A model of the informal economy with an application to Ukraine

Transcription

1 A model o the inormal economy with an alication to Ukraine imon Commander 1 EBRD One Exchange quare London EC2A 2JN, United Kingdom Commander@ebrd.com Natalia sachenkova 2 aculty o Business and Law, Kingston University London Kingston Hill, Kingston uon Thames, urrey, KT2 7LB, United Kingdom Phone: +44() ax: +44() n.isachenkova@kingston.ac.uk Yulia Rodionova 3 University College London, chool o lavonic and East Euroean tudies, 16 Taviton treet, LONDON WC1H BW, United Kingdom y.rodionova@ucl.ac.uk This version 21 etember 29 4 Abstract The size o the inormal economy has grown sharly in many transition countries, articularly in the ormer oviet Union. To rovide a better understanding o this henomenon, our aer develos a model o how the structure o labour comensation inherited rom central lanning aects labour allocation. Using a anel dataset on individuals and households rom the Ukraine Longitudinal Monitoring urveys (ULM) or 23 and 24, we irst quantiy the size o the Ukraine inormal economy. We then write down a model o an economy with state and rivate sectors and ormal and inormal work, where all sectors can emloy both ull- and art-time workers. Private irms choose whether to be ormal - and ay ayroll taxes - or stay inormal, subject to some robability o detection or evading ayroll tax. This setting allows us to derive the imact o social beneits, as well as demand shocks and detection rates, on the allocation o emloyment across dierent labour market states. Predictions rom our model are then tested econometrically using the ULM data. Keywords: inormal sector, labour mobility, structure o labour non-monetary comensation, Ukraine JEL Classiication: J2, P2. 1 irst author; Commander@ebrd.com. 2 Corresonding author; n.isachenkova@kingston.ac.uk. 3 E- mail: y.rodionova@ucl.ac.uk. The authors are listed in the alhabetical order. 4 We are grateul or helul comments to uren Basov, Randolh Bruno, Xiao Chen, Hsueh-Ling Huynh, tanislav Kolenikov, Anna Lukyanova, Alexander Muravyev, ohia Rabe-Hesketh, ergei Roschin, Andrei Tolstoiatenko, Tolga Yuret, and to the articiants in the research seminars and conerences held at EBRD, London Business chool, Oxord University, and LRT-HE. Earlier drats o the aer were resented at an ZA-EBRD Conerence on Labour Market Dynamics, Role o nstitutions and nternal Labour Markets, held at the University o Bologna, May , and at a LRT-HE s seminar in the sring o 29. We thank EBRD and UK DD or suort. 1

2 1. ntroduction A substantial inormal economy - deined as emloyment in irms that do not ay ayroll tax - is a characteristic o many develoing countries. The last decade has seen a strong growth o the inormal economy or most transition countries, accounting or between 35-44% o GDP in the countries o the ormer oviet Union (U) and between 21-33% in Central and Eastern Euroe (Bernabe, 1999; chneider and Enste, 2; Yoon et al, 23; Krstic, 23; Commander and Rodionova, 25). Attemts to understand why there has been such growth have mostly emhasized the role o ublic olicy. n articular, high ayroll tax rates may have raised the incentive or creating inormal jobs. n addition, where the business environment has been roblematic inormal sectors have tended to be relatively large. Yet in the transition economies, it is interesting to note that crosscountry estimations o the size o the inormal economy ind no robust association between size and the extent o reorm, whereas the latter is clearly highly correlated with the quality o the business environment. This suggests that the size o the inormal economy is not simly an issue or the early or slow reormers. While existing models o the inormal economy tend to be organised around two distinct sectors ormal and inormal - a deining eature o the transition country context has been the revalence o multile job-holding, whereby individuals combine ormal sector emloyment with inormal sector emloyment. Recent research has highlighted the imortant consequences or creating incentives or multile job-holding o the dynamic interlay between such actors as control regimes in state sector irms, the structure o comensation and the level o outside oortunities articularly unemloyment beneits made available to searated workers 5. Accordingly, to relect this comlexity and incororate the henomenon o multile jobholding, this aer uses a multi sector model o labour allocation, including a mixed ormal/inormal sector. Based on this ormalisation, we show the relevance to the inormal economy growth o the inheritance o central lanning, articularly the structure o labour comensation where the rovision o housing subsidies, health and child care and other social beneits was common. The aer roceeds by irst quantiying the size o the inormal economy based on unique individual and household data rom the Ukraine Longitudinal Monitoring urveys (ULM) or 23 and 24. We then describe transitions o workers between the three sectors: ormal, inormal and ormal/inormal. We ind that, over the course o , the share o emloyment in Ukraine s inormal sector has jumed rom around 1-16% to between 17-23%. This increase is ar higher i eole involved in agricultural roduction or their own use are also 2

3 considered. econd, based on an analytical model o an economy with state and rivate sectors and involving ormal, inormal and ull and art-time work, we are able to derive testable roositions as to how the reallocation o labour is aected by such actors as the amount o non-monetary comensation, the extent to which such beneits are subsidized by the state, and the robability o the irm being detected evading ayroll tax. The anel element o the ULM dataset is then used to assess the emirical validity o the main rediction rom our theoretical model. Emloying state-o-the-art econometrics, such as mixed logit, our aer rovides imortant irst evidence on the signiicant eects o nonmonetary beneits on the static allocation o labour across the three sectors. n addition, we analyse the dynamics using multinomial logit or transitions o workers between the sectors, which also conditions on the detection robability, involuntary temorary unemloyment (when the worker remains on the irm s ayroll but receives no wage), and occuation. Our overall conclusion oints to the imortant attaching role social beneits lay in determining the mix o ormal/inormal sector emloyment. The aer is organised as ollows. n ection 2, we quantiy the inormal sector in Ukraine using various emloyment measures and micro-data rom the ULM. ection 3 resents a model o the inormal sector and ormulates hyotheses or emirical testing. ection 4 gives a brie descrition o the econometric aroach and reorts estimation results. ection 5 concludes. 2. The inormal sector: evidence rom Ukraine Existing estimates or Ukraine o the size o the inormal economy, derived rom hysical inuts, such as electricity consumtion, or based on money demand unctions, not only indicate that a large inormal sector came into existence near the start o transition in 1992 but show that in the 199s the country aears to have had one o the sharest rates o increase in the inormal economy. Johnson et al (1998) estimated that inormal activity accounted or around 16% o GDP in 1989/9, rising to over 47% by 1994/95, while Lacko (1999) had a yet higher estimate o around 54% at the latter date. chneider (25) laced the inormal sector share o GDP at around 53/54% in The extent o inormal emloyment is comarable with the estimates or the U countries during the same time eriod. or instance, chneider and Enste (2) using hysical inuts suggest that by 1995, the inormal economy accounted or between 35-44% o GDP in the U (see also Bernabe, 1999; Yoon et al, 23; Commander and Rodionova, 25). The extent o inormal emloyment has also been higher than a reorted in Krstic (23) or Central and Eastern Euroe level o 21-33% o GDP. most transition economies are indeed develoing economies, then the raid growth in the size o the inormal sector may simly relect convergence. However, it may also relect some o the articular institutional and other eatures o the transition economies. Exlanations or why this has been the 5 or examle, Rein, riedman and Worgotter (1997). 3

4 case in Ukraine have mostly emhasised the slow and artial nature o reorms and the continuing high level o ayroll taxation. ndeed, throughout the transition, the ayroll tax rate has remained above 4%.Yet, aggregate measures give little or no sense o what constitutes the inormal economy and how that may have changed over time. ndeed, measurement error is a challenge or research and can only be adequately addressed by adoting a cross-level ocus and using household-level data containing observations over time. Given the shortcomings o the aroaches relying on aggregate measures, this aer makes use o the comrehensive and reresentative samle o Ukrainian households, created in two rounds o the ULM. The irst round was imlemented or 456 households and 8641 individuals in 23 with a retrosective art covering some but not all items o the questionnaire or the years, 1986, 1991, A second round was comleted in late 24 and covered 35 households and 721 individuals. The reerence eriod or the second round was 23 and 24. The ULM data rovide extensive inormation on household income and exenditure, as well as individual-level inormation about emloyment status, working hours, earnings, non-monetary (social) beneits and other comonents o income. Based on this data we are able to ut together a number o estimates o the size o the inormal economy as measured by the ercentage o total emloyment - or 1991, 1997, 23 and 24. Measure 1 in the irst column o Table 1 reorts the share o emloyment in inormal activity outside o agriculture. This comrises individuals with an unregistered job as well as those who are sel-emloyed or have a second job or are involved in occasional sulementary work. Broadening the concetualisation o the inormal economy, Measure 2 adds those involved in nonagricultural household roduction and sale o agricultural goods on a secondary basis. Measure 3 urther augments the estimates by including all individuals involved in agricultural roduction or their own use. Not surrisingly the size o the inormal economy is signiicantly aected by whether agriculture is included. On Measure 3, inormal emloyment accounted or over 66% o emloyment at its eak in 24. By contrast, when excluding individuals involved in agricultural activity, the inormal economy share droed to 17%. Because agriculture is largely an untaxed art o the economy in most develoing countries, to ensure comarability, we ocus our attention on the second - and signiicantly narrower - measure o the extent o inormal activity - which includes only those with secondary agricultural outut or sale. This measure gives an inormal emloyment share o 16% in 1991 rising to 26% in The share alls substantially in 23 beore rising again to 23% in 24. Turning to emloyment distributions across ormal and inormal sectors, in 1997 nearly three quarters o workers had jobs solely in the ormal sector. A urther 2% were emloyed in inormal work only, whereas about 6% held multile jobs, articiating in both ormal and inormal work. 4

5 By 23-4, ormal emloyment remained roughly constant at between 75-8% while the share o inormal emloyment ranged between 7-15%. The share o multile job holders was comarable ranging between 9-12%. Table 2 shows transitions across the three emloyment states or 23-4, calculated or 2824 individuals with comlete records in both years. Amongst individuals with ormal emloyment in 23, 9% did not change their status and urther 4-6% moved to either inormal or multile job holding. A substantial roortion o multile job holders switched entirely into the inormal economy. As regards the associated structure o labour comensation, and in articular the intensity o use by state-owned and rivatised irms o non-monetary or social beneits, such as housing subsidy, rovision o health and child care and other services, the ULM data indicate that in both 23 and 24 around 36% o individuals in the samle were in receit o non-monetary comensation. t is notable that more than 5% o multile jobholders in the samle received social beneits. 3. A model o the inormal economy We take the economy to be oulated by three tyes o irms: state-owned, rivate ormal and rivate inormal irms. All tyes can emloy both ull-time and art-time labour. We now outline the otimization roblem or each tye o the irm tate sector irms ull-time emloyees in the state sector receive monetary wages and also non-monetary or social beneits. Part-time emloyees receive only non-monetary beneits. tate irms ay ayroll taxes or their ull-time emloyees but not or their art-timers. Part-time emloyees working in the state sector can also work in the rivate sector and receive a wage. That wage will, however, be discounted by the robability o detection or not aying taxes i they work inormally. We model state- or insider-run irms by analogy with trade unions 6. n the context o develoed market economies, these irms have oten been modelled as either maximizing wages, or maximizing utility with resect to both wage and emloyment, subject to a zero constraint on roits. n the Ukraine context, we assume that, instead o maximizing wages, state-owned irms maximize emloyment (i.e., they reer not to ire existing workers), setting wages consistent with their emloyment objective. This can be modelled as the state irm icking the largest ull-time emloyment ossible subject to a zero-roit constraint. Clearly, the resulting wage-emloyment combination is ineicient. 6 ee, or instance, arber (1987). 5

6 Alternatively, state-owned irms can be assumed to ick a combination o emloyment and wages to maximize rents subject to a zero-roit constraint. n this case, the wage-emloyment solution will be eicient. n terms o the validity o both assumtions, the existing research (see, or instance, Commander et al (1993, 1996a, 1996b) and Commander and Tolstoiatenko (1997) ind suort or both tyes o irm behaviour, with more roitable enterrises adoting joint maximization with resect to wages and emloyment, and less well erorming irms choosing to otimize with resect to emloyment only with resultant labour hoarding. Anticiating the results rom emirical tests described in the next section, the joint wage-emloyment maximization hyothesis modelled below is strongly suorted by our indings on the comarative statics o labour allocation among states. Additional ( not reorted here) results that suort our general conclusion about the relevance o social beneits and wage dierences to understanding the inormal economy, come rom the redictions o a distinctly dierent theoretical ramework develoed in this study to analyse the case o state irms maximising emloyment only, that have been tested by using multinomial logit and the ULM inormation on workers transitions between the three emloyment sectors. 7 We assume that the state (ormal) sector is oulated by identically skilled risk-neutral workers, who can combine ormal sector emloyment with inormal sector emloyment. We label the workers who do this ormal/inormal 8. The utility o the state irm is given by U ( N N, w ) = N w ( 1 ) + N ( θw ( 1 ϕ) + ( 1 θ) w ( 1 τ )),, + τ Mb (1) where: θ = share o art-time emloyees who work in the inormal rivate sector; 7 The emloyment maximisation only model and a set o corresonding emirical results rom multinomial logit are available rom the authors uon request. The estimation results suggest that among those who hold jobs in the ormal sector, enjoying greater beneits increases the chances o their taking an additional job in the ormal-inormal sector but decreases the likelihood o moving out o the ormal sector comletely and taking u a ull-time job in the inormal sector. t has also to be noted that the model resents the case o inter-sector, but not inter-irm, transitions. n other words, workers kee their jobs in the ormal sector irm and take u an additional job in the inormal sector outside the irm, which shits them into the mixed ormal/inormal sector without changing their main emloyment irm. We checked whether this assumtion is satisied in our samle. ndeed, it turns out to be the case: out o 154 movers rom ormal to ormal/inormal only 3 changed main jobs (irms) in 24. However, even i it was not the case, we could easily get around this by assuming not a single ormal irm, but a measure one o identical insider-dominated irms. 8 riebel and Guriev (25) study the eect o emloyer concentration on the attaching role o social beneits and regional worker mobility. 6

7 b = N social or non-monetary beneits rovided by the state sector; = ull-time emloyment in the state sector; N = art-time emloyment in the state sector; M = state sector s total emloyment; w = gross state sector wage; τ = income tax (art o ayroll tax) levied on the ormal sector emloyees ay; τ = ϕ = ayroll tax aid by the emloyer on the ormal sector emloyees ay; robability o detecting the non-aying ayroll tax emloyer; w ( w ) = art (ull)-time wages in the inormal rivate sector and, w ( w ) = art (ull)-time wages in the ormal rivate sector. Due to the otential resence o labour market rigidities, we assume that both ormal rivate and inormal rivate art-time exected wages, w and w, are increasing unctions o the net (atertax) state sector wage, but that the total net state sector wage is not necessarily equal to its total exected multile job sector counterart: ( ) w 1 τ = g( w ( 1 τ )), g '(.) > (2) and w ( 1 ) = z( w ( 1 τ )), z' (.) >. ϕ (3) We also suose that the state sector's total emloyment is ixed at M - in other words, the state sector does not hire or ire, it only moves workers between ull-time and art-time emloyment 9 : N + N M. (4) = With a quadratic roduction unction 1 and assuming substitutability o art-time or ull-time labour - albeit with some eiciency loss δ [,1] - the irm's zero-roit constraint can be written as 9 This assumtion holds or the ULM data in 23 and We use a quadratic roduction unction as it satisies the assumtions we make about the two tyes o labour (ull- and art-time) being substitutes as well as decreasing returns to scale. 7

8 N ( 1 δ ) + M = M (1 s) b + N w (1 + τ ), (5) where s = the subsidy rate rovided by the government to cover the cost o roviding social or non-monetary beneits, τ = the rate o ayroll tax that the irm ays on its ull-time emloyees, and is the outut rice. We assume that in the state irm workers maximize rents (which in our case o linear utility means maximizing the total wage bill) with resect to wages and ull-time (ormal sector only) emloyment, subject to a zero-roit constraint. n this case, the otimization roblem o the state sector irm looks as ollows: MaxU with resect to subject to ( N, N, w ) = N w ( 1 τ ) + N ( θw ( 1 ϕ) + ( 1 θ ) w ( 1 τ )) + Mb (6) ( N, w ), N ( 1 δ ) + M = M (1 s) b + N w (1 + τ ). (7) Grahically, condition 7) inverted U-shaed lines. ( could be reresented in ( w ) N, sace as a set o The eicient combination o ( w ) MRT N, will be ound at the oint where MR = MRT : ( 1 δ ) ( 1 δ ) π w (1+ τ ) w (1+ τ ) = N 2 N (1 δ ) + M 2 N (1 δ ) + M dw = π = = dn w N (1+ τ ) N (1+ τ ), (8) MR U N w dw = = U = dn w ( 1 τ ) θw ( 1 ϕ) + ( 1 θ) w ( 1 τ ) N ( 1 τ ), (9) ign sign ( MR) = { w ( τ ) ( θw ( 1 ϕ) + ( 1 θ ) w ( 1 ))}. (1) 1 τ Then solving or N ; we can ind art-time state sector emloyment N rom N = M N. (11) 8

9 We also have conditions (2) and (3) on wages, and assume that art-time rivate sector wage (either ormal or inormal) is a roortion δ o its ull-time counterart, e.g. w P = δw. (12) This gives us the suly o art-time labour to the rivate sector Private sector irms Private irms can choose whether to be in the ormal sector and ay ayroll taxes or be in the inormal sector, by comaring the relative ay-os to both states, V and V. While rivate inormal irms do not ay ayroll tax but ace the robability o being detected - ϕ, with the corresonding ine, rivate ormal sector irms ay the ayroll tax on both ull- and art-time labour. 11 Both rivate inormal and rivate ormal irms maximize roit subject to the suly o art-time labour rom the state sector, as well as conditions (2) and (3) or wages, and the condition or equilibrium in the market or art-time labour N + N = N. (14) We assume that the constraint on the suly o ull-time labour or rivate irms is not binding normal rivate sector irms the irm chooses to be inormal, it receives an exected ayo o V N max[( 1 )( N N ){ N ( b) N w N w N } (1 (1 ) N = ϕ ϕ ). (15) The irm aces the ollowing otimization roblem Max[( 1 ϕ){ N + ( b) N w N w N }] (1 (1 ϕ)). (16) with resect to ( N, N ) and subject to (2), (3) and (14) ormal rivate sector irms the irm chooses to be ormal, its ayo is given by 11 Note that this eature o the model does not lay a role in econometric tests with the data available to us, due to the very small number o observations on workers with art time emloyment in the rivate ormal sector. 9

10 V = max[ N + ( b) N w N (1 + τ ) w N (1 + τ )]. (17) The irm maximizes roit Max[ N + ( b) N w N (1 + τ ) w N ]. (18) with resect to ( N, N ) and subject to the constraints given by to (2), (3) and (14) Comarative statics To exlore the roerties o the otimization solution or state-owned irms jointly maximizing with resect to wage and emloyment, we vary assumtions about the ratio between the net wage in the ormal sector and the exected wage in the ormal/inormal sector and consider the ollowing three cases. Note that the wage ratio relects by how much insiders in state-owned irms avour emloyment in the ormal sector over emloyment in the mixed ormal/inormal sector. Case 1: uose that the net wage in the ormal sector is greater than the exected wage in the ormal/inormal sector w ( ). (13) ( τ ) > θw ( 1 ϕ) + ( 1 θ ) w ( 1 ) 1 τ n this case, the indierence curves o the state irm s insiders are negatively sloed. The otimal tangency oint ( N, w ) OPT will lie to the right o the zero-sloe oint o the iso-roit (zeroroit) curve 12. The value o the marginal roduct is less than the marginal (wage) cost. Case 2: The net wage in the ormal sector is less than the exected wage in the ormal/inormal sector w ( ). (14) ( τ ) < θw ( 1 ϕ) + ( 1 θ ) w ( 1 ) 1 τ n this situation, the indierence curves o the state irm s insiders are ositively sloed. The otimal tangency oint ( N, w ) OPT will lie to the let o the zero-sloe oint o the iso-roit (zero-roit) curve 13. The value o the marginal roduct is greater than the marginal (wage) cost. 12 The ormal sector irm is in the diminishing marginal roduct art o the iso-quant. 13 The ormal sector irm is in the increasing marginal roduct art o the iso-quant. 1

11 Case 3: The net wage in the ormal sector equals the exected wage in the ormal/inormal sector w ( τ ) = θw ( 1 ϕ) + ( 1 θ ) 1 w (1 τ ). (15) The indierence curves o the state irm s insiders are horizontal. The otimal tangency oint N, w OPT will be at the zero-sloe oint o the iso-roit (zero-roit) curve. The value o the marginal roduct is equal to the marginal (wage) cost. These three ossible outcomes have dierent imlications or our main question o interest: the dierences in the imact that social beneits have on the emloyment in the ormal, ormal/inormal and inormal sectors. We can now sign the eects on emloyment in the various sectors and states o a change in subsidies (s), beneits (b), the ayroll tax rate (τ ), the detection robability (ϕ ) and rices ( ). Column b o the below table summarizes the signs or the exected directions o the relationshis in the model. or Case 1, as social beneits increase rom b to b 1 > b, the zero-roit curve shits downwards, and the otimum ( N, w ) OPT shits in, resulting in lower wages and lower ormal sector emloyment, N, and higher ormal/inormal emloyment. N s b M τ φ N N N N N We call this roerty the "attaching" roerty o social beneits, in the sense that desite lower exected wages in the ormal/inormal sector, a higher level o social beneits leads to an inlow o workers into that sector. A higher subsidy (Column s in the above table) would have an oosite eect to that o an increase in beneits, leading to a higher ormal sector emloyment, N. A ositive shock to aggregate demand (a higher ) would also lead to higher ormal sector emloyment, but higher wages as well. Note that we have assumed that workers who seek multile jobs in both ormal and inormal sectors can ind emloyment there. ince these ormal/inormal workers are substitutes, with some eiciency loss, or those who are emloyed solely in the inormal sector, the inlow o workers rom the ormal to the ormal/inormal sector will reduce inormal sector 11

12 emloyment, as well as exected wages in the ormal/inormal and inormal sectors. 14 A higher robability o detection (columnϕ ) will increase ull-time state sector emloyment and rivate ormal emloyment, but will reduce inormal and ormal/inormal emloyment. Cases 2 and 3 roduce drastically dierent results. n Case 2, as social beneits increase and the zero-roit curve shits down, the otimal oint ( N, w ) OPT shits down to the right o the old otimum, so that ormal sector emloyment is higher, while ormal sector wages are lower than beore and ormal/inormal emloyment decreases. Higher rices bring about lower ormal sector emloyment but also higher wages. Case 3 oers much the same results 15. There are imortant testable roositions to be gleaned rom the irst-order conditions. irst, when the ratio between ormal wage to exected ormal/inormal wage is less than unity, the sloe o the insiders indierence curve is ositive, indicating that ull-time ormal emloyment within the irm is an economic bad or the insiders, and that their utility is decreasing in it, or, conversely, that insiders utility is increasing in art-time emloyment (naturally, as both grous receive equal beneits). o when the iso-roit curve shits u due to an exogenous decrease in beneits or an increase in the subsidy to beneits, imlying that higher emloyment can be sustained or the same zero roit, the insiders will shed their bad consumtion good (ull-time labour) and acquire more o the good one (art-time labour). n contrast, when the ormal sector wage is high relative to the mixed ormal/inormal sector, a decrease in beneits (and/or an increase in the subsidy to beneits) shits the iso-roit curve u leading to a higher consumtion o the reerred tye o emloyment, so that ull-time emloyment o insiders will go u and mixed sector emloyment will all. The oosite haens when beneits increase. 4. Testing the model with Ukrainian data We now test the model s roositions regarding the imact o changes in the structure and inancing o comensation on labour allocation across the three sectors. We use micro data rom the ULM, exloiting inormation on emloyment by sector. We use a mixed multinomial logit model o sector choice that integrates correlated errors arising rom reeated measurements on workers A mixed multinomial logit estimation o sector choice 14 As or the shit rom the ormal/inormal and the ormal into the inormal sector, we do not model it exlicitly but we exect social beneits to lay an attaching role. We also exect a higher subsidy to social beneits and a ositive shock to aggregate demand to have a similar eect. 15 Detailed results are available on request. 12

13 The revealed choice between J alternative sectors or work y it is observed or individual worker i on occasion t. The choice set contains just three alternatives. These are: (i) being emloyed in the inormal sector; (ii) being emloyed in the ormal sector; and (iii) holding multile jobs in the ormal/inormal (mixed) sector. Associated with each alternative sector j is a j robability o being emloyed in this sector, π it. n our aer, redictors o labour allocation (sector choice) include the ocal indeendent variables - the count o social (non-monetary) beneits, the resence o subsidy to beneits, and wages - and the control variables that cature economic and socio-demograhic characteristics o individual i and contextual actors oerating on the irm and sector levels. Table 3A describes the main indeendent variables and the control variables and Table 3B gives their descritive statistics. The redictors reside in a matrix o exlanatory variables X it =(x it 1,, x it J ), with x it j being a column vector associated with the robability j π it. We use an extension o the multinomial logit model, where the resonse robabilities j π it deend on the nonlinear transormations o the linear unction, X β + u, and where because o it j i heterogeneity between individuals, individual-seciic random intercets u i account or intraindividual correlation caused by multile observations or individual i. (see, or instance, Rabe- Hesketh and krondal, 26). With a redictor vector x it j that includes a constant term, and with the last among j = 1,..., J alternatives as the reerence category, the conditional robability o a articular choice j can be written as ollows: ( α + X β + u ) ex j j it j i Pr( y, ) ex( ). it = j X it ui = π it = (16) J 1 α + X + u k = 1 j itβk i The eect o x im (the mth characteristic or individual i) on the logit o choice j relative to choice k ( i.e. on the log-odds) is obtained as the contrast β β. The random eects u i are assumed to be indeendent and identically distributed according to a normal distribution. Note that in our seciication, the vector u i allows or random variation in intrinsic reerences across individuals with resect to their choice o emloyment sector but remains constant over time and between alternatives or work. n the model above (see ection 3), the eect o social beneits deends on the ratio o exected ormal to ormal/inormal wages in both sign and magnitude. The sign is deendent uon the wage ratio being more or less than unity. To test this roosition, we add to the model 13 jm km

14 an interaction term between the wage ratio dummy and the social beneits variable, where the wage ratio dummy takes the value o one i the relative wage is greater than one. 16 We exect to see a negatively-signed coeicient on this interaction variable. We also include a second interaction term that is constructed as the roduct o the redicted wage ratio and the beneits variable. This second interaction term is exected to show the imact o changes in the wage ratio on the relative magnitude o the eect o social beneits on sector choice. Our conjecture is that taking u work in the ormal/inormal sector will overall be ositively (negatively) inluenced by the rovision o social beneits (subsidies to beneits), and by whether workers have exerienced comulsory leave (temorary lay-os), which is an indicator o the level o activity in the irm 17. Using the ULM data, we now create a sub-samle comrised o 616 individual-years with comlete records or 23 and 24 or the variables listed in Table 3. Table 4 shows the estimation results when the regression coeicients reresent log-odds ratios. A ositive coeicient or an indeendent variable imlies higher odds o observing an individual being in the destination sector j rather than in the ormal/inormal (mixed) sector that is taken as the reerence category. To allow or the ossibility o a non-linear relation between the amount o social beneits and emloyment sector choice, we categorize the count variable or social beneits by using our categories (Table 3). or all the categories o the categorised variable or beneits, the main eects are ositive and signiicant (Table 4). This result is robust and holds or various seciications, suggesting that non-monetary beneits are an imortant actor aecting sector choice. The reerence or being in the ormal sector over the ormal/inormal sector is ositively related to the amount o beneits er se. However, the total eect o social beneits also deends on the wage ratio between the ormal sector and ormal/inormal sector. As redicted by the model, wage ratios greater than unity are associated with a higher level o beneits leading to the ormal/inormal sector being chosen with a higher robability. To exlore this rediction urther, we comute the economic eects or the beneits and the beneits-wage ratio dummy interactions. Because available statistical sotware that handles quantiication o such eects or categorical variables is restricted to binary logit models, 18 we estimate a binary logit or choice as an aroximation o the irst equation o Table 4 so that we can comute the total eect o beneits on the robability o being in the mixed, ormal/inormal sector. We can quantiy the eects in a dummy-by-dummy interaction and hence use the our-category reresentation or the 16 Note that a continuous scale is assumed or the beneits variable in the interaction term. 17 t is a roxy or an aggregate demand shock. 18 Note that in the mixed logit ramework, such eects are nonlinear and deend on the realised values o the other covariates (Mitchell and Chen, 25). 14

15 beneits variable. We estimate a seciication where the categorical variable or beneits is interacted with the wage ratio dummy. The ull set o estimation results rom binary logit is available uon request, while Table 5 dislays the corresonding results rom a mixed logit model that has the same set o indeendent variables. The results o the binary logit estimation are consistent with, and comlement, the mixed logit results o Table 4 19 and Table 5 2. O articular interest are the total eects o each o the beneits grous and subsidy. We ind that while all beneit dummies are ositive and signiicant, or beneits16_grou2 (one or two beneits) and beneits16_grou4 (our to six beneits) the interaction dummies are insigniicant. However, in the case o having three beneits, both the main eect or beneits16_grou3 and the interaction term are signiicant. or the wage ratio dummy equal to zero, the beneits16_grou3 (three beneits) dummy decreases the robability o choosing the ormal/inormal sector by 6 er cent, while or the wage ratio dummy equal to one, the beneits16_grou3 dummy actually increases this robability by 1 er cent. This result suorts the rediction o our model. We conclude that or high enough ormal sector wages, social beneits lay an attaching role in the mixed (ormal/inormal) sector, thus also inluencing the comosition o the urely inormal sector and ormal emloyment. ubsidy to beneits is insigniicant 21 or the ormal sector work alternative, but tends to reduce inormal emloyment. A higher redicted relative ormal wage increases the robability o working in the ormal sector. t also increases the robability to be emloyed in the inormal sector since higher ormal emloyment leads to lower mixed emloyment Conclusion The growth o an inormal economy has been a eature o many transition countries. This aer looks articularly at the case o Ukraine. Relying on the data rom the ULM, we started by estimating the size and comosition o the inormal sector over the eriod rom , and tried to shed some light on individual and irm-level actors behind emloyment choices through a novel analytical model o the irm-level utility maximization by insiders. Our aroach, while building u on the standard models o trade unions behaviour, brings in a number o original eatures that describe an 19 The wage ratio and the wage ratio interacted with continuous beneits are both excluded rom the binary logit seciication due to multicollinearity. 2 n this estimation, we also control or resondent s age, age squared, gender, educational attainment and settlement tye. 21 n the binary logit seciication corresonding to that in Table 5, subsidy to beneits actually turns ositive and signiicant. n articular, at the median level o the other covariates, subsidy raises the robability o being in the ormal sector by 4 er cent. 22 Comulsory leave was always insigniicant and was droed. Emloyer size was not included due to endogeneity concerns, as data limitations recluded the inclusion o lagged variables. Note also that these results are robust to 15

16 economy with a strong inheritance rom the lanned economy, such as the imortance o nonmonetary beneits in workers comensation, labour hoarding in the orm o unaid temorary comulsory leave and state subsidies to beneits. When the model s redictions are tested on the ULM anel data using mixed multinomial logit, we ind that, in terms o labour allocation among ormal, mixed (ormal/inormal) and inormal sectors at a given oint o time, non-monetary comensation lays an attaching role in the mixed ormal/inormal sector or a high enough redicted ormal/mixed wage ratio. Our indings conirm, i only at the sectoral level, revious results by Commander and chankerman (1997) and riebel and Guriev (25) on the imortant workerattaching role o social beneits at the irm level. We also ind or the static case o the model that the emirical results oint to the joint maximization o the irm s utility with resect to both emloyment and wages. n the dynamic setting, the results rom a multinomial logit analysis o transitions between sectors again suggest the same imortant attaching role o non-monetary beneits. However, the otimization roblem here tends to be with resect to emloyment only. This is one uzzle that merits urther analysis. One ossible exlanation or this could be the underlying otimization olicy o irms rom which transitions take lace. or the mover starting in the ormal sector, her decision to make a transition may relect the resonse to her current irm s olicy to otimize with resect to only emloyment. inally, our results have imlications or olicies aiming at imroving eiciency o labour allocation in transition countries. Although our indings relate to Ukraine, we susect that they may generalize beyond its borders, in articular to the other countries o the ormer oviet Union. adding the wage ratio term to the set o regressors. The results are available on request. 16

17 Reerences Bernabe, abine, A roile o inormal emloyment: the case o Georgia. nternational Labour Oice, Geneva. Commander, imon, Rodionova, Yulia, 25. The inormal sector in Ukraine: its size, structure and dynamics. EBRD and London Business chool, mimeo. Commander, imon, Tolstoiatenko, Andrei, A model o the inormal economy in transition. World Bank, mimeo. Commander, imon, chankerman, Mark, Enterrise restructuring and social beneits. Economics o Transition, 5, Commander, imon, umana Dhar, Yemtsov, Ruslan, 1996a. How Russian irms make their wage and emloyment decisions. n: Commander, imon, an, Qimiao, chaer, Mark, E. (Eds.), Enterrise Restructuring and Economic Policy in Russia. Washington, D.C.: The World Bank, Commander, imon, Lee, Une, J., Tolstoiatenko, Andrei, 1996b. ocial beneits and the Russian industrial irm. n: Commander, imon, an, Qimiao, chaer, Mark, E. (Eds.), Enterrise Restructuring and Economic Policy in Russia. Washington, D.C.: The World Bank, Commander, imon, Liberman, Leonid, Yemtsov, Ruslan, Wage and emloyment decisions in the Russian economy: an analysis o develoments in Working Paer No ED. Washington, D.C. arber, Henry., The analysis o union behaviour. n: Ashenelter, O., Layard, R., and Card, D. (Eds.), Handbook o Labor Economics, Ch.18, Elsevier riebel, Guido, Guriev, ergei, 25. Attaching workers through in-kind ayments: theory and evidence rom Russia. World Bank Economic Review, 19, Johnson, imon, Kauman, Daniel, chleier, Andrei, The unoicial economy in transition. Brookings Paers on Economic Activity, 2, December. Krstic, Gorana, 23. Emirical asects o the ormal and inormal labour market in the ederal Reublic o Yugoslavia. PhD, University o ussex. Lacko, Maria, 2. Hidden economy an unknown quantity? Economics o Transition, 8, Mitchell, Michael N. and Chen, Xiao, 25. Visualizing main eects and interactions or binary logit models. tata Journal, 5, Rabe-Hesketh, ohia, krondal, Andrei, 26. Multilevel modelling o comlex survey data. Journal o Royal tatistical ociety Association, 169, Part 4, Rein, Martin, riedman, Barry, L., Worgotter, Andreas, Enterrise and social beneits ater communism. Cambridge University Press, Cambridge. chneider, riedrich, Enste, Dominik H., 2. hadow economies: size, causes and consequences. Journal o Economic Literature, 38, Yoon, Yang-Ro, Reilly, Barry, Krstic, Gorana, Bernabe, abine, 23. A study o inormal labour market activity in the C-7. World Bank, mimeo. 17

18 Tables: Table 1: Ukraine: inormal sector emloyment in 1991, 1997, 23 & 24 (% o all working) Measure 1: share o emloyment in inormal activity outside o agriculture Measure 2: Measure 1 lus individuals involved in non-agricultural household roduction and sale o agricultural goods on a secondary basis Measure 3: Measure 2 lus individuals involved in agricultural roduction or their own use Table 2: Transition matrix or (%) N obs = 2824 ormal only ormal / normal normal only ormal only ormal/normal normal only

19 Table 3A: Deinition o variables Variable name Emloyer characteristics and wage dierences across sectors ocial beneits (count o beneits1-6; reerence category Beneits_grou1 = No social beneits) Beneits16_grou2 (1 or 2 beneits) Beneits16_grou3 ( 3 beneits) Beneits16_grou4 (4-6 beneits) Variable descrition Dummy = 1 i resondent receives one or two tyes o beneits Dummy = 1 i resondent receives three tyes o beneits Dummy = 1 i resondent receives our to six tyes o beneits ubsidy to beneits Emloyer size (reerence category Emloyer size 1 = 1-9 emloyees (micro irms) Emloyer_size eole Emloyer_size eole Emloyer_size 4 25 eole and more Dummy = 1 i the enterrise is a budgetary enterrise, or a state enterrise, or a local municial enterrise, or a state or collective arm Dummy = 1 i the enterrise has 1 to 49 emloyees Dummy = 1 i the enterrise has 5 to2 49 emloyees Dummy = 1 i the enterrise has >25 emloyees Ratio o redicted ormal sector wage to redicted ormal/inormal sector wage Wage Ratio Dummy Beneits x (Predicted ormal sector wage/ Predicted ormal/inormal sector wage) Ratio o redicted wages constructed using Heckman estimation. Dummy =1 i (Predicted ormal sector wage/ Predicted ormal/inormal sector wage) >1 Beneits variable treated as continuous times the Ratio o redicted wages constructed using Heckman estimation. Beneits Wage Ratio Dummy Emloyment sector Beneits variable treated as continuous times Wage Ratio Dummy Categorical variable = 1 i resondent has ormal emloyment, = 2 i resondent has both ormal and inormal emloyment, = 3 i resondent has inormal emloyment only. (continued on next age) 19

20 Table 3A: cont. Worker characteristics socio-demograhics and measures o access to, and level o education; revious unemloyment, occuation Age emale ettlement Tye (reerence category Village) PGT (small settlement o town tye) mall or medium-sized town Large city Caital city Education (reerence category Education1 = diloma o high school /general secondary education) Education2 Education3 Previous unemloyment (i.e. whether was temorarily laid o) Comulsory leave Occuation / Job Tye (reerence category being manager or roessional ) Technician Clerks ervice worker killed agricultural worker Artisan Plant oerator Elementary occuation Armed orces Variable descrition Age o resondent, in years Dummy = 1 or emale resondent Dummy = 1 i resondent lives in PGT Dummy = 1 i resondent lives in small or medium-sized town Dummy = 1 i resondent lives in a large city Dummy = 1 i resondent lives in the caital city Dummy = 1 i resondent has incomlete roessional higher education, or bachelors, masters, or candidate o sciences degree Dummy = 1 i resondent has comleted grades 1-6, 7-9, 1-11 or received a PTU diloma Dummy = 1 i resondent exerienced comulsory leave in the current emloyment Dummy = 1 i resondent works as technician Dummy = 1 i resondent works as clerk Dummy = 1 i resondent works as service worker Dummy = 1 i resondent works as skilled agricultural worker Dummy = 1 i resondent works as artisan Dummy = 1 i resondent works as lant oerator Dummy = 1 i resondent has an elementary occuation Dummy = 1 i resondent serves in armed orces 2

21 Table 3B: Descritive statistics or the main indeendent variables Variable Mean td Median ubsidy_(to)_beneits Overall Beneits_cat == Overall Beneits_cat == Overall Beneits_cat == Overall Beneits_cat == Overall Emloyer size na Overall na na 24 Wage ratio : Ratio o redicted ormal sector wage to redicted ormal/inormal sector wage Overall Wage ratio beneits Overall Wage ratio dummy.2.4. Overall Beneits Wage ratio dummy.4 1. Overall Gender.5.5 Overall Age Overall amle requencies or the settlement tye variable Village PGT mall town Medium-sized town Large town Caital city Overall

22 Table 4: Estimation results or sector choice rom mixed logit, with beneits entering the Beneits Wage Ratio interactions as a continuous variable 23 Deendent variable: ector or Work Choice Reerence category: Work in ormal/normal ector ndeendent variables: Equation : Work in ormal ector Equation : Work in normal ector Emloyer characteristics and wage dierences across sectors: ocial beneits (relative to Beneits_grou1 = No social beneits) Beneits16_grou2 (1 or 2 beneits) 1.28 *** *** (.272) (.382) Beneits16_grou3 ( 3 beneits) *** *** (.385) (.56) Beneits16_grou4 (4-6 beneits) 1.15 ** *** (.51) (.899) ubsidy to beneits dummy *** (.173) (.342) Predicted ormal sector wage/ Predicted ormal/inormal sector wage ***.841 * (.43) (.48) Beneits (Predicted ormal sector wage/ Predicted ormal/inormal sector wage) (.197) (.295) Wage Ratio Dummy: 1 i (Predicted ormal sector wage/ Predicted ormal/inormal sector wage)>1 droed droed Beneits Wage Ratio Dummy *** *** (.16) (.262) Comulsory leave Not included Not included Emloyer size (relative to Emloyer size 1 = 1-9 emloyees, i.e. micro irms) Not included Not included Emloyer_size 2 (1-49 eole) Emloyer_size 3 (5-249 eole) Emloyer_size 4 (25 eole and more) Year dummy or *** 1.71 *** (.64) (.48) Constant *** *** ( ) (95.465) Variance o random intercets (.387) No. o observations Reorted: coeicients (log odds ratios), robust standard errors in arentheses. igniicance levels: * - 1%, ** - 5%, *** - 1%. Weighted by samle (oulation) weights. 23 n this estimation, we also control or resondent s age, age squared, gender, educational attainment and settlement tye. 22

23 Table 5: Estimation results or sector choice using mixed logit, with beneits entering the Beneits Wage Ratio interaction as a categorical variable 24 Deendent variable: ector or Work Choice Reerence category: Work in ormal/normal ector ndeendent variables: Equation : Work in ormal ector Equation : Work in normal ector Emloyer characteristics and wage dierences across sectors: ocial beneits (relative to Beneits_grou1 = No social beneits) Beneits16_grou2 (1 or 2 beneits) 2.24 *** *** (.292) (.372) Beneits16_grou3 ( 3 beneits) 1.26 *** *** (.257) (.467) Beneits16_grou4 (4-6 beneits) *** (.31) (.634) ubsidy to beneits dummy *** (.173) (.342) Predicted ormal sector wage/ Predicted ormal/inormal sector wage droed droed Beneits (Predicted ormal sector wage/ Predicted ormal/inormal sector wage) droed droed Wage Ratio Dummy: 1 i (Predicted ormal sector wage/ Predicted ormal/inormal sector wage)> (.323) *** (.336) Beneits16_grou2 Wage Ratio Dummy *** *** Beneits16_grou3 Wage Ratio Dummy Beneits16_grou4 Wage Ratio Dummy (.629) *** (.56) droed (.842) (.984) droed Comulsory leave Not included Not included Emloyer size (relative to Emloyer size 1 = 1-9 emloyees, i.e. micro irms) Not included Not included Emloyer_size 2 (1-49 eole) Emloyer_size 3 (5-249 eole) Emloyer_size 4 (25 eole and more) Year dummy or *** *** (.4) (.68) Constant *** *** (.97) ( ) Variance o random intercets 2.78 (.23) No. o observations Reorted: coeicients (log odds ratios), robust standard errors in arentheses. igniicance levels: * - 1%, ** - 5%, *** - 1%. Weighted by samle (oulation) weights. 24 n this estimation, we also control or resondent s age, age squared, gender, educational attainment and settlement tye. 23