AUTHOR ACCEPTED MANUSCRIPT FINAL PUBLICATION INFORMATION Firing Costs and Flexibility : Evidence from Firms' Employment Responses to Shocks in India The definitive version of the text was subsequently published in Review of Economics and Statistics, 95(3), 2013-07 Published by MIT Press THE FINAL PUBLISHED VERSION OF THIS MANUSCRIPT IS AVAILABLE ON THE PUBLISHER’S PLATFORM This Author Accepted Manuscript is copyrighted by World Bank and published by MIT Press. It is posted here by agreement between them. Changes resulting from the publishing process—such as editing, corrections, structural formatting, and other quality control mechanisms—may not be reflected in this version of the text. You may download, copy, and distribute this Author Accepted Manuscript for noncommercial purposes. Your license is limited by the following restrictions: (1) You may use this Author Accepted Manuscript for noncommercial purposes only under a CC BY-NC-ND 3.0 IGO license http://creativecommons.org/licenses/by-nc-nd/3.0/igo. (2) The integrity of the work and identification of the author, copyright owner, and publisher must be preserved in any copy. (3) You must attribute this Author Accepted Manuscript in the following format: This is an Author Accepted Manuscript by Adhvaryu, Achyuta; Chari, A. V.; Sharma, Siddharth Firing Costs and Flexibility : Evidence from Firms' Employment Responses to Shocks in India © World Bank, published in the Review of Economics and Statistics95(3) 2013-07 CC BY-NC-ND 3.0 IGO http://creativecommons.org/licenses/by-nc- nd/3.0/igo © 2018 World Bank FIRING COSTS AND FLEXIBILITY: EVIDENCE FROM FIRMS’ EMPLOYMENT RESPONSES TO SHOCKS IN INDIA Achyuta Adhvaryu, A. V. Chari, and Siddharth Sharma* Abstract—A key prediction of dynamic labor demand models is that firing in unobserved determinants of employment. Arguably, the restrictions attenuate firms’ employment responses to economic fluctua- tions. We provide the first direct test of this prediction using data from latter condition does not obtain in cross-country or even India. We exploit the fact that rainfall fluctuations, through their effects on within-country time-series variation in employment protec- agricultural productivity, generate variation in local demand within districts tion policies (however, see Heckman & Pages, 2004, for over time. Consistent with the theory, we find that industrial employment is more sensitive to shocks where labor regulation is less restrictive. Our some evidence that labor reforms in Latin America may be results are robust to controlling for endogenous firm placement and vary considered to have been exogenous). across factory size in a pattern consistent with institutional features of Indian Being able to credibly attribute differences in outcomes to labor law. differences in labor regulation is obviously a general problem for any study of the effects of labor policies. An additional I. Introduction concern for a study such as ours is the identification and mea- surement of fluctuations. Because the source of fluctuations A N old insight from labor economics is that firing costs reduce the extent of employment adjustment to eco- nomic shocks: during a downturn, firing costs reduce the is typically not observable or directly quantifiable, previous empirical studies have inferred the magnitude of fluctuations from changes in observable quantities such as output or sales. number of layoffs, while during an upturn, hiring is curbed For example, Bentolila and Saint-Paul (1992), in their study because of the possibility of having to lay off workers of the effects of the introduction of flexible labor contracts in the future (Oi, 1962; Nickell, 1986; Hamermesh, 1993). in Spanish manufacturing, measure shocks by the change in Employment inflexibility (from the firm’s perspective) and log sales of a firm, which they then relate to employment its possible negative effects on average as well as aggregate responses. Similarly, Abraham and Houseman (1993) relate output, employment, and wages is therefore the price of job (aggregate) employment to output in their comparison of security provisions, and this is the basis of a great deal of pol- employment dynamics in the United States and Germany. icy debate surrounding draconian labor laws that have been This approach is problematic for at least two reasons. First, enacted in many countries (as documented, for example, by fluctuations in aggregate or firm-level output can reflect either Botero et al., 2004).1 unobserved demand or cost shocks (or both), and the cor- In this paper we provide the first direct test (to our knowl- responding change in employment can be expected to be edge) of the prediction that the magnitude of employment different in each case. Second, this method cannot satisfacto- responses to shocks should vary negatively with the degree of rily distinguish between fluctuations that are foreseeable and employment protection. Obtaining a credible test of this pre- those that are inherently unpredictable.2 The key innovation diction is difficult for a number of reasons. First, we require of this paper is its utilization of a well-defined and mea- a setting where there is variation across space or time in the surable source of fluctuations that are strictly unpredictable extent of employment protection, with the added requirement in nature, exogenous to the labor regime, and comparable that this policy variation does not simply reflect variation across the units of study; this is the precise sense in which we think of our test as being direct. This approach avoids Received for publication May 17, 2010. Revision accepted for publication the problems associated with defining fluctuations in terms April 16, 2012. * Adhvaryu: Yale University; Chari: RAND Corporation; Sharma: World of endogenously determined variables. Bank. Our setting is rural India, where agriculture exists along- We thank Abhijit Banerjee, Fran Blau, Kaushik Basu, Jim Berry, side industry. Differences in employment protection laws Prashant Bharadwaj, Peter Brummund, Fernanda Lopez de Leon, Annemie Maertens, Russell Toth, and seminar participants at the NEUDC, PAC-DEV, across the states of India (and over time) provide variation and the development workshops at Cornell and Yale University for helpful in firing costs in the industrial sector. To obtain a plausible comments. We are also grateful to Seema Jayachandran, Pinar Keskin, and shock variable, we measure rainfall fluctuations that affect Rohini Pande for sharing data with us. The views expressed in this paper are those of the authors and do not reflect the views of the RAND Corporation agricultural yield. In this particular context, rainfall shocks or the World Bank. are ideal for a number of reasons: (1) they plausibly give 1 The effects of job security provisions on average and aggregate outcomes rise to labor supply or output demand shifts for local indus- are, however, theoretically ambiguous. Because firing restrictions reduce hiring as well as firing, the average level of employment (for a given firm) tries through their effect on agricultural yields; (2) they are may either increase or decrease (Bentolila & Bertola, 1990, suggest that unpredictable in nature and therefore not likely to induce for realistic parameter values, higher firing costs may actually raise average employment). The effects on aggregate levels of output and employment are also indeterminate once we account for the effects of these restrictions on 2 The distinction is potentially important because, depending on the entry and exit (however, see Hopenhayn & Rogerson, 1993, for calibrations structure of adjustment costs, it may be optimal to smooth foreseeable that suggest a negative overall effect of a tax on job destruction). Finally, fluctuations in advance, so that the resulting variation in employment is as Basu, Fields, and Gupta (2008) argue, job security provisions can even of a different character than in the case of unpredictable shocks. Indeed, result in a lower level of wages, hurting the very constituency they are meant the relation between employment and leads and lags of aggregate output is to protect. likely to be very different in the two cases. The Review of Economics and Statistics, July 2013, 95(3): 725–740 © 2013 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology 726 THE REVIEW OF ECONOMICS AND STATISTICS anticipatory smoothing of employment (which is important literature tends to focus on either the agricultural sector or the for our purposes because our data are not disaggregated industrial sector in isolation, our results highlight the close enough at the temporal level to identify such anticipatory relation between the two—in particular, our finding of the smoothing); (3) they are temporary and recurring and there- significance of local demand for the factory sector may be fore factor into the forward-looking decisions of firms; (4) surprising and should be treated as a caveat against thinking they are exogenous to the labor regime, and are not caused of formal sector products as being bought and sold in national by employment changes in the industrial sector or by any rather than regional markets. other factors that may affect employment; and (5) we are The remainder of the paper is organized as follows. Section able to provide evidence that the measured rainfall fluctu- II describes labor regulations in India, section III describes ations represent comparable shocks across labor regimes. the data, section IV describes the empirical strategy, section The empirical strategy is then to test whether these shocks V describes the results, and section VI concludes. induce larger factory employment responses in states that have enacted proemployer legislation. II. Labor Regulation in India Our results provide a confirmation of the prediction that industrial employment should be more flexible in proem- The basis of labor regulation in India is the Industrial Dis- ployer regions. We first confirm that rainfall fluctuations putes Act (IDA) of 1947, which sets out the regulations do indeed have a impact on local agricultural production governing employer-worker relations and the legal proce- and incomes (but not agricultural wages). Importantly for dures to be followed in the case of labor disputes in the our identification strategy, these effects are not differential factory sector. The IDA was passed by the central govern- across labor regimes. We then document that high (low) ment, and in its original form, it applied equally to all states. rainfall increases (decreases) industrial employment, indi- But since India is a federal democracy, with both the central cating the operation of a demand effect through agricultural and state governments having jurisdiction over labor legisla- incomes. Furthermore, as predicted by theory, the induced tion, the act has since been amended by state governments. change in employment is indeed significantly greater in These amendments have caused the states to differ markedly proemployer states. We also find that the responses to these in their labor regulations. local shocks appear to be concentrated among industries The IDA covers several aspects of industrial disputes, such that are likely to be dependent on local demand, consistent as unfair labor practices, strikes and lockouts, and layoffs and with our interpretation of the shocks as representing demand retrenchments. It calls for the setting up of special bodies (tri- fluctuations. bunals, boards of conciliation, and labor courts, for example) We address the endogeneity of labor regulations in two to arbitrate disputes in the industrial sector, while specifying ways: we verify the robustness of the results to the inclusion their composition and extent of authority. Of specific interest of a set of controls that may be plausibly correlated with labor for us are sections V-A and V-B of the IDA, which describe regulation, and we use the fact that labor regulations apply the regulations pertaining to layoffs and retrenchments. The only to factories above a size threshold: if our results reflect regulations in section V-A cover industrial establishments in the effects of labor regulation, then the responses to rainfall which more than “fifty workmen on an average per working shocks should not vary across labor regimes for unregulated day have been employed in the preceding calendar month” factories, and this is indeed what we find. (section 25-A, chapter V-A, IDA; see Malik, 1997). This Our focus in this paper is primarily on the test of the section asserts the right of workers who have been laid off or hypothesis that firing costs reduce employment flexibility, but retrenched to adequate compensation. Specifically, workers it is natural to ask how reduced flexibility translates into out- who have been on the rolls for at least a year are entitled puts, profits, and intensity of use of nonlabor inputs. Although to compensation at 50% of their regular wage for each day we have less confidence in the accuracy of measurement of that they are laid off (up to a maximum of 45 days). Work- nonlabor variables in the factory data, there is some weak ers who are to be retrenched are to be given one month’s evidence that the average change in outputs and profits due notice and are eligible for compensation from the employer to shocks is no greater for factories in prolabor regions. Taken equal to 15 days’ average pay for each year of completed at face value, this finding suggests that the latter are able to service. Section V-A also limits closure of undertakings by compensate for the lack of employment flexibility by adjust- requiring notification of the government at least 60 days prior ing along other margins. We are, however, unable to find any to closure. Furthermore, all workers thereby dispossessed of differential adjustment of other observable inputs, leaving the jobs are to be compensated according to the compensation possibility that adjustments may take the form of hiring and for retrenched workers. firing casual (temporary) workers. Section V-B lays out some special provisions that apply Our employment results are a striking confirmation of the only to industrial establishments employing at least 100 hypothesis that job security provisions in India have con- workers.3 This section is more draconian: it requires that strained labor adjustment on the part of firms. Our paper 3 In the original IDA, this section applied only to establishments with more also ties into a wider literature that seeks to understand the than 300 workers, but this threshold was subsequently revised by the central workings of the rural economy in India. Whereas the existing government in 1982. FIRING COSTS AND FLEXIBILITY 727 Table 1.—Data Sources Source Years Variables Annual Survey of Industries (ASI), conducted by the Central Statistical 1988, 1991, 1994 Factory employment, fixed capital, output, Organization of India raw material and fuel expenditures State × three-digit industry panel constructed from ASI, from Aghion et al. (2008) 1980–1997 Factory employment Terrestrial Air Temperature and Precipitation: Monthly and Annual Time Series, 1950–1999 Rainfall shock version 1.02 (Center for Climatic Research, University of Delaware) Besley and Burgess (2004), based on state-level amendments to the Industrial 1949–1995 Labor regulation Disputes Act of India India Agriculture and Climate Data Set, updated using statistics published by the 1980–1997 Agricultural yields Directorate of Economics and Statistics, Ministry of Agriculture, India Consumer Expenditure Survey, conducted by the National Sample Survey 1987, 1993, 1999 Household per capita expenditure Organization of India no workers may be laid off or retrenched without the prior Table 2.—ProWorker, ProEmployer and Neutral States in ASI District-Level Data permission of the government. Closure of establishments requires an application to be filed with the government at ProWorker Neutral ProEmployer least ninety days before the proposed closure. The penalty Gujarat Bihar Andhra Pradesh for violating the regulations in V-B includes a prison term Maharashtra Haryana Karnatakaa Orissa Karnatakaa Rajasthan of up to a year or a fine of five thousand rupees in the case West Bengal Madhya Pradesh Tamil Nadu of illegal closure (or both) and prison term of up to a month Punjab and a fine of 1,000 rupees in the case of illegal layoff or Uttar Pradesh a retrenchment. Karnataka switches from neutral to proemployer in 1987–1988; Classifications are based on adding the number of proworker laws and subtracting the number of proemployer laws passed in each state; these The IDA does not cover temporary or casual workers, so classifications hold between 1987 and 1994, inclusive. For details, refer to section II. in principle, firms could work around the provisions in V-A and V-B by using casual labor. We do not have any data on the extent of casual labor and are therefore unable to iden- tify whether it is indeed being substituted for formal labor. respectively. A state’s labor regulation regime in any year However, as Fallon and Lucas (1993) note, the vigorous oppo- was then obtained as the sum of these scores over all preced- sition of labor unions, as well as the restrictions imposed on ing years. Based on this cumulative score, Besley and Burgess the use of contract labor by the Contract Labor Regulation (2004) classified four states—Gujarat, Maharashtra, Orissa, and Abolition Act of 1970, are likely to significantly curtail and West Bengal—as proworker in 1988. Six states—Andhra this channel of avoidance of labor regulation. Pradesh, Karnataka, Kerala, Madhya Pradesh, Rajasthan, and Tamil Nadu—were categorized as proemployer. Six others— III. Data Assam, Bihar, Haryana, Jammu and Kashmir, Punjab and Uttar Pradesh—were classified as neutral with respect to We combine a few data sources: data on labor regulation; labor laws. These categorizations are summarized in table 2. manufacturing outcomes and district and state-industry lev- We followed this scheme of cumulating the Besley- els of aggregation; and agricultural production, agricultural Burgess scores to categorize the states as proworker, proem- wages, and district per capita expenditure. These sources are ployer, or neutral in each year of our study. Since there were summarized in table 1. few labor law amendments after 1987, this classification remains identical to the original Besley-Burgess classifica- A. Labor Regulation tion for 1988 throughout our study period. The only exception is Karnataka, which switched from being neutral to being The basis of industrial labor regulation in India is the IDA proemployer between 1987 and 1988. of 1947, which sets out the legal procedures to be followed We also report results using three alternative measures of in the case of labor disputes such as layoffs, retrenchments, regulation suggested by Ahsan and Pages (2008). The first and strikes in a factory. The IDA was passed by the central measure is based on the critique by Bhattacharjea (2006) government, but has since been extensively amended by state of Besley and Burgess’s, coding. In addition, Ahsan and governments, causing Indian states to differ markedly in their Pages distinguish between sections of the IDA that specif- labor regulations. ically relate to layoffs and sections that address the dispute Besley and Burgess (2004) read all state-level amendments settlement process between employers and workers. Follow- made to the IDA during 1958 to 1995 in sixteen major Indian ing Ahsan and Pages, we refer to the former as employment states (from Malik, 1997). Each amendment was coded as protection legislation (EPL) and the latter as dispute settle- being either proworker, neutral, or proemployer, depend- ment (DS) legislation. All three measures are constructed ing on whether it lowered, left unchanged, or increased an in a similar way to Besley and Burgess’s measure: by cod- employer’s flexibility in hiring and firing factory workers, ing amendments to the IDA and cumulating the labor score 728 THE REVIEW OF ECONOMICS AND STATISTICS Table 3.—Baseline Summary Statistics by ProWorker, Neutral, and ProEmployer States District-level Panel Proworker Neutral Proemployer Number of districts (in 1988) 78 186 80 Labor regulation strictness measurea 1.79 0 −1.54 (0.76) (0) (0.50) Rainfall shock −0.03 −0.12 0.09 (0.68) (0.67) (0.62) % agrarian employment in district 44.48 45.57 46.04 (16.31) (15.66) (15.53) % landless in district 22.40 16.67 19.89 (13.56) (11.87) (13.09) Average capital-to-output ratio in district 1.56 1.60 1.09 (2.76) (4.96) (1.66) % employment in industries linked to agriculture 43.26 43.82 46.65 (31.25) (33.93) (32.31) States are classified as proworker, proemployer, or neutral based on adding the number of proworker amendments and subtracting the number of proemployer amendments passed after Indian independence in 1947. Rainfall shock variable = 1 if annual rainfall is less than the 20th percentile of historical distribution, = 0 if less than the 20th and more than the 80th percentile of historical distribution, and = −1 if more than the 80th percentile of historical distribution. Share of agrarian employment, percent landless, capital-to-output ratios, and share employed in agriculture-related industries are measured in 1988. a This measure is from Besley and Burgess (2004). over preceding years. In contrast with the remaining mea- which constitute the sixteen largest Indian states and account sures, there are only proworker and neutral states in the EPL for nearly 95% of India’s population. To arrive at district- coding. level estimates of factory sector employment, revenue, input costs, fixed capital, and wages, we used the survey weights B. The Industrial Sector to aggregate unit (factory) level data from three rounds of ASI. Our final district data set has 1,042 observations across Manufacturing establishments in India are broadly classi- three years: 1987, 1990, and 1994.4 Tables 3 and 4 summarize fied as either factories or informal enterprises, where the dis- characteristics of the districts in our sample and the industrial tinction is based on a cutoff defined in terms of employment. sector outcome variables we use, respectively. The summary According to the Factory Act, a manufacturing establishment statistics are grouped by proworker, proemployer, and neutral that uses any form of power (such as electricity, steam, or states. diesel) to drive machinery is a factory if it employs at least Our rainfall data are from the Center for Climatic Research ten workers. A manufacturing establishment that does not at the University of Delaware.5 The rainfall measure for a use power is a factory if it employs at least twenty workers. latitude-longitude node (on a 0.5◦ latitude by 0.5◦ longitude Since factories alone are subject to industrial entry and labor grid) combines data from twenty nearby weather stations regulation laws such as those laid out in the IDA, our data using an interpolation algorithm based on the spherical ver- set on manufacturing establishments pertains to the factory sion of Shepard’s distance-weighting method. We matched sector. these rainfall data to districts by calculating the grid point The source of our data on factories is the Annual Survey nearest to the geographic center of a district. of Industries (ASI), a cross-sectional, national survey and Previous research on India suggests that while low rainfall census of factories that is conducted annually by the Central hurts agricultural production, excess rainfall helps.6 Our pri- Statistical Organization of India. The ASI has two parts: a mary measure of the rainfall shock (Rainshock ) is therefore census of all factories employing 100 workers or more and constructed in such a way that higher values indicate lower a survey that randomly samples about a quarter of all other amounts of rainfall. Rainshock is equal to 1 when the annual registered factories. The data are not a panel at the factory district rainfall is less than the 20th percentile of the district’s level due to the unavailability of factory identifiers, but the historical average, 0 when it is between the 20th and 80th combined data from the ASI census and survey sections are percentiles, and −1 one when it is above the 80th percentile fully representative of all factories in India and can be used to estimate industrial sector aggregates at regional levels by weighting the factory-level data by the inverse of the sampling 4 The number of districts in India increased during the study period when probabilities. new districts were created. To be consistent, we have used the 1988 district definitions throughout the paper, and any new districts created between 1988 and 1994 have been merged back to their parent districts. Going by the 1988 C. District-Level Data Set: Factories, Rainfall Shock, district definitions, we have data on 344 districts in 1987, 347 in 1990, and Agricultural Production, and Household Expenditure 351 in 1994. The total number of districts varies by year because the ASI does not stratify sampling by districts and is therefore not guaranteed to The majority of our regressions examine the effects of labor cover all districts. 5 This is the Terrestrial Air Temperature and Precipitation: Monthly and regulation and rainfall shocks on the industrial sector at the Annual Time Series (1950-99), Version 1.02. spatial level of districts, the primary administrative unit in 6 Jayachandran (2006) finds similar results for the effects of excess rainfall India. Our district-level data set covers nearly 360 districts, on agricultural yields. FIRING COSTS AND FLEXIBILITY 729 Table 4.—Outcome Variables by ProWorker, Neutral, and ProEmployer States Proworker Neutral Proemployer District-level employment outcomes Number of workers 24,406.92 8,806.28 19,560.38 (44,328.34) (14,086.25) (28,465.96) Man-days (thousands) 10,088.15 3,429.88 7,078.49 (18,998.96) (5,836.59) (9,937.37) State-industry level employment outcomes Number of workers 5,016.81 2,678.27 3,807.37 (14,090.93) (7,619.89) (13,920.15) Number of employees 6,543.20 3,444.01 4,674.66 (16,638.14) (9,349.44) (14,867.71) Other district-level outcomes Agricultural production 99,917.92 79,458.38 92,412.49 (72,989.70) (62,873.04) (76,355.40) Monthly per capita expenditures 320.34 308.41 346.62 (134.76) (129.99) (124.58) Capital stock at close of business year 421.36 135.41 239.09 (790.44) (334.22) (568.89) Value of materials used in production 813.67 253.60 422.67 (1,839.71) (463.95) (697.14) Value of electricity used in production 35.58 13.63 22.18 (59.49) (30.28) (28.96) Value of fuel used in production 80.60 28.27 44.09 (145.01) (60.82) (58.77) Value of total output 1,274.213 396.03 661.96 (2,842.30) (728.12) (1,058.61) Value added 243.96 71.51 120.16 (576.19) (146.71) (207.96) Profits 64.60 21.63 30.62 (218.02) (63.19) (87.78) States are classified based on adding the number of proworker amendments and subtracting the number of proemployer amendments passed after Indian independence in 1947. Agricultural production is a weighted sum in which agricultural output for each crop (in kg) is weighted by the crop’s average price from 1950 to 1987 (in INR/kg). Per capita expenditures are in 1999 INR. Capital stock, materials, fuel, total output, value added, and profits have been converted to thousands of 2009 U.S. dollars. (this is identical to Jayachandran’s 2006 definition of rainfall with data on wheat and rice cultivation area by state, also shocks).7 obtained from Ministry of Agriculture publications. Before examining the relationship between Rainshock , Data on average household per capita expenditure in dis- labor laws, and factory employment, we show that Rainshock tricts are based on consumption expenditure surveys con- is associated with drops in agricultural production, wages, ducted by India’s National Sample Survey Organization and district mean per capita expenditure. Our data on agricul- (NSSO) in 1987, 1993, and 1999. These cross-sectional, tural production and wages of agricultural laborers are from nationally representative household surveys are a standard an updated version of the district-level India Agriculture and source of poverty measurement in India. In estimating district- Climate Data Set. This data set was originally compiled for level averages, households were weighted by the inverse of the years 1957/58 to 1986/87 by James Robert E. Evenson the sampling probabilities.9 and James W. McKinsey Jr. using statistics published by the Directorate of Economics and Statistics (within the Indian D. State- and Industry-Level Data Set Ministry of Agriculture). These data have been updated to 1996 using more recent issues of the same government pub- In some of our regressions, we use state-industry-level lications.8 We measure district annual agricultural production panel data, which are also drawn from multiple rounds of by a constant price-weighted sum of the district output of all ASI. We are hesitant to attempt to distinguish among indus- major crops, where the individual crop prices are fixed at their tries in the district data set due to the small sample size at average value over 1957 to 1987. We supplement these data the district level. Fortunately, the ASI is designed to estimate manufacturing sector outcomes by industry at the state level, with every state and industry group surveyed as an individual stratum. This stratification by industries is at the three-digit 7 This definition of shocks seems appealing because adjusting the num- ber of workers in the face of small fluctuations is an unlikely event in as level of the ISIC classification of industries, which yields a regulated an environment as we are considering. Nonetheless, we have also high level of disaggregation. experimented with a continuous shock measure, which is the negative of We aggregated our district-level data on rainfall to the state the deviation of annual rainfall from the district’s historical average, nor- malized by the historical standard deviation of rainfall in the district. The level by taking simple averages of district rainfall within each results (available on request) are qualitatively similar to the ones we report using the discrete shock measure, but they are less precise. 9 We are grateful to Rohini Pande and Petia Topalova for sharing with us 8 Yield data updates were compiled by Rohini Pande and Siddharth their district-level estimates based on the NSSO Consumption Expenditure Sharma. Surveys. 730 THE REVIEW OF ECONOMICS AND STATISTICS state. Our state-level rainfall shock measure is analogous to A. District-Level Regressions the district-level measure and is defined in terms of deviations from the historical state averages of rainfall. There was no Exploiting variation in rainfall across districts over time, need to modify the labor regulation dummies since they were we first measure the impact of rainfall shocks by regressing already defined at the state level. We then merged these with district outcomes on a rainfall shock measure (Rainshockjt ) state- and industry-level factory data constructed by aggre- for district j and year t . The regressions control for macro- gating unit-level annual ASI data using sampling weights and shocks with year fixed effects and for time-invariant regional three-digit industry codes. The resulting data set is an annual variation with district fixed effects. For outcome x , our base panel covering 130 industry groups across thirteen states over specification is thus seventeen years (1980–1997).10 xjt = αRainshockjt + ρj + ρt + jt , (1) IV. Empirical Strategy where ρj and ρt denote district and year fixed effects, respec- The basic empirical analysis is derived from a simple tively. The coefficient α estimates the average effect of the model of labor adjustment to exogenous shocks in the face rainfall shock on the district outcome xjt . Since Rainshock is of linear adjustment costs. The model is based on Bertola constructed to take on higher values the lower the amount of (1990). Because the theory is standard, we have relegated it rainfall, a negative estimate of α would mean that low rainfall to the appendix. The essential intuition is that firing costs fac- has a negative effect on xjt . tor into the forward-looking employment decisions of firms, The theory suggests that the response of the industrial creating a wedge between the wage and the marginal prod- sector to shocks depends on industrial labor regulation. uct. In the face of positive shocks to the environment, the Accordingly, our key regressions estimate how the effect of firm finds that the effective wage is higher than the actual rainfall shocks varies across districts with different labor reg- wage and therefore curtails its hiring (relative to the case ulation regimes by interacting Rainshockjt with the labor law with no firing costs). When faced with a negative shock, dummies: the firm finds the effective wage to be lower than the actual xjt = αRainshockjt + β(Rainshockjt × Proworkerjt ) wage, and this curtails its layoffs (relative to the case with no firing costs). Firing costs therefore restrict the firm’s adjust- + δ(Rainshockjt × Proemployerjt ) ments to exogenous shocks. In the context of our empirical + ρj + ρt + jt . (2) analysis, fluctuations represented by rainfall shocks should induce smaller employment adjustments in more regulated As described earlier, districts are either Proworker, Proem- environments.11 ployer, or Neutral, depending on the cumulative value of the A potential complication in the empirical exercise is that Besley-Burgess labor law index in their state. Thus, β and rainfall fluctuations create opposing effects on employment: δ measure the effect of rainfall shocks on Proworker and on the one hand, good (bad) rainfall increases (decreases) Proemployer districts, respectively, relative to that in Neu- agricultural incomes and hence demand for local industrial tral districts. For example, suppose that the average effect of goods, but on the other hand, good (bad) rainfall may increase rainfall shocks, as measured by α in equation (1), is negative. (decrease) agricultural demand for labor and represent a nega- Then a negative estimate of δ would imply that the decrease tive (positive) labor supply shock for local industry. However, in xjt due to low rainfall is larger in Proemployer districts as the model clarifies that if rainfall fluctuations create compara- compared to Neutral districts. If α in equation (1) is estimated ble wage and price shocks across labor regimes, the net effect to be positive, then a negative estimate of δ would imply that of price and wage changes on employment is magnified in relative to Neutral districts, the increase in xjt due to low lower firing cost regimes: this is the hypothesis being tested. rainfall is lower in Proemployer districts. Key to this test is the comparability of measured fluctuations We estimate equation (2) for several outcome variables. We across space and time, which we establish in the following examine the direct effect of rainfall shocks by looking at how sections. district agricultural production, farm wages, and household per capita expenditures decline when the rains fail. Then our 10 The state-industry data on factories were used by Aghion et al. (2008). main set of estimations examines the impact of rainfall shocks We are grateful to the authors and the American Economic Review for and labor regulation on employment in the factory sector. making these data publicly available. Finally, we look at other industrial sector outcomes such as 11 A caveat is that we are not able to deal explicitly with multiestablishment firms. In principle, a multiestablishment firm could avoid firing any workers input costs, wages, revenue, and profits. by moving them from districts facing a negative demand shock to districts The coefficients β and δ are intended to capture how with a positive demand shock. This could therefore bias downward the effect responses to rainfall shocks vary across districts with different of labor regulations on employment flexibility. We think this is an unlikely scenario, given the extremely low rates of geographical mobility of labor in labor laws, holding all other district characteristics constant. India, the high degree of correlation of rainfall fluctuations within a state, One concern with our interpretation of the coefficients is that and our sense, given the available information, that there are not many labor regulation might be correlated with other factors that, multiestablishment firms in the data. On average, only about 3% of ASI respondents report being part of a multiestablishment firm having another determine rainfall impacts on the local economy or factories factory in the same state. responses, to the rainfall shock. For example, since workers FIRING COSTS AND FLEXIBILITY 731 might lobby the government for proworker regulation, states outcomes are more precise in these data. Therefore, one of our with more nonagricultural employment (and thus presumably first robustness checks is to replicate the main district-level a larger blue-collar lobby) may have enacted more proworker results on the state and industry panel. legislation. But less agricultural areas are also less likely to be dependent on rainfall. Another possibility is that factories’ response to shocks varies by their capital intensity and that C. Robustness Checks Using the District and State-Industry labor laws are correlated with the average labor intensity of Panels factories. In Section V, we present the main results on employ- Jayachandran (2006) addresses such concerns by includ- ment responses, as well as a variety of supporting results that ing relevant area characteristics and their interactions with demonstrate the consistency and robustness of our empirical rainfall shock as controls. Following a similar strategy, we findings: (1) exploiting the fact that larger firms (in partic- control for the interaction of Rainshockjt with key baseline ular, the IDA specifies two employment size cutoffs) are characteristics of districts, such as the percent of total employ- subject to more draconian firing costs; (2) testing for differ- ment that is in the agrarian sector and in food-based sectors, ential responses to shocks across industry types classified by and the average capital-to-output ratio in industry. We also their a priori susceptibility to local demand; (3) robustness to interact the rainfall shock variable with the following state- the inclusion of fixed effects, which control for the potential level variables measured in 1980 (before the first year of our selection of firms into states based on their level of flexibility data): the number of years that hard-left parties have held a in response to shocks; and (4) robustness to the alternative majority in the state legislature (this variable is reported to measures of labor regulation described in section CA. be an important correlate of manufacturing growth by Besley & Burgess, 2004), the total cultivated area in the state, and the percentages of cultivated area sown with rice and wheat. V. Results The cultivation variables are intended to account for differ- A. Effects of Rainfall Shocks on Agricultural Production, ences across states in the agricultural responses to rainfall Agricultural Wages, and Expenditures fluctuations. Rice and wheat are the two major crops grown in India and are also known to have different requirements in We begin by testing our premise, which is that the factory terms of rainfall. In addition, wheat cultivation is somewhat sector is affected by rainfall shocks through their effects on concentrated in the neutral states. the local population. To test whether poor rainfall induces a negative shock to local demand, we examine the impact of B. State and Industry Panel Regressions rainfall shocks on agricultural production and expenditures. The results of these regressions, which control for district and We replicate our main results on the differential responses year fixed effects (thus exploiting changes within districts to rainfall shocks across labor regulation regimes using the over time), are reported in table 5. Columns 1 and 2 capture state-industry panel. The basic specification is now the main effects of rainfall shocks on the value of agricultural production and per capita monthly expenditures. We see large xskt = αRainshockst + β(Rainshockst × Proworkerst ) declines associated with a rainfall shock in both variables, + δ(Rainshockst × Proemployerst ) indicating that the mechanism of rainfall shocks’ effects on + ρsk + ρt + jt , (3) the factory sector through local demand could be at play. Next, we test whether rainfall shocks may also induce where ρsk denotes a fixed effect for industry k in state s. a labor supply effect: low (high) rainfall, as it reduces This is analogous to the district-level regressions that measure (increases) the productivity of agricultural laborers, would how the response to rainfall shocks varies by labor law. xskt drive down (up) the agricultural wage and move workers into measures the outcome in state s, three-digit industry group k , (out of) the industrial sector. We test for this mechanism by and year t . The rainfall shock Rainshockst is measured at the measuring the impact of rainfall shocks on the agricultural state level by averaging the rainfall in districts within every wage, and find, as reported in column 3, that the effect is state and as in the district-level regressions, it is interacted weak and not statistically significant.12 with state labor law dummies. Thus, the interpretation of β Finally, in columns 4 to 6 of table 5, we verify that and δ is similar to that in the district-level specification. The the effects of rainfall shocks on agricultural production, regressions control for state-industry and year fixed effects. Clearly, compared to the district data, the local rainfall 12 A decline in agricultural production could also affect the industrial sec- shock is less precisely measured in these state-level data. tor through its effect on the price of agricultural outputs, which are often used as intermediate inputs into production in the industrial sector (e.g., But the state-industry panel adds to our analysis in several cotton sold to textile mills). Because we do not have price data for these ways. With data stretching over a period of seventeen years at inputs, we cannot test for this channel directly. However, when we regress annual frequency, the state-industry panel offers substantially the value of materials used in production on the rainfall shock main effect, we find a small and insignificant negative coefficient (results not shown more variation in rainfall over time. Second, since the ASI here), indicating that at least the value of intermediate materials used does is stratified by state and industry, estimates of factory sector not decline in response to rainfall shock. 732 THE REVIEW OF ECONOMICS AND STATISTICS Table 5.—Effect of Rainfall Shock on Agricultural Production and Household Expenditures District Panel Agricultural Per Capita Agricultural Agricultural Per Capita Agricultural Production Expenditures Wage Production Expenditures Wage Dependent variables (1) (2) (3) (4) (5) (6) Rainfall shock −8, 462 −14.28 −0.220 −18, 393.23 1.66 0.23 (1, 748)∗∗∗ (3.741)∗∗∗ (0.159) (7, 183.168)∗∗ (14.660) (0.462) Rainfall shock × (β) Proworker state −1, 854.27 0.66 −0.25 (4, 823.374) (11.422) (0.359) (δ) Proemployer state 4, 039.27 15.03 −0.18 (4, 025.022) (11.569) (0.367) Fixed effects District + Year District + Year Ho: δ − β = 0 5,894 0.0752 14.37 (3,710) (0.314) (9.744) Number of observations 4,398 1,071 4,398 4,398 1,071 4,398 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1. Robust standard errors are reported in parentheses below the coefficient estimates and allow for correlation in the error term within districts. States are classified as proworker, proemployer, or neutral based on adding the number of proworker amendments and subtracting the number of proemployer amendments passed after Indian independence in 1947. Rainfall shock = 1 if annual rainfall < 20th percentile of historical distribution, = 0 if > 20th and < 80th percentile of historical distribution, and = −1 if > 80th percentile of historical distribution; “Agricultural production” and “Per capita expenditures” are as defined section IIIC. “Agricultural wage” is the daily wage rate in INR. Specifications in columns 4–6 include rice and wheat area per acre and total cultivated area in the state in 1980 interacted with rainfall shock. Table 6.—District-Level Results: Effect of Rainfall Shock on Industrial Employment by Labor Regulation Strictness District Panel Number of Man-Days Number of Man-Days Workers (thousands) Workers (thousands) Dependent Variables (1) (2) (3) (4) Rainfall shock −582.0 −250.0 −15, 445.075 −7, 635.478 (403.5) (145.4)∗ (14, 778.097) (5, 466.523) Rainfall shock × (β) Proworker state −512.621 59.315 (1, 625.653) (577.387) (δ) Proemployer state −2, 851.241 −855.226 (1, 512.609)∗ (506.933)∗ Fixed effects District + Year Ho: δ − β = 0 −2,339 −914.5 (1,164)∗∗ (435.5)∗∗ Number of observations 1,042 1,042 1,000 1,000 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1. Robust standard errors are reported in parentheses below the coefficient estimates and allow for correlation in the error term within districts. States are classified as proworker, proemployer, or neutral based on adding the number of proworker amendments and subtracting the number of proemployer amendments passed after Indian independence in 1947. “Rainfall shock” = 1 if annual rainfall < 20th percentile of historical distribution, = 0 if > 20th and < 80th percentile of historical distribution, and = −1 if > 80th percentile of historical distribution. Specifications in columns 3 and 4 include rice and wheat area per acre and total cultivated area in the state in 1980, a dummy for whether the three-digit industry was delicensed by 1988, factory sector wages in 1988, the ratio of fixed capital to workers in 1988, percent of 1988 employment in food-based sectors and in the agrarian sector, percent of 1988 employment that was landless, the ratio of capital to output in 1988, and cumulative years (in 1988) since 1957 hard-left parties majority state legislature, all interacted with rainfall shock. farm wages, and household expenditures do not differ sys- Table 6 reports results from the district-level panel. First, tematically across proworker and proemployer states. Such columns 1 and 2 report the average impact of rainfall shocks differentials would suggest that our key estimates reflect on two measures of employment: workers and man-days. some spatial variation in the economic shocks induced by The estimated impacts are large (in relation to the district rainfall that just happens to be correlated with differences and state-industry means) and negative: for example, moving in the labor law regime. Columns 3 to 6 argue against this from the 80th to the 20th percentile of the historical rain- concern: we cannot reject the hypothesis that the response of fall distribution generates a decrease of 500,000 man-days agricultural outcomes and per capita expenditures to rainfall (column 2). shocks is the same across proworker and proemployer states. We then interact the rainfall shock variable with dummies for proworker and proemployer states; the results are reported in columns 3 and 4. In addition to district and year fixed B. Effects of Rainfall Shocks on Employment by Labor effects, these regressions control for interactions of rainfall Regulation Strictness shock with a set of baseline characteristics. These include the total cultivated area and its percentage devoted to rice and to As described in Section IV, we use both district-level and wheat in the state in 1980, percentage of 1988 employment the state-industry-level panels to test the theoretical predic- in food-based sectors and in the agrarian sector, percentage tion that the employment response to shocks should be larger of 1988 employment that was landless workers, and the ratio the lower the firing costs. Tables 6 and 7 report our main of capital to output in factories in 1988. results on the differential response of industrial employment For both employment outcomes, we can reject the hypoth- to shocks across proworker, neutral, and proemployer states. esis that the response to shock is equal across proemployer FIRING COSTS AND FLEXIBILITY 733 Table 7.—State-Industry Panel Results: Effect of Rainfall Shock on Industrial Employment by Labor Regulation Strictness State-industry panel Number of Number of Number of Number of Number of Number Workers Employees Workers Employees Workers Employees Dependent Variables (1) (2) (3) (4) (5) (6) Rainfall shock −151.3 −186.1 −170.925 −218.959 −780.93 −874.456 (48.92)∗∗∗ (56.81)∗∗∗ (141.542) (170.965) (464.247)∗ (533.400) Rainfall shock × (β) Proworker state −33.709 −61.501 −41.188 −77.518 (108.452) (127.510) (110.772) (130.446) (δ) Proemployer state −261.125 −300.622 −271.326 −319.141 (106.779)∗∗ (125.449)∗∗ (110.563)∗∗ (129.501)∗∗ Fixed effects State × 3-digit State × 3-digit Rainfall shock × 3-digit industry code industry code + Year industry code + Year + State × 3-digit industry code Ho: δ − β = 0 −227.4 −239.1 −230.1 −241.6 (116.7)∗ (132.3)∗ (111.3)∗∗ (126.6)∗ Number of observations 24,374 24,374 24,374 24,374 24,374 24,374 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1. Robust standard errors are reported in parentheses below the coefficient estimates and allow for correlation in the error term within state × 3-digit industry code. States are classified as proworker, proemployer, or neutral based on adding the number of proworker amendments and subtracting the number of proemployer amendments passed after Indian independence in 1947. Rainfall shock = 1 if annual rainfall < 20th percentile of historical distribution, = 0 if > 20th and < 80th percentile of historical distribution, and = −1 if > 80th percentile of historical distribution; specifications in columns 3–6 include rice and wheat area per acre and total cultivated area in the state in 1980, a dummy for whether the three-digit industry was delicensed by 1980, factory sector wages in 1980, and cumulative years (in 1980) since 1957 hard-left parties majority state legislature, all interacted with rainfall shock. and proworker states. Further, the point estimate on the dif- type. The results are presented in columns 5 and 6 of table 7. ference between the two interaction coefficients shows that Compared to columns 3 and 4, there is virtually no difference the employment response is larger in proemployer states. In in the estimates of the differential impact of rainfall shocks terms of magnitudes, proemployer districts shed about 2,300 across proemployer and proworker states. This gives us some more workers than proworker districts do, which translates confidence that our findings are not driven by firm selection into a 9 percentage point difference in response. into labor regimes. Columns 1 to 4 in Table 7 report the analogous results Finally, in table A1, we report the main employment using the state-industry panel, the outcomes being the num- results using the three alternative measures of labor regula- ber of employees and workers.13 The regressions include the tion described in section IIIA. We find that that the estimated interactions of rainfall shock with a set of state and industry effects are qualitatively as well as quantitatively comparable baseline characteristics, including the total cultivated area to the results obtained in tables 6 and 7, although the results and its percentage devoted to rice and to wheat in the state in using Bhattacharjea’s coding and the dispute settlement cod- 1980 and a dummy for whether an industry was delicensed ing are more precise than the results using the employment by 1980. In addition, they include state-by-industry fixed protection coding. effects. The results are fully consistent with those from the district panel: proemployer state-industry groups shed about 230 more workers than their proworker counterparts, a 7% C. Effect of Shocks on Employment by Factory Size point difference in response. Together, these results constitute our main test of the theoretical predictions from the canonical In this section, we exploit the particulars of the regula- labor demand model laid out in the appendix. tion set forth in the IDA related to the extent of firing costs A potential concern is that firms’ location decisions across for large factories. As described earlier, the IDA regulation states are nonrandom and may be correlated with labor regu- stipulates that larger factories will face higher firing costs. lation regimes as well as the way in which these firms adjust In particular, factories with employment lower than 50 face to shocks. For example, if the least flexible firms—those that no firing costs, factories with employment between 50 and require the most labor adjustment in times of shock—locate 100 must compensate workers who are retrenched, and fac- where there are weak worker lobbies (which often generate tories with employment greater than 100 workers must file proemployer amendments), the differential response across each layoff with the government, who then has the power to proworker and proemployer states would tend to overstate deny the factory the ability to retrench. Accordingly, we par- the effect of labor regulations on the average firm. tition our data by these size cutoffs before aggregating to the Our conjecture is that in part, such inherent differences district level, thus creating a panel data set disaggregated by in flexibility would be determined by the industry to which a both district and factory size category. This data set has three factory belongs. Therefore, controlling for differential effects observations per district-year, corresponding to small (fewer of rainfall shocks by industry should control for the potential than 50 workers), medium (50–100 workers), and large (more selection of flexible firms into proworker states, to the extent than 100 workers) factories. that this flexibility is encapsulated by NIC codes for industry In table 8, we use this data set to test if the differen- tial employment response across proworker and proemployer 13 The former includes supervisory and managerial employees. states is largest for large factories, which is to be expected if 734 THE REVIEW OF ECONOMICS AND STATISTICS Table 8.—Triple Interaction Tests of Effect of Rainfall Shock on Employment by Factory Size District Panel Number of Man-Days Workers (thousands) Dependent Variables: (1) (2) Rainfall shock × (ζ1 ) Proworker state × large factory 887.530 386.890 (1, 036.535) (392.724) (ζ2 ) Proemployer state × large factory −1, 062.329 −248.158 (904.782) (277.437) (ζ3 ) Proworker state × medium factory 786.504 324.94 ∗∗ (394.378) (160.249)∗∗ (ζ4 ) Proemployer state × medium factory 205.405 112.081 (302.347) (108.586) (β1 ) Proworker state −538.913 −150.358 (606.160) (229.872) (β2 ) Proemployer state −354.361 −122.865 (566.411) (216.420) Fixed effects District × size + Year Response of small firms across labor regimes [β2 − β1 ] 184.522 27.49 (494.09) (195.69) Response of medium firms across labor regimes [(ζ4 + β2 ) − (ζ3 + β1 )] −396.55 −185.37 (343.39) (122.15) Response of large firms across labor regimes [(ζ2 + β2 ) − (ζ1 + β1 )] −1, 765.31 −607.56 (919.59)∗ (323.00)∗ Diff-in-diff for large firms relative to small [ζ2 − ζ1 ] −1,950 −635.0 (1163)∗ (412.6) Diff-in-diff for medium firms relative to small [ζ4 − ζ3 ] −581.1 −212.9 (446.0) (176.7) Number of observations 3,000 3,000 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1. Robust standard errors are reported in parentheses below the coefficient estimates and allow for correlation in the error term within districts. States are classified as proworker, proemployer, or neutral based on adding the number of proworker amendments and subtracting the number of proemployer amendments passed after Indian independence in 1947. “Rainfall shock” = 1 if annual rainfall < 20th percentile of historical distribution, = 0 if > 20th and < 80th percentile of historical distribution, and = −1 if > 80th percentile of historical distribution. “Large” indicates factories with > 100 workers, “medium” indicates factories with > 50 and < 100 workers, and “small” indicates factories with < 50 workers. The excluded category is “small”. Specifications include rainfall shock × size and labor regulation × size interactions. Specifications also include rice and wheat area per acre and total cultivated area in the state in 1980, a dummy for whether the three-digit industry was delicensed by 1988, factory sector wages in 1988, the ratio of fixed capital to workers in 1988, percent of 1988 employment in food-based sectors and in the agrarian sector, percent of 1988 employment that was landless, the ratio of capital to output in 1988, and cumulative years (in 1988) since 1957 hard-left parties majority state legislature, all interacted with rainfall shock. Bold text emphasizes results corresponding to specific hypothesis test of interest. the estimated differential captures the impact of labor reg- effect is statistically significant only in large firms. Specif- ulations. To do this test, we estimate a model with triple ically, the difference in the response of large firms across interactions (rainfall shock by labor regulation by size), con- proemployer and proworker states [(ζ2 + β2 ) − (ζ1 + β1 )] is trolling for district by size rainfall shock by size, and year negative and significant at the 10% level. fixed effects (s indicates size index and r indicates rainfall Next, we test the null hypotheses that the differences-in- shock index): differences relative to the small factories are zero: ζ2 − ζ1 = 0 and ζ4 − ζ3 = 0. In words, the null ζ2 − ζ1 = 0 states xjqt = ζ1 (Rainshockjt × Proworkerjt × Largejqt ) that labor regulations moderate the employment response to rainfall shocks in the same way for both large and small + ζ2 (Rainshockjt × Proemployerjt × Largejqt ) factories. We can statistically reject the null for employ- + ζ3 (Rainshockjt × Proworkerjt × Mediumjqt ) ment in the large versus small comparison (column 1). This + ζ4 (Rainshockjt × Proemployerjt × Mediumjqt ) difference-in-differences estimate for large versus small is + β1 (Rainshockjt × Proworkerjt ) negative, as expected. The same holds for the medium versus small comparison, though the difference-in-differences is not + β2 (Rainshockjt × Proemployerjt ) statistically significant. + ρjq + ρrq + ρt + jt , D. Effect of Shocks on Employment by Industry Type where q indexes the sector (small, medium, or large), ρjq is a district-sector specific fixed effect, and ρrq is a rain-shock- Next, we use the state-industry panel introduced ear- sector specific effect. As before, the omitted categories are lier to test whether the differential employment response small factories and neutral states. to rainfall shocks varies by industry type. In particular, we The regressions indicate that the employment response group industries based on their susceptibility to local demand to rainfall shocks is larger in proemployer (as compared to shocks. If, as the results from table 5 suggest, rainfall shocks proworker) states among medium and large factories but not affect the factor sector predominantly through their effects small factories. This is consistent with the absence of man- on local demand for the goods that factories produce, then dated firing costs for small factories. Further, the differential industries that are more dependent on local demand should FIRING COSTS AND FLEXIBILITY 735 Table 9.—Triple Interaction Tests of Effect of Rainfall Shock on Employment by Industry Type State-Industry Panel Number of Number of Number of Number of Workers Employees Workers Employees Dependent Variables: (1) (2) (3) (4) Rainfall shock −285.1 −387.5 −189.976 −245.168 (114.7)∗∗ (138.5)∗∗∗ (137.696) (167.316) Rainfall shock × (ζ1 ) Proworker state × Dummy for NIC code in 200–299 18.757 67.147 (137.216) (168.642) (ζ2 ) Proemployer state x Dummy for NIC code in 200–299 −390.469 −387.504 (264.746) (294.352) (ζ3 ) Proworker state −45.336 −96.510 (106.501) (136.068) (ζ4 ) Proemployer state −79.633 −120.829 (120.268) (141.052) Dummy for NIC code in 200–299 −51.02 −10.40 51.502 67.701 (87.60) (99.96) (59.395) (75.405) Fixed effects State × 3-digit industry code + Year Response of 300–400 NIC industries across labor regimes −34.397 −24.319 [ζ4 − ζ3 ] (104.683) (136.150) Response of 200–300 NIC industries across labor regimes −443.523 −478.970 [(ζ2 + ζ4 ) − (ζ1 + ζ3 )] (244.668)∗ (266.424)∗ Diff-in-diff for 200–300 NIC industries relative to 300-400 NIC −409.2 −454.7 [ζ2 − ζ1 ] (286.6) (322.0) Number of observations 24,374 24,374 24,374 24,374 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1. Robust standard errors are reported in parentheses below the coefficient estimates and allow for correlation in the error term within state × 3-digit industry code. States are classified as proworker, proemployer, or neutral based on adding the number of proworker amendments and subtracting the number of proemployer amendments passed after Indian independence in 1947. “Rainfall shock” = 1 if annual rainfall < 20th percentile of historical distribution, = 0 if > 20th and < 80th percentile of historical distribution, and = −1 if > 80th percentile of historical distribution. Specifications in columns 3–6 include rice and wheat area per acre and total cultivated area in the state in 1980, a dummy for whether the three-digit industry was delicensed by 1980, factory-sector wages in 1980, and cumulative years (in 1980) since 1957 hard-left parties majority state legislature, all interacted with rainfall shock. Bold text emphasizes results corresponding to specific hypothesis tests of interest. exhibit a larger differential response across proworker and labor regulation regimes. However, a triple interactions test proemployer states. (rainfall shock by labor regulation by NIC code grouping) Based on the NIC code, we split industries into two groups: cannot reject the null that this difference across the two between NIC codes 200 and 299, inclusive, and between NIC industry groups is zero. codes 300 and 399, inclusive. NIC codes 200 to 299 corre- spond to industries whose focus is agricultural and natural E. Effects of Shocks on Output and Profits industrial products, such as food products and beverages, tex- tiles, paper, wood, and leather products. NIC codes 300 to 399 Finally, we examine whether firms in proworker states describe the more technological and heavy industries like were able to adjust their output to the same extent as those in chemicals and pharmaceutical, metal products, machinery, proemployer states in response to shock and whether the con- and electronics. This is an admittedly rough categorization. straints imposed by firing costs have on impact on proworker Although we expect the first group to be more dependent firms’ profits more than proemployer firms. To test these on local demand, both industry groups are likely to con- hypotheses, we again employ the district-level panel and tain traded and nontraded goods industries. Another caveat run regressions of the form described in equation (2), which is that rainfall shocks might also affect factories through include district and year fixed effects. higher prices of locally produced raw materials. Since NIC The results are reported in table 10, with the dependent groups 200 to 299 are more likely to be tied to the local variables being the value of total output, value added, and supply of agricultural raw materials, both the demand and profits. They suggest that there is no differential change raw material shocks are expected to be stronger for this across proworker and proemployer states in these outcomes. group. We should note, however, that the coefficients on the inter- The results are reported in table 9. The differential response action terms in table 10 are large relative to the mean values across proworker and proemployer states is more sizable for of these variables in table 4. Hence, the evidence for no dif- industries tied to local demand (NIC 200 to 299). Specif- ferential change is weak; that is, while we cannot reject that ically, the difference between proemployer and proworker the differentials between proworker and proemployer are 0, responses is estimated to be −443.5 for NIC groups 200 to we also cannot reject that they are very large. 299 and only −34.9 for NIC groups 300 to 399, and only the Taken at face value, the results on total value added and former is statistically significant. These results are consistent output are consistent with our evidence on the equal effects with the demand or input channel interpretations of the effects of rainfall shocks across proworker and proemployer states of rainfall shocks on the factory sector, as well as with our on agricultural production and household expenditures: we basic premise that shocks induce differential responses across find large declines in these (consistent with rainfall shocks’ 736 THE REVIEW OF ECONOMICS AND STATISTICS Table 10.—Effect of Rainfall Shock on Output and Profits by Labor Regulation Strictness District-Level Panel Total Output Value Added Profits Total Output Value Added Profits Dependent Variables: (1) (2) (3) (4) (5) (6) Rainfall shock −38.86 −14.23 −8.495 −597.533 −238.728 −137.519 (36.02) (7.958)∗ (5.895) (1, 080.098) (232.376) (145.620) Rainfall shock × (β) Proworker state −34.791 10.651 9.540 (125.066) (27.377) (21.007) (δ) Proemployer state 13.173 8.072 11.046 (100.574) (23.064) (18.599) Fixed effects District + Year District + Year Ho: δ − β = 0 47.96 −2.579 1.506 (92.73) (22.24) (14.95) Number of observations 1,042 1,042 1,042 1,000 1,000 1,000 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1. Robust standard errors are reported in parentheses below the coefficient estimates and allow for correlation in the error term within districts. States are classified as proworker, proemployer, or neutral based on adding the number of proworker amendments and subtracting the number of proemployer amendments passed after Indian independence in 1947. “Rainfall shock” = 1 if annual rainfall < 20th percentile of historical distribution, = 0 if > 20th and < 80th percentile of historical distribution, and = −1 if > 80th percentile of historical distribution. Specifications in columns 4–6 include rice and wheat area per acre and total cultivated area in the state in 1980, a dummy for whether the three-digit industry was delicensed by 1988, factory-sector wages in 1988, the ratio of fixed capital to workers in 1988, percent of 1988 employment in food-based sectors and in the agrarian sector, percent of 1988 employment that was landless, the ratio of capital to output in 1988, and cumulative years (in 1988) since 1957 hard-left parties majority state legislature, all interacted with rainfall shock. “Value added” is the pecuniary value of total output minus intermediate inputs; capital stock, materials, fuel, total output, value added and profits have been converted to thousands of 2009 U.S. dollars. Bold text emphasizes results corresponding to specific hypothesis tests of interest. Table 11.—Effect of Rainfall Shock on Nonlabor Inputs, Wages and Labor Intensity by Labor Regulation Strictness District-Level Panel Man-Days Value of Number of Wage per per worker Output per Capital Materials Fuel Electricity Factories worker (thousands) Worker Dependent variables: (1) (2) (3) (4) (5) (6) (7) (8) Rainfall shock −542.315 −176.485 33.557 60.413 −1, 378.69 0.004 0.170 0.089 (697.882) (749.014) (86.060) (41.867) (518.152)∗∗∗ (0.002)∗ (0.122) (0.053)∗ Rainfall shock × (β) Proworker state −16.060 −63.414 −7.668 −4.917 −9.394 −0.000 0.003 −0.008 (93.969) (84.223) (8.898) (5.748) (53.123) (0.000) (0.017) (0.007) (δ) Proemployer state −28.481 −19.850 3.376 −1.528 −67.577 0.000 −0.003 −0.004 (67.904) (65.353) (7.843) (4.222) (43.873) (0.000) (0.017) (0.006) Fixed effects District + Year Ho: δ − β = 0 −12.42 43.56 11.04 3.388 −58.18 0.000227 −0.00614 0.00370 (61.31) (61.01) (9.840) (5.641) (39.06) (0.000152) (0.00953) (0.00382) Number of observations 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 ***p < 0.01; **p < 0.05; *p < 0.1. Robust standard errors are reported in parentheses below the coefficient estimates and allow for correlation in the error term within districts. States are classified as proworker, proemployer, or neutral based on adding the number of proworker amendments and subtracting the number of pro-employer amendments passed after Indian independence in 1947. “Rainfall shock” = 1 if annual rainfall < 20th percentile of historical distribution, = 0 if > 20th and < 80th percentile of historical distribution, and = −1 if > 80th percentile of historical distribution; specifications in columns 4–6 include rice and wheat area per acre and total cultivated area in the state in 1980, a dummy for whether the three-digit industry was delicensed by 1988, factory-sector wages in 1988, the ratio of fixed capital to workers in 1988, percent of 1988 employment in food-based sectors and in the agrarian sector, percent of 1988 employment that was landless, the ratio of capital to output in 1988, and cumulative years (in 1988) since 1957 hard-left parties majority state legislature, all interacted with rainfall shock. “Capital” is the value of fixed capital stock at close of business year. “Materials,” “Fuel”, and “Electricity” are annual values of intermediate inputs used. “Wage per worker” is the wage bill (total amount paid to workers in 2009 USD) divided by number of workers. “Man-days per worker” is the number of man-days divided by the number of workers; “Value of output per worker” is value of total output in thousands of 2009 USD divided by number of workers. Bold text emphasizes results corresponding to specific hypothesis tests of interest. effects through a demand channel) but no differential decline If firms were able to adjust along these margins well enough, across labor regulation regimes.14 we might not measure an effect on profits. The finding of no differential change in profits is surpris- To test this hypothesis, we use as dependent variables the ing in the sense that we might expect there to be a larger value of capital and intermediate inputs—in particular, mate- profit decline in proworker states, which were constrained rials, fuel, and electricity. The results are reported in columns by firing costs from adjusting the level of employment opti- 1 to 4 of table 11. For all four outcomes, we find no differen- mally. Several potential arguments could explain the results tial adjustment across proworker and proemployer states in on profits. response to shock. These results indicate that nonlabor inputs We might expect that the constraints imposed on firing are not declining more intensively in proworker states to costs by labor regulation could generate adjustment along mitigate the impact of employment adjustment constraints.15 other margins of inputs. If this were the case, we might see The second hypothesis is related to differential attrition of that, for example, capital or intermediate inputs would adjust factories across labor regulation regimes. We might find no more intensively in proworker states than proemployer states. effects on profits if the firms with the largest negative profits 15 We note that as is the case with output and profits, the evidence for 14 We also verify that the results presented in table 10 are robust to the using no differential change in intermediate inputs is weak: the coefficients on the alternative measures of labor regulation. These results are available on intermediate inputs in table 11 are large compared to the mean values of request. these variable in table 4. FIRING COSTS AND FLEXIBILITY 737 are going out of business (dropping out of the sample) more (from column 6 of table 11) that reductions in the wage bill intensively in proworker states in response to the shock. To are equal across proemployer and proworker states. test this hypothesis, we look directly at responses in the num- ber of factories in the district to rainfall shock. Column 5 of VI. Conclusion table 11 reports the results. We find no evidence of differen- tial declines in the number of factories across proemployer Job security provisions, although politically popular, have and proworker states in response to shock. been the focus of intense academic debate. The job security The third hypothesis we examine for the lack of differential they confer needs to be weighed against reduced flexibil- declines in profits is that the industrial wage declined more ity in hiring and firing, which has been found to have intensively in proworker states than in proemployer states.16 negative impacts on aggregate outcomes. In this paper, we If this were the case, we would expect that in proworker states, have devised a novel test of the fundamental hypothesis firms would see a greater reduction in the wages per worker that employment protection laws attenuate the employment than that seen by firms in proemployer states. To test this responses of firms to external shocks. We exploit a setting hypothesis, we examine wages per worker. The results are that exhibits variation in labor regulation as well as a mea- reported in column 6 of table 11. The results show that wages surable source of unpredictable shocks. Our setting is rural per worker do not decline differentially across proworker and India, where rainfall fluctuations create demand and wage proemployer states. shocks for local industries and labor regulation varies tem- The fourth hypothesis related to the nonresult on profits porally as well as spatially. Our results provide a striking is that since firms in proworker states are constrained from confirmation of the theory: rainfall shocks change industrial adjusting the number of workers to the optimal extent, they employment by shifting the demand for industrial prod- may choose instead to adjust the labor intensity of their cur- ucts, and the employment adjustment is more pronounced in rent workforce, for example, the number of hours per day regions where labor regulations are less restrictive. We also or the number of days per month each worker puts forth in examine the responses of factories that were exempt from labor. If this were the case, we should observe that proworker the regulation and find that there is no differential adjustment states use workers differentially more intensively in times of across labor regimes, consistent with our interpretation that shock. We use two outcomes to test this hypothesis: man-days the differential responses for nonexempt factories are indeed per worker and value of total output per worker (the second attributable to labor regulation. of which is used under the assumption that a worker more Because we are looking at employment adjustment in the intensively utilized will produce more output). The results of formal manufacturing sector (the only part of the economy these regressions are reported in columns 7 and 8 of table 11. subject to the labor laws in question), this is only one part Again, we find no evidence of differential adjustment across of the full picture. To understand the overall effects of labor proworker and proemployer states. laws on employment and job security requires more compre- Finally, there are two hypotheses for which, due to data hensive data. Reduced job creation and destruction rates may constraints, we have no test, which might explain the results seem to imply longer unemployment spells, and possibly dis- on profits. First, the prices of nonlabor inputs might be adjust- proportionately so for certain segments of the labor force. We ing differentially across proworker and proemployer states. believe this is a promising line of inquiry for future research. If we saw a larger reduction in the price of nonlabor inputs In the Chilean context, Montenegro and Pages (2004) use in proworker states, profits fluctuations due to shocks may household survey data and find that job security provisions equalize across labor regulation regimes. We believe this and minimum wage requirements confer positive benefits on explanation is unlikely to be the only one, given that if older and skilled workers, as well as male workers, but that the prices of nonlabor inputs changed differentially across these benefits are achieved at the expense of young, unskilled, proworker and proemployer states, we would likely see a or female workers. These costs of labor regulation are likely differential change in the use of those inputs as well. The to be magnified when labor is not very mobile. Jayachandran results from columns 1 to 4 of table 11 seem to refute this (2006) shows that agricultural productivity shocks in rural claim. India create large changes in the wage when labor is immo- Second, if proworker firms were more intensively laying bile and incomes are near subsistence level, a finding that off casual laborers (part-time laborers who are not accounted may be related to the inability of the manufacturing sector to for in the data) during periods of shocks, these firms might absorb workers. be able to achieve a commensurate reduction in “effective” Although we have not touched on the issue in this paper, employment as the reduction seen for firms in proemployer one may wonder how employment adjustment plays out in an states. This explanation would also be consistent with the fact economy in which there is a large, unregulated informal sec- tor coexisting with a smaller, regulated (but more productive) formal sector. In fact it has been conjectured that labor regula- 16 In fact, we might suspect the opposite if we believe that worker lobbies tions, in as much as they apply only to the formal sector, tend in proworker states are better able to bargain for wage stability. In this case, we would expect the wage to fall more intensively in response to shock in to encourage informality. In the Indian context, this could proemployer states. account for the preponderance of small firms. In fact, the vast 738 THE REVIEW OF ECONOMICS AND STATISTICS majority of nonagricultural workers in India are employed in Jayachandran, S., “Selling Labor Low: Wage Responses to Productivity the informal sector. An interesting hypothesis, and one that Shocks in Developing Countries,” Journal of Political Economy, 114 (2006), 538–575. we propose to test in the future, is that the informal sector Malik, P. L. Industrial Law (Lucknow, India: Eastern Book Company, serves as a buffer for the formal sector: when the latter sheds 1997). workers, the informal sector soaks up the extra workers. In Montenegro C., and C. Pages, “Who Benefits from Labor Market Regula- tions? Chile 1960–1998,” in J. Heckman and C. Pages (Eds.) Law this case, the employment adjustment in the informal sector and Employment (Chicago: University of Chicago Press, 2004). would be the mirror image of employment adjustment in the Nickell, S. J., “Dynamic Models of Labor Demand,” in O. Ashenfel- formal sector. ter and R. Ledyard (Eds.), Handbook of Labor Economics, Vol. 1 (Amsterdam: Elsevier North Holland, 1986). There is a sizable literature on another aspect of job secu- Oi, Walter, “Labor as a Quasi-Fixed Factor,” Journal of Political Economy rity provisions, namely their effect on aggregate employment 70 (1962), 538–555. and output. Fallon and Lucas (1993) estimated labor demand to show that the increased stringency of job security pro- APPENDIX visions in India after 1982 resulted in a large reduction in employment. Similar findings are reported in Besley and Model Burgess (2004), based on comparing employment and out- We outline a partial-equilibrium model based on Bertola (1990) that put across labor regimes in India. Aghion, Burgess, Redding, formalizes the key intuition of the paper. To keep the model simple, we and Zilibotti (2008) have extended the analysis to show that do not directly introduce an agricultural sector or specify a labor supply the effect of labor regulations on aggregate employment and equation. Instead, we consider the labor demand of a price-taking firm that is subject to exogenous shocks to the wage and output price, the shocks output has been greater in more regulated product markets. being assumed to flow from productivity shocks to agriculture. Overall, the negative effects of firing restrictions are many, The model is set in continuous time. Consider an infinitely lived price- and need to be weighed against the employment stability that taking firm that uses only labor to produce its output according to an increasing, concave production function f (L ). The firm discounts future they confer. profits at the constant rate r . There are two possible states of the world, denoted by G (good, or high, rainfall) and B (bad, or low, rainfall). The associated prices and wages in these states are given by pG , wG , pB , and wB respectively. Suppose that the state is currently B at time t . The transition to the G REFERENCES state follows a Poisson process with constant rate of arrival θB . Similarly the Abraham, K., and S. Houseman, “Labor Adjustment under Different Institu- transition from state G to state B is a Poisson process with constant arrival tional Structures: A Case Study of Germany and the United States,” rate θG . We model employment protection in terms of a simple firing cost: NBER working paper 4548 (1993). hiring workers is frictionless, but firing workers is assumed to entail a cost of Adda, J., and R. Cooper, Dynamic Economics (Cambridge, MA: MIT Press, c per worker. This linear specification of adjustment costs is convenient for 2003). our purposes but not strictly necessary. However, because our data cannot Aghion, P., R. Burgess, S. Redding, and F. Zilibotti, “The Unequal Effects be used to distinguish between different adjustment cost specifications, we of Liberalization: Evidence from Dismantling the License Raj in stay with the linear specification here while remaining agnostic about the India,” American Economic Review 98 (2008), 1397–1412. exact form. Ahsan, A., and C. Pages, “Are All Labor Regulations Equal? Evidence from In what follows, we will consider a stationary policy for the firm such Indian Manufacturing,” IZA discussion paper 3394 (2008). that the firm employs LG workers whenever the state is G and LB workers Basu, K., G. Fields, and S. Gupta, “Labor Retrenchment Laws and Their whenever the state is B.17 We will assume that pG > pB , corresponding to Effect on Wages and Employment: A Theoretical Investigation,” the assumption that high rainfall tends to raise demand for the industrial mimeograph, Cornell University (2008). good. The wage rates in the two states are unrestricted, although we may Bentolila, S., and G. Bertola, “Firing Costs and Labor Demand: How Bad assume without loss of generality that wB < wG , reflecting the possibil- Is Eurosclerosis?” Review of Economic Studies 57 (1999), 381–402. ity that poor rainfall reduces the labor demand in agriculture and thereby Bentolila, S., and G. Saint-Paul, “The Macroeconomic Impact of Flexible increases the labor supply to industry. For concreteness, we will assume Labor Contracts, with an Application to Spain,” European Economic that the price of output and the wage in the different states are such that Review 36 (1992), 1013–1047. LG > LB (i.e., the demand effect outweighs the wage effect, as will turn out Bertola, G., “Job Security, Employment and Wages,” European Economic to be true in the data). Review, 34 (1990), 851–886. The choice of L is analogous to investment in an asset whose return is Besley, T., and R. Burgess, “Can Labour Regulation Hinder Economic Per- stochastic. Since the policy is stationary, we need only define the value of formance: Evidence from India,” Quarterly Journal of Economics the asset in the two states of the world, G and B. Let VG and VB denote these 119 (2004), 91–134. two values. Given the assumptions on the transition probabilities, we can Bhattacharjea, A., “Labour Market Regulation and Industrial Performance use the standard asset equation to write: in India—A Critical Review of the Empirical Evidence,” Centre for Development Economics, Delhi School of Economics, working rVG = pG f (LG ) − wG LG + θG [VB − VG − c(LG − LB )], (A1) paper 141 (2006). rVB = pB f (LB ) − wB LB + θB [VG − VB ]. (A2) Botero, J., S. Djankov, R. L. Porta, F. L. de Silanes, and A. Shleifer, “The Regulation of Labor,” Quarterly Journal of Economics 119 (2004), 1339–1382. Upon transitioning to state B from state G, the firm chooses LB to solve Fallon, P. R., and R. Lucas, “Job Security Regulations and the Dynamic Demand for Industrial Labor in India and Zimbabwe,” Journal of max VB − c(LG − LB ). (A3) Development Economics, 40 (1993) 241–275. Hamermesh, D., Labor Demand (Princeton NJ: Princeton University Press, The first-order condition is simply ∂ VB ∂ LB = −c. 1993). Heckman, J., and C. Pages, Law and Employment (Chicago: University of 17 The profit function of the firm is bounded, continuous, and concave. Chicago Press, 2004). Hopenhayn, H., and R. Rogerson, “Job Turnover and Policy Evaluation: A The stationarity of the shocks and the presence of discounting can then be General Equilibrium Analysis,” Journal of Political Economy, 101 used to establish the optimality of a stationary policy. For a simple proof in (1993), 915–938. the discrete-time case, see Adda and Cooper (2003). Table A1.—Main Results Using Alternative Labor Regulation Measures District Panel State-Industry Panel Bhattacharea’s Employment Protection Dispute settlement Bhattacharjea’s Recoding Employment Protection Dispute Settlement Recoding of Besley & Legislation (Ahsan & (Ahsan & of Besley & Legislation (Ahsan & (Ahsan & Burgess (2004) Pages, 2008) Pages, 2008) Burgess (2004) Pages, 2008) Pages, 2008) Labor regulation measure used: (1) (2) (3) (4) (5) (6) Rainfall shock −14, 602.59 −9,110 −20, 034.91 −216.55 −304.32 −169.16 (14, 976.48) (11,840) (18, 954.27) (111.88) (132.41)∗∗ (101.18)∗ Rainfall shock × (β) Proworker state −667.45 400.3 1532.05 −0.001 113.58 −93.15 (1, 442.68) (834.3) (2, 198.52) (0.004) (95.23) (94.27) (δ) Proemployer state −2, 662.96 −1, 538.87 −215.46 −190.41 (1, 054.48)∗∗∗ (638.71)∗∗ (106.76)∗∗ (95.54)∗∗ Fixed effects District + Year State × 3-digit industry code + Year Ho: δ − β = 0 −1, 995.51 −400.3 −3,070.91 −215.46 −113.58 −97.26 (1, 588.09) (834.3) (2335.90) (106.76)∗∗ (95.23) (113.58) Number of observations 1,000 1,000 1,000 24,374 24,374 24,374 FIRING COSTS AND FLEXIBILITY The dependent variable in all regressions is the number of workers. ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1. Robust standard errors are reported in parentheses below the coefficient estimates and allow for correlation in the error term within districts for specifications in columns 1–3, and within state × 3-digit industry code for specifications in columns 4–6. “Rainfall shock” = 1 if annual rainfall < 20th percentile of historical distribution, = 0 if > 20th and < 80th percentile of historical distribution, and = −1 if > 80th percentile of historical distribution; please refer to the Appendix for definitions of the alternate labor regulation definitions used; for controls used in specifications reported in columns 1–3, please refer to footnotes in table 4, for controls used in specifications reported in columns 4–6, refer to footnotes in table 7. Bold text emphasizes results corresponding to specific hypothesis tests of interest. 739 740 THE REVIEW OF ECONOMICS AND STATISTICS On transitioning to state G from state B the firm chooses LG to solve These equations capture the intuition that adjustment costs create a wedge between the firm’s marginal revenue product and the wage. The max VG . (A4) effective wage is therefore higher than the actual wage during good times and lower during bad times. It is easy to see that an increase in the fir- The first-order condition is ∂ VG = 0. These first-order conditions, along ing cost c reduces employment in the high-rainfall state G and increases ∂ LG employment in the low-rainfall state B. Put differently, fluctuations repre- with equations (A1) and (A2), imply that ∂ ∂ VB LG = ∂ VG ∂ LB = 0. sented by rainfall shocks will induce smaller employment adjustments in Using the asset-pricing equations, we also have more regulated environments. This is the hypothesis we will proceed to test. ∂ VB 1 ∂ VG As we noted in the text, shocks represented by rainfall fluctuations = pB f (LB ) − wB + θB , ∂ LB r + θB ∂ LB plausibly create opposing effects on industrial labor demand, through ∂ VG 1 ∂ VB the demand and labor supply channels. The model outlined here clari- = pG f (LG ) − wG − cθG + θG . fies that it is the net effect on labor demand of these wage and price ∂ LG r + θG ∂ LG shocks that is magnified in lower firing cost (i.e., more flexible) regimes. That is, if the net effect of good rainfall is to increase (decrease) indus- The first-order conditions, together with the fact that ∂ VB ∂ LG = ∂ VG ∂ LB = 0, trial employment, then we should expect to observe a greater increase then imply (decrease) in employment in regions where labor regulations are less stringent. This conclusion is valid as long as rainfall shocks represent pB f (LB ) = wB − (r + θB )c, identical demand and labor supply fluctuations across different labor pG f (LG ) = wG + cθG . regimes.