Can Social Programs Reduce Producitivity and Growth? A Hypothesis for Mexico

Transcription

1 INTERNATIONAL POLICY CENTER Gerald R. Ford School o Public Policy University o Michigan IPC Working Paper Series Number 37 Can Social Programs Reduce Producitivity and Growth? A Hypothesis or Mexico Santiago Levy January 2007

2 Can Social Programs Reduce Productivity and Growth? A Hypothesis or Mexico Santiago Levy* January 2007 * This paper was written or the Eight Global Development Conerence organized by the Global Development Network in Beijing, China, January 12-19, I would like to thank Dr. Ernesto Zedillo or inviting to participate in this event. Responsibility or the views expressed here is mine.

3 ABSTRACT Social programs can reduce productivity and growth as they inadvertently generate perverse incentives or workers and irms. The core hypothesis is that these programs segment the labor market, tax ormal salaried employment and subsidize inormal salaried and non-salaried employment. Larger than optimal sel-employment and employment by inormal irms lowers aggregate labor productivity. In turn, dierences in the cost o labor produce dierences in returns to capital across irms, some ormal legally hiring salaried workers and some inormal illegally hiring salaried workers. Given the cost o credit, higher labor costs or ormal irms distort the allocation o investment in avor o the inormal sector; this investment is distributed in many small irms that may ail to exploit advantages o size as a result o irms strategies to evade social security contributions. This lowers the average productivity o capital causing dynamic productivity losses. The analytical argument is linked to empirical evidence indicating that dierences in labor and capital productivity between sectors and irms contribute to explain dierences in productivity growth across countries, on one hand; and to evidence suggesting a negative association between productivity and inormality, on the other. A subsidiary hypothesis is that social programs are partly inanced by reducing public investment rather than raising taxes, limiting the expansion o growth-promoting public inrastructure. The paper suggests that social programs that lower total actor productivity together with the eects o lower public investment partly account or Mexico s lackluster growth and productivity perormance in the context o intensiied international competition and the erosion o the advantages o the North American Free Trade Agreement.

4 I. Introduction Mexico s low growth and reduced competitiveness has become a critical area o concern. In the last decade the average annual growth rate o per capita GDP was 2.1%; and or the last ive years 0.4%. Productivity growth has also been slow relative to other countries; 1 as a result, the country s competitiveness has allen. This lackluster perormance has generated a search or explanations and appropriate policy responses, all the more puzzling as it occurs ater important reorms in the last twenty years aimed at making the economy more competitive and eicient; and ater beneiting rom unusually avorable international conditions, particularly over the last ew years. Given Mexico s openness to trade explanations have centered in the perormance o sectors producing key non-traded intermediate inputs or in the behavior o labor and capital markets. Attention has ocused on the high costs o transport; the monopolistic eatures o the telecommunications sector; the low rates o lending to irms by commercial banks; the high cost and uncertain supply conditions associated with public monopolies in the energy sector; the rigidities in the labor market derived rom onerous iring and hiring regulations; and the relatively low levels o education o the work orce. Without diminishing the merit o these actors or weighing their relative importance, I argue that there is another source o low growth and stagnant productivity overlooked so ar: an incentive structure associated with social programs that induce workers and irms into low productivity jobs and investments, respectively. At irst sight it may seem counterintuitive to argue that social programs can reduce productivity and growth, as a large literature argues that a healthier and more educated labor orce is a positive actor or growth; urthermore, an equally large literature argues that a country with a more equitable income distribution will more likely experience higher and sustained growth rates. 2 And clearly, in a country like Mexico with a very unequal income distribution and an important share o the population living in poverty, social programs are indispensable to redistribute income and promote equality o opportunities. 1 Bassi et al. (2006) ind that average product per worker in Mexico grew at an average annual rate o 1.4% between 1995 and 2004, substantially below the rates observed in OECD countries. Duval (2006) estimates a similar average rate o 1.3% or the period 1988 to An excellent summary o these arguments is World Bank (2005).

5 I do not dispute these assertions; in act, I strongly coincide with them. Eradicating poverty and increasing equity are not only valuable goals in their own right, but in my view pre-requisites or sustainable growth in Mexico. My point is that the combination and eatures o the programs chosen to deliver health, pension, housing and other beneits to workers de acto generate a perverse incentive structure that work against the long run interests o workers, particularly those with low incomes. The challenge, thereore, is not to remove these programs but to reorm them so they go hand-in-hand with increasing productivity and aster growth, not against. It is a question o means, not ends. I ocus on the programs by which workers receive health insurance, housing loans and pensions, principally, based on their labor status. There is no discussion o education or other social programs that provide beneits independently o a particular orm o participation in the labor market. 3 I center attention on the division o social security programs or salaried workers, on the one hand, and social protection programs or non-salaried workers, on the other; on the bundled or unbundled nature o beneits in each case; and on the dierences in their methods o inancing. I show that these eatures are key determinants o the eective cost o labor aced by irms hiring salaried and non-salaried workers, and thus o irms proitability. This paper can be interpreted as contributing to the literature arguing that microeconomic distortions in actor markets are determinants o aggregate productivity and growth; see Banerjee and Dulo (2004). Its novelty lies in pointing out that social programs can segment workers and irms into a ormal and an inormal sector, and that this segmentation is important not only rom the social point o view --as similar workers receive dierent social beneits-- but also rom the point o view o growth and productivity. 4 More particularly, 3 I also do not discuss poverty reduction programs directly targeted on poor households, although i these programs conditioned beneits on the labor status o poor workers similar results obtain; see Levy (2006a). 4 Analyses o segmentation in the labor market usually ocus on minimum wages or, o greater relevance or Mexico, on the eects o regulations regarding iring and hiring o workers. But labor regulations and social security regulations go hand-in-hand. They are either observed or evaded jointly by irms and workers. From a quantitative point o view, however, the impact o social security regulations exceeds that o labor regulations. In Mexico the costs o the ormer are approximately 30% o wages, while the contingent costs o the latter are estimated by Heckman and Pages to be around 3.2% o wages (2004, table 3, p. 30). In any case, rom my perspective labor regulations should be considered social programs since their purpose is to protect workers, just as social security provisions. The analysis here extends to incorporate the eects o these regulations; see Levy (2006a). However, I ocus on social security because in Mexico there is an explicit attempt by the government, backed by increasing budgetary resources, to practically replicate social security beneits or inormal workers through a parallel system o social protection generating incentive problems that add to the ones associated with the better-known labor regulations. 2

6 the paper tries to identiy the links between inormality and actor productivity. This matters greatly or Mexico or a simple but powerul reason: the majority o irms and workers are in the inormal sector. I social programs are a cause o inormality, and i the microeconomic distortions associated with inormality reduce productivity and growth, then social programs can harm productivity and growth, more so i they are expanding rapidly as is the case in Mexico. The paper is organized as ollows: section II describes relevant characteristics o Mexico s social programs and key eatures o its labor market. Section III exploits the standard two-sector static general equilibrium ramework to link these characteristics to the distribution o employment and labor productivity by sector. It then extends the set-up to introduce evasion o social security laws and explores how this behavior aects the productivity o labor and the allocation o investment. Section IV briely surveys papers that give guidance as to the magnitude o these eects. Section V concludes with remarks on the relevance or Mexico o social policy reorm to increase productivity and accelerate growth. II. Social Security and Social Protection II.1 Beneits and Coverage Mexico provides social beneits to workers based on labor status. On one hand, social security is a right o salaried workers only, and an obligation o irms only with respect to the salaried workers they hire. 5 Coverage o social security involves a wide set o beneits: (i) health insurance, (ii) day-care services or children; (iii) lie insurance; (iv) disability pensions; (v) work-risk pensions; (vi) sports and cultural acilities; (vii) retirement pensions; and (viii) housing loans. These beneits have two relevant characteristics: they are paid out o workers and irms contributions; and they are bundled in the sense that workers and irms must pay or all o them, regardless o whether workers desire all or only a subset. 6 5 Articles 12 and 13 o the Social Security Law and article 20 o the Federal Labor law make a key distinction between salaried and non-salaried workers, where the ormer are deined as those perorming subordinated work or a boss (irm) in exchange or a wage. The words subordinated and wage are very relevant here, as they leave open whether workers engaged with irms under risk-sharing or eort-eliciting contracts receiving a commission or proit-sharing are subordinated or not, and are receiving a wage or not. I they are not, as is the established legal doctrine in Mexico, then irms hiring workers with non-wage contracts are not obligated to ailiate their workers to social security, although rom a strictly economic point o view these are also proit maximizing irms and the commissions paid are the equivalent o wages (except that they are based on output or other measurable characteristic, as opposed to hours worked which in these cases may not be observed). 6 Henceorth social security coverage is equated with ailiation with IMSS (Instituto Mexicano del Seguro Social). 3

7 On the other hand, the sel-employed and workers with non-salaried relations with irms are excluded rom social security. However, these workers beneit rom various health, housing, day care and, more recently, pension beneits --grouped here under the label o social protection programs-- which dier rom social security in two relevant senses: one, beneits are not bundled as workers can voluntarily access one (say, a health program) without necessarily accessing another one (say, a housing program). And two, beneits are paid out o general revenues and not with workers or irms contributions. 7 Associating coverage o social security with ormality and lack thereo with inormality the labor orce divides into a ormal and an inormal sector, with ormal/inormal workers receiving social security/social protection beneits. In principle, this division should coincide with salaried and nonsalaried workers. But this is not so, since as a result o evasion o social security laws there are salaried workers with social protection beneits that add to inormal employment. Table 1 describes the magnitudes o each in Note that 36% o salaried workers evade social security laws. Table 1 Mexico s Labor Force, 2005 (thousands o workers) Number Share I. Formal 13, I.1) > 3mw 6, I.2) 3mw 7, II. Inormal 26, II.1) non-salaried > 3mw 2, II.2) non-salaried 3mw 16, II.3) salaried (evasion) 7, III. Total 40, Source: Levy (2006b). Notes: The table excludes public sector workers; mw=minimum wage, so that > 3mw reers to workers earning more than 3 times the mw, taken here as the cut-o point between workers with high and low wages. 3 mw is close to the mode o the wage distribution o workers registered in IMSS. II.2 Mobility o Workers A critical eature o Mexico s labor market is large scale mobility o workers between the ormal and the inormal sector. Here I summarize the results o an exercise where the length o stay in the ormal sector --measured by ailiation with IMSS-- o nine million individual workers is ollowed rom 7 These beneits may be perect or imperect substitutes or those oered by social security; see Levy (2006b). 4

8 July 1997 to July Interestingly, only 11.6% o low wage workers were enrolled with IMSS or nine complete years over the nine-year period, as opposed to 42.4% o high wage workers. At the other end, o all low wage workers enrolled in IMSS in 1997 almost 18% were in the ormal sector or only one year over the last nine; this contrasts with 7.6% o high wage workers. All in all, the average low wage worker who in 1997 was enrolled in IMSS had social security coverage or only 4.3 out o the nine years during which he could have been covered, or 48% o his working time; the corresponding average or high wage workers is 6.5 years, or 72% o his working time. More generally, workers in Mexico have spells o ormality with social security coverage, and spells o inormality with social protection coverage; these transits are more requent the lower is the worker s wage. These results are conirmed using a dierent data base, the Encuesta Nacional de Empleo Urbano, which is also a panel data set that ollows workers ormal-inormal transits (although or only a year). According to this data in % o low wage workers who began the year in the ormal sector ended it in the inormal; and 10% who started the year in the inormal sector ended it in the ormal (Levy (2006b)). Calderon-Madrid (2006) computes transition matrices or Mexican workers between ormality and inormality or years o high and low GDP growth (1997, 2001 and 2005) and inds similar results, suggesting as well that the phenomenon is not a result o the business cycle. 9 On the other hand, transits rom ormality and inormality into open unemployment are very small, and spells o open unemployment are very short; see the evidence in IDB (2004). Note there is no unemployment insurance in Mexico. In sum: there appear to be no substantive barriers or workers to enter the ormal sector and be covered by social security. 10 The ormal-inormal dichotomy is a characterization o the legal status o workers at a point o time, not a permanent separation o individual workers into two mutually exclusive sub-sets. The dichotomy, urther, is less useul or lower wage workers. For 8 The exercise begins in July 1, 1997 when the current Social Security Law came into eect. There were 9 million workers ailiated with IMSS then o which 3 million earned more than 3 mw ( high ) and 6 million 3 mw or less ( low ). The exercise measures number o years in ormality, given that the worker was in IMSS in Workers with 56 years o age or more in 1997 are excluded rom the data base, so departures rom ormality are not due to retirement; when the worker is not in the ormal sector he is in the inormal, openly unemployed, has dropped out o the labor orce or migrated abroad; see Levy (2006b). 9 See Gong et al. (2004), Navarro and Schrimp (2004), Kaplan et al. (2005), and Bosch and Maloney (2006). Duryea, et. al (2006) study nine middle income countries and ind that labor mobility is highest in Mexico. 10 Evidence also suggests that minimum wages are not binding; see Maloney and Nuñez (2004) and Bell (1997). 5

9 these workers the problem is not accessing a ormal job; the problem is that or a variety o reasons there is very erratic permanence in ormality. Transits between ormality and inormality raise undamental questions or Mexico s social policy. It is not the purpose o this paper to pursue this issue. But it is nonetheless relevant because it signals that social security is not working according to design, particularly or low income workers; and because under these circumstances the productivity o all workers is reduced. II.3 Budgetary Resources Graph 1 shows government subsidies or social security and social protection programs since the start o the current Social Security Law in Resources or social protection include various ederal health and housing programs, principally, and exclude state or municipal resources or similar programs thus underestimating total resources or these programs. Graph 1 Federal Subsidies or Social Security and Social Protection Programs (thousands o millions o 2006 pesos) Source: Levy (2006b). Resources or 1997 are not shown as the Social Security Law started in the middle o that year; resources or 2006 are in accordance with the approved budget and may vary by the end o the year. 11 Resources or social security reer only to government subsidies or worker s retirement pensions and health insurance; they exclude the costs o pensions o the transition generation rom the previous Social Security Law, as these are not beneits received by workers that are currently in the labor market. These costs derive rom the change in the pay-as-you-go retirement pension system to a deined-contribution pre-unded system. 6

10 Graph 2 shows total available resources or public health services to ormal and inormal workers. The irst includes contributions by irms and workers and government subsidies or social security health services. The second includes ederal and state resources or social protection health programs operated by ederal and state agencies. Graph 2 Public Resources or Health Services (thousands o millions o 2006 pesos) Seguridad Social Protección Social Source: Levy (2006b). Graphs 1 and 2 depict two simple but powerul acts: (i) in the period ederal subsidies or social protection programs have grown much aster than or social security (110 vs. 21% in real terms) and since 2003 are larger in absolute terms; and (ii) even including social security contributions by irms and workers as o 2003 there are more resources to provide health services to workers in the inormal sector than in the ormal sector. I end this section noting that social protection programs are no longer considered transitory by the government, until economic growth allows more workers to ind a ormal job with social security coverage. Beneits rom some o these programs are now legally similar to social security entitlements 7

11 and their joint delivery, although not legally bundled as with social security, is pursued as an explicit policy goal. 12 III. Social Programs and Productivity III.1 Static General Equilibrium This section describes a ramework to identiy the eiciency eects o social security and social protection programs. The total labor orce L is divided into salaried and non-salaried employment, with social security and social protection coverage, or into ormal and inormal employment, L and L i, respectively. Assume these are all unskilled low wage workers. In the absence o evasion (see below), inormal employment includes two types o workers: the sel-employed and those with a non-salaried labor relation with a irm. The sel-employed might own a productive asset (say, a sewing machine, a hal hectare o land, a truck, a mixer to make ruit juices or sale in city streets) and may exploit that asset on their own in a small selowned irm (perhaps with one or two assistants); critically, the amount o time they devote to work with their own assets depends on the wages they could earn working as salaried or non-salaried employees or somebody else (which measures the opportunity cost o their own time). Alternatively, the selemployed might own no productive assets at all but still work on the own (parking or washing cars in the street, say). Inormal workers, however, can also borrow working capital and sell products in the streets (contraband, newspapers, candy, and so on). Or they can work or a irm selling its products door-to-door or in the streets (cosmetics, ood, lottery tickets). Because their eort cannot be monitored (as hours worked are variable and the work-place is mobile), when employed by a irm these workers receive a commission and not a wage. As a result, they are non-salaried. But in a labor market characterized by mobility across sectors, the level o these commissions depends on the wages and beneits in salaried employment as the latter measure the opportunity cost o their time. The key point is that the decisions o workers with or without productive assets to 12 In 2006 a Presidential Decree created the National Council or Social Protection with, among others, the ollowing purpose: To guarantee the observance o social protection policy, insuring the unctional integration o beneits in health, housing and savings or retirement, among others, that the Federal Government oers to the population lacking social security coverage. My translation o article 4 o the Decree published in the Federal Register (Diario Oicial de la Federación) on February 27,

12 devote part or all o their time to various orms o non-salaried work depends on the earnings they can obtain in these occupations and the beneits derived rom social protection programs vis-à-vis the wages and beneits o social security; see Levy (2006a) or more discussion. Firms divide between ormal and inormal, respectively hiring salaried and non-salaried workers. I initially take the number o irms and the capital stock as given in each sector and assume that there is no evasion o social security laws. As a result, all irms hiring salaried workers register them with IMSS and are ormal irms, while all the sel-employed or workers in a nonsalaried contractual relation with a irm together constitute the inormal sector. Firms produce output according to: (1a) Q = Q (K, L ) and (1b) Q i = Q i (K i, L i ) where K, K i is the (ixed) capital stock in each sector. At this point it is convenient to think o (1a) and (1b) as irm-level production unctions, although irms are not indexed as their numbers are ixed in each case. Let T be the costs o social security beneits and w the ormal sector wage so ormal irms cost o hiring a worker are given by (w + T ). 13 Formal irms hire workers up until the point where the value o their marginal product, MPL, equals (w + T ). For various reasons, however, ormal sector workers might not attach a value to social security beneits equivalent to their costs. 14 Let β [0,1] be workers valuation coeicient o social security beneits so that their utility rom a ormal job is (w + β T ). Clearly, i workers value social security ully, so that β =1, ormal irms labor costs equal ormal workers utility; conversely, i workers do not ully value these beneits, so that β < 1, there is a wedge between what the ormal irm pays and what the worker receives. This wedge is equal to (1 β ) T and is exactly equivalent to a tax on salaried employment (not on labor). In the inormal sector there are no social security costs. Firms (including irms owned by the sel-employed) hire workers so that the MPL i is equal to w i, where this wage should be interpreted as equivalent to the commissions paid 13 T is usually expressed as a raction o the wage but is easier to express it in absolute terms. Note that T includes all other costs associated with hiring a ormal worker: labor taxes and, more importantly, the contingent costs o hiring and iring workers derived rom labor regulations; see Heckman and Pages (2004). 14 Levy (2006b) discusses these reasons in the case o Mexico. 9

13 or non-salaried labor; or to the remunerations obtained by the sel-employed net o the quasi-rents on their own productive assets (i they own any); or to the remunerations earned by the sel-employed parking or washing cars or in any other occupation requiring only labor inputs. At the same time, inormal workers get beneits rom social protection programs whose costs per worker are T i, valued by them atβ iti with βi [0,1]. Note, critically, that neither selemployed workers nor workers in non-salaried relations with irms pay or T i, while T is paid by workers and irms in the ormal sector. So while it may be the case that β T > βiti, so that workers value more social security than social protection beneits, it is also the case that the ormer are costly and the latter ree. O course, it may also be the case that β T < βit i. Letting p w stand or the exogenously given output prices, the equilibrium in this economy is represented by: 15 (2a) p w Q / L - (w + T ) = 0 (2) (2b) p w Q i / L i - w i = 0 (2c) w i + β iti = w + β T (2d) L i + L = L where (2a) and (2b) are the proit-maximizing conditions or ormal and inormal irms, (2c) is the utility maximizing condition or workers, and (2d) the equilibrium condition in the labor market. Graph 3 depicts the distribution o employment with the demand or ormal labor D drawn rom the let hand side and the demand or inormal labor D i rom the right hand side, in the standard two-sector general equilibrium set-up. D and D i result rom the aggregation o individual irm demand curves or labor in each sector and in the case o D i includes the demand or sel-employment. The key properties required or the analyses are that they be negatively-sloped and shit exogenously with output prices. 15 To ease notation I do not index goods by sector or irm. Output can be thought o as a Hicks-composite o many goods whose relative prices are ixed. In a more elaborate model one could dierentiate goods across sectors or let goods be imperect substitutes. The key assumption is that output prices are exogenous (say because all goods are traded). Attention is centered here in the market or actors. 10

14 Graph 3 Social Security with a Formal and an Inormal Sector What is the impact o social programs? Consider irst the case where social security is ully valued and there are no social protection programs. This is obtained by solving (2) with β =1 and T i = 0. In this case the equilibrium is at point A, with L * workers employed in the ormal sector and L i * (not drawn, but equal to (L - L *)) in the inormal. Workers in the inormal sector receive a wage o w i *, while workers in the ormal sector get w *. However, when the value o the social security beneits received by ormal workers is considered, worker s utility is the same across sectors. Note that the MPL in the ormal sector, (w * + T ), is equal to the MPL in the inormal, w i *, so that the allocation o labor in the economy is eicient. A well unctioning social security system, interpreted here as β = 1, maximizes output at world prices and the productivity o labor (and workers wages). Note as well that when β = 1 inormal employment is the same with or without social security, as is the level o output o inormal irms. Evidently, inormality in the sense o lack o social security coverage is inevitable in a legal ramework that excludes non-salaried labor rom the 11

15 obligations o social security. As a result, some level o inormality will always be present because in any economy there are many valid reasons or nonsalaried employment relations: because irms and workers ind it proitable to engage in such relations or risk-sharing or eort-eliciting reasons; and because many workers might also ind it proitable to work on their own. 16 Inormality in this context is as eicient as ormality, except that inormal workers consume a dierent bundle o goods than ormal workers as a result o the existing legal ramework: ormal workers reely dispose o their wage w * and are orced to consume beneits worth to them T, while inormal workers reely dispose o their wage w i * and may or may not use part o their wage to purchase health insurance, save or a retirement pension, and so on. Consider now the solution to (2) when β < 1, so that social security is not ully valued, but assume still that there are no social protection programs (so T i = 0). Note in graph 3 that at (w * + β T ) < w i * workers in the ormal sector are less well-o than in the inormal. This induces workers to move out o the ormal sector with ormal employment alling rom L * to L (and inormal employment increasing). But with ewer ormal workers w inevitably increases, rom w * to w. A key point here, however, is that ormal irms still have to incorporate into their labor costs the ull amount o social security contributions, T, regardless o whether workers value them ully or not; inevitably, their labor costs increase. This induces ormal irms to move rom point A to point C, where MPL is equal to (w + T ). On the other hand, i more workers are now in the inormal sector, they can only be employed i their wages all rom w i * to w i ; this allows inormal sector irms to increase their employment rom point A to G, where the MPL i = w i, and induces more workers to be sel-employed. Note that at w and w i workers are indierent between ormal and inormal employment because (w + β T ) = w i. In this equilibrium the productivity o ormal workers is (w + T ), while that o inormal workers is w i. Since w and w i moved in opposite directions, it is now the case that ormal workers are more productive than inormal workers. But this is not the result o them being more educated; nor is it the result o a barrier to entry into ormal employment. The dierences in productivity between similar workers are caused by an undervalued social security system. Note also that since ormal irms ace higher labor costs they are less competitive than beore; that is why they employ ewer workers and their output is less. 16 Because they have entrepreneurial abilities or because they own productive assets (including land). 12

16 Another characteristic o this equilibrium is that now inormal employment results not only rom a legal design that ex-ante excludes nonsalaried workers rom social security coverage, but also rom a social security system that does not work well. The result is not surprising: i irms and workers are being orced to pay or something, but consider that they are receiving less than that, they will turn to non-salaried inormal activities. The act that GDP and labor productivity all as a by-product o their behavior is something that they are probably unaware o, nor care about. The shit rom [w *, w i *] to [w, w i ] impacts quasi-rents on capital. In the ormal sector the uture low o quasi-rents on K and the price o capital goods alls. The opposite happens in the inormal one; productive assets in the inormal sector become more valuable: the hal-hectare o low quality land exploited by a rural worker; the mixer or stove used by urban workers to make ood or sale in city streets; or the sewing machine used by workers in their own home to make clothing or sale in a tiangis (market in nahuatl). At β = 1 these assets would probably not be used; at β < 1 they are valuable. Graph 4 depicts what happens to wages, labor productivity and the distribution o workers between sectors when social protection programs are introduced, so that (2) is now solved with T i > At points G and C workers in the ormal sector would be receiving (w + β T ), while workers in the inormal (w i + β i T i ) > (w + β T ). But this makes ormal workers worse o than inormal workers, inducing the ormer to move to the inormal sector. As the supply o labor to the ormal sector alls wages there increase rom w to w. As a result, ormal irms labor costs are higher and they reduce ormal employment, which alls rom L to L. However, the additional workers in the inormal sector can only be employed there i wages all rom w i to w i. 17 The government quite understandably introduces these programs because it wants to provide social beneits or workers excluded rom social security. These programs might be considered transitory and imperect substitutes or social security until economic growth extends social security coverage to all workers, as was the case in Mexico or a long time; or may be considered permanent and improved in scope to become closer substitutes or social security, as has occurred more recently. But the attempt to correct or the design laws and operational problems o social security with these programs generates three problems. One, these programs will urther reduce social security coverage as they depress salaried employment. Two, they are not bundled as is social security so while health insurance coverage is extended, or instance, the number o workers orced to save or retirement pensions alls (as is the number o workers covered by disability or workrisk insurance). And three, they induce workers into lower productivity jobs. This last point is the one emphasized in this paper, but rom the point o view o social policy it is important to note that social protection programs extend protection to workers along some dimensions but reduce it along others. 13

17 Graph 4 The Labor Market with Social Security and Social Protection What is the eect o social protection programs on labor productivity? In the ormal sector the productivity o workers is (w + T ) as irms are at point K. In the inormal it is just w i as irms are at point M. The result is that the productivity dierence widens (contrast distance KM vs. distance CG). Dierently put: social protection programs generate productivity losses additional to the ones created by an undervalued social security system. Note that output and quasi-rents o ormal irms all more. The intuition is simple: β < 1 acts as a tax on salaried employment in the ormal sector; T i > 0 worsens things as it acts as a subsidy to inormal employment. The shit rom [w, w i ] to [w, w i ] resulting rom T i > 0 distorts urther the price o productive assets, alling/increasing more in the ormal/inormal sector. The extensive margin to exploit low-quality land expands, and simple (and oten old) capital assets like sewing machines, trucks, and mixers and so on are put to use rather than put to rest. I highlight a result rom graph 4 that may be initially surprising but that is not really so: with social protection programs inormal and ormal workers are both better-o. Despite the act that wages in the inormal sector all, they now receive beneits that they were not getting beore; and because there are 14

18 ewer ormal sector workers, their wages increase. So it is the case that (w + β T ) = (w i + β i T i ) > (w + β T ) = w i. It is noteworthy that workers are better o despite that act that they are less productive and the economy is less eicient. This happens because there are beneits that (apparently) nobody is paying or: the cost o social protection programs. 18 Since the MPL in each sector is (w + T ) and w i, respectively, using (2c) I obtain: (3) MPL MPL i = [(1 - β )T + βi T i ] > 0 This expression identiies the two components o the productivity loss: the irst is the output loss in the ormal sector caused by incompletely valued social security, and it is greater the lower is the value o β. The second is the output loss in the ormal sector caused by social protection, and it is greater the higher is the value o β i. Note that β and β i have opposite eects: improving the valuation o social security augments productivity as ormal employment increases; but improving social protection lowers productivity by inducing over-employment in the inormal sector. The total productivity loss is the value o GDP at world prices oregone because there are too many workers employed in low productivity jobs in the inormal sector, and too ew workers employed in high productivity jobs in the ormal sector. This is: (4) Annual GDP Loss = p w { [Q *(.,L *) + Q i *(.,L i *)] [Q (.,L ) + Q i (.,L i )] } L * = p w L '' which using (3) can be approximated by: [ Q / L Q / L] dl i i (5) Annual GDP Loss [(1 - β )T + βi T i ] Δ L where Δ L(T, T i, β, β i ) is the dierence between L * and L in graph 4 and consists o two components: the shit rom L * to L as a result o β < 1, and the shit rom L to L as a result o T i > O course, these programs must be paid or, but this is done rom general taxes or rom other sources like oil rents; their costs are not internalized by workers and irms in the inormal sector. 15

19 As seen, in a context where the capital stock is ixed the interaction o social security and social protection programs operates through a wedge in the cost o labor to irms across sectors. It is useul to summarize this in δ = (cost o labor in the ormal sector)/(cost o labor in the inormal sector). Since the cost o inormal labor to irms is simply w i, using (2c) again I ind: (6) δ = ( w + T)/( w + β T βit i ) Expression (6) serves to map the range o values or δ given possible combinations o social security and social protection programs. At one extreme, i social security is ully valued and there are no social protection programs, δ =1. In this context, a labor market segmented by social programs has no eects on labor productivity. The other extreme occurs when social security is not valued at all while social protection programs are present and ully valued, so that δ = ( w + T )/( w T i ). I summarize this in: (7) δmax = ( w + T)/( w Ti) > 1= δmin How large could δ max be? To answer this question assume the ormal wage is unity. In Mexico T is approximately 35% o w so without social protection programs δ max = With these programs the wedge is higher. I we assume their beneits are hal o social security δ max = 1.35/ = Even i beneits are one ourth o ne obtainsδ max = 1.35/ 0.92 = 1.47, so the dierence in the cost o labor would be close to 50%. As shown in graphs 1 and 2 in Mexico both T and T i > 0 so unless workers attach no value to social protection beneits, on one hand, and ully value social security beneits, on the other, δ > 1 somewhere in the range given by (7) Social security contributions are 29.5% o the wage; contingent costs o hiring are 3.2%; and state labor taxes 2%, so T = 34.7%. The calculation ignores ederal labor taxes and proit-sharing to workers which could increase T. 20 As noted, resources or social protection programs have increased rapidly; this would moveδ towardδ max. On the other hand, IMSS (2006, ch. X) shows that the availability o inrastructure or social security health services has been alling steadily over the last twenty-ive years: the number o hospital beds per beneiciary has allen rom 1.85 in 1980 to 0.83 in 2005 and the number o doctors oices in irst level health clinics rom 0.60 to This would move β away rom unity and δ towardδ. max 16

20 III.2 Dispersion in the Cost o Labor The discussion so ar has assumed that all workers have homogeneous valuations o social programs, i.e., that: (8a) β l = β and (8b) β l i = β i l L There are many reasons why this may not be so, however. Workers risk-aversion or time-discount rates may dier making some value health insurance or savings or retirement more than others, or example. But even i all workers had equal preerences, there are practical reasons why social security and social protection programs will be valued dierently by identical workers. These have to do with the quality o services provided given variations in the availability o health, day care and housing inrastructure across regions. This implies that even i statutory beneits are the same or all, the real value o these beneits is not. Under these circumstances workers with equal abilities and preerences sel-select into dierent jobs some preerring inormal ones (where there are better social protection services), and some ormal (where there are better social security services). And, evidently, to the extent that social protection services improve relative to social security services --see graphs 1 and 2-- this sel-selection process reduces the supply o labor to the ormal sector and increases it to the inormal. Alternatively, i some regions o the country have more social security inrastructure (say, the North) while other regions have more social protection inrastructure (say, the South), then all else equal inormal employment will be proportionately less in the North. Using (8) in (6) generates a continuum o δ s that can be ordered between δmin and δ max rom the worker who most values social security relative to social protection to the worker who least values it. Equilibrium wages will depend on the number o workers o each kind. The important point is that workers o equal abilities, characteristics and preerences will receive dierent wages. I do not pursue this here or reasons o space, but clearly this will be associated with a dispersion o productivity losses in the economy. III.3 Evasion and Inormal Salaried Employment The wedge in the cost o ormal and inormal labor when δ > 1 induces irms and workers to evade social security laws, generating salaried employment without social security coverage, labeled here L i. 21 Whether this occurs or not 21 Because I ocus on the incentives o social programs I reer to evasion o social security laws only, but this behavior should be interpreted more broadly involving evasion o labor regulations as well as taxes and 17

21 depends on the gains rom evading, which are positive when β < 1 even o there are no social protection programs (T i = 0), on one hand; and the penalties associated with evading the Law, on the other hand. Let F > T be the ine imposed on a irm hiring salaried workers but not registering them with IMSS; and let λ [0,1] be the probability o being ined. It is natural to assume that λ increases with the level o evasion, measured here by the number o workers hired by irms but not registered with IMSS. On their part, workers accept salaried employment without social security beneits only i the irm compensates them with a higher wage than w ; call this w i, the wage paid to salaried workers by irms evading social security. At the same time, evading workers receive social protection beneits as I assume these programs cover all workers without social security, regardless o whether this is because they are non-salaried or because they are salaried but illegal workers (as is the case in Mexico). Since the worker values social security beneits in β T it ollows that (w i + β i T i ) = (w + β T ) or workers to accept salaried employment without social security. The (expected) cost o labor to inormal irms hiring salaried workers, on the other hand, is (w i +λ F). Equilibrium in the labor market with evasion o social security is given by: (9a) p w Q ( L + L )/ L ( w + T ) =0 i (9b) p w Q ( L + L )/ L [ w + λf + ( λ( L )/ L ). F. L ] = 0 i i i i i i (9) (9c) p w Q / L w = 0 i i i (9d) w i + β i T i = w i + β i T i (9e) w i + β i T i = (w + β T ) (9) λ λ( ) = ; λ ' > 0 ; λ (0) = 0 L i (9g) Li + Li + L = L registration with various authorities. But note that the risks o evasion are highly correlated. I employment increases the risk o detection rom IMSS inspectors increases. But more employment means more output, and selling more output increases the risk o detection by authorities rom the Finance Ministry. In turn, the gains rom evasion are not only savings on social security costs but also on taxes and other costs o ormality. 18

22 Inormal employment (L i + L i ) now results rom a legal design that excludes non-salaried workers rom social security; rom a social security system that does not work well; rom social protection programs; and rom the response o irms and workers to these circumstances through evasion o social security. The latter is not a minor issue in Mexico since as can be seen rom table 1 L i /(L + L i ) = Dierently put, in the absence o evasion in 2005 social security coverage would have been 55% higher than what it was! Depending on the mix o legal and illegal employment irms hiring salaried workers can be classiied as ully ormal (L > 0, L i = 0), ully inormal (L = 0, L i > 0), and mixed (L > 0, L i > 0). Aside rom all other parameter values, this depends critically on the probability that the irm is detected, λ. Equation (10) suggests a plausible orm or λ : 0 i (L + L i ) < L _ (10) λ (L i ) = L α i i (L + L i ) [L, L] ; α > 1 _ 1 i (L + L i ) > L _ where L is perhaps 9/10 workers, and L 50. The combination o (9) and (10) generates a distribution o irms by size o employment: ully inormal irms with up to ten salaried workers (here labeled micro and small irms); mixed irms hiring between 9/10 and 50 workers combining registered and unregistered workers (here labeled medium size irms); and ully ormal irms with 50 workers or more registering all their workers (large irms). 22 Table 2 shows the number and the distribution o registered irms with IMSS by number o salaried workers. Table 3 shows the distribution o irms by number o salaried workers employed according to the Economic Census O course, there may be other reasons that can make a small irm be ully ormal (or at least mixed), like having access to government procurement programs or to subsidized credit programs. 23 The tables are indicative since they are not ully comparable. IMSS igures are or 2005 while Census igures are or Moreover, the Census varies its deinition o small/medium/large irms depending on whether they are in commerce, services or industry. Levy (2006b) discusses these dierences but they do not aect the central message: there is a very large number o irms with ew salaried workers each evading IMSS. 19

23 Table 2 Firms and Salaried Workers Registered with IMSS, 2005 Workers by irm No. o irms % No. o workers % , ,967, , ,444, or more 38, ,551, Total 803, ,966, Source: Levy (2006b). Table 3 Firms and Salaried Workers Registered in the Economic Census, 2004 Type o irm No. o irms % No. o workers % Small 2,849, ,829, Medium 114, ,037, Large 41, ,814, Total 3,005, ,681, Source: Levy (2006b). Despite methodological dierences in the deinition o irm size, the contrast is sharp and revealing. At one extreme there are approximately 40,000 large irms that account or about 38% o total salaried employment (about 7.8 million workers), and or about 70% o all ormal employment (about 8.5 million workers). At the other extreme there are about 2.8 million micro and small irms that account or 33% o salaried employment (about 6.8 million workers); but only around 650,000 irms with up to 10 workers are registered with IMSS accounting or 15% o ormal employment (about 2 million workers). I conclude that evasion o social security (or illegal salaried employment receiving social protection beneits) is concentrated mostly in small irms, although a non-negligible amount occurs in medium size irms. III.4 Marginal Labor Costs, New Investments and Factor Productivities Depending on production unctions and parameter values (9) and (10) admit a large number o solutions. In some cases inormal salaried employment will co-exist with ormal salaried employment in the same industry; in some others the whole industry might consist only o inormal irms; and yet in some others only ormal irms might be present. In parallel, the wage rates given by (9) determine as well the level o non-salaried and selemployment. To gain urther insights, in this and the next sub-section I shit to a partial equilibrium analysis and ignore non-salaried and sel-employment in the 20

24 inormal sector. I analyze a simple case with only two irms hiring salaried labor, one ully ormal and one ully inormal, both producing the same good. 24 Assume wage rates in both sectors are given. Assume as well both irms have access to the same technology in the sense o knowing the same inormation about production blueprints. (In act, both irms might not be ar away rom each other.) Note now rom (9a) and (9b) that (w + T ) > (w i + λ F). Since average labor costs are lower or the inormal irm, the ormal irm will have a higher capital/labor ratio. It is critical to highlight, on the other hand, that the response o labor costs to changes in employment diers between irms. To see this, contrast the marginal costs o hiring one more worker (or MCL, marginal cost o labor). From (9a) and (9b) again and assuming constant wage rates: (11a) MCL = w + T so that MCL / L = 0 (11b) MCL i = w + λf + λ( L )/ L. F. L so that MCL / L > 0 i i i i This result is important: ormal irms can expand employment at constant labor costs, while inormal irms that initially have a cost advantage because they are evading social security contributions ace increasing (expected) labor costs. This because hiring more workers increases the probability o being ined, and the ine is imposed on all workers hired not only on the last worker hired that caused the irm to be caught. There is, so-to-speak, an endogenous limit to the expansion o inormal irms; their cost advantage is rapidly eroded with size. 25 Inormal irms will be, on average, smaller than ormal irms. Dierences in labor costs will inluence investment decisions. I want to make two points. One, social programs increase the relative proitability o investments in inormal vs. ormal irms. And two, investments in the inormal sector are biased in the direction o creating new irms rather than expanding i i 24 Think o products like jeans, shoes, toys or shirts produced in a maquiladora plant and the backyard o a house, respectively. L i 25 This depends on the steepness o λ ( ). Note too that the advantage o inormality is rapidly lost because size not only increases expected labor costs but also the probability o being ined by the tax authorities. This could be modeled as inormal irms acing de acto an output price that is inversely proportional to output even i they produce traded goods. Dierently put, even though they are small irms they ace a negatively marginal revenue product curve and not a horizontal demand curve. The pressure or smallness would now come rom both the output price and the cost o labor. 21