Indian Labor Regulations and the Cost of Corruption Tables

Table 3: Tau vs Strikes and Lockouts strikes per capita (1) tau -0.272∗ (0.145) (2) tau -0.196 (0.148) mandays lost due to strikes per capita (3) tau (4) tau -0.119 (0.157) -0.148 (0.159) lockouts per capita (5) tau (6) tau -0.0544 (0.146) -0.0915 (0.143) mandays lost due to lockouts per capita (7) tau (8) tau -0.0527 (0.145) -0.0995 (0.139) log of net state domestic product pc -0.400 (0.234) -0.493∗ (0.238) -0.506∗∗ (0.235) -0.515∗∗ (0.235) share of employment in manufacturing 2.548 (3.644) 4.482 (4.171) 3.831 (3.995) 3.754 (3.913) Constant Observations 0.721∗∗∗ (0.187) 18 4.352∗ (2.204) 18 0.646∗∗∗ (0.212) 17 Standard errors in parentheses Only including Major Indian States ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 23 5.017∗∗ (2.272) 17 0.620∗∗∗ (0.198) 18 5.191∗∗ (2.221) 18 0.618∗∗∗ (0.197) 18 5.291∗∗ (2.222) 18 Table 4: Tau vs Enforcement of Regulations percent of factories inspected (1) tau 0.448∗ (0.230) convictions under FA per factory (2) tau (3) tau (4) tau (5) tau (6) tau (7) tau (8) tau -0.0914 (0.411) prosecutions under SEA per capita -0.122 (0.357) fines under SEA per capita 0.323 (0.342) prosecutions per inspection -0.134 (0.254) fines per inspection under SEA 2.598 (3.557) cases disposed per inspection under SEA -9.616 (16.45) cases disposed per cases prosecuted under SEA 1.058 (1.094) log of net state domestic product pc -0.0180 (0.921) -0.648 (1.265) -1.545∗∗ (0.606) -2.220∗∗ (0.783) -1.572∗∗ (0.537) -1.977∗∗ (0.677) -1.471∗∗ (0.600) -1.610∗∗ (0.500) share of employment in manufacturing -2.361 (7.120) 4.091 (12.61) 4.873 (4.855) 5.550 (4.542) 4.525 (4.883) 7.130 (5.277) 4.486 (4.848) 2.514 (5.340) Constant 0.970 (8.791) 10 6.706 (11.88) 9 15.42∗∗ (5.835) 13 22.20∗∗ (7.767) 13 15.72∗∗ (5.150) 13 20.08∗∗ (6.986) 13 12.60 (8.218) 13 16.57∗∗∗ (4.802) 13 Observations Standard errors in parentheses Only including Major Indian States (except UP) for which data exist. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 To briefly summarize the results so far, when regressing τ against regulatory substance, enforcement or industrial disputes, one only observes correlations for certain specific types of measures of those phenomena. In contrast, in our remaining analysis, we will demonstrate that our measures of τ are strongly and robustly correlated with corruption, almost regardless of how it is measured. Table 5 reports the results of regressing τ against corruption as measured in a 2005 Transparency International (TI) Survey.38 Column 1 includes all states for which there is data, while the remaining columns include only the 18 largest Indian states. Column 3 adds controls for state GDP per capita, share of manufacturing 38 The TI corruption measure is based on a survey of perceptions and experience regarding corruption in the public sector. in employment and some others, while Column 4 adds the aggregate measure of regulatory reform from Dougherty (2009). With no exceptions, the coefficient on the TI corruption score is consistently significant and very large in magnitude: a one standard deviation increase in a state’s corruption score is associated with a .5 standard deviation increase in τ . In particular, the fact that the coefficient remains significant in Column 4 even after controlling for Dougherty’s measure of regulatory reforms suggests that the relationship between τ and corruption is at least partly independent from the relationship with the regulations themselves. In what follows we will use the TI corruption score as our primary measure of corruption. One might be concerned, however, that the TI measure may be flawed as it is the result of individuals’ perceptions (it has been argued by some that the perception of corruption is an unreliable indicator for actual corruption). Therefore, we also regress τ against an alternative measure of corruption that is not perception based to check for robustness of the relationship between τ and corruption: Table 6 reports the results of τ regressed against the percent of a state’s available electricity that was lost in transmission and distribution in 2005. This variable has been used by other researchers as a proxy for corruption and poor state governance, and has the virtue of being a concrete and objective measure that does not depend on perceptions Kochhar, Kumar, Rajan, Subramanian, and Tokatlidis (2006). As with the TI Corruption Score, the correlations between τ and this alternative measure of corruption are significant and large in magnitude regardless of sample or controls - including, again, the addition of the Dougherty measure of regulatory reform in Column 4. To make sure that the results are not driven by the actual transmission of electricity, we control for per capita electricity available in Column 5 - which does not affect the results. Table 5: Tau vs Transparency International Corruption Score (3) tau 0.574∗∗∗ (0.153) (4) tau 0.512∗∗∗ (0.139) log of net state domestic product pc -0.381∗ (0.191) -0.446∗∗ (0.172) share of employment in manufacturing 7.362∗ (4.007) 6.444∗ (3.569) share of privately owned establishments -2.662 (4.805) -2.057 (4.258) share of registered establishments 1.455 (1.106) 0.665 (1.048) TI Corruption Score (1) tau 0.422∗∗ (0.157) (2) tau 0.583∗∗∗ (0.145) -0.296∗ (0.142) Dougherty measure (all reforms) Constant Observations States Included 0.557∗∗∗ (0.153) 20 All 0.644∗∗∗ (0.138) 18 Major 5.494 (4.847) 18 Major * p¡0.10, ** p¡0.05, *** p¡0.01, Standard errors in parentheses 5.976 (4.291) 18 Major Table 6: Tau vs Transmission and Distribution Losses (1) (2) (3) (4) tau tau tau tau electricity 0.318∗ 0.648∗∗ 0.663∗∗ 0.566∗∗ transmission and distribution losses (0.165) (0.244) (0.253) (0.237) (5) tau 0.708∗∗ (0.268) log of net state domestic product pc -0.477∗ (0.222) -0.536∗∗ (0.205) -0.459∗ (0.229) share of employment in manufacturing 7.651 (4.812) 6.515 (4.439) 6.817 (5.094) share of privately owned establishments -2.148 (5.677) -1.440 (5.200) -1.980 (5.824) share of registered establishments 1.513 (1.307) 0.644 (1.284) 1.577 (1.343) -0.317∗ (0.172) Dougherty measure (all reforms) Electricity available (GWH) Constant Observations States Included 0.119 (0.182) -8.42e-09 (0.163) 35 All 0.643∗∗∗ (0.163) 18 Major 5.936 (5.746) 18 Major 6.322 (5.253) 18 Major 5.610 (5.910) 18 Major * p¡0.10, ** p¡0.05, *** p¡0.01, Standard errors in parentheses Although the state-level correlations between τ and corruption appear to be robust, the regressions lack exogenous variation and are subject to the concern that our measures of corruption may be correlated with omitted variables that are also correlated with τ . To partially address these concerns, we attempt to take advantage of State X Industry level heterogeneity. In particular, inspired by Novosad and Asher (2012), we use 2005 World Bank Enterprise Survey (WBES) data to create an industry-level measure of “dependence on government bureaucracy”. Specifically, Indian firms in the 2005 WBES were asked how many times in a year they had an inspection or other required meeting with a government official. Averaging the firm-level responses by industry, we classify industries according to their average number of visits with officials (i.e., their dependence on government bureaucracy). If some industries have more meetings with officials, and if corruption takes place during some of these meetings, we would imagine that the costs of corruption would be highest for firms in those industries with the highest dependence on bureaucracy and in those states that have the highest levels of corruption. That is, we would expect that the interaction between industry level dependence on bureaucracy and state level corruption is positive. If found to be the case, it would be harder to argue that the result is due to the presence of omitted variables. The hypothesis is tested in Table 7. To do so we generate our measures of τ at the State X Industry level39 and interact each of our state level measures of corruption with a) the industry average number of visits from officials and b) the industry average duration of visits from officials. We include the interaction with average duration of visits as a placebo test: it is not clear that the duration of an inspection should be positively or negatively correlated with corruption.40 We then regress our State X Industry measures of τ against the covariates including interaction terms. Our prior is that the interaction of corruption with duration of visits should be less significant than the interaction with number of visits. Indeed, this is mostly what we observe. The interaction between our measures of corruption and the number of visits is at least weakly significant (at the 10% level) for one of the two measures, while the interaction between corruption and average duration of visits is never significant. To summarize our results from these investigations, we find: 1. a correlation between τ and certain aspects of the substance of regulations as measured in Dougherty (2009), 2. a nonexistent or inconclusive relationship between τ and measures of the labor environment and enforcement of regulations, and 3. a strong and robust relationship between τ and two distinct measures of corruption. Although none of these results can be said to be causal, we find them suggestive of a relationship between corruption and high labor costs. Next, we turn our attention to the question of how and why greater corruption would lead to higher labor costs for firms. To this end, in the following subsection we outline a simple theoretical framework to elucidate the potential connection. 39 Industries here are categorized according to their groupings in the World Bank Enterprise Surveys, which distinguishes 24 distinct industry categories. Examples include “auto components”, “leather and leather products”, and “food processing”. 40 In particular, corruption may lead to longer inspections if the process of extracting the bribe takes time, or it may lead to shorter inspections if corruption obviates the need to carry out the actual inspection. Table 7: Tau vs State Level Corruption Interacted with Industry Level “Dependence on Regulation” (with Industry FEs) (1) (2) (3) (4) tau tau tau tau 2005 TI 0.132∗ 0.140∗ Corruption Score (0.0668) (0.0666) electricity transmission and distribution losses number of inspections 0.0583 (0.114) 0.127 (0.0829) duration of inspections corruption score X num inspections 0.130 (0.0945) -0.00610 (0.124) -0.109 (0.143) 0.101∗ (0.0545) corruption score X duration of inspections 0.0513 (0.0691) electricity TDLs X num of inspections 0.0112 (0.0848) electricity TDLs X duration of inspections log of Net State Domestic Product pc Constant Observations 0.0493 (0.110) -0.0698 (0.107) -0.185∗∗∗ (0.0364) -0.185∗∗∗ (0.0374) -0.197∗∗∗ (0.0428) -0.194∗∗∗ (0.0427) -0.244 (0.187) 189 0.0235 (0.442) 189 -0.235 (0.187) 189 -0.217 (0.498) 189 * p¡0.10, ** p¡0.05, *** p¡0.01, Standard errors clustered at the State Level, Industry FEs included. 6.2.1 A Theoretical Framework for Understanding Corruption Between Inspectors and Firms We find it helpful to distinguish between two types of corruption that could take place between inspectors and firms: collusion and extortion. Collusion takes place when inspectors allow firms to avoid the costs of complying with regulations in exchange for bribes. However, poor state governance (which here would imply an inability to control corruption) would then lead to lower costs for firms, as greater corruption would make it easier to avoid the full costs of regulation.41 However, what we observed in Section 6.2 was a robust positive correlation between effective costs (τ ) and poor governance/corruption. To explain this phenomenon, we need a model of extortion. In this section, we will sketch the intuition for such a model. A fuller (but still simple) model of extortion and bribery is provided in Appendix 3. Let us start with the observation that all firms reporting at least 10 employees fall under the jurisdiction of certain regulations. Imagine, now, that the regulations are so complex so as to make it impossible (or prohibitively costly) for any firm to be fully in compliance with all aspects of the law as written.42 Then, an inspector can, at any time, choose to subject a firm under his jurisdiction to a penalty e, which may include financial (e.g.: fines) and/or non-financial elements (e.g.: harassment, time needed to defend claims of violations, etc). We can think of the extent of the penalty (e) as a function of state governance: properly functioning governments hire and motivate inspectors to pursue substantive violations rather than minor ones, while inspectors in corrupt or dysfunctional governments can get away with threatening to impose high penalties for even minor technical violations if a bribe is not paid (i.e.: extortion). In such an environment, firms reporting 10 or more employees (and hence under the jurisdiction of the inspector) may face a choice between exposing themselves to the penalty, e, or paying a bribe, b. Assume that inspectors face no costs or benefits from imposing the penalty on the firm, but naturally benefit from receiving the bribe. There is thus a surplus to be had from paying/receiving the bribe b and avoiding the penalty. If the inspector and firm Nash Bargain over the surplus with bargaining weights α and β, respectively, the problem is the following: max (b)α (e − b)β b α ∗ e. The cost born by the firm is The solution is for the firm to pay a bribe b = α+β therefore increasing in α, the bargaining weight of the inspector, and in e, the maximum penalty to which the firm can be subjected. It is reasonable to imagine that this maximum level of extortion, e, is roughly proportional to the size of the firm, so that e = e0 ∗ n, where n is the number of workers in the firm and e0 is the per worker level of extortion. In that α case the bribe per worker, nb , is equal to α+β ∗ e0 .43 41 See, for example, a model of corruption such as the one in Khan, Khwaja, and Olken (2014). This does not not require much imagination. As we mentioned in Section 2, many of the laws have components that are antiquated, arbitrary, contradictory and confusing. That the laws may be impossible to fully comply with is suggested by some of the anecdotes we provide in Appendix 2 as well as the following observation, which we re-quote: “Rules under the Factories Act, framed in 1948, provide for white washing of factories. Distemper won’t do. Earthen pots filled with water are required. Water coolers won’t suffice. Red-painted buckets filled with sand are required. Fire extinguishers won’t do... And so on” (TeamLease Services, 2006). 43 Again, we provide some support for the claim that bribes are proportional to the number of workers 42 This framework can be embedded into the firm’s choice of true and reported employment as modeled in Section 4.3. In particular, the firm now faces a choice between reporting employment greater than or less than 10, where reporting less than 10 allows it to avoid the costs of bribery, and reporting greater than 10 exposes it to the bribery costs. In that framework, τ corresponds to nb , and is therefore increasing in the bargaining power of the inspector (α) and the corruption level of the state (e0 ). In this way, we can make sense of the empirical results above in terms of this basic framework. Again, a more fully fleshed out model that explicitly incorporates features missing here (such as an appeals process and inspector types) is provided in Appendix 3. 6.3 Possible Consequences of τ In the subsections above, we tried to argue that our estimated costs (τ ) are most likely due, not only to the substance of the regulations themselves, but also to high levels of corruption. In this subsection we will indicate possible consequences of high values of τ . Again, the results cannot be given a causal interpretation, but we find them compelling nevertheless. In what follows we use two distinct measures of τ : one which is created using all the enterprises in a state, regardless of economic sector (τ ) and another which is created using only the enterprises engaged in manufacturing (τmanuf ). Table 8 displays the results of employment growth in the manufacturing sector between 2010 and 2005 at the State Level regressed against our two measures of labor market distortions (τ and τmanuf ) as well competing measures (BB and Dougherty). For each of the four measures, we observe its performance as a predictor of future employment growth in registered manufacturing as well as its correlation with employment growth in unregistered manufacturing. Interestingly, in the regressions of employment growth in registered manufacturing against τ and τmanuf , the coefficient on τ is negative and at least weakly significant, while the coefficient for employment growth in unregistered manufacturing is positive - significantly so in the case of τmanuf . This result makes sense: we should expect higher costs to negatively effect the sectors to which the costs apply - in this case the registered sector, since that is under the ambit of labor regulations while the unregistered sector is not. If these correlations reflect a causal chain, it would mean that high levels of regulator costs and corruption (as measured by τ ) are pushing employment from the registered to the unregistered sector. Also included in Table 8 are the results of employment growth in manufacturing regressed against the BB and Dougherty measures. Neither regressor has a coefficient that with anecdotal evidence from ipaidabribe.com in Appendix 2. is statistically significant or of a meaningful magnitude.44 Putting aside the considerable caveat that none of these results has the virtue of exogenous variation, it would appear to be the case that our measures of labor market distortions do a better job of predicting future employment growth (or the lack thereof) than the established alternatives. This is also true when considering future growth in manufacturing productivity rather than employment, as shown in Table 9. Higher levels of τ are associated with slower growth of productivity in the registered manufacturing sector (less so in the unregistered sector) Table 8: Manufacturing Employment Growth (2005 - 2010) vs Tau and Other Measures tau (1) reg manuf -0.0240 (0.0176) (2) unreg manuf 0.00197 (0.0233) tau (manuf) (3) reg manuf (4) unreg manuf -0.0471∗∗ (0.0217) 0.0623∗∗ (0.0256) Besley-Burgess measure (regs) (5) reg manuf (6) unreg manuf -0.00525 (0.00731) 0.00979 (0.0142) Dougherty measure (all reforms) (7) reg manuf (8) unreg manuf 0.0226 (0.0130) -0.0143 (0.0159) log of net state domestic product pc 0.00312 (0.0178) 0.0189 (0.0214) 0.0107 (0.0145) 0.0192 (0.0161) 0.00413 (0.00863) 0.0140 (0.0168) 0.0212 (0.0154) 0.0136 (0.0195) share of employment in manufacturing -0.393 (0.258) 0.00558 (0.329) -0.708∗∗ (0.258) 0.435 (0.325) 0.0194 (0.186) -0.559 (0.362) -0.515∗ (0.245) 0.0525 (0.323) Constant 0.0969 (0.173) 18 -0.182 (0.209) 17 0.0372 (0.139) 18 -0.229 (0.152) 17 0.0209 (0.0825) 15 -0.0675 (0.160) 15 -0.0861 (0.147) 18 -0.131 (0.182) 17 Observations * p¡0.10, ** p¡0.05, *** p¡0.01, Standard errors in parentheses, Only including Major Indian States 44 One might argue that it is not quite fair to regress growth between 2010 and 2005 on a regressor that uses data only up until 1997 (as is the case for the BB measure). However, a) we have duplicated these results using from growth from 1997 to 2002 and the results are the same, and b) the Besley Burgess measure from Aghion et al. (2008) should be largely the same in 2005 due to the lack of state level reforms between 1997 and 2005. Table 9: Manufacturing Productivity Growth (2005 - 2010) vs Tau and Other Measures (1) reg manuf -0.0321∗∗ (0.0146) tau (2) unreg manuf 0.00522 (0.0239) tau (manuf) (3) reg manuf (4) unreg manuf -0.0512∗∗ (0.0181) -0.0567∗ (0.0275) Besley-Burgess measure (regs) (5) reg manuf (6) unreg manuf -0.00266 (0.0122) -0.00372 (0.0154) Dougherty measure (all reforms) (7) reg manuf (8) unreg manuf 0.0167 (0.0122) 0.00973 (0.0166) log of net state domestic product pc -0.0160 (0.0148) -0.00902 (0.0220) -0.00478 (0.0121) -0.0120 (0.0174) -0.00455 (0.0143) -0.00723 (0.0182) 0.00438 (0.0145) -0.00793 (0.0204) share of employment in manufacturing 0.206 (0.214) 0.0392 (0.338) -0.154 (0.215) -0.349 (0.350) 0.453 (0.310) 0.526 (0.392) 0.0728 (0.230) 0.0104 (0.338) Constant 0.141 (0.144) 18 0.124 (0.214) 17 0.0454 (0.116) 18 0.198 (0.163) 17 -0.0208 (0.137) 15 0.0530 (0.174) 15 -0.0678 (0.138) 18 0.119 (0.190) 17 Observations * p¡0.10, ** p¡0.05, *** p¡0.01, Standard errors in parentheses, Only including Major Indian States 6.4 Inter-temporal Variation: A Final Puzzle As we have noted, all of the analysis above uses data only from the 2005 Economic Census. However, the discontinuity observed in the 2005 data does not appear in the 1990 or 1998 ECs. Explaining the puzzling inter-temporal variation is a main goal of our future work. Perhaps the simplest explanation is that the data quality of the EC improved between 1998 and 2005. Indeed, it is intuitive that the presence of a downshift will be harder to discern upon the introduction of measurement error. This explanation has received anecdotal support from our meetings with the Directors of the state Directorates of Economics and Statistics, who largely claimed to be more confident in the results of the 2005 EC than in the results of previous rounds. One reason for this could be that the 2005 EC was the first wave in which ICR (“intelligent character recognition”) was used to read in the raw data. This technology should have gone a long way in alleviating the considerable amount of error that comes from manual interpretation and typing. But one might also be concerned in the other direction: that somehow the distortion observed in 2005 is anomalous and is perhaps due to something like a change in the EC’s enumeration practices. Based on our interviews with officials and enumerators in charge of collecting these data, however, we believe this is unlikely. While we have discovered that enumeration practices have changed slightly over the years, we have not discovered changes that could have produced the specific patterns we observe. For example, unlike the previous Figure 7: Variation in the distribution of establishment size across time: waves, the 2005 Economic Census included an “address slip” that was meant to be filled out for establishments with 10 or more workers. It is conceivable that enumerators, in an effort to avoid the extra work of filling out the address slip, preferred to misrepresent the number of workers for establishments with more than 10 workers. However, while this could reasonably explain why there are fewer 10, 11 and 12 person establishments, we find it hard to understand how this kind of phenomenon could explain why there are also fewer 30, 40 and 50 person establishments (see section 4.3 for more on this). Furthermore, in post-enumeration checks done in West Bengal, Bihar and Tamil Nadu, this kind of misrepresentation was not found to be in occurrence. Nor does the culprit seem to be changes in the regulations themselves. Indeed, none of the regulations which we assume are responsible for the discontinuity in 2005 were greatly changed in the right direction between 1990 and 2005. There were reforms on the margin (reflected in the Dougherty (2009) measures), and these do seem to be correlated with our measures of τ , but most of the changes have gone in the direction of loosening regulations and thus cannot explain why τ would have increased over time. Enforcement has also changed to some degree, but we have not yet found evidence that it can explain the variation in our data (see the interstate analysis above). Another possible (but ultimately unlikely) explanation is that changes in the competitive environment, particularly related to the increased exposure to international markets and competition, are responsible. The period in question (1998 to 2005) saw heavy reductions in protective barriers from foreign competition - particularly through the elimination of nontariff barriers. However, in our preliminary analysis (not yet reported) we do not find a strong link between trade liberalization and τ , and such link as exists goes in the opposite direction. A final possibility is that changes in the availability of contract labor cause the discontinuity to show up in 2005. Indeed, there was a large speed-up in the use of contract labor over the period,45 however state-level changes in the fraction of contract labor in registered manufacturing are not robustly correlated with τ . The absence of such a correlation is not necessarily evidence that such a link does not exist, but it does not give support to the hypothesis either. We have been able to share this intertemporal paradox with a number of experts regarding these issues in India but have not yet been able to find a watertight explanation. This continues to be a priority in our ongoing work. 7 Conclusion Our goals in this paper are 1) to document the effect of size-based labor regulations on the misallocation of resources across firms via the employment decisions of business enterprises, 2) to estimate the net costs of the set of regulations that become binding when establishments choose to employ 10 or more workers, and 3) to shed some light on the source of these costs by demonstrating that corruption in the form of harassment bribery may play a large role in making Indian regulations costly. To the best of our knowledge, this paper is the first to provide cost estimates of regulations in India (particularly non-IDA regulations), the first to analyze the effects of regulations without using necessarily subjective evaluations of state-level amendments to labor laws, and the first to provide evidence for the link between regulatory costs and corruption. To accomplish these tasks, we use the 2005 Economic Census of India, an uncommonly used dataset which is uniquely suited to our task because it includes the entire universe of non-farm enterprises. In our investigation, we find a significant level shift down in the natural log of the probability mass of establishments with 10 or more workers. Adapting a method from Garicano et al. (2013), we interpret this as evidence of substantial per-worker costs of operating above the 10 worker threshold. At the all-India level, we find that operating at or above the 10-worker threshold is associated with a 35% increase in the unit cost of labor as modeled. Furthermore, we observe a great deal of variation in our estimated costs by state, industry and ownership type. We estimate the highest (lowest) costs for privately owned firms (government-owned firms) and firms in construction (public administration and defense). Exploring this variation reveals that Indian states with the highest costs also have the 45 In the registered manufacturing sector, the share of contract labor in total labor increased from 15% to 26%. Prior to this period, its growth was markedly slower. highest levels of corruption and poor governance (as measured through two distinct indices), and that firms in industries with high bureaucratic dependence are exposed to particularly high costs if they are also in highly corrupt states. This analysis suggests that the size of regulatory costs may have as much to do with how regulations are implemented and who implements them, as with the content of the specific labor and industrial laws themselves. We hope that these findings will help shift the present debate away from arguments over the pro or anti-labor stance of regulations and towards arguments about clarity, bureaucracy and the proper enforcement of regulations.