N=379) I II III IV V Number working-age members Squatter*program Squatter*program periods Squatter*program* tenure Squatter*program* working-age members (Squatter*program* working-age members) (6.305) (6.31) (6.99) (6.96) (6.001) (5.50)* (4.495)** (4.594)** (6.065)* (5.11)* (5.117)** (4.798)* (4.859)* (5.396)** (4.817)* (0.550)* (0.538)* (0.538)* (0.56)* (0.535) (11.840)* (11.797)* (11.866)* (11.717)* (11.481)* (1.31)* (1.308)* (1.309)* (1.307)* (1.5)* Impied program effect (N=4, T=15) (8.393) (8.154) (8.044) (8.484) (8.65) region*program interactions Yes Yes No No Yes region*sqatter interactions Yes No No No Yes neighborhood (custer FE) characteristics Yes Yes Yes No Yes detaied famiy composition No No No No Yes 65

67 Appendix G: Definition of Variabes BABIES: Number of househod members aged -4. EDCATION OF HH HEAD: Four dummies indicating whether or not househod head competed primary schoo, high schoo (common or technica), non-university post-secondary, and university. FIRST OWNERS OF DWELLING: Househod members buit or were first to reside in dweing. INHERIT: From question, How did you obtain this property? respondent answered, By inheritance. INVADED PROPERTY: From question, How did you obtain this property? respondent answered, By invasion. INVADED DWELLING: From question, What type of ownership caim does the househod have with respect to the dweing? respondent answered, Invaded or de facto. LITERACY OF HH HEAD: Househod head can read and write. LOT SIZE: Property size in square meters. Interviewers instructed to verify if possibe size recorded on property deed/tite. MARITAL STATS OF HH HEAD: Dummy indicator of whether or not househod head reports either being married or having a domestic partner. MEMBERS: Tota number of peope who habituay ive and share meas in the househod. NEIGHBORHOOD HAS FAMILY PLANNING PROGRAM: Someone in custer reports participating in or benefiting from a famiy panning program. NEIGHBORHOOD HAS CHILD WELFARE PROGRAM: Someone in custer reports participating in or benefiting from Vaso de Leche, Wawa Wasi, PRONEI, or pubic day care program. NEIGHBORHOOD HAS FOOD ASSISTANCE PROGRAM: Someone in custer reports participating in or benefiting from schoo unch, community kitchen, PANFAR, food-for-work, church kitchen, or food donation program. NEIGHBORHOOD HAS SCHOOL ASSISTANCE PROGRAM: Someone in custer reports participating in or benefiting from iteracy program, schoo insurance program, or INFES. NEIGHBORHOOD HAS GENERAL SOCIAL ASSISTANCE PROGRAM: Someone in custer reports participating in or benefiting from FONCODES, VAN, or other type of assistance program. NEIGHBORHOOD HAS [ ] FACILITY: Anyone in custer unit reports that neighborhood currenty has primary schoo/secondary schoo/job training center/bank/paved road/bus stop/soccer fied/pubic phone/post office/market/commissary/cinema/heath center reasonaby in reach. NEIGHBORHOOD HAD [ ] TWO YEARS AGO: Anyone in custer unit reports that neighborhood had current nearest [ ] faciity for at east two years. NEIGHBORHOOD WALKING INDICATOR TO [ ]: Whether or not anyone in neighborhood reports waking to [ ] faciity. NEIGHBORHOOD AVERAGE WALKING DISTANCE TO [ ]: The mean commuting time to [ ] among peope in custer who caim to wak there. NEIGHBORHOOD LACKS INFRASTRCTRE: From question: Which services does community ack? Answer: Pubic infrastructure (roads/bridges). NEIGHBORHOOD HAS NO MNICIPAL SERVICES: How does HH generay dispose of trash? Municipa trash services. Year of municipa service? Pre NEIGHBORHOOD HAS NO PBLIC WATER SYSTEM: From question: Which services does community ack? Answer: Pubic water system. PERCENT IMMIGRANTS: Percentage of HH members who report being born in the province. PERCENT MALE: Percentage of HH members age 5-70 (working-age) who are mae. [INDOOR] PLMBING: From question, Where does this HH get water? Answer: Faucet inside home. From which year pubic system? Before PROGRAM: More than one househod in custer reports having a COFOPRI tite. PROGRAM PERIODS: One of four program periods determined by custer year of program entry. Year of entry defined as the custer minimum year of COFOPRI tite greater than 199. If minimum year ess than 199, second minimum year is used. Lengths of program periods described in Section 3.. SCHOOL-AGED KIDS: Number of chidren ages 5-10 (primary schoo age). SENIORS: : Number of househod members over age 70. SHOCK: Househod reports experiencing economic shock unreated to titing program during past year. 66

68 SQATTER: A househod is identified from my data as a squatter househod in the foowing manner: Treatment Group squatter =1 if: (1) current squatter, identified by (a) tite type = No tite ; or (b) tite type = Other documents and registered = no ; or (c) tite type = Lega settement and registered = no, OR () pre-program squatter, identified by tite type = COFOPRI, and (a) od tite type = No tite or (b) od tite type = Other documents and previousy registered = no or (c) od tite type = Lega settement and previousy registered = no Contro Group squatter =0 if: (1) current tite-owner in non-program area, OR () pre-program tite-owner in program area TENRE: Residentia tenure is defined as the eariest year in which a resident reports having ived in the house. Data comes from information coected on each househod member, incuding which year they moved (or were born) into house. WALKING INDICATOR TO [ ] FACILITY: Whether or not househod reaches [ ] faciity by waking. WALKING DISTANCE TO [ ] FACILITY: HH waking time to reach [ ] faciity. 67

69 1 Introduction Strengthening economic institutions is widey argued to foster investment in physica and human capita, boster growth performance, reduce macroeconomic voatiity and encourage an equitabe and efficient distribution of economic opportunity (Acemogu et a., 00; North, 1981). As one of the basic roes of institutions and fundamenta to a economic transactions, codifying and protecting property rights is seen in many academic discussions as requisite for economic deveopment and poverty reduction. 1 Among poicy-makers as we, property titing is increasingy considered one of the most effective forms of government intervention for targeting the poor and encouraging economic growth (Baharogu, 00; Binswanger et a, 1995). Despite the consensus on the importance of institutiona factors for economic performance, there is a shortage of reiabe estimates of the infuence of property reforms on a range of market outcomes. This paper studies the impact of property rights on abor markets in deveoping countries by anayzing househod abor suppy responses to exogenous changes in forma ownership status. In particuar, I assess the vaue to a squatter househod of increases in tenure security associated with obtaining a property tite in terms of hours of abor suppy gained and improved efficiency of abor aocation between home and market work and between chid and adut abor. An obstace to measuring the infuence of tenure security is the potentia endogeneity of ownership rights. I circumvent the probem by using data from a dramatic natura experiment in Peru, in which a nationwide program issued forma property tites over a five-year period to more than 1. miion urban househods. This approach in arge measure breaks the ink between 1 See, generay, Demsetz (1967), Achian and Demsetz (1973) and Sheifer et a. (001). Direct evidence of this is provided by Micei et a. (001), who anayze the extent of endogeneity of forma agricutura property rights in Kenya. 1

70 tenure security and income and heps isoate the causa effect of property titing on market outcomes. Athough no pane data are avaiabe on program participants, extensive crosssectiona data were coected on past and future tite recipients mid-way through the program, generating a natura set of comparison groups composed of treated and yet-to-be-treated househods. The Peruvian titing program constitutes the first arge-scae urban property rights reform that has occurred in the deveoping word, and its impact has impications for many deveoping countries in which urban squatting is a widespread phenomenon. An important contribution of this paper is the specific focus on non-agricutura househods and the vaue to urban residents of increased ownership security. In deveoping countries, arge proportions of urban and rura residents aike ack tenure security. Yet, presumaby because of historic interests in agricutura investment and reated poitics of and reform, the majority of both academic and poicy attention to property rights reform has centered on rura househods tenure insecurity. Nevertheess, in most of the deveoping word, the popuation and particuary the impoverished popuation is increasingy urban. 3 Though advocates of urban property reform cite many of the same benefits to and titing for nonagricutura as for farm househods, the reationship between tenure security and economic efficiency is ikey to be distinct in the urban setting. In particuar, as wi be addressed in this paper, there is cause to beieve that urban empoyment eves are particuary sensitive to the degree of residentia formaization. In this manner, the paper aso contributes to the iterature by examining a unique aspect of the wefare gains to property titing: the effect of improvements in tenure security on abor suppy and abor aocation decisions within the househod. The fundamenta consequence of 3 In Latin America and the Caribbean, for instance, the popuation shifted between 1950 and 000 from 41% to 75% urban (nited Nations, Word rbanization Prospects: The 1999 Revision, 000).

71 successfu residentia formaization is a reduction in the househod s ikeihood of forced eviction by the government or expropriation by other residents. As ong as untited househods expend their own human resources in an effort to soidify informa caims to and, the acquisition of a property tite has direct vaue in terms of freeing up hours of work previousy devoted to maintaining tenure security through informa means and securing forma rights. As the foowing quote iustrates, there is ampe anecdota evidence that urban squatters are commony constrained by the need to keep a famiy member at or cose to home to protect against residentia property invasion: I go to work, and my mother ooks after the house, says Aejandrina Matos Franco, who ses cassettes on the street in Lima and who worries that peope coud seize her house when she is away. (Conger, 1999) In addition, the ega process of acquiring forma property tites traditionay invoved substantia monetary and time costs. 4 Both factors ceary raise untited househods abor needs for production of home security and in turn the opportunity cost of empoyment outside the home. As a resut, untited househods make constrained decisions in aocations of eisure, home production, and the amount of chid reative to adut abor. To study these reationships, I impement a quasi-experimenta empirica strategy using cross-section micro-data from a survey of past and future beneficiaries of the Peruvian titing program. Two sources of variation in program infuence are used to isoate the effect of titing: neighborhood program timing and program impact based on prior househod ownership status. In particuar, staggered regiona program timing enabes a comparison of househods in neighborhoods aready reached by the program with househods in neighborhoods not yet reached. Meanwhie, variation in pre-program tenure security aows residents not subject to 4 According to one report, In Peru, the process of getting a deed from the bureaucracy invoved 07 steps divided among 48 government offices, took an average of 48 months to compete, and was too expensive for sma property owners. (Economist, 1995) 3

72 changes in security to serve as a quasi-contro group for residents who experience reativey arge changes as a resut of the program. The fact that the program targeted neary a untited househods regardess of househod demand for forma property rights aso enabes a broader exporation of heterogeneity in response to the program. Heterogeneity in the demand for property tites has been shown to depend heaviy on factors which contribute to the cost of maintaining informa rights. 5 For this purpose, both residentia tenure a proxy for informa tenure security and househod size are used as indicators of the reative vaue of a property tite for a given househod. Given that overa de facto property rights are observed to increase with residentia tenure, the vaue of a property tite and therefore the program impact shoud be ower for househods with onger residentia tenure (De Soto, 1986). Likewise, since (for a given property size) househods with more aduts have greater capacity to provide home security, the tenure security vaue of a forma tite shoud be ower for arger famiies. Severa interesting findings emerge. My estimates of eary program impact suggest that househods with no ega caim to property spend an average of 16. hours per week maintaining informa tenure security, refecting a 17% reduction in tota househod work hours for the average squatter famiy. Aso, househods are 47% more ikey to work inside of their home. Thus, the net effect of property titing is a combination of an increase in tota abor force hours and a reaocation of work hours from inside the home to the outside abor market. My estimates further support the predictions that informa property rights and househod size infuence the home security demands facing an untited househod. For a abor suppy measures, the effect of obtaining a property tite is decreasing in residentia tenure and in the number of working-age househod members. Finay, for househods with chidren, urban and titing is associated with a 5 In fact, heterogeneity in the demand for property tites is modeed expicity in Micei et a. (001). 4

73 8% ower probabiity of chid abor force participation. The resuts are particuary convincing in ight of a number of possibe downward biases. The next section of the paper reviews the theoretica and empirica iterature on and rights in deveoping countries. The third section describes the titing program in greater detai. The fourth section presents a mode of househod abor suppy in which, under very genera conditions, tota abor suppied to the outside market unambiguousy rises with an increase in forma property rights, and both abor hours in home production and chid abor unambiguousy fa. The fifth section describes the empirica mode and discusses the identification strategy for program effect. The sixth section presents resuts and robustness checks. The seventh section discusses ong-run predictions and the eighth section concudes. Reated Literature There exists a wide body of iterature demonstrating the positive infuence of property institutions on market outcomes. Severa macroeconomic anayses have shown a reationship between economic deveopment and cross-country variation in institutiona strength, which encompasses property institutions (Knack et a., 1995; Mauro,1995; Ha et a.,1999; Rodrik,1999). In the microeconomic iterature, the ink between property rights and wefare enhancement has generay been confined to three channes estabished in a semina paper by Besey (1995) that expores the benefits of ownership rights for agricutura househods. These are: increased tenure security and greater investment incentives, ower transactions costs and gains from trade in and, and greater coatera vaue of and and improved credit access. The reationship between and rights and abor markets has been mentioned ony in the context of 5

74 residentia mobiity and abor market adjustment, a coroary impication of higher transaction costs in rea estate (Yao, 1996; Word Deveopment Report, 000; Moene, 199). Empirica estimates of the vaue of property tites in agricutura settings corroborate these predictions. Studies such as Aston et a. (1996), Lopez (1997) and Carter and Ointo (1997) ink and tites with improved credit access, whie many authors incuding Feder (1998), Besey (1995), Banerjee et a. (00) and Aston et a. (1996) provide evidence that ack of property tite indeed affects agricutura investment demand. 6 In urban settings, the vaue of property tites has been measured far ess often and empirica work has focused primariy on rea estate prices. A major contribution is a paper by Jimenez (1984), invoving an equiibrium mode of urban squatting in which it is shown that the difference in unit housing prices between the non-squatting (forma) sector of a city and its squatting (informa) sector refects the premium associated with tenure security. The accompanying empirica anaysis of rea estate markets in the Phiippines finds equiibrium price differentias between forma and informa sector unit dweing prices in the range of 58%, and greater for ower income groups and arger househods. Consistent with the agricutura investment iterature, Hoy and Jimenez (1996) find that and tites are aso associated with greater oca pubic goods provision in squatter communities in Indonesia. A separate ine of research on property institutions reates to the roe of informa or de facto property rights. A number of authors such as Carter (1994, 1996) and Gaa and Razzaz (001) note that, in many settings, informa institutions arise to compensate for the absence of forma property protection. Thus, ega enforcement constraints are binding ony insofar as they correspond to rea tenure insecurity. Lanjouw and Levy (00) find that eves of informa 6 Other work, such as Migot-Adhoa et a. (1998) and Kimuyu (1994) detect itte impact of and titing on investment. The mixed resuts are commony attributed to the difficuty of addressing the endogeneity of tite status. 6

75 property rights vary greaty in urban communities in Ecuador, and de facto tenure security varies systematicay with observabe househod characteristics such as sex of househod head and ength of residence. In addition, their paper demonstrates that the vaue of a forma tite can be overestimated by rea estate price differentias when non-transferabe informa rights are ignored. In my paper, the concept of informa rights is further extended to comprise not ony exogenous househod characteristics, but aso home security investment choices. 3 Project Background This paper examines the effects of the Peruvian government s recent series of ega, administrative and reguatory reforms aimed at promoting a forma property market in urban squatter settements. Peru's informa urban settements grew out of the massive urban-rura migration that occurred over the ast haf-century as a resut of the coapse of the rura economy (due in part to a faied and reform program) and the growth of terrorism. The existence of extensive barren and owned by the state on the perimeters of major cities aong with an impicit housing poicy during the 1980s that aowed squatter settements on unused government ands ed to an extended era of urban migration, often in the form of organized invasions by squatters from the same area of emigration (Oórtegui, 001). 7 It is estimated that in 1997, a quarter of Peru s urban popuation ived in margina squatter settements in peri-urban areas and many more untited residents occupied inner-city neighborhoods (Word Bank, 1997b). 8 Prior to the reforms, obtaining a property tite for a Peruvian househod was neary impossibe due to heavy bureaucratic procedures and prohibitive fees. As described in the initia 7 Invasion of privatey-owned property was aowed by aw if the and had been unused for a period of four years. The aw has since changed (in 1990) so that invasions of private property are not aowed under any circumstances. 8 See Appendix A for a country map of the untited popuation and properties targeted for formaization. 7

76 project report: Peru s traditiona system of titing and registration is compex, inefficient, expensive prohibitivey so for poor peope and prone to rent-seeking. Fourteen different agencies are invoved in the generation of each tite, the courts have rarey been abe to vaidate these tites as the aw requires (Word Bank, 1998a). 9 Due to acute housing shortages and ack of ega transparency, tenants strugged not ony with the government but aso among themseves to secure residentia properties. The common faiure of the government to defend or even recognize informa tenure rights in individua disputes gave rise to rent-seeking behavior in the form of invasions of untited and (Oórtegui, 001). In 1991, a Peruvian non-governmenta organization embarked on an innovative property titing project in the capita city of Lima whose goa was the rapid conversion of informa property into securey deineated and hodings by the issuing and registering of property tites (Word Bank, 1998b). Between 199 and 1995, roughy 00,000 tites were issued at an extremey ow cost, convincing the government and a growing internationa audience of the potentia for efficiency gains from urban property formaization (Word Bank, 1998a). In 1996, under the auspices of the pubic agency COFOPRI (Committee for the Formaization of Private Property) and Decree 44: Law for the Formaization of Informa Properties, the Peruvian government estabished a nationa property registry based on the eary mode to formaize the remaining properties in Lima and extend the program to seven other cities. 10 Just as in the piot project, impementation of the nationa program invoved area-wide titing by neighborhood, which was presumed to foster, through community participation and 9 In his groundbreaking study of the underground economy, economist Hernando de Soto documented the same phenomenon: In The Other Path, de Soto and aids concuded that to get tite to a house in an informa settement whose permanence the government had aready acknowedged took 78 steps from one agency aone, and ten other agencies aso required approva (Rosenberg, 000). 10 According to the Word Bank Project Appraisa Document (1998), target cities were chosen according to a formua based on city size, density of informa settements, and distance from commercia centers, measures indicating the ikey ease and cost of formaization and the expected poverty impact. 8

77 education, a demand for formaization, reduce the unit cost of formaization, and rapidy generate a minimum critica mass of beneficiaries (Word Bank, 1997c). Whie the od process of acquiring a property tite was prohibitivey sow and expensive, the new process was free and extremey rapid. Once a oca property registration system was set up, oca program officias were trained, and the city s target areas were propery identified and mapped, severa project teams simutaneousy entered neighborhoods starting from different points in the city. 11 To be eigibe for program participation, tite caimants were required to verify residency predating 1995, and had to ive on eigibe pubic properties. 1 As a resut of the reforms, by December 001 neary 1. miion of the country s previousy unregistered residents became nationay registered property owners, affecting approximatey 6.3 miion of the roughy 10 miion untited residents iving in the range from just above to beow the poverty ine. 13 In the ream of iterature on the economic benefits of tenure security, the Peruvian experience provides a unique research opportunity for many reasons. Briefy, the nationa formaization pan constitutes a one-of-a-kind natura experiment wordwide in terms of providing neary cost-free improvements in ownership security on such a arge scae. Furthermore, unike many arge-scae government programs, the titing efforts took pace at an 11 In campaigns of two months each, project teams entered 50 to 70 neighborhoods encompassing roughy 30,000 to 35,000 pots. Within a neighborhood, teams spent five to seven weeks estabishing residentia caims and deineating properties before conferring state-registered property tites onto a eigibe residents. The registration process for these tites took an additiona period of one to six months. 1 Ineigibe properties incuded archeoogica sites and food panes, among other exceptions see page for a description. In the COFOPRI data, 9.4% of samped househods are ineigibe according to reported ength of residence, and an additiona 10% remain untited after severa years of program operation. 13 Though the grant period is not yet over unti December 00, thus far, 1.64 miion ots have aready been formaized and 1.1 miion tites granted, the vast majority of which took pace between 1998 and 000. Whie no residents who previousy possessed registered municipa tites are incuded in this figure, it is uncertain what fraction of this number had ocay registered saes documents before the nationa reforms as these househods were incuded in the government s definition of untited, though in reaity the program simpy transferred such tites to the nationa registry. In my paper, the term squatter refers ony to househods with no saes or judicia tites prior to the reforms, which is estimated to be 37% of the target popuation. 9

78 extremey rapid pace, which faciitates program evauation by eiminating much of the need to consider time trends that coud obscure the independent effects of program participation. At the same time, in the absence of pane data on participating househods, the fact that program timing was staggered proves to be an asset for evauation purposes. A survey of 750 urban househods was conducted in March 000 midway through program impementation. Because the sampe was drawn from the universe of a target popuations for eventua program intervention, the data contain a number of househods in neighborhoods in which the program has not yet entered. 4 Conceptua Framework 4.1 Tota Househod Labor Suppy This section presents a househod production mode that formaizes the intuition that, in a setting of incompete property rights, the standard abor-eisure choice wi be infuenced by househod demand for security of property. There are three principa mechanisms by which it is assumed that tenure insecurity removes individuas from the abor force. First, individuas in untited househods are constrained by the need to provide informa poicing, both to deter prospective invaders from targeting individua properties and to participate in community enforcement efforts to protect the neighborhood boundaries. 14 If prospective squatters seek out abandoned and, signaing that the property is occupied may deter conficts over and or property boundaries. Second, reducing the probabiity of government eviction at the community eve may require a critica mass of individuas squatting on neighborhood and, particuary in eary stages of community formation. As a resut, socia norms may evove at the community eve such that 14 In a reated sense, it is reasonabe to assume that untited househods fee a greater threat of robbery given that it is more costy for them to rey on oca aw enforcement in addition to the fact that househods that do not have ega rights to a residence may have ess ega caim to property inside the home. 10

79 househods that do not spend time squatting on neighborhood and, which is good for the entire neighborhood, are punished by other community members. Finay, househods may attempt to increase tenure security through forma channes by going through administrative steps to acquire and rights. In addition, greater tenure security may encourage househod members to work on account of an increase in the vaue of consumption of immobie assets such as housing infrastructure. As discussed at the end of the section, the entire set of predictions from the foowing theoretica mode aows me to test empiricay whether the abor suppy response to improvements in tenure security is driven in part by a change in the security vaue of eisure. I capture the infuence of these incentives on abor suppy in a simpe variation of the basic agricutura househod mode The main innovation is the incorporation of a tenure security function, s ( ), into the utiity function, such that both eisure and home production enter househod utiity through two separate channes: through their respective consumption and production vaues and through their effect on home security. 15 Furthermore, the security vaue of time at home is sensiby modeed as a househod pubic good, such that individua utiity depends on the eisure and home production hours of a other members via s ( ). In this framework, utiity, given a set of househod characteristics ψ and resource endowment E, is an increasing function of per capita eisure, consumption, and home security, and home security is determined by the foowing three parameters: tota hours of househod time at home (time spent by famiy members protecting property), an exogenous parameter, θ, which refects the 15 As opposed to modes of joint production in the vein of Gronau (1977), I assume incompete substitution between market goods and home security due to the absence of an outside market for home security protection. 11

80 househod s eve of forma property rights, and a summary measure, τ, which refects the degree of informa or de facto rights the househod has acquired. For tractabiity, I make the foowing set of assumptions. First, the househod is assumed to maximize per capita eisure and not the eisure of individua members. Given that this mode is concerned with the effect of θ on tota househod abor, ignoring the second stage of the househod decision probem in which eisure is aocated across individua members is inconsequentia to the centra resuts. Second, there is no outside abor market for the provision of home security. Assuming a missing abor market for property protection is easiy justified by an incompete contracts argument (there is risk invoved in empoying non-members to guard property), athough a more compicated mode woud have this market depend on θ. 16 Furthermore, whie the mode does not expicity incude hired security, there is room to incorporate the existence of a back market for property protection into τ. Fourth, as opposed to modes of joint production such as Graham and Green (1984), in this mode eisure and home production hours are assumed to be perfect substitutes for the hours an individua spends on property protection. 17 Finay, this is a unitary househod mode, and it is assumed that a househod members face a common wage, w. Let N be the number of househod members, and i be eisure, x i consumption, h fi abor hours in home production, and h oi outside abor hours of househod member i, and L = N i i= 1 N, H f = h fi, H o = h oi, X = i= 1 N i= 1 N x i i= 1, X x =, N L =, Z = H f + L. N 16 Additionay, extension of this mode to a more compicated setting in which there is an imperfect (as opposed to nonexistent) market for the provision of home security is inconsequentia under the uniform wage assumption. 17 Whie this assumption might seem unreasonabe in ight of the fact that eisure time which contributes to home security is constrained reative to eisure which can be spent inside or outside of the home, incorporating a jointness function which measures the psychic vaue of home reative to market production does not change the comparative statics of the mode. 1

81 Labor hours of househod members are divided between work at home ( H f ) and work in the outside market ( H ). Time spent at home (Z ) is divided between work at home ( H f ) and o eisure (L). The vaue of abor at home is given by the production function q H ), whie the ( f vaue of work outside the home is the market wage w. 18 Househod utiity is then given by: ( x,, s; ψ, E), where s = s ( Z, θ, τ ). Here ( ) and s ( ) are twice continuousy differentiabe, concave, and increasing in each argument. 19 Whie the tenure security function impies that the production of home security is determined purey by exogenousy given and rights (θ and τ ) and the amount of time spent in the home, s ( ) coud easiy be extended to incude other househod inputs such as secure ocks and doors. The parameter θ can be thought of either as a binary indicator of a egay registered property tite, or ese a more nuanced parameter which refects the eve of forma ega recognition of a househod s tenure status (eve of efficiency of court systems, eves of poice cooperation, etc.). The choice variabes for the househod are: H, H, X, L and s. The constraints to the f o maximization probem are: s = s( H f + L, θ, τ ) px = wh + q T = L + H o + H f o ( H f L, H, H, X 0 o f ) 18 Incorporating a market for hired abor in home production does not affect the mode s predictions. Inseparabiity in this mode comes from the ack of substitutabiity of househod members in the production of security, not q( ). 19 I assume that security inputs ( Z,θ and τ ) are substitutes in production, and make corresponding assumptions on the cross-partia derivatives of s( ). 13

82 where q ( ) satisfies decreasing margina productivity ( q > 0, q < 0). Then, normaizing prices to one, the househod s optimization probem can be written: max ( ( w* Ho + q( H f )), ( T Ho H f ), s( T Ho, θ, τ)) Ho, H f N N This yieds the foowing necessary first-order conditions for an interior soution ( H > 0; H > 0; H + H T ) : 1 o f o f < w 1 * x = * + S * sh N N o (4.1.1) q f * = (4.1.) H x Equation estabishes that, at the optimum, househods equate the margina vaue of an additiona hour of outside abor with the margina utiity of eisure. Equation 4.1. states that they aso equate the margina utiity of eisure with the margina vaue of an additiona hour of work at home. For each househod invoved in both home and market work, the soution to this set of equations impicity defines demand functions for abor hours in the outside market and in home production which depend on θ, w, and τ : * f = * f * o = * o H H ( w, θ, τ), H H ( w, θ, τ) Assume that 0, 0, 0. Then tota differentiation yieds the foowing xs x inequaities for vaues of w, θ, and τ corresponding to inner optima: s H f θ < 0 H and o > 0 θ. 0 For the remainder of the anaysis, househod characteristics and resource endowment are assumed to be fixed and omitted from the arguments of the utiity function. 1 The boundary conditions and guarantee that (H f + H o )<T and that at east 0 x x 0 one of H f and H o is stricty positive. It is shown on the foowing page that the corner soutions H f =0 and H o =0 do not affect the aggregate predictions of the mode. Note that this incudes the additivey separabe case, as we as the case in which the vaue of consumption is rising in tenure security. 14

83 For househods invoved in both types of abor, an increase in forma tenure security decreases work hours at home and increases work hours in the outside market. At the corner H soution H o =0, f H 0 and o H 0, and at the corner soution H f =0, f H = 0 and o > 0. θ θ θ θ Thus, in aggregate, strengthening forma property rights decreases work hours at home and increases hours outside the home. Detais of the comparative statics are provided in Appendix B. Intuitivey, this refects the fact that an exogenous increase in the eve of forma property rights corresponds to a decrease in the househod s need to spend time on home security, thereby owering the opportunity cost of outside abor force hours. 3 In the empirica anaysis, data imitations prevent me from separating empoyment hours inside and outside of the home. With respect to the net effect of a property tite on tota empoyment hours, my mode predicts that househods with zero home production hours ex ante (H f = 0) wi increase tota househod abor hours by some positive amount in response to stronger forma property rights. For househods with any amount of abor hours devoted to a home business, the net effect on tota hours is ambiguous. Whie the eve of outside work hours wi unambiguousy rise for househods invoved in both types of production, the resuting change in average houry earnings arising from the difference between wages earned in the externa abor market and the margina productivity of abor in home production wi generate both income and substitution effects. The net change in tota abor hours, H + H ), wi ( f o depend on the reative sizes of these effects. In the empirica section, due to the fact that ony 3 It is important to note at this point that I have ignored the consumption vaue of home security via its infuence on the market price of tradabe assets, which has a potentia income effect on abor suppy that coud counteract the impication stated in equation (4.1.). This is justified by two considerations: first, rea estate markets are often nonexistent in these settings; second, for the purposes of estimating a abor suppy effect, the possibe income effect of increases in home security which is being ignored biases downwards the effect on abor suppy. Hence, any finding of an effect is a ower bound on the impact of the program. 15

84 5% of househods are invoved in home production, the program effect on househods working outside the home is presumed to dominate the possibe negative effect on househods with home businesses. Thus, I predict ex-ante that a titing program wi be associated with an increase in tota empoyment hours. At the same time, I wi expore the effect on househods with home businesses by studying the probabiity that a househod uses their residence as a source of economic activity. Since work hours inside the home are predicted to fa unambiguousy, so shoud the percentage of househods that spend any time working at home. Two auxiiary impications foow from the mode. First, the effect of a change in forma property rights on abor suppy is decreasing in househod eve of informa property rights, τ : * H f θ τ > 0 H and o < 0 θ τ * Second, given average consumption eve x, the effects are decreasing in the number of working-age househod members, N. 4 * H f θ N > 0 H and o < 0. θ N * The intuition behind the famiy size effect is that, the more famiy members iving in a househod, the more ikey it is that someone chooses to stay at home independent of security considerations, thus arge househods are ess distorted by the need to keep watch over the residence. These predictions wi motivate me to test empiricay whether the effect of acquiring a forma property tite on abor suppy differentiay impacts househods of different sizes and with different engths of residentia tenure. 4 Given that members of extended famiies often divide their time between househods, some authors treat N as continuous peope hours instead of a discrete number of peope. The same resut can be proven for discrete N. 16

85 In addition, testing the entire set of predictions generated by this mode aows me to rue out the possibiity that the reationship between tenure security and the utiity of consumption is responsibe for the entire effect of property rights on abor suppy. As mentioned earier, greater tenure security may encourage househod members to work on account of an increase in the vaue of consumption of immobie assets. Whie this scenario has simiar impications for tota househod abor suppy, the second-order impication of the mode with respect to househod size appies ony to the case in which househod members time at home contributes to home security. 4. Labor Suppy of Chidren An extension of the mode, aso detaied in Appendix B, incorporates differences in the househod suppy of adut and chid abor when ony aduts contribute to home security provision. This extension formaizes the intuitive idea that, if aduts have a comparative advantage in the provision of home security, in the absence of secure property rights, chidren wi substitute for aduts in the abor market. In this case, whie tota househod abor hours rise with an increase in forma rights, chid abor hours wi actuay fa. For simpicity, in the foowing description I ignore the roe of home production, though the resuts hod under very genera conditions when production is incuded. Here, N A and N C are the number of adut and chid househod members, respectivey, A and C are per capita adut and chid eisure, L A and LC are tota adut and chid eisure and T A and TC are tota adut and chid time endowments. In this setting, the househod s maximization probem is: max (x, A, C, x A, C, s( L A, θ, τ )) such that wa ( TA LA) + wc ( TC LC ) = X The first-order conditions corresponding to each empoyed adut member i and chid member j are: 17

86 wa 1 = + * + = 0 A i x s sl (4..1) A A N N A wc 1 = + * = 0 Cj x (4..) N N C C From these conditions it can be shown that, for a interior optima, C > 0and A < 0. θ θ In househods in which chidren are abor force participants, chid abor hours wi fa and adut abor hours wi rise with an increase in tenure security. For a other househods, adut abor hours wi aso rise and chid abor hours wi remain at zero. Thus, given a positive amount of ex-ante chid abor, the aggregate number of chid abor hours wi unambiguousy fa, whie the number of adut abor hours rises with an increase in forma property rights. 5 Whie the theoretica mode deas with changes in abor suppy at a fixed wage rate, the empirica mode wi capture changes in actua empoyment eves, which are functions of both suppy and demand. Given the size of the program, it is reasonabe to anticipate genera equiibrium effects on the wage rate. However, because increased abor suppy wi decrease the market wage, as ong as eisure is a norma good such effects woud ony bias downward the estimated program effect. Thus, the actua abor suppy response to titing is presumaby higher than what can be measured with changes in working hours. 5 Athough this mode focuses on optima abor aocation, the income effects that foow from reaxing the househod s time constraint provide a pausibe aternative expanation for a decrease in chid abor with an increase in forma rights, and one that has been proposed by other authors. In particuar, a decrease in chid abor woud foow from the uxury and substitution axioms of the Basu and Van (1998) mode of chid abor suppy, in which chidren can substitute for aduts in the abor market and a famiy wi send chidren to the abor market ony if the famiy s income from non-chid abor sources fas beow some threshod amount. 18

87 5 Data and Estimation Methods 5.1 Data Set My empirica anaysis of househod abor suppy responses to changes in forma property rights reies on the COFOPRI baseine survey data. The sampe universe for the survey was a residences in non-incorporated urban and peri-urban settements identified in the 1993 census of the eight cities targeted by the titing program. The data consist of 750 househods distributed across a eight program cities. The survey was stratified on city, with custer units of ten househods randomy samped at the neighborhood eve within cities. The number of custers drawn from each city was based on the city s share of eigibe residents. The survey instrument cosey mirrors the Word Bank Living Standards Measurement Survey (LSMS) in content, and therefore contains a wide variety of information on househod and individua characteristics. In addition, there are five modues designed to provide information on the range of economic and socia benefits associated with property formaization. 5. Identification Strategy To study the impact of receiving a property tite on househod abor suppy, I expoit variation in the year in which the COFOPRI program entered a neighborhood to compare househods in program neighborhoods that have aready been reached by the survey date to househods in ate program neighborhoods. The first step in cassifying program timing was to identify whether or not a neighborhood had been reached by the time of the survey. The survey data do not directy identify program neighborhoods, nor can this variabe currenty be constructed by matching geographic identifiers to COFOPRI office data. Instead, a observations within a survey custer are assigned a program entry vaue of one if more than 19

88 one househod in the custer reports owning a COFOPRI tite. 6 Custers in which no househod or ony one househod have a COFOPRI tite are assumed to be those in which the program has not entered, athough it is generay impossibe to separate the neighborhoods in which the program wi never enter from those which wi be treated eventuay. Nonetheess, such neighborhoods share the key feature of no expected program effect. 7 A breakdown of program and non-program neighborhoods by region is provided in Appendix C. Not every squatter househod that the program reaches is granted a COFOPRI tite by the time of the survey. Reasons that househods may be excuded incude: the househod cannot prove residence prior to 1995; the househod beongs to a cooperative association; the residence ies on an archeoogica site, food pane, mining site or private property; and ambiguous or disputed ownership caims. nfortunatey, none of the above information is coected in the survey. 8 Since the househods in the treated neighborhoods may or may not actuay have received a government tite, this is an intent-to-treat (ITT) anaysis. The second step in cassifying variation in program timing was to identify the year in which the program entered. The effect of the program is presumed to increase over time in a fashion anaogous to a dose response measure from the experimenta design iterature for three reasons: First, titing an entire neighborhood can be a engthy procedure, such that the percentage 6 There is ceary some measurement error in this method of identifying treated neighborhoods. In particuar, it is possibe that neary a residences in the custer were not given tites athough the program did in fact enter the neighborhood. To address this, I aso estimate the mode excuding seven custers in which a samped househods had registered municipa property tites prior to the program, making it impossibe to observe whether or not the program entered. In none of my anaysis does excuding these 69 househods affect the estimate of program effect. 7 Incuding custer units with ony one reported COFOPRI recipient as non-program neighborhoods does not affect the resuts. Since it is extremey unikey that ony one househod is tited in a program neighborhood severa months into the program, such neighborhoods are ikey to refect either misreported tite data or recent program entry. If ony one househod has actuay received treatment, effectivey the neighborhood is at this stage untreated and neighborhood effects shoud not be observed. 8 According to anecdota evidence from program administrators, disputed caims within famiies or between neighbors are the most common reason that tite distribution is deayed for an untited househod in a treated neighborhood (Caros Gandofo, persona interview, Lima, August 9, 000). 0

89 of tited househods within a treated neighborhood increases (at a decreasing rate) over time. Secondy, househod abor suppy takes time to adjust. Finay, it is pausibe that confidence in the vaue of a COFOPRI tite is increasing over time. For purposes of exporing the program effect over time, year of program entry was defined as the eariest reported COFOPRI tite year within the custer. 9 Dynamic response was restricted to be inear in four time periods: January 1999 June 000, January 1997 December 1998, January 1995 December 1996, and January 199-December This division corresponds to three major waves of program expansion: From 199 to 1995, 00,000 tites were granted by the Institute of Liberty and Democracy as part of a piot project prior to COFOPRI; the first wave of COFOPRI tites was initiated in 1995 in Lima and Arequipa; and beginning in 1997 the program expanded into six other cities. 30 Furthermore, these intervas were consistent with the observed reationship between subjective statements on tenure security and years since program entry, as is reported for squatters in the city of Lima in Appendix D. 31 Athough target areas for wide scae economic deveopment programs are never randomy seected, these data have the advantage that a sampe members ive in areas that wi eventuay be targeted for program intervention, increasing confidence in the comparabiity of treated and untreated househods. Furthermore, the universa nature of the treatment and the participation rues of the program generay rue out concern over individua seection bias that 9 Due to the fact that not a househods were given property tites right away and because of measurement error in tite year reporting, househods in the same custer who had received a COFOPRI tite did not necessariy report the same tite year. When the minimum reported tite year fe beow the first regiona tite year according to program data, the second owest tite year was assigned to the custer. 30 This region-specific pattern of intervention makes it important to incude city dummies in regression estimates of program effect. 31 The tabe in Appendix D reveas a tota change in average reported tenure security for residents of Lima of roughy 0.6 points on a four-point scae. The tabe aso iustrates that, whie newer househods have consistenty ower perceived tenure security than more estabished famiies, the change in perceived tenure security foows the same approximate trajectory over time since titing program for both groups. 1

90 coud arise even if program pacement were random. Nonetheess, there is sti potentia for program timing bias, in which areas seected for eary program participation are different from the rest. If program timing is not randomy assigned to neighborhoods conditiona on observabes, a comparison of pre- and post-program neighborhoods wi produce a biased estimate of program effect. The infuence of non-random city timing is easiy resoved by incuding city fixed effects in the regression estimates. 3 A more compicated source of program timing bias concerns the order in which project teams entered neighborhoods within cities. Empirica evidence that this is not a reevant compication is provided from a comparison of eary and ate neighborhood characteristics prior to the program. Tabe 1 reports district eve poverty indicators from the Peruvian Ministry of Economics and Finance based on 1993 census data. The ast row reports the genera poverty indicator constructed from a weighted mean of eight district-eve measures, reported in the rows above: rates of chronic manutrition, iiteracy, fraction of schoo-aged chidren not in schoo, residentia crowding, adequacy of roofing, and the proportion of the popuation without access to water, sewerage, and eectricity. 33 Not ony is the genera poverty index simiar across program and non-program neighborhoods in 1993, but the differences in a eight base indicators reported in the rows above are sma and insignificant, and vary in sign across indicators. The observed simiarity between program and non-program neighborhoods in a 3 The ony information on the ordering of cities comes from a vague statement in the Word Bank Project Report (#18359), which specifies that the order was designated in advance according to ease of entry. As far as neighborhood program timing, there appears to have been no specific agorithm in the program guideines. The COFOPRI office caim ony that order was subject to geographica situation, feasibiity to become reguarized, dweers requests, existing ega and technica documents, and inkages with other institutions invoved in the existing obstaces (Yi Yang, 1999). 33 Higher vaues of the index refect higher poverty. For a detaied description of how the FONCODES indicator was constructed, see Schady (00).

91 range of poverty measures is strong evidence against a obvious sources of endogenous neighborhood program timing within cities. Further evidence that program timing was independent of neighborhood economic deveopment comes from a visua inspection of the entry patterns of the titing program in Lima, the ony program city in which a four waves of program expansion are represented. Figure 1 pots the basic progression of and titing through districts in Lima as reported in my sampe. In genera, program activity begins in the city center (during the ILD period), then moves to the perimeter of the city and graduay spreads back into the city center. The spatia pattern of poverty in Lima according to 1993 poverty indicators appears entirey unreated to program timing patterns. According to the corresponding poverty map in Figure 1, Wave 3 ( ) and Wave 4 (1999+) program activity takes pace in districts that span the entire range of poverty eves (1-4). Wave 1 ( ) activity, which took pace in the center of the city, covers districts spanning poverty eves -4, whie Wave ( ) takes pace in districts ranging in poverty eve from 1-3. Worth noting is the fact that when the government took over the titing program during Wave, program activity in Lima was initiated simutaneousy for poitica reasons in each of the three regions of peri-urban settements, shown by the white squares on the map. Thus, in waves and 3, program activity is spread across districts from the Southern, Northern, and Eastern Cones of Lima. Whie the avaiabe information on program timing suggests that is was argey exogenous to the economic environment of neighborhoods, without precise knowedge of the formua for neighborhood timing I cannot safey assume random assignment to treatment nor accuratey specify a seection on observabes mode. Hence, cautious quasi-experimenta anaysis cas for an estimation strategy that is robust to potentia seection on unobservabes. 3

92 To reduce the roe of endogenous program timing, my identification strategy makes use of a comparison group of non-beneficiary househods. In a framework anaogous to differencein-difference (DID) estimation, I compare the difference in abor suppy of potentia program beneficiary and non-beneficiary househods in neighborhoods that the program has reached to the difference in neighborhoods that have not been reached. The simpe idea underying this distinction is that the tenure security effect of titing disproportionatey (or soey) benefits househods with weak ex ante property caims, for whom the demand for tenure security is high. 34 To capture this, I make use of detaied survey data on past and present property tites to construct a binary indicator of whether or not a househod had a tite at the start of the titing program. Those who do not are abeed squatters, whie the term non-squatter refers to househods with pre-program tites. 35 Whie the abor suppy of squatters may systematicay differ from that of non-squatters due to any number of unobservabe factors, identification of program effect wi be robust as ong as this behavior is constant across program and non-program regions. To address the possibiity that it is not, I take two additiona steps. First, I contro for a arge set of observabe househod and neighborhood characteristics in an effort to capture exogenous differences in househod types between program versus non-program areas. Nonetheess, the conditiona independence assumption wi sti be vioated if there exist patterns across program and nonprogram neighborhoods in a reevant unobserved characteristic that affects the economic 34 There were severa ways a househod might have obtained a property tite in the era before the recent titing effort. First, there was aways the engthy and costy option of foowing the officia bureaucratic process for obtaining and registering a municipa property tite. Second, there were a handfu of past isoated attempts at property reform in which interim titing agencies were set up by munic ipa governments in an effort to incorporate some proportion of informa residents (De Soto, 1986). Finay, on a number of occasions, mayora and presidentia candidates were known to distribute property tites in an effort to win voter support prior to an eection (Yi Yang, 1999). 35 Throughout this paper, squatter wi refer to househods acking property tites prior to the program. 4

93 environment of squatters differenty than non-squatters. As a further step, I expoit two sources of predicted variation in the impact of the treatment on different househods types. As impied by the mode of Section 4.1, I expect the impact of receiving a tite to be decreasing in both the number of working age members and the eve of informa property rights. This aows me to additionay estimate modes that test for predicted heterogeneity in response to the program according to househod size and residentia tenure.. Residentia tenure is used as a summary measure of a househod s eve of informa property rights. This stems from the assumption that househods with onger community membership can rey more heaviy on community enforcement, documented in studies on informa property protection such as Lanjouw and Levy (00) and De Soto (1986). Furthermore, aside from refecting community ties, ength of residence coud enter positivey into home security by owering the househod s uncertainty about eviction ikeihood. Because both househod size and residentia tenure are highy correated with poverty but in opposite directions, the dua restriction that program effect be increasing in househod size and decreasing in residentia tenure heaviy reduces concerns over program timing bias by eiminating the possibe confounding roe of any unobservabe trends that are correated with househod poverty. 36 In order for a regiona trend in some unobservabe determinant of abor suppy to be mistakeny attributed to the program, its infuence woud have to be decreasing in both residentia tenure and househod size, and hence no such factor coud be correated with poverty in either direction. Tabe provides descriptive statistics on the sampe popuation, aowing an informa check for random assignment of program timing. As the means in the tabe indicate, there is 36 Correations between a 3-eve poverty index and househod size and ength of residence verify these patterns in the COFOPRI baseine survey data. 5

94 variation in some demographic characteristics across program and non-program regions. Namey, sampe househods in program areas on average have smaer dweings (fewer rooms), are more ikey to have eectricity, and have higher nativity rates (percentage of members born in province). However, whie statisticay significant differences exist across program and nonprogram areas, no statisticay significant differences in differences are observed between squatters and non-squatters in program and non-program areas (coumn 3). This finding supports the use of non-squatters as a comparison group. 5.3 Regression Mode The basic estimate of program effect is obtained from the foowing OLS regression: L i = ß 0 + ß 1 (N) + ß (N) + ß 3 (squatter) + ß 4 (program) + ß 5 (program*squatter) + a X i +e i, (5.3.1) where L i refers to some measure of househod abor suppy; N is number of househod members; squatter refers to a househod with no pre-program property tite; program indicates whether the househod ives in a neighborhood that has been reached by the program; and X i is a vector of demographic contros. The coefficient on the interaction between program and squatter, ß 5, is the estimated program effect, which provides a measure of the conditiona (on X i ) average difference in time worked by ex-squatters in program areas versus non-program areas. The incusion of contros for squatter and program fixed effects corresponds to a standard DID empirica specification. The second estimate incorporates a gradient of the program effect over time. 6

95 L i = + ß 6 (program periods) + ß 7 (program periods*squatter) (5.3.) Here, the variabes of interest are the interactions between the dummy variabes for squatter househod and program entry, ß 5, and between the squatter dummy and the number of periods since the titing program entered, ß 7. Together, these pick up any differentia patterns in abor suppy of squatters reative to non-squatters that are consistent with the neighborhood s years of program experience. The combination of these interactions, ß 5 + ß 7 (mean # program periods), is the estimated average program effect. This can be interpreted as the margina change in the amount of abor suppied by the average squatter househod in a program neighborhood for each additiona period with a property tite. 37 Additiona variation in program response by residentia tenure and househod size is captured by the foowing modes: L i = + ß 8 (tenure) + ß 9 (tenure*squatter) + ß 10(tenure*program) + (5.3.3) ß 11 (tenure*program*squatter) L i = + ß 1 (N*squatter)+ß 13 (N*program) + ß 14 (N*squatter) + ß 15 (N*program) + (5.3.4) ß 16 (N*program*squatter) + ß 17 (N*program*squatter) The variabe tenure in equations and refers to the number of years a househod has ived in a residence, which is used as a summary measure of househod informa rights and corresponds to τ in the theoretica mode. In (5.3.3), the average program effect is 37 The vaidity of the inear constraint on the program effect across periods of program entry is tested by running unconstrained versions of the regressions for a outcome measures, presented in Appendix E. In these modes, instead of the interaction term squatter*(program period), four dummy variabes are incuded corresponding to each period of program entry such that the sope of the program effect is not constrained to be inear over time. The coefficient estimates revea a strikingy consistent trend of increasing program effect over number of periods since the titing program began, supporting the use of a inear restriction. For a outcomes, adjusted Wad tests fai to reject the hypothesis that the differences between program periods are equa (and therefore that the sope of the program effect is inear). Furthermore, the estimates in Appendix E revea the necessity of aowing for a eve effect of the program that is arger than the period-to-period program effect for a outcomes except in-home work. 7

96 captured by [ß 5 + ß 7 (mean # program periods) + ß 11 (mean residentia tenure)], whie in (5.3.4) the estimated average program effect is [ß 5 + ß 7 (mean # program periods) + ß 11 (mean residentia tenure) + ß 16 (mean househod size) + ß 17 (mean HH size) ]. The quadratic term in (5.3.4) captures the idea that eisure hours are ikey to be correated across househod members, such that the ikeihood that any househod member is at home in a given moment is increasing with famiy size at a decreasing rate. A estimates are adjusted to account for the sampe custers and strata, the standard errors derived from the Huber-White robust estimator for the variancecovariance matrix. 38 The set of regressors contained in X i is common to a regressions in the empirica section, and incudes contros for the number of working-aged househod members, city fixed effects, ot size and residentia tenure, as we as a constant. In addition, X i incudes the foowing demographic contros: sex, age, education and degree eve of househod head; number of househod members, number of schoo-age chidren, number of babies (ages 4), fraction of aduts that are mae, fraction of aduts that are immigrants (born outside of province), and number of members age 70 and oder; size of property, househod residentia tenure, whether indoor pumbing, whether the property was acquired by invasion, and whether the property was inherited; whether dweing ies within waking distance of nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market, and this indicator interacted with waking time to each ocae; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and whether each of these existed two years ago, and whether neighborhood has government schoo, chid, food or genera socia assistance program. A regressions aso incude a set of dummy interactions between cities and program entry, and between cities and pre-program tite status. The incusion of these interactions absorbs 38 For a description of the technique used to estimate standard errors, see Chapter. of Deaton (1998). 8

97 potentia regiona variation in program impementation and regiona differences in informa property institutions that coud be driving reative differences in program impact between tited and untited residents. It is arguabe that the incusion of such a wide set of demographic contros amounts to over-controing. However, as detaied in Appendix F, a of the proceeding resuts are robust to the excusion and incusion of a wide variety of right-hand-side variabes. For a outcomes in Section 5, coefficient estimates from regressions with no demographic contros are presented aongside the saturated modes. 5.4 Endogeneity Concerns With respect to the choice of right-hand-side variabes, whie an effort was made to incude principay time-invariant househod characteristics, there remain many sources of potentia endogeneity in the set of regressors. Most notaby, endogenous migration of househod members, fertiity and housing investment are a behaviors arguaby correated with tenure security. The robustness of regression estimates to a wide range of specifications provides genera evidence against the roe of endogeneity bias (see Appendix D). With respect to investment, increased credit opportunities among post-program squatters shoud ony bias downward the estimated program effect, given that greater abiity to smooth income has the potentia to ower the margina utiity of wage income, thereby reducing the opportunity cost of eisure. Furthermore, credit has the potentia to increase educationa investment, an additiona pu factor reducing empoyment hours in post-program areas. Nonetheess, in order to minimize endogeneity concerns, ony ot size and underground residentia infrastructure are incuded 9

98 among the characteristics of the residence, both of which are reasonaby beieved to be reativey time-invariant. 39 The potentia endogeneity of credit access generates one notabe compication in interpreting the home business outcome ony. Namey, it is possibe that the untited are sufficienty credit constrained to be unabe to cover the fixed cost of moving a business from inside to outside the home (this woud appy to non-sef-empoyed as we if abor force participation invoved a high enough fixed cost of participation). However, this is inconsistent with corresponding sampe data on business oans, as we as evidence from four separate studies of credit effects of COFOPRI, in which property tites were found to have no significant effect on residents access to business credit (Fied and Torero, 00; Cockburn, 000; Kagawa, 001; Torero, 000). Individua sampe seection arising from househod migration is unikey to be a reevant compication in this anaysis due to the fact that is was widey known that new residents were ineigibe for a property tite. Migration of individua househod members, however, coud compicate the anaysis if non-random migration rates differentiay atered famiy composition of treatment and contro groups. The principa evidence that this is not the case comes from direct comparisons of treatment and contro group data on residency of househod members, recent migration of past members, number of working-age members, and age and sex of househod head, none of which revea significant differences in famiy composition. As fertiity is potentiay infuenced by changes in tenure security, chidren under age two are excuded from right-hand-side measures of famiy size. 39 A 000 study of a sampe of COFOPRI participants by Kagawa reveaed that residentia eves of subterra infrastructure, and in particuar the pubic water connection system, does not systematicay vary with neighborhood reguarization (Kagawa, 000). 30

99 A fina source of potentia endogeneity bias arises in a experimenta and quasiexperimenta settings in which participants are aware of treatment. In particuar, program timing woud not identify the treatment effect of obtaining a tite if the contro group adjusted their behavior in anticipation of treatment. Anecdota evidence from COFOPRI office personne suggests that there was much uncertainty as to the timing and choice of program ocations, making it is unikey that househods woud fee confident in advance that the program woud eventuay enter their vicinity. 40 More importanty, this behavior woud ony bias downward the estimated program effect in my mode. The ony possibiity for upward biases is an Ashenfeter dip response of future program participants, in which squatters spend disproportionate time safeguarding property when the program is about to enter. Whie possibe, there is no intuitive nor anecdota reason to expect demand for invasions to rise in anticipation of the program. 6 Empirica Resuts 6.1 Program Effect on Tenure Security The theory of Section 4 posits that obtaining a property tite affects househod abor suppy by increasing tenure security. Naturay, if becoming a tited property owner does not change househods perceived probabiity of eviction, there wi be no expected program effect. Survey data on househod perceptions of eviction ikeihood are therefore informative for verifying the presumed reationship between tite acquisition and tenure security before continuing with the anaysis. The foowing indicators are expored: whether the househod reported experiencing a change in tenure security with the acquisition of a property tite, whether eviction is considered very ikey and whether eviction is considered very unikey. Indeed, 40 Interview with Caros Gandofo, COFOPRI Office, Lima, Peru, August

100 according to the simpe DID estimates in Tabes 3a 3c, the data provide evidence of a basic program effect that is consistent with the variations in program entry and groups of beneficiaries described above. Squatters in program neighborhoods report significanty higher current eves of home security (3a, 3b) and changes in tenure security associated with property tites (3c). Thus, it is reasonabe to concude that the program indeed ed to significant increases in tenure security. 6. Reduced-form Estimates of Effect on Labor Suppy Strong evidence of a corresponding program effect on househod abor suppy comes from a visua comparison of pre-program squatter and pre-program tited househods in program and non-program neighborhoods. Figure pots the distribution of annua abor force days per househod worker by these four sub-sampes. 41 The density marked by squares, which corresponds to squatters in neighborhoods not yet reached by the program, is visiby distinct from the densities corresponding to the two groups of residents in program areas and aso from that of the tited residents in non-program areas. Two important patterns are worth noting: First, among non-squatters, the empoyment hours distribution of residents across program regions is very simiar, whereas among squatters the distributions depend heaviy on whether or not the program has entered. 4 Second, not ony are the work patterns of the comparison group reativey constant across program and non-program areas, but they are aso simiar to the work patterns of pre-program squatters after the program has entered. These reguarities end confidence to the use of non-squatters as a comparison group. The program effect interpretation of such a picture is that the titing program eads squatter househods to shift outward their distribution of work 41 Whie my empirica estimates wi focus on weeky and not annua hours worked, the patterns refected in Figure is usefu in providing the cearest iustration of my identification strategy. 4 In fact, the hours distribution of squatters in program areas stochasticay dominates that of squatters in non-program areas. 3

101 hours to reach that of tite-hoders, as woud occur if ack of tenure security were responsibe for the empoyment hours differentia. To further expore this pattern, a inear regression framework is needed to contro for househod, neighborhood and regiona determinants of abor suppy which, if unbaanced, coud confound measures of program impact. Tabes 4 6 present the coefficient estimates of interest from modes (5.3.1) (5.3.4) of Section 5.3. Coumn 1 reports resuts from the sparsest regression, which constrains the program effect to be constant across househod type and time since titing, whie coumns, 3 and 4 aow the program effect to vary by time since program entry, ength of residence and famiy size, cumuativey. The outcomes of interest are tota househod weeky hours of work, tota househod annua months of work, and fraction of househod members in the abor force. 43 Weeky hours of work refer to ast week s empoyment, and is constructed from survey questions on the number of days and mean hours per day worked ast week asked of a househod members who report having worked during the past week. Working-age members who are not in the abor force and those who are in the abor force but report not having worked ast week are assigned empoyment hours vaues of zero. Annua months of work is constructed from survey questions on the number of months worked of the ast tweve, asked of a househod members who report having worked during the past year (which 43 In tota, 99 househods are dropped from the anaysis due to missing abor suppy information (a househod is considered to have missing weeky hours data if it has one or more members who both report having worked ast week and have positive reported vaues of either hours worked per day or days worked per week and missing vaues of the other variabe), 31 househods have missing data on property size and/or oca eementary schoo faciities, 0 househods are excuded in two custers in which program entry does not match institutiona data on regiona program timing, and 8 househods are excuded because a members are reported as over the age of 80, eaving a tota of 59 househods. Due to the survey design, information on daiy and houry work time was incompete (but not missing) for 69 individuas who reported not working in the ast week but working over the ast tweve months. For such individuas, ony the number of months out of the year worked was asked, and not days a week or hours a day worked. For the weeky hours variabe, these individuas are assigned vaues of zero for days worked ast week and hours worked ast week. For the annua hours estimates in Figure, predicted vaues of hours and days a week were assigned to these observations based on a vector of househod and individuas characteristics. No predicted vaues, however, were used in the regression or probit estimates. 33

102 incudes a those who worked ast week). 44 Labor force participation is measured as the fraction of working-age househod members who report either having worked, had a temporary absence from the abor force or searched for a job during the past week. In coumn 1 of Tabe 4 the margina effect impied by the estimated coefficient on the interaction term between squatter and program is roughy 13.4 hours per week. In coumn, which aows the program effect to increase with time since the program began, the margina effect impied by the estimated coefficient on the interaction term between squatter and program periods is roughy 14.5 hours per week, whie the fixed effect is -1.7 hours but insignificant. This impies a tota program effect of roughy 16. hours per week for the median squatter househod with two periods of property rights. For the average househod without a property tite, this impies a 17% increase in tota househod abor suppy per week or around two days of fu-time work. The ong-run or steady state effect of the program refected in the estimated effect on househods with the maximum number of program periods, is an average increase of 45 hours of empoyment per week across the entire target popuation of squatters roughy the same as one fu-time worker being added to the abor force. For new househods and househods with few working-age members, the program effect is even arger. The estimates in coumn 3, in which the program effect is aowed to vary by residentia tenure, provide evidence that newer residents increase abor hours more in response to an increase in tenure security. In the regressions that account for differences according to househod years of residence, the estimated program effect rises to.6 hours per week for the average squatter famiy with 15 years of residentia tenure a 3% increase in househod abor suppy. Furthermore, aowing the program effect to vary by residentia tenure 44 nfortunatey it is impossibe to combine months and hours responses to create a summary measure of annua abor suppy without using predicted vaues of weeky hours for peope who worked ast year but not ast week. 34

103 accounts for the negative coefficient on the main effect of the program in coumn. When the program effect is aowed to vary by famiy size, we observe even stronger evidence that the impact of the titing program on abor suppy is concentrated among househods with few potentia workers. In coumn 4, when both sources of variation in treatment response are taken into account, both the eve effect and the dose effect of the program become significant. Athough the estimated effect on the average squatter househod fas to 1.3 hours and becomes insignificant, the estimates indicate that the size of response depends heaviy on househod type. Thus, sma famiies and famiies with few years of residence account for the majority of the program effect captured in coumns 1 3. As mentioned in Section 4, the coumn 4 patterns of heterogeneity in program response according to residentia tenure and famiy size provide additiona evidence that unobservabe factors are not biasing the resuts. Whie poverty and program effect shoud decrease with residentia tenure, poverty and program effect move in opposite directions with respect to househod size. Thus, any unobserved heterogeneity between eary and ate program neighborhoods that is correated with poverty eve coud not be responsibe for both patterns of variation in program infuence. To expore in more detai how the program response varies by househods size and residentia tenure, Tabe 4a presents the estimated program effect for a range of househod types. At east two things are worth noting from this chart. First of a, the program effect does not appear to kick in unti more than a year after the first tite is distributed. This coud be driven by the fact that titing within a neighborhood takes an estimated eight months to compete, such that a disproportionate number of househods in the most recent program regions are sti untited 35

104 by the time of the survey. 45 Aternativey, this coud refect an adjustment ag necessary for househods to either re-optimize abor suppy or to ascertain the increase in tenure security associated with their newy acquired and tite. Secondy, the program effect fas with famiy size ony for househods with ess than five workers. As shown above, the quadratic function estimating the program effect according to famiy size reaches a minimum at five working-age members, at which point the estimated program effect remains we above zero. This is inconsistent with the mode of Section 4, in which, as ong as it is significant, the program effect fas with househod size, in which case the minimum of the quadratic function shoud not be bounded away from zero additiona hours. Instead, the resuts above suggest a mode in which either desired eisure time per capita fas (equivaenty, desired consumption per capita increases) with househod size, or ese the demand for tenure security increases with househod size (as opposed to the mode s assumption that security demands are independent of number of members, controing for ot size, residentia tenure, and forma rights). Most ikey, the second association is responsibe on account of unobserved heterogeneity in househod type correated with househod size. As reported in Tabe 5, the effect of the titing program on tota househod months of work tes a simiar story to the estimates on weeky hours. The measured program effect on househod annua empoyment months is approximatey.9 months, sighty ess than the month effect impied by the weeky hours estimates. Differences between the sizes of the program effects reported in Tabe 4 and Tabe 5 argey refect the extent to which reductions in abor suppy driven by tenure insecurity are due to shorter average work weeks versus extended 45 Time estimate reported in a mimeo on the program procedure distributed by the COFOPRI office in Lima. 36

105 periods of unempoyment or non-participation. Thus, the combined estimates suggest at east some increase in the number of abor force participants. Indeed, Tabe 6 reveas that added workers account for a significant portion of the change in famiy abor suppy resuting from the titing program. When the same regressions are run on househod abor force participation rates, we observe an impied 6 7 percentage point increase in the number of working-age househod members who are empoyed or searching for work (coumns 1 and ). With an average 49% abor force participation rate among squatter househods with four working-age members, an effect of this size woud be accompished if one in every four househods that obtains a property tite adds a worker to the abor force (5%/4 = 6.5%). Even if every such added worker worked fu time (48 hours per week), additiona abor force participants coud not account for the entire impied program effect on hours. This suggests that average hours of the empoyed are aso higher among program participants. As evidenced in Figure, a rough comparison of average hours per worker reveas a difference in the average number of empoyment hours of workers in program areas and non-program areas of around 5 hours per week. In the average two-worker (four-member) famiy, this accounts for approximatey two-thirds of the program increase in hours. Tabe 7 decomposes by gender the program effect on hours to study separatey the impact of titing on work hours of adut men and women. The regressions in coumns 1 and 4 of Tabe 7 are identica to the coumn 1 and 4 regressions in Tabe 4 except that further contros for famiy composition are incuded (number of adut men, adut women, boys and girs aged 1 16, and chidren aged 5 11). Furthermore, to reduce the dimensionaity of the program effect for anaytica purposes, in a Tabe 7 regressions the program effect is constrained to be constant 37

106 over time. 46 The estimates reported in coumns 1 3 indicate that changes in mae empoyment account for the majority of the program effect on hours. In coumn, we see that higher mae hours account for 10.3 out of the impied tota program effect of 1.9 hours. Meanwhie, the difference in femae hours (coumn 3) is sma and insignificant. However, not surprisingy, femae hours are much more eastic than mae hours. Athough the mean effect of acquiring a property tite on hours worked by women is cose to zero and insignificant for the average famiy, when the program effect is aowed to vary with famiy size and residentia tenure, we observe that the effect on femae hours depends heaviy on househod type. For instance, in famiies with ony two working-age members and ten years of residence, the impied program effect on femae abor is 18. additiona hours per week and statisticay significant. This is equivaent to one in three women joining the abor force fu time. In contrast, as observed in coumn 5, the average program effect on mae hours does not depend on either famiy size or ength of residence. 6.3 Effect on Chid Labor Force Participation As motivated by the mode of in Section 4, an increase in forma property rights is predicted to generate a decrease in the amount of chid empoyment if chidren have a comparative advantage in market work reative to home security. The next set of estimates ooks for an effect of property titing on chid abor force participation. In the sampe, ony 8.% of a househods report reguar abor force participation (excuding unpaid domestic work) by chidren 46 When working hours of men and women are regressed separatey on the eve effect and the does response, (program periods)*squatter, it appears that femae hours change initiay but do not rise over time, whie mae hours change ess initiay but graduay increase with additiona years post-program. The discrete change in femae hours suggests that femae workers are ikey to be new abor force entrants, whereas men are more ikey to be od abor force participants increasing hours of work over time. 38

107 between the ages of five and 16. This fraction coud easiy underreport the actua eve of work hours by chidren, as househods might be reuctant to admit to chidren working or not consider irreguar empoyment of chidren when answering survey questions. Yet, whie this number is ow, it is not ceary underreported. According to Internationa Labor Office estimates, 4.1% of a Peruvian chidren aged 6 14 were economicay active in Though the rate shoud be higher for the reativey poor househods in my study, it is aso true that urban househods have ower rates of chid abor than do rura househods in Peru (Ray, 000). To study the effect of urban property titing on chid abor, I estimate a probit mode where the dependent variabe is a dummy indicator of whether or not any househod members under age 16 are reported as working more than five hours per week. I estimate a binary mode rather than modeing the margina effect on chid abor hours due to the fact that the majority of famiies report no chid abor hours, necessitating a imited dependent variabe mode with more stringent functiona form assumptions. Tabe 8 reports the coefficients and margina effects from the probit estimates with a fu set of contros. 47 Coumn 1 estimates the program fixed effect on chid abor, where the coefficient on the interaction term is anaogous to the DID strategy in a inear framework. Coumns and 3 decompose the program effect across househods of different sizes, first aowing the program effect to change ineary with househod size, then by measuring the program effect on ony the smaest 85% of househods. Whie the first coumn shows no average program effect on the probabiity of chidren working, when the effect is aowed to vary by famiy size, we observe a significant effect of property titing on househods with fewer than four working-age members. As reported in coumn, for househods with three working-age members, the impied margina effect of 47 In an effort to avoid mistaking young domestic workers for chidren, I excude singe mae -headed househods. Incuding these househods owers the point estimate of program effect sighty but the estimate remains significant. 39

108 property titing is arge (.4 percentage points, where the mean is 7.8%) and significant. For arger famiies, the effect is cose to zero and insignificant. This is consistent with the theoretica predictions and with the estimates of Tabe 4: if famiies with more than four working-age members are unconstrained by the need to keep famiy members at home, neither shoud they have incentive to send chidren to work in pace of aduts. To estimate the average program effect on constrained househods and aso test for potentia non-inearities in the famiy size effect, I aso run the same mode excuding the argest 13% of househods in the sampe, those with more than six working-age members. Coefficients from this mode are presented in coumn 3. When famiies with many potentia workers are excuded, we observe that obtaining a property tite reduces the average ikeihood of chidren entering the abor market by. percentage points. According to this estimate, the impied program effect on chid abor force participation among famiies with 1 6 working-age members amounts to a reduced ikeihood of roughy 8%. Whie the estimated impact of property titing on the probabiity of chidren working is compeing, the mechanism by which property rights reduce chid abor is ambiguous. If chid eisure is a norma good, the prediction woud aso foow from the income effect of an increase in adut wage earnings due to added work hours. Both expanations are consistent with past research on the determinants of chid abor force participation in Peru, in which it was found that chid empoyment eves are responsive to changes in the adut mae wage (Ray, 000). Whie property titing does not necessariy generate an increase in the adut wage, an anaogous resut shoud arise from a change in the opportunity cost of adut eisure, which in this mode is the wage minus the security vaue of eisure I observe no significant effect of titing on the probabiity of chid schooing. This is consistent with evidence from past studies on chid schooing and empoyment in Peru, which found schooing eves to 40

109 6.4 Effect on Rate of In-home Work The fina question addressed in this paper is whether or not members of a househod participate in market work at home. In the sampe, 4.3% of househods report running a business from home. 49 Whie a genera cass of modes of househod production treat abor suppy decisions as separabe from production decisions, in my mode, in-home work has the additiona feature of increasing tenure security and thereby reducing the househod demand for eisure. Thus, in the absence of a property tite, the mode impies that the decision to run a business from home is determined jointy with decisions about the tota number of hours worked by househod members. According to the predictions of Section 4, the margina vaue of in-home work fas when forma property rights are secured and there is no onger a security incentive to stay at home. As a resut, newy unconstrained decision-makers wi have incentive to more efficienty aocate resources by moving production outside of the home or finding work with an outside empoyer. The nature of this reationship between business investment and and titing is a surprising departure from the rura context, in which and tites are hypothesized to promote investment in home production (Besey, 1995). Given the amount of attention paid to increasing credit access via and titing programs, it is interesting to note that investment demand in the urban case may actuay fa with increases in tenure security if increased worker mobiity causes the rate of sef-empoyment to fa. The probit estimates presented in Tabe 9 support the theoretica prediction. In coumn 1, the margina effect impied by the coefficient on the interaction term between squatter and be unresponsive to chid abor due to the country s high percentage of working chidren who are aso enroed in schoo (Ray, 1999). 49 The exact survey question is: Do you participate in some economic activity within your home or use part of your property as a source of economic activity? 41

110 program periods is a 7.6 percentage point reduction in the ikeihood of owning a home business for the average squatter househod, though the estimate is not significant. However, when the program effect is aowed to increase over time, the impied program effect rises and becomes significant. In coumn, the impied margina change in the ikeihood of working inside the home fas by 11.6 percentage points for the average squatter famiy with two program periods impying a reduction in the rate of home business activity of approximatey 47%. Interestingy, as shown by the coefficient estimates in coumn 3, the program effect on in-home work does not appear to depend heaviy on famiy size or residentia tenure, a possibe indication of omitted variabes bias or other specification error. As an additiona test of variation in program response in which the covariates are not assumed to be constant across househod types, I run the probit estimate separatey for househods iving on properties acquired by invasion of first resident (3% of sampe) versus non-invaded properties (purchased, inherited, or acquired by some other transfer). 50 Househods on invaded properties generay suffer from acute tenure insecurity, and are therefore presumed to have higher demand for a property tite. Coefficients from this estimate are reported in the ast coumn of Tabe 9. As expected, the effect of obtaining a property tite on the decision to operate a home business is much more severe for the sampe of invaded properties and insignificant for a other residents. In fact, the estimated coefficient for reativey insecure househods is more than six times the size of the coefficient for a other residents. The sub-sampe of invaded residences aso exhibits the famiiar pattern of program response by residentia tenure and househod size. In contrast to the differentia effects on in-home work, the program effect on 50 The expected trends were aso observed in comparisons between other sub sampes, incuding mae versus femae househod heads and househods with and without chidren. 4

111 abor hours does not differ substantiay according to whether the househod was acquired by invasion. 51 This combination of abor suppy and business ocation responses suggests a more compicated mode of househod abor suppy and tenure security. The differentia impact of the home business resut for famiies on invaded properties is consistent with a story in which very insecure househods whose security needs require a arger amount of time spent at home and/or very poor househods with itte disposabe income fee particuary constrained by the amount of foregone earnings home security provision entais. Given the aternative to work inside the home, such famiies choose to reduce tota work hours ony up to a point after which it is more beneficia for househod members to shift production inside the home rather than substitute eisure for outside work hours. This woud expain why the home business effect is ony observed among very insecure househods, whereas the abor hours effect is universa Robustness Checks To end support to the previous set of estimates, I use propensity score matching based on the probabiity of residing in a program neighborhood to construct a comparison group of untited residents of non-program areas. Propensity score matching reduces bias created when 51 Nonetheess, the tota program effect on invaded househods is substantiay arger than it is for noninvaded househods, since this popuation experiences both an increase in hours as we as a shift from production inside the home to production outside of the home. 5 Further evidence of the home business effect is provided by a comparison of the average ages of home businesses before and after the titing program. If the impications of Tabe 10 on in-home work fit the proposed mode, not ony shoud the frequency of home businesses be ower but aso the average age of home businesses shoud be higher among squatters after the titing program. Given that househods on invaded properties appear to account for the vast majority of the estimated program effect on rates of inhome work, the estimates are run separatey on the sub sampes of invaded and non-invaded househods. Indeed, we observe that home businesses ocated on invaded properties are an average of 6.3 years oder in program regions than in non-program regions. For home businesses ocated on non-invaded properties, the age difference is sma, positive and insignificant. 43

112 the inear mode underying regression adjustment is incorrect. For comparabiity with the OLS estimates, the same covariates are used to derive the predicted probabiity of a neighborhood being reached by the program in a probit estimate. As reported in Tabe 10, average treatment effects based on kerne matching on the predicted z-score repicate the pattern of program effects found in the OLS estimates in both magnitude and pattern of program impact according to househod size. 53 The estimated abor suppy response to obtaining a property tite is 1.3 additiona hours of work, a 5.1 percentage point increase in the fraction of working-age househod members in the abor force, and a 9.1 percentage point decrease in the ikeihood of running a business from home. When the abor hours resut is broken down by househod size, the estimated average effect of a property tite is 14. weeky hours among househod with ess than four members, 7. hours among househods with 4 5 members, and insignificant among househods with more than five potentia workers. As an additiona robustness check, I run identica estimates on the sub sampe of househods that are ineigibe for receiving a tite on account of having moved into their current residence post Ceary, if property tites are responsibe for the observed change in abor suppy, we shoud observe no program effect among ineigibe residents. 54 Indeed, there is no measurabe program effect on ineigibe househods (in fact, the estimated program effect is negative, though insignificant), which is particuary compeing given that newer househods tend to have very ow tenure security. Finay, the previous resuts are robust to severa aternative definitions of squatter. For instance, atering the definition of squatter to incude househods with unregistered municipa tites actuay increases sighty the predicted effect. In addition, excuding Lima from the anaysis produces the same pattern of coefficients, but with 53 Nearest neighbor and stratified matching produce a simiar pattern of outcomes. 54 Whie ineigibe residents coud serve as a contro group, too few (9.4%) are identifiabe in the data. 44

113 much arger standard errors. The mean ITT effect is smaer, which is accounted for by the ower rate of titing in newer program areas. 7 Long-Run Predictions Given the size of short-run effects of the COFOPRI program, it is interesting to consider the scope of impact of the nationwide titing effort. Based on the previous estimates, what is the change in abor suppy that a program neighborhood woud experience once a eigibe squatters have been tited and the tota increase in perceived tenure security and adjustment ags have occurred? Two chaenges arise in predicting the average treatment effect on tited househods after severa periods with a tite from the current set of estimates. In particuar, the resuts presented in Tabes 4 6 underestimate the ong-term impact on abor suppy because both the rate of titing and the impact of titing within a program neighborhood presumaby increase over time. The first compication arises from the ITT nature of the identification strategy, which is anaogous to non-compiance in experimenta data. The fact that it is impossibe to observe in pre-program neighborhoods whom among the eigibe woud prompty receive a tite upon program entry makes it necessary to incude a eigibe recipients in an ITT anaysis. A disadvantage of this strategy is that it fais to isoate the program effect on the househods that actuay received a tite through the government program and therefore underestimates the ongterm impact of residentia formaization. Whie 74.% of squatters in tited neighborhoods did in fact receive a registered government property tite by the time of the survey, the incusion of the remaining 5.8% of untited program participants biases downward the estimate of program effect. The second compication is that the previous resuts average the short-term effect on newy tited househods with the ong-term effect on househods tited many years ago. If the 45

114 security gains from receiving a tite increase after a tite is granted, the program effect wi aso increase in the ong-run among the 74.% who are aready tited. To generate ong-run predictions, it is necessary to isoate the treatment effect on the treated and to isoate ong-term from short-term gains. One method of isoating the average treatment effect on the treated is to assume that the program infuence is concentrated excusivey among tite recipients. This amounts to using the program as an instrument for whether or not the househod acquires a tite. Because IV attributes the measured ITT effect to ony those who actuay received treatment, it is equivaent to scaing the ITT estimates of program effect by the rate of titing that occurs through the program, a standard method of obtaining a correction factor for non-compiance in experimenta data. 55 These estimates are presented in coumns 1 and of Tabe 11. Whereas the ITT program effect associated an increase of 13.4 hours per week with program intervention, rescaing by the number of tited impies an average treatment effect of 17.8 empoyment hours per week and an 8.4 point increase in the fraction of abor force participants among househods that are actuay granted a tite. 56 However, ony if the benefits of titing are reaized immediatey wi these numbers accuratey approximate the ong-range impact of neighborhood titing among the remaining eigibe househods. This estimate wi sti be biased downwards if the infuence of receiving a tite does not kick in immediatey and therefore some tite recipients are not affected by the 55 For consistency with the previous estimates, IV is appied within the difference-in-difference framework, such that receiving a tite is instrumented with the interaction term between squatter and program neighborhood and the rate of titing among non-squatter househods is controed for by incuding a fixed effect for squatter househods among the right-hand side variabes. See Newhouse and McCean (1997) for a detaied description of IV in the context of difference-in-difference anayses. 56 Extrapoating these gains to future tite recipients aso requires that eigibe untited househods are simiar in type to the tited (permitting ignorabe non-compiance). nfortunatey, the compiers and noncompiers are ikey to be inherenty different with respect to abor suppy outcome, inducing nonignorabe non-compiance, making IV best interpreted as average treatment effects on compiers. 46

115 program by the time of the survey. For this reason, an arguabe improvement over assuming the program effect is concentrated among tited househods is to assume the program effect is concentrated ony among those tited househods that aso report experiencing a change in tenure security. The mode in Section 4 assumes that property tites encourage peope to work by increasing perceived tenure security and thereby decreasing the margina security vaue of eisure. If this is truy a necessary condition for the titing program to affect abor suppy, then a more pausibe excusion restriction is that the program ony operates through changes in security among tited househods. 57 In support of this assumption, the data provide direct evidence of a strong first stage: tabes 3a 3c demonstrate arge concomitant increases in perceived tenure security. Among the 74% of eigibe squatters that were tited, 81% report a change in tenure security associated with the program tite (that is, 81% more than the program/non-program difference reported among non-squatters). As a resut, the IV estimates in coumns 3 and 4 predict that titing efforts that are successfu in making peope fee more secure wi ead to an average abor suppy gain of 4 weeky hours per househod and 11. point increase in the fraction of househod abor force participants. Assuming that a tite recipients eventuay fee more secure, this provides an estimate of the ong-run effect on eigibe househods. However, this cacuation probaby sti underestimates ong-term gains since it assumes that the program effect is imited to a one-time improvement in tenure security (athough it does not necessariy happen right away). In fact, there is reason to beieve that perceptions of tenure security increase graduay over time since tite is granted. In other words, the group of tite 57 Ceary, there are other potentia expanations for the observed positive correation between property tites and abor suppy. For instance, the reative vaue of eisure versus empoyment coud be higher for untited househods due to fewer work opportunities or more incentive to participate in community organizations. In fact, data coected on community organization participation reveas that househod days spent participating in community organizations increases with the acquisition of a tite, further evidence that househod members were ex-ante constrained to stay inside the home. With respect to empoyment opportunities, there is no anecdota evidence that home ownership directy affects empoyment offers. 47

116 recipients who report having aready experienced a change in tenure security are ikey to be a mix of househods that have experienced sma improvements in a short time with househods that have experienced arge improvements over many years. One straight-forward method of isoating the ong-term program effect is to imit the sampe to househods in eary program areas. In coumns 5 and 6, the same IV estimates are run excuding the subpopuation of recent program neighborhoods, or those in which the program entered within the ast 16 months. 58 Consistent with the notion of a agged impact of titing, these estimates are consideraby arger rising from 4 to 38 hours per week, whie abor force participation rises to 1 percentage points. 59 Here, the same excusion restriction appies as in coumns 3 and 4 the program ony infuences tited househods that report changes in tenure security associated with the tite. The benefit of the estimates in coumns 5 and 6 is that it is more pausibe that after at east 16 months with a tite, househods have had sufficient time to become convinced of its security vaue and to adjust their behavior. Excuding the newest program areas generates a more convincing estimate of the average treatment effect on compiers aso because there is ikey to be an anticipatory positive effect of the program on those who are waiting in ine for a tite in new program areas, whereas the househods which have sti not received a tite in ate neighborhoods can be assumed to be ineigibe for a tite due to unobservabe factors. If the program has some degree of positive impact on non-tited squatter househods in program neighborhoods, the scaed ITT estimate wi be biased downwards and underestimate the effect 58 Sampe size restrictions prevent me from isoatin g ony very eary program neighborhoods. For instance, a pre-1996 neighborhoods are concentrated in Lima, reducing the externa vaidity of corresponding predictions. 59 Coincidentay, the OLS estimates excuding ate treatment neighborhoods are identica to the IV estimate of the average treatment effect on a tited househods, or 3.8 hours. 48

117 of titing. 60 If a non-tited househods were ineigibe, this woud not be an issue, however, as evidenced by the growing rate of titing in program neighborhoods over time, this is not the case. In fact, in a survey question in which untited househods were asked whether or not they expected to receive a tite, haf of squatters in program neighborhoods said that they expected a tite in the next tweve months. Finay, using the eary neighborhoods to predict ong-run responses reduces the potentia roe of non-ignorabe non-compiance. Whie househods ate to receive a tite may have systematicay different abor suppy responses than those tited eary, under the restriction that eary program neighborhoods have had time to reach a of the eigibe, the eary program estimates incorporate this potentia heterogeneity into the estimated average treatment effect on compiers. For these reasons, the eary neighborhood program response of coumns 5 and 6 arguaby constitutes a reasonabe ower-bound estimate of the ong-term impact of titing efforts on future neighborhoods. The estimates predict that, once a eigibe squatter househods have been tited for at east 16 months, the average increase in abor suppy attributabe to the program wi be in the order of 38 hours per week. This is consistent with a scenario in which untited househods commony keep one working-age member at home fu-time to protect property. This estimate is aso in the same range as the predictions of the ITT effect on Period 1 househods: According to coumn of Tabe 4, neighborhoods treated in the first program wave experience an increase of around 45 hours per week as a resut of the program, or approximatey 40% more work hours per househod. 60 Aternativey, if the program had a negative abor suppy effect on non-treated househods in treatment neighborhoods, the IV estimate woud overestimate average treatment effects on compiers. There is, however, no reason to beieve that untited househods in treatment neighborhoods fee ess secure as a resut of the program. 49

118 8 Cost-Benefit Anaysis In my sampe, 37% of eigibe government tite recipients in non-program areas are squatters by the strict definition used in my paper. By the end of the program 1. miion tites were granted by the COFOPRI program, suggesting that the above ong-run predictions appy to approximatey 447,000 househods in Peruvian cities, or around ten percent of the country s popuation. This is equivaent to reaxing the time constraint tenure insecurity paced on neary haf a miion workers. In contrast, the cost to the government of nation-wide titing amounts to an estimated $66 per tite, around 0% of which is recovered from user fees and property taxes. 61 The additiona cost to the government of maintaining a nationa property registry in terms of abor hours is margina empoyment figures from pubic registry offices have actuay faen since the consoidation of the oca registries so it is reasonabe to assume that the majority of the program cost comes from the initia mapping and titing process. 6 Thus, it is safe to say that the ong-term benefit fows per househod in wages far exceed the net cost of government titing per househod, which is roughy haf the monthy minimum wage. 63 From a socia accounting perspective, the difference in abor hours expended by househods reative to governments to soidify property caims amounts to societa dead weight oss, and attests to the efficiency of pubic institutions in providing tenure security services. In a compete cost-benefit anaysis, this wefare gain shoud be considered in addition to capita gains resuting from the change in the vaue of property, the ony benefit fow typicay considered in 61 Project costs reported in the cost-benefit anaysis section of the Project Appraisa Document (Word Bank, 1998). 6 There is no indication that enforcement costs have risen, as evidenced by the number of court cases and poice expenditures. 63 Given the possibiity of genera equiibrium effects on the wage, a ower bound estimate of the ong term wage gains per househod wi equa the minimum wage mutipied by the additiona time spent protecting property in the absence of a tite discounted over time by the time it takes to increase de facto rights (which is we over a month) 50

119 project vaue assessments. This is reassuring from the perspective of project appraisa given that capita gains projections based on rea estate price differentias wi overstate increases in househod wefare in the presence of non-transferabe de facto tenure rights (Lanjouw and Levy, 001). Furthermore, the wefare enhancements from capita gains wi not be reaized by the househod unti the residence is sod or mortgaged. My estimates, on the other hand, demonstrate that the benefit fow to squatter househods from a nation-wide titing program in terms of the vaue of hours gained aone we surpasses the costs to the government of project impementation amost immediatey. 9 Concusions This paper has presented new evidence on the vaue of forma property rights in urban squatter communities in deveoping countries. By studying the reationship between the exogenous acquisition of a property tite and househod abor suppy, I have provided empirica support for the anecdota evidence that untited squatters commony attain informa rights by taking time off work to participate in such activities as guarding their property, participating in community groups and fiing administrative caims for formaization. My resuts indicate that the cost of maintaining informa rights via remova from the abor force and distortions in optima househod abor suppy decisions is substantia. There are three major findings. First, unike empoyment responses to most wefare programs, which tend to invove an income effect that potentiay removes peope from the abor force, government property titing programs appear to have the opposite impact on empoyment eves. Second, urban property titing is associated with a significant decine in the fraction of househods that use their residence as a source of economic activity. This finding, which inks property rights to ower rates of business 51

120 investment, aso departs from the property rights iterature in other settings. Furthermore, property tites appear to reduce the househod demand for chid abor in the majority of househods by amost one-third. Whie eary program effects are noteworthy, the ong-run impications of the titing program are particuary striking. In the survey data, many of the treated househods are sti awaiting ega documents. The ITT estimates of program impact on househods tited very eary on suggest that over time, as a househods are actuay reached by the titing program and receive ega ownership rights, newy tited househods wi increase weeky abor force hours by an average of 45 hours per week or an increase in average weeky househod hours equivaent to one fu-time worker. This prediction is supported by IV estimates of the effect of the program on those househods who have actuay been tited for at east 16 months, which predict a 40% gain in abor force hours Addressing this gap in the iterature is important at this juncture for severa reasons. In recent years, a handfu of poicy initiatives have arisen to address tenure insecurity among untited urban residents of deveoping countries. 64 Whie cost-benefit anayses universay suggest that governments are more efficient suppiers of property rights, these caims tend to ignore actua quantifications of the immediate cost to househods of individua property protection, which appears to be substantia (Barber, 1970; Word Bank, 1998). As the resuts of this study indicate, accuratey measuring the return to property formaization requires adequate attention to the cost of informaity. In addition, understanding empoyment responses to property formaization may be critica to understanding and anticipating other market responses to areawide titing programs. For instance, higher empoyment coud be an important channe for 64 In particuar, the Word Bank has sponsored a number of projects aimed at promoting forma property institutions in urban sums wordwide. For an overview, see Land, Security, Property Rights and the rban Poor: Twenty Five Years of Word Bank Experience. Word Bank Briefing Note 8,

121 increasing access to credit, whie the income effect of increases in earnings coud simutaneousy ower demand for credit. Simiary, greater abor mobiity from increased tenure security coud encourage the deveopment of rea estate markets (as opposed to the other way around). Finay, given the evidence on the roe of institutiona causes underying bad macroeconomic performance, these resuts have potentia impications for genera understanding of abor market frictions in deveoping countries. In particuar, in settings characterized by a arge amount of residentia informaity, distortions resuting from informa urban property protection may constitute an important obstace to abor market adjustment and economic growth. 53

122 10 References [1] Acemogu, Daron, Simon Johnson and James Robinson (001). Coonia Origins of Comparative Deveopment: An Empirica Investigation. American Economic Review 91: [] Acemogu, Daron, Simon Johnson and James Robinson (00). Reversa of Fortune: Geography and Institutions in the Making of the Modern Word Income Distribution. Forthcoming, Quartery Journa of Economics. [3] A Matter of Tite, Economist, December 9, 1995: [4] Achian, A.A. and H. Demsetz (1973). The Property Rights Paradigm. Journa of Economic History 33: [5] Aston, Lee J., G.D. Libecap and B. Mueer (1999). Tites and Land se: The Deveopment of Property Rights on the Braziian Amazon. niversity of Michigan Press, [6] Angrist, J.D., G.W. Imbens, and D.B. Rubin (1996). Identification of Causa Effects sing Instrumenta Variabes. Journa of the American Statistica Association 91: [7] Baharogu, Deniz (00). Word Bank Experience in Land Management and the Debate on tenure Security. Word Bank Housing Research Background Land Management Paper, Juy 00. [8] Banerjee, Abhijit, Pau Gerter and Maitresh Ghatak (00). Empowerment and Efficiency: Tenancy Reform in West Benga. Journa of Poitica Economy 110(): [9] Basu, Kaushik, and P. H. Van (1998). The Economics of Chid Labor. American Economic Review 88(3): [10] Besey, Tim (1995). Property Rights and Investment Incentives: Theory and Evidence from Ghana. Journa of Poitica Economy 103(5): [11] Binswanger, Hans and Kaus Deninger (1999). The Evoution of the Word Bank s Land Poicy: Principes, Experience, and Future Chaenges. Word Bank Research Observer 14(): [1] Binswanger, Hans, Kaus Deninger, and Gershon Feder (1995). Power, Distortions, Revot, and Reform in Agricutura Land Reations, Handbook of Deveopment Economics, Voume IIIB, Amsterdam: Esevier. [13] Byamugisha, Frank (000). The Effects of Land Registration on Financia Deveopment and Economic Growth: A Theoretica and Conceptua Framework. Word Bank Working Paper. [14] Caderon Cockburn, J. (1998). Reguarisation of rban Land in Peru. Land Lines, May 1998, Lincon Institute of Land Poicy, Cambridge, Massachusetts,.S.A. [15] Carter, Michae and Keith Wiebe (1994). Tenure Security for Whom? An Econometric Anaysis of the Differentia Impact of Land Titing Programs in Kenya. In S. Migot-Adhoa and J. Bruce (eds.) Land Tenure Reform in Sub-Saharan Africa, Kenda/Hunt Press. [16] Carter, Michae and Pedro Ointo (1996). Getting Institutions Right for Whom? The Weath Differentiated Impact of Property Rights Reform on Investment and Income in Rura Paraguay. Working Paper. [17] Carter, Michae and Eduardo Zegarra (000). Land Markets and the Persistence of Rura Poverty in Latin America: Conceptua Issues, Evidence and Poicies in the Post- Liberaization Era. Chapter 8 in A. Vades and R. Lopez (eds.) Rura Poverty in Latin America, MacMian Press. [18] COFOPRI Office, mimeo, June

123 [19] Coe, Wiiam E. (1996). Labor Migration and rban Empoyment in Deveoping Countries: The Impact of Popuation Growth and Property Institutions. In John Adams and Anthony Scaperanda, eds. The institutiona economics of the internationa economy. Boston; Dordrecht and London: Kuwer Academic: [0] Conger, Lucy (1999). Entited to Prosperity, rban Age Magazine, The Word Bank Group, Fa 1999 Issue. [1] Deaton, Angus (1998). The Econometrics of Custered Sampes. Chapter. of The Anaysis of Househod Surveys: A Microeconomic Approach to Deveopment Poicy. Johns Hopkins niversity Press. [] Demsetz, H. (1967). Toward a Theory of Property Rights. American Economic Review 57: [3] De Soto, Hernando (1990). The Other Path. Harper and Row Pubishers, Inc., New York, New York. [4] Dufo, Esther (1999). Schooing and Labor Market Consequences of Schoo Construction in Indonesia: Evidence from an nusua Poicy Experiment. American Economics Review 91(4): [5] Eastery, Wiiam (001). The Lost Decades: The Deveoping Countries Stagnation in Spite of Poicy Reform. Journa of Economic Growth 9: [6] Eickson, Robert C. (1991). Order Without Law: How Neighbors Sette Disputes, Cambridge: Harvard niversity Press. [7] Feder, Gershon, Tongroj Onchan, Yongyuth Chaamwong and Chira Hongadarom. (1988). Land Poicies and Farm Productivity in Thaiand, Batimore: Johns Hopkins niversity Press, for the Word Bank. [8] Fied, Erica and Maximo Torero (00). Do Property Tites Increase Credit Access among the rban Poor? Evidence from Peru. Research Program in Deveopment Studies Working Paper #3, Princeton niversity. [9] Gaa, A. and O. Razzaz (001). Reforming Land Markets. Poicy Research Working Paper 616, Word Bank. [30] Graham, J. and C. Green (1984). Estimating Parameters of a Househod Production Function with Joint Production. The Review of Economics and Statistics 66(): [31] Gronau, Rubin (1977). Leisure, Home Production, and Work The Theory of Aocation of Time Revisited. Journa of Poitica Economy 85: [3] Gruber, Jonathan and Brigitte Madrian (1997). Empoyment Separation and Heath Insurance Coverage. Journa of Pubic Economics 66(3): [33] Jacoby, H. (1993). Shadow Prices and Peasant Famiy Labor Suppy: An Econometric Appication to the Peruvian Sierra. Review of Economic Studies 60(4): [34] Johnson, Simon, John McMian, and Christopher Woodruff. (00). Property Rights and Finance. American Economic Review, forthcoming. [35] Jones, Eric L. (1981) The European Mirace: Environments, Economies and Geopoitics in the History of Europe and Asia, Cambridge niversity Press, Cambridge K. [36] Hoy, M. and E. Jimenez (1991). Squatters' Rights and rban Deveopment: An Economic Perspective. Economica 58: [37] Jaan, Jyotsna and Martin Ravaion (1998). Are There Dynamic Gains from a Poor-Area Deveopment Program? Journa of Pubic Economics 67(1): [38] Jimenez, E. (1984). Tenure security and urban squatting. Review of Economics and Statistics 66:

124 [39] Jimenez, E. (1985). rban squatting and community organization in deveoping countries. Journa of Pubic Economics 7: [40] Kagawa, Ayako (000). Poicy Effects and Tenure Security Perceptions of Peruvian rban Land Tenure Reguarisation Poicy in the 1990s. Workshop Paper, ESF/N-AERS Internationa Workshop, Leuven and Brusses, Begium, 3-6 May 001. [41] Knack, Stephen and Phiip Keefer (1995). Institutions and Economic Performance: Cross- Country Tests using aternative measures. Economics and Poitics 7: [4] Lanjouw, J. O. and Phiip Levy (00). ntited: A Study of Forma and Informa Property Rights in rban Ecuador. The Economic Journa 11(48): [43] Lopez, R. and C. Romano (1997). Rura poverty in Honduras: Asset Distribution and Liquidity Constraints. Working Paper. [44] McCean, Mark and Joseph P. Newhouse (1997). The Margina Cost-effectiveness of Medica Technoogy: A Pane Instrumenta Variabes Approach. Journa of Econometrics 77: [45] Micei, Thomas J., C.F. Sirmans, and Joseph Kieyah (001). The Demand for Land Tite Registration: Theory with Evidence from Kenya. American Law and Economics Review 3(): [46] Moene, Kar (199). Poverty and Landownership. American Economic Review 8(1): [47] North, Dougas C. (1981). Structure and Change in Economic History, W.W. Norton & Co., New York. [48] North, Dougas C. and Richard P. Thomas (1973). The Rise of the Western Word: A New Economic History, Cambridge niversity Press, Cambridge K. [49] Oórtegui, Ingrid G. Informa Setters in Lima, ESF/N-AERS Internationa Workshop, Leuven and Brusses, Begium, 3-6 May 001 [50] Ray, Ranjan (000a). Anaysis of Chid abor in Peru and Pakistan. Journa of Popuation Economics 13: [51] (000b). Chid Labor, Chid Schooing and their Interaction with Adut Labor. The Word Bank Economic Review 14(): [5] Rosenberg, Tina. Looking at Poverty, Seeing ntapped Riches, Editoria Observer, The New York Times, October 1, 000. [53] Rosenbaum, Robert, and Donad Rubin (1983). The Centra Roe of the Propensity Score in Observationa Studies for Causa Effects. Biometrika 70(1): [54] Schady, Norbert (00). Picking the Poor: Indicators for Geographic Targeting in Peru. Word Bank Working Paper, Number 477. [55] Singh I., L. Squire and J. Strauss (1986). Agricutura Househod Modes: Extensions, Appications and Poicy, Word Bank, Washington, D.C. [56] Torrero, Maximo (1999). Estudio de a Oferta,, Demanda y Fuentes de Credito Informa, Documento Supementario de Estudio: Perfi de a Demanda y Oferta de Credito Forma y Informa, paper prepared for the COFOPRI office, Grupo de Anaisis para e Desaroo, November [57].S Department of Labor (1998). By the Sweat and Toi of Chidren: Voume V: Efforts to Eiminate Chid Labor Bureau of Internationa Labor Affairs, Washington DC, Avaiabe onine at: [58] Wasmer, E. and Y. Zenou (1999). Does Space Determine Search Frictions? A Theory of rban nempoyment, CEPR Discussion Paper

125 [59] Wasmer, E. and Y. Zenou (00). Does City Structure Affect Job Search and Wefare? Forthcoming, Journa of rban Economics. [60] Word Bank Deveopment New Archives, Peru s rban Poor Gain Access to Property Markets, February, 000. [61] Word Bank (1997a). The Lega and Institutiona Framework, ANNEX A1, rban Property Rights Project, Word Bank Interna Paper, Washington D.C. [6] (1997b). Socia Context, ANNEX A1, rban Property Rights Project, Word Bank Interna Paper, Washington D.C.,.S.A. [63] (1997c). Socio-Economic Assessment, ANNEX A3, rban Property Rights Project, Word Bank Interna Paper, Washington D.C. [64] (1997d). Impementation of the Nationa Formaization Pan, ANNEX A5, rban Property Rights Project, Word Bank Interna Paper, Washington D.C. [65] (1998a). Project Appraisa Document, Report No.1845PE, Peru - rban Property Rights Project, Washington D.C. [66] (1998b). Project Information Document, No. PID653. Peru - rban Property Rights Project, Washington D.C. [67] Yao, Yang (1996). Three Essays on the Impications of Imperfect Markets in Rura China, Ph.D. dissertation, niversity of Wisconsin. [68] Yi Yang, Zoia Z. (1999) COFOPRI, an Experience of Land Tenure Reguarization in Informa Settements in Perú: Reguarisation process case study at the Sau Cantora Settement. Paper prepared for the Advanced Internationa Training Programme, Housing and Deveopment, Lund Institute of Technoogy Schoo of Architecture. 57

126 Tabe 1: FONCODES Poverty indicators, 1993 No program Program t Water Roofing Eectricity Sewerage Fraction chidren enroed in schoo Literacy Residentia crowding Manutrition Overa poverty index Note: Means weighted by city fraction of entered neighborhoods. Source: Peruvian Ministry of Economics and Finance 58

127 Tabe. Sampe Means (N=668) (N=08) (1a) (1b) (1c) (a) (b) (c) (3) Program No Program t Program No Program t t Femae head of HH Mean age of HH member Age of HH head HH size (# members) Number of rooms in dweing Pre-program squatter househods Pre-program tited househods Lot size (m) Highest grade head Residence acquired by invasion Age of dweing HH adut iteracy rate Pumbing Light Municipa service (water) HH monthy expend (S/) Whether HH saves Number of members moved/eft HH Number of members born in province Notes: Coumns 1c and c report the t-statistics of the difference between coumns 1a and 1b, and a and b. Coumn 3 reports the t-statistic of the difference in difference. 59

128 Tabes 3a-3c: Evidence of Program Effect on Perceived Tenure Security Tabe 3a: Large change in tenure security with ast tite No Program Program Difference (not yet entered) (entered) Difference-indifference Difference-indifference Not squatter (N=191) (0.01) (0.019) (0.03) Squatter ** (N=559) (0.000) (0.09) (0.037) (0.045) Tabe 3b: Do you consider dweing currenty at risk of eviction/invasion? No Program Program Difference (not yet entered) (entered) Not squatter (N=191) (0.011) (0.013) (0.017) Squatter ** (N=559) (0.03) (0.019) (0.030) (0.035) Tabe 3c: Do you consider dweing currenty very secure from eviction/invasion? No Program Program Difference (not yet entered) (entered) Difference-indifference Not squatter (N=191) (0.01) (0.00) (0.04) Squatter ** (N=559) (0.06) (0.030) (0.040) (0.046) * Significant at the 0.05% eve. ** Significant at the 0.01% eve. Notes: Standard errors in parentheses. Ony eigibe HHs (according to residentia tenure) incuded. Change in tenure security in Tabe 3a comes from survey question: "Did the ast property document you obtained affect the security of your residence?" asked ony of househods with property documents. Data in Tabe 3b and 3c based on responses to survey question, "How secure do you consider your property?" Respondents coud report: (1) Very secure, I do not beieve that it wi be taken; () Secure; (3) Not so secure, I beieve that in any moment it coud be taken; (4) Not at a secure, I beieve that it is very probabe that at some moment it wi be taken. Tabe 3b cassifies househods as "at risk" if they answer (3) or (4). 60

129 Tabe 4: Tota Househod Weeky Hours in Labor Force (N=379) (1) () (3) (4) (a regressions incude demographic contros, city*program years, and city*initia rights) (5) no demog. contros Number working-age members Squatter*program Squatter*program periods Squatter*program* tenure Squatter*program* working-age members (Squatter*program* working-age members) Impied program effect: N=4, T=15 Impied program effect: N=3, T=15 Impied program effect: N=3, T= (3.37)** (3.37)** (3.36)** (6.45) (4.934)** (6.49)* (1.1) (16.69) (6.04)* (7.19)* (5.8)* (5.7)** (5.37)** (6.0)** (0.57)* (0.56)* (0.6) (11.66)* (11.89)* (1.31)* (1.36)* (6.49)* (6.55)** (7.03)** (7.98) (8.65) (6.47)** (7.04)* (8.48)** (7.97)** (8.5)** Impied program effect evauated at N number of working age HH members, T years of residentia tenure and median number of program periods (). * Significant at the 0.05% eve. ** Significant at the 0.01% eve. Notes: OLS regression, dependent variabe is HH tota weeky work hours. Standard errors in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns 3-5 incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing hours or days vaues for working members are excuded.demographic contros incude: sex, age, iteracy and degree eve of HH head; # HH members, # of schoo-age chidren, # of babies, fraction mae, fraction immigrants, and # members 70 and oder; whether indoor pumbing, whether property acquired by invasion, and whether inherited ot; whether dweing ies within waking distance and this indicator interacted with waking time to nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and for the ast two years, and whether neighborhood has schoo, chid, food or genera socia assistance program. 61

130 Tabe 4a: Variation in Program Response according to Househod Type (Outcome: Househod weeky empoyment hours) Residentia tenure=10 years Househod size (# working-age) 1 program period program periods 3 program periods 4 program periods workers 3 workers 4 workers 5 workers (9.81) (8.71) (10.64) (14.43) (9.41) (7.97) (9.80) (13.64) (10.59) (9.3) (10.9) (14.47) (11.61) (10.69) (1.31) (15.69) Residentia tenure=15 years Househod size (# working-age) 1 program period program periods 3 program periods 4 program periods workers 3 workers 4 workers 5 workers (8.69) (7.45) (9.65) (13.73) (8.17) (6.47) (8.63) (1.84) (9.41) (7.98) (9.81) (13.66) (10.45) (9.43) (11.4) (14.88) * Significant at the 0.05% eve. ** Significant at the 0.01% eve. 6

131 Tabe 5: Tota Househod Annua Months in Labor Force (N=379) (1) () (3) (4) (a regressions incude demographic contros, city*program years, and city*initia rights) (5) no demog. contros Number working-age members Squatter*program Squatter*program periods Squatter*program* tenure Squatter*program* working-age members (Squatter*program* working-age members) Impied program effect: N=4, T=15 Impied program effect: N=3, T=15 Impied program effect: N=3, T= (0.59)** (0.59)** (0.60)** -1. (0.94)** (1.9) (.590) (3.410) (5.80) (5.44) (1.180) (1.16)* (1.10)* (1.15)* (0.1) (0.1) (0.13) (.6) (.30) (0.6) (0.6) (1.9) (1.7)* (1.31)** (1.51) (1.64) (1.0) (1.3) (1.59)** (1.50) (1.53) Impied program effect evauated at N number of working age HH members, T years of residentia tenure and median number of program periods (). * Significant at the 0.05% eve. ** Significant at the 0.01% eve. Notes: OLS regression, dependent variabe is HH tota annua months of work. Standard errors in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns 3-5 incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing hours or days vaues for working members are excuded. Demographic contros incude: sex, age, iteracy and degree eve of HH head; # HH members, # of schoo-age chidren, # of babies, fraction mae, fraction immigrants, and # members 70 and oder; whether indoor pumbing, whether property acquired by invasion, and whether inherited ot; whether dweing ies within waking distance and this indicator interacted with waking time to nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and for the ast two years, and whether neighborhood has schoo, chid, food or genera socia assistance program. 63

132 Tabe 6: Fraction of Househod Members in Labor Force (N=379) (1) () (3) (4) (a regressions incude demographic contros, city*program years, and city*initia rights) (5) no demog. contros Number working-age members Squatter*program Squatter*program periods Squatter*program* tenure Squatter*program* workingage members (Squatter*program* workingage members) Impied program effect: N=4, T=15 Impied program effect: N=3, T=15 Impied program effect: N=3, T= (0.011)** (0.011)** (0.011)** (0.018)** (0.017)* (0.08)* (0.050) (0.065) (0.156) (0.173) (0.01)* (0.01)** (0.00)** (0.0)** (0.003) (0.003) (0.003) (0.060) (0.065) (0.005)* (0.006) (0.08)* (0.07)* (0.07)** (0.09) (0.031) (0.034)* (0.038)* (0.035)** (0.038)* (0.041)* Impied program effect evauated at N number of working age HH members, T years of residentia tenure and median number of program periods (). * Significant at the 0.05% eve. ** Significant at the 0.01% eve. Notes: OLS regression, dependent variabe is percentage of working-age HH members who are either empoyed or searching for a job. Standard errors in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns 3-5 incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing hours or days vaues for working members are excuded. Demographic contros incude: sex, age, iteracy and degree eve of HH head; # HH members, # of schoo-age chidren, # of babies, fraction mae, fraction immigrants, and # members 70 and oder; whether indoor pumbing, whether property acquired by invasion, and whether inherited ot; whether dweing ies within waking distance and this indicator interacted with waking time to nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and for ast two years, and whether neighborhood has schoo, chid, food or socia assistance program. 64

133 Tabe 7: Gender Distribution of Househod Weeky Hours (N=379) Number working-age members Number adut men Number adut women Squatter*program Squatter*program* tenure Squatter*program* working-age members (Squatter*program* working-age members) (1) () (3) (4) (5) (6) Tota Hours Men Women Tota Hours Men Women (3.44) (.0) (1.93) (6.13) (4.01) (3.03) (5.79)** (4.01)** (5.78)** (4.05)** (4.74)** (.74)** (4.69)* (.69)** (6.17)* (4.6)* (4.1) (4.74)** (17.51) (15.41)** (0.54) (0.38) (0.35)** (11.50)* (8.46) (7.31)** (1.7)* (0.96) (0.80)** Mean Program Effect SE (6.17)* (4.6)** (4.1) (7.88) (5.38) (5.35) * Significant at the 0.05% eve. ** Significant at the 0.01% eve. 65

134 Tabe 8: Whether any Househod Member Age 5-16 Works (1) () (3) A househods with chidren ages 5-16 (N=1557) Househods with <6 members (N=150) Number boys age 1-16 Number girs age 1-16 Number chidren age (0.157) (0.157)* (0.15)** (0.160) (0.160) (0.03) (0.14) (0.14) (0.155) Squatter*program (0.76) (0.619)* (0.300)* Squatter*program* working-age members 0.80 (0.10)* Mean program effect on HH with 3 potentia workers SE (0.7) (0.34)* (0.30)* Margina effect * Significant at the 0.05% eve. ** Significant at the 0.01% eve. Notes: Binomia probit estimation, dependent variabe is a dummy indicator of whether HH members ages 5-16 report working more than 5 hours/week. Standard errors are in parentheses. A regressions contro for city, size of property and residentia tenure of HH. In addition, coumns a and b incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing hours or days worked vaues for working members are excuded. Demographic contros incude: sex, age, iteracy and degree eve of HH head; # HH members, # of schoo-age chidren, # of babies, fraction mae (of working-age members), fraction immigrants, and # members 70 and oder; whether indoor pumbing, whether property acquired by invasion, and whether inherited ot; whether dweing ies within waking distance and this indicator interacted with waking time to nearest primary schoo, secondary schoo, bus stop, pubic phone, and pubic market; and whether neighborhood has oca bus stop/market/pubic phone/primary and secondary schoo currenty and for the ast two years, and whether neighborhood has schoo, chid, food or genera socia assistance program. 66

135 Tabe 9: Whether Residence Source of Economic Activity (N=97) (1) () (3) (4) (demographic characteristics, city*program years, and city*initia rights) (5) (no demog) (6) (7) Not invaded Invaded Squatter*program (0.178) Squatter*program periods Squatter*program* tenure Squatter*program* working-age members (Squatter*program* working-age members) Impied program effect: N=4, T= (0.091)* (0.137) (0.164) (0.090) (0.105) (0.178)* (0.01) (0.016) (0.186) 0.01 (0.00) (0.178) (0.18)* (0.183)* (0.1)* (0.181) Impied margina change Impied program effect evauated at N number of working age HH members, T years of residentia tenure and median number of program periods (). * Significant at the 0.05% eve. ** Significant at the 0.01% eve. Notes: Binomia probit estimate, dependent variabe is whether residence used as source of economic activity. Standard errors in parentheses. A regressions contro for city, size of property and residentia tenure of HH, and coumns 3-4 incude a reevant intermediate interactions of HH tenure and size. Robust standard errors account for sampe custering and stratification. Ineigibe HHs (residentia tenure pre-1995) and HHs with missing vaues for working members are excuded.demographic contros incude: sex and age of HH head; # HH members, # of schoo-age chidren, # of babies, percentage mae and percentage immigrants; whether indoor pumbing, whether property acquired by invasion and whether inherited ot; and whether neighborhood has municipa services, eectrica infrastructure, whether oca bus stop/market/commissary/primary and secondary schoo two years ago, whether neighborhood has schoo assistance program, custer average waking distance to oca primary schoo, and custer average waking distance to bus stop. 67

136 Tabe 10: Propensity Score Estimates: Kerne Matching Estimator (N=536) Mean of matched treated Mean of matched contros Average treatment effect HH Weeky Hours in Labor Force ** (4.35) -3 working-age members * (6.77) 4-5 working-age members * (3.35) 6-7 working-age members (0.0) Fraction of HH in Labor Force Whether Home Business ** (0.00) * (0.044) * Significant at the 0.05% eve. ** Significant at the 0.01% eve. Notes: Propensity score estimated as probit mode, where dependent variabe is whether or not program enters neighborhood. Gaussian kerne, bandwidth 0.06, bootstrapped standard errors. 68

137 Tabe 11: Instrumenta Variabes Estimates Late program neighborhoods excuded (N=346) Number working-age members Registered property tite (Instrument=program ) Whether change in tenure security (Instrument=program) Weighted compier average treatment effect HH Weeky Hours Fraction of HH in Labor Force (1) () (3) (4) (5) (6) 1.11** ** 11.83** ** 10.81* ** (3.36) (0.011) (3.46) (0.011) (4.19) (0.013) 17.95* * (8.6) (0.037) HH Weeky Hours Fraction of HH in Labor Force HH Weeky Hours Fraction of HH in Labor Force 3.95* -0.11* 37.83* -0.10* (11.88) (0.051) (15.04) (0.06) 17.95* 0.084* 3.95* 0.11* 37.83* 0.10* (8.6) (0.037) (11.88) (0.051) (15.04) (0.06) * Significant at the 0.05% eve. ** Significant at the 0.01% eve. Notes: Property tite and change in tenure security with tite instrumented with interaction between program area and squatter. Change in tenure security indicator comes from survey question: "Did the ast property document you obtained affect the security of your residence?" asked ony of househods with property documents. Set of regressors in a coumns corresponds to OLS regressions from Tabes 4 and 5. Robust standard errors account for sampe custering and stratification. Ony eigibe HHs (residentia tenure pre-1995) incuded. 69

138 Figure 1: Timing of program intervention and poverty across districts in Lima 70