Report No. 47193-GT Guatemala Investment Climate Assessment (In Two Volumes) Volume II: Background Notes on Productivity June 26, 2008 Finance and Private Sector Unit Poverty Reduction and Economic Management Unit Latin America and the Caribbean Region Document of the World Bank TABLE CONTENTS OF SECTION 1. METHODOLOGICAL NOTES ................................................................... 1 1. REPLACEMENT STRATEGY FOR MISSINGPRODUCTION F U N C T I O N VARIABLES ...................... 1 2. ENDOGENEITY PROBLEMS ............................................................................................................. 6 SECTION 2. PRODUCTIVITY RESULTS ........................................................................ 8 SECTION 3. INTERPRETATIONS OF THE PRODUCTIVITY RESULTS ...............40 1. SECURITY COSTS AND THEIREFFECT ON PRODUCTIVITY ........................................................ 40 2. EFFECTSICBLOCKVARIABLES PRODUCTrVITY........................................................... OF ON 42 3. GENDER IMPACT ............................................................................................................................ 43 4. DIFFERENCES INSAMPLING(2003VS. 2007DATI) .................................................................. 44 Table 1.1: Observations available for regression analysis in Guatemala by Industry and size and percentage o f observations lost ............................................................................................. Table 1.2: Observations available for regression analysis in Guatemala by Industry and region and percentage of observations lost .. Table 1.3: Extended Table 2.1: General Info Table 2.2: Investment climate (IC) and control (C) variables .... Table 2.3: Investment climate (IC) and control (C) variables (c Table 2.4: Investment c Table 2.5: Investment climate (IC) and control (C) variables Table 2.6: Total number o f observations before and after cleaning missingvalues and outliers in production function CpF)variables .......................................................................................................................................................... 14 Table 2.7: Representativeness of production function variables before and after cleaning for missing values and outliers ............... ........................... ............. ...... Table 2.8: Total number o f observations and response rate o f I C and C variables inthe original sample ..........15 16 Table 2.9: Correlation matrix among productivity measures.... Table 2.10 Correlation Table 2.11: I C A elastic Table 2.12: I C A elas industry,region and size .................. Table 2.13: Percentag Aggregate Productivity Table 2.14: I C impact on the decomposition by inputs of the efficiency term of the Olley and Pakes decomposition inlogs ........................................................................................................................................................... 24 Table 2.15: Two-stage least squares (2SLS) estimation o f employment equation. Table 2.16: Two stage least squares (2SLS) estimation o f real wages equation ...... Table 2.17: T w o stage least squares (2SLS) estimation of probability o f exporting equation ................................ 27 Table 3.2: Percentage of female workers instaff, average by industry Table 3.3: Percentage o f female workers instaff, average by size ........................................... Table 3.6: Differences inSampling (2003 vs. 2007) ...................................................................................................... 44 i Figures Figure 2.1: Olley and Pakes Decomposition in Levels by Industry and Region of Aggregate Productivity ................................................................................... 29 ..................................................... ....................... ......................................................................................... Figure 2.5: Relative I C effects by groups of variab efficiency (mixed O&P decomposition and simulations of a 20% improvement inI C and C variables) .............30 Figure 2.6: Relative I C effects on aggregate productivity @fixed O&P decomposition).. Figure 2.7: Relative I C effects on average productivity (Ahxed O&P decomposition) ........................ Figure 2.8: Relative I C effects on efficiency @fixed O&P decomposition)........................................... Figure 2.9: Relative I C effects by groups of variables on average productivity (Decomposition in Logs); b y sue ........................................... ............................ Figure 2.10: I C A Percentage Absolute Contrib Figure2.11: Relative IC effects on average log-e Figure 2.12: Relative I C effects on average log-r ................................................................... 35 Figure2.13: Relative IC effects on the probab Figure 2.14 Relative I C A effects on the probab Figure 2.15: Relative I C effects by groups of va Figure2.16: RelativeI C A effects by groups ofvariables on average log-realwages; by size ................................ 37 Figure 2.17: Relative ICA effects by groups of variables on the probability of exporting; b y size Figure2.18: Relative ICA effects by groups ofvariables on the probability of rece Figure 2.19: Managers' perceptions; percentage of firms that considers each one of the a severe obstacle to firms' economic performance ................................................. This report was preparedby Stefka Slavova andJorge Peiia. ii Section 1. METHODOLOGICAL NOTES 1. REPLACEMENTSTRATEGY FORMISSINGPRODUCTIONFUNCTIONVARIABLES The sample size o f the Guatemala Enterprise Survey used inproductivity regressions (see Appendix 2) i s augmented by approximately 20% through replacing missing information on sales/workers/capital/materials with the corresponding location-sector-size averages (Tables 1.6 and 1.7) in Appendix 2. Such replacement i s problematic in cross-sectional data as it could be biasing the estimates. Incomplete data i s a common problem that standard econometric and statistical methods have nothing to say about or how to solve it, and constitutes a continuous source o f problems for researchers. The problem of having too many missing values in a dataset may bias the representativeness of the data, causes losses o f efficiency in regression analysis and, like inthe case o f Enterprise Surveys, implies losing a large number of very expensive interviews, both inpecuniary and time terms. The Guatemala ICA is not an exception. Table 1.1 shows the distribution of the missing values by industry and size. From this table it i s clear that for instance in the food sector we would have lost 26.3% o f observations for small firms, while after replacing missing values, we only lose 5%. This problem i s common to all sectors and sizes, and it i s clear that the pattern o f missing values does not follow a clear correlation by size of firms or industries; all the industries and sizes suffer from this problem. Table 1.1: Observations available for regression analysis inGuatemala by Industry and size and percentageof observations lost Total Without replacing 116 17.1 93 21.8 53 23.2 262 20.1 With replacing 136 2.9 117 1.7 65 5.8 318 3.0 1 The same can be said about the pattern of missing values by industries and regions o f Table 1.2. Again, the pattern of missing values i s uncorrelated with the distribution o f industries and regions. Both firms located in Guatemala City and firms located inthe rest o f the country refusedto answer the productivity questions of the I C survey. For instance we lost 21% o f `food' firms located in Guatemala City and 21.4% of firms of the same sector located in the rest o f the country, while after replacing missing observations we only lost 4.8% o f the food firms in Guatemala City and 3.6% inthe rest o f the country. Table 1.2: Observations available for regression analysis inGuatemala by Industry There are a number o f approaches to deal with the problem of missing data. These approaches may be grouped into two different families o f methods: maximum likelihood and multiple imputation, see Allison (2001) and Little and Rubin (1987) for a review. The objective of these methods i s not to augment the sample size, but to maintain the sample representativeness and to gain efficiency inthe estimation. Our method of imputing missing data, which we call ZCA method, shares the expectation step of the Expectation-Maximization (EM) algorithm proposed in the seminal paper o f Dempster, Laird and Rubin (1977), method that, within the maximum likelihood approaches, has been widely applied in several scientific fields, see McLachlan and Krishnan (1997). In particular, the replacement strategy used in the ICA of Guatemala departs from the expectation o f the production function variables conditional on the industry, region and size the corresponding observation belongs to, in other words we replace the missing value with the expectation o f the distribution of the variable, conditional on the information on industry, region and size according to the equation: E(JiIDT,i 7OI,i7DS,i=PO +P T ,J DT,i+PT,JOI,i +PT ,.I DS,i = L, ` 9 7 (1) Y, L,M and K represent output, labor, materials and capital and DT,DI and Ds are time, industry and size dummies respectively. Estimated values to replace incomplete data are given by 2 The ICA method has the advantage o f imputingmissing data without a population model, which i s the main advantage over EM algorithm.' Imputation of missing data without a population model i s the main characteristic of the second family o f approaches: the multiple imputation methods. Our strategy is, in fact, a general multiple imputation method in which we assume that each imputed variable can be represented as a linear function of other variables (dummies of industry, region and size) and therefore the fitted values can be used to replace missing data. The second condition that needs to hold for the multiple imputation method to work well i s that all the variables, including those replaced and those used to replace them, have normal distributions. Although these are strong assumptions the multiple imputation method seems to work well even when the variables have distributions that are manifestly not normal, see Schafer (1997). Therefore, under these assumptions our method leads to a consistent estimation o f the ICAparameters, but at the same time it can be argued that a more efficient method can be used. Notice that by imputing missing values we are modifying the population distribution of replaced variables. In particular, if the two conditions mentioned in the previous paragraph hold, the sample average of the modified distribution of the variable converges to the population expectation. Unfortunately, this does not hold for the case of the standard deviation. With the replacement strategy we are reducing the variability of the distribution and therefore any statistical inference will be based in downward-biased standard errors. To correct for this, a plausible and elegant solution is to do a re-sampling a given number of times and to replace the missingdata ineach sample so we can obtain a distribution of the estimators of interest under different replacements o f the missing data, and thus use the bootstrap standard errors to do statistical inference. As Escribano et al. (2008) show, the replacing strategy proposed introduces enough variability in the distribution which allows making correct statistical inferences. The variability comes from the number o f industriesused multiplied by the number of regions and sizes. When the two assumptions mentioned above do not hold, our replacement strategy i s no longer consistent. Very little can be said about the asymptotic distributions o f the estimators obtained under such circumstances. In general, in these cases we can treat our replaced variables as variables measured with error. Thus, the parameters obtained from the regression analysis would be consequently downward biased, and the magnitude of the bias will depend on the standard deviation of the error term relative to the standard deviation of the variable and the proportion of replaced values. Another question that needs to be taken into account i s the nature of the mechanism that generates the missing data. Our replacement strategy leads to unbiased results when the pattern of missing values i s completely missing at random (CMAR), missing at random (MAR) (no relation with the population model), and inthe exogenous sampling selection, in which the pattern of missing values depends on the explanatory variables of the 1The EM algorithm imputes missingdata conditional o n a given population model, and therefore chooses the candidate values to replace the missing cells that maximize the likelihood fmction conditional o n a vector of parameters of that model. 3 population model. To see it, let us suppose the next extended production function as in Escribano and Guasch (2005 and 2008): yi, =a,+aLlj, +aMmj,+aKki,+P,ic,,, +P2i~2,it +...+P,ic,,, +qc,,, +62~2,it+...+6jcj,it+ui, (3) where y, 1, m and k represents output, labor, materials and capital all in logs, ic are the investment climate variables and c are other control variables. Let the pattern o f missing values for each observation i at moment t be given by sit, where sir=O if missing value and 1otherwise. So what we observe is: Ifthe pattern of missingvalues isM.A.R,then the necessary conditions for equation (4) to be identified are E(s,u,,)=0 E[(si,Jit)(si,uil)] E[(sitJituil)] 0 = = J =l,m,k,ic,,...,ic,,c, ,...,ck Andinthe case of exogenoussample selection we need that That is, for the identification condition in this case to hold we need to control for any exogenous variable affecting the pattern o f missing values, and this i s the way we shall proceed in the estimation o f the productivity equations. Note that in these cases the complete case (deletion of observations with any missing value) also leads to unbiased I C A parameters, although at the cost o f losing efficiency and in some cases the representativeness of the original sampling frame. If the pattern of missing values is endogenously determined (it is correlated with output ('y)in equation (4)) and there is self-selection in our model, the replacing strategy may lead to inconsistent estimates. Inthese cases one has to implement the Heckman modelto correct for self-selection, since OLS applied either on the complete case or on the sample with replacement estimates biasedparameters. Table 1.3 offers a review o f the different methods that can be used to estimate the parameters of equation (4). The first column uses our method or the ICA method, the second column i s the same method with bootstrap standard errors with 1500 replications. The third column i s the complete case and the last column i s the Heckman model. There are no great differences in the standard errors between column 1 and 2, which supports the idea that the ICA method can be used to make inferences .when there is enough variability by industry, region and size. The complete case of column 3 leads to slightly 4 different production function and ICA parameters, what poses some doubts about the representativeness o f the complete case, notice that in this column the significance of IC variables i s generally reduced, although there are no significant changes in the magnitude o f the parameters. Finally, Heckman's Lambda in the last column i s not significant, indicating that the selection model does not make sense. Anyway, even in this case the Heckman model leadto similar results as the ICA method. Table 1.3: Extendedproductionfunction of Guatemala ard Complete Heckman case selection model [0.014] [0.002] i0.002 -0.097 -0.084 -0.06! [0.050] rO.0651 [0.054] r0.054 ter from public sources (b) 0.002** 0.002 0.002*** 0.003**: [0.001] [0.002] [0.001] [0.001 -O.OlO*** -0.010 -0.011*** -0.011*: [0.006] [0.007] [0.006] [0.006 -0.073 -0.073 -0.179 -0.16~ [0.116] rO.1161 [0.128] [0.125 0.208 0.208*** 0.262" 0.25) [O. 1511 [0.049] [0.155] [0.176 0.524*** 0.524*** 0.386*** 0.405 ** : [O. 1241 [0.053] [0.109] [0.098 -0.097 -0.097 -0.046 -0.03~ [0.0791 [0.151] [0.084] [0.086 nager's time in bureaucratic issues (a) -0.026** -0.026*** -0.017' -0.019*' rking capital financed by informal sources ngcapital financed by non-bank 0.004 0.004 0.005** 0.001 cia1institutions [0.003] [0.021] [0.003] [0.006 mmy for checking or saving account 0.166 0.166 0.142 0.16: [0.102] [0.115] [0.119] [0.118 mmy for credit line 0.161* 0.161*** 0.126 O . l l r [0.057] 10.0821 [0.085 5 I C A method. Bootstrap standard Complete Heckman ICAmethod errors, 1500 case selection model repetitions. Dummyfor I S 0 Certification (b) 0.297' 0.297* 0.499*** 0.422*** [0.176] [0.160] [0.182] [0.154] Percentage of female workers instaff (b) -0.002 -0.002* -0.002 -0.002 [0.001] [0.0011 [0.0021 [0.002] Training to non-production workers (b) 0.002 0.002** 0.002 0.002* [0.001] [0.001] [O.OOl] [0.001] Share of importedinputs (b) 0.003** 0.003 0.002* 0.003** [O.OOl] [0.0021 [0.0011 [0.001] Percentage of unionized workforce -0.021** -0.021*** -0.019* -0.015 [0.009] [0.007] [0.011] [0.015] Dummyfor FDI 0.25 0.250*** 0.111 0.084 [0.190] [0.024] [0.196] [0.175] Share of exports 0.006** 0.006*** 0.006** 0.006*** [0.002] [0.002] [0.002] [0.002] Dummy for large firm 0.446*** 0.446*** 0.441** 0.447*** [0.169] [0.035] [0.175] [0.161] Constant 1.845** 1.845 1.346 1.441*** [0.928] [0.986] [0.753] Observations 318 318 263 277 R-squared 0.891 0.891 0.91 Heckman's Lambda (inverse of Mill'sratio) 0.46C [0.408] 2. ENDOGENEITY PROBLEMS Our regressions of TFP in IC variables might suffer from endogneity problems, i.e. the use o f explanatory variables which might be correlated with the error term and thus result in biased and inconsistent estimates. Some of the literature emphasizes the potential endogeneity of access to finance variables to firm productivity and other measures o f firmperformance. Endogeneity i s yet an unresolved issue in econometrics. In fact it i s difficult to test whether a variable i s endogenous or not, and too often one has to rely in economic intuition and to make aprioristic assumptions on the plausible exogeneity o f the variables. Moreover, in the context of IC data it i s difficult to use Hausman-type tests to check for the presence of endogeneity given the low power o f these tests and the highrejection rate of the nullhypotheses. A possible solution to the endogeneity of IC variables i s to use the industry-region-size averages. While this endogeneity correction has been proven to work well when there are no industry-region-size (I-R-S) processes correlated with the error term, sometimes it i s difficult to get good instruments o f crude plant-level IC variables. Some o f the explanatory variables in table 2.11 in Appendix 2 are in this form. For other variables their I-R-S average i s not a good instrument and we have to rely on the crude variable. Note that excluding these variables from the regression, although they were endogenous, 6 i s not a good solution provided we are modeling the expectation o f TFP on the whole investment climate firms are facing and the exclusion o f any relevant variable may result in a omitted variables problem which is another serious problem that causes biases inthe rest of the parameters o f the model. Our models should be interpreted in terms of conditional expectations. What we are doing i s modeling the conditional expectation o f productivity given all the firm-level information we have on the investment climate which has an effect on firms' performance and efficiency. This means that the coefficients o f IC variables cannot be interpreted in causal terms; the term reverse causality does not make sense in our model, the data lack the optimal properties to do Granger-causality inference. Rather, the coefficients should be interpreted as marginal effects on the conditional expectation o f productivity. Whether this effect i s driven by a simultaneous effect i s something we a priori do not know and a more thorough analysis should be made to provide more insights into this. A possible solution would be to model TFP and finance in a simultaneous model and estimate it by 2SLS or 3SLS techniques. Nevertheless this solution should imply an enormous system o f simultaneous equations. We prefer our model for its parsimony and simplicity. Therefore, even though it could be true that the effect o f finance variables may be driven by simultaneous forces, the relative contribution on productivity of this group of variables holds interms of conditional expectations although it does not do so in terms of causal relations. 7 Section 2. PRODUCTIVITYRESULTS Table 2.1: GeneralInformation at Plant Leveland Production FunctionVariables' General Industrial classification (a) Food; (b) Apparel and Textiles; (c) Other manufacturing. Informationat PlantLevel Regional classification (a) Guatemala City; (b) Rest of the country ~ ~~ Production Sales Used as the measure of output for the production function Function estimation. Sales are defined as total annual sales in 2005. The Variables series are deflated by using the Producer Price Index (PPI), base 2002. Employment Total number of permanent and temporary workers. Total hours worked per Total number of employees multiplied by the average hours year worked per year. Materials Total costs of intermediate and raw materials used inproduction (excluding fuel). The series are deflated using the Producer Price Index (PPI), base 2002. Capital stock Net book value of machinery and equipment. The series are deflated using the Producer Price Index (PPI), base 2002. User cost of capital The user cost of capital is defined interms of the opportunity I costmachinery of using capital; it is defined as 15% of the net book value of and equipment. Labor cost I Total expenditures on personnel. The series are deflated using the Producer Price Index (PPI), base 2002. Dependent Exports Dummy variable that takes the value of 1if exports are greater Variables in than 10%. Regression Foreign Direct Dummy variable that takes the value of 1if any part of the Equationsand Investment capital of the firm i s foreign. Linear Wages Real wage is defined as the total expenditure on personnel Probability (deflated by the Producer Price Index, base 2002) divided by the Models total number of permanent and temporary workers. Employment Total number of permanent and temporary workers. 8 Table 2.2: Investment climate (IC) and control (C)variables Descriptionof the variable 9 Table 2.3: Investment climate (IC) and control (C) variables (continued) Blocksof IC Variable name Descriptionof the variable Variables Redtape, Sales reported for tax Percentageof total annual sales that a typical firmoperating inplant's sector reports corruptionand purposes for tax purposes. crime Workforce reported for tax Percentageof total workforce that atypical firm operating in plant's sector reports purposes for tax purposes. Dummy for conflicts with Dummy taking the value of 1if the plant has conflicts with clients with a third party clients involved. Dummy for conflicts in Dummy taking the value of 1ifthe plant has conflicts with clients with a court courts involved (conditional on having conflicts with clients with a thirdparty involved). Weeks to judgment Number of weeks that took the court to deliverjudgment inthe latest conflict with clients (conditionalon having conflicts with clients with a third party involved). Dummy for security Dummy taking the value of 1if the plant has security expenses. Security cost Security expenses as apercentage of total annual sales. Dummy for crime Dummy for gifts to obtain a Giftsexpected or requestedto obtain aconstruction permit,conditional on applying construction permit for a construction permit. Wait for an operating license Days to obtain a main operating license (conditional on applying for an operating license). Dummy for gifts for Giftsexpected or requestedto obtain anoperating license, conditional on applying operating license for an operating license. 10 Table 2.4: Investment climate (IC) and control (C) variables Blocks of IC variables Variable name Descriptionof the variable Finance and Largest shareholder Percentage of firm's capital owned by the largest shareholder. corporate Initialinvestment: private Percentage of the investment needed to start operations received from governance banks private commercial banks. Initial investment: public banks Percentage of the investment needed to start operations received from state-owned banks and/or government agencies. Purchases paidbefore delivery IIPercentage of annual purchases paid for before delivery. IPurchases paid on delivery I Percentage of annual purchases paid for on delivery. I Purchases paid after delivery Percentage of annual purchases paid for after delivery. Sales paidbefore delivery Percentage of annual sales paid for before delivery. Sales paid on delivery Percentage of annual sales paid for on delivery. Sales paid after delivery Percentage of annual sales paid for after delivery. Working capital financed by Percentage of firm's working capital financed with internal funds. internal funds Working capital financed by Percentage of firm's working capital financed by private commercial private banks banks. Working capital financed by Percentage of firm's working capital financed by state-owned banks. state-owned banks Working capital financed by Percentage o f firm's working capital financed by family/friends. family/friends Working capital financed by Percentage of firm's working capital financed by non-banking financial non-bank financial institutions institutions. Working capital financed by Percentage of firm's working capital financed with credit from suppliers. credit from suppliers Working capital financed by Percentage of firm's working capital financed by informal sources. informal sources New fixed assets financed by Percentage of investments in new fixed assets financed with internal internal funds funds. New fixed assets financed by Percentage of investments in new fixed assets financed by private private banks commercial banks. New fixed assets financed by Percentage of investments in new fixed assets financed by state-owned state-owned banks banks. New fixed assets financed by Percentage of investments in new fixed assets financed by family/friends. - familylfriends New fixed assets financed by Percentage of investments innew fixed assets financed by non-banking non-bank financial institutions financial institutions. New fixed assets financed by Percentage of investments in new fixed assets financed with credit from credit from suppliers L suppliers. New fixed assets financed by Percentage of investments in new fixed assets financed by informal informal sources sources. Checking or savings account Dummy taking the value of 1if the plant has a checking or savings account. Owner of the land Percentage of the land on which the plant operates owned by the firm. Dummy for credit line Dummy that takes the value of 1if the firm has access to a credit line or overdraft facility. Dummy for loan Dummy that takes the value of 1 if the firm has access to a loan. Dummy for loan with collateral Dummy that takes the value of 1 if the firm has access to a loan which requires collateral (conditional on having a loan). Value of collateral Value of collateral as a Dercentaee of the loan value (conditional on - having a loan with collateral). Dummy for debt Dummy taking the value of 1if the number of rejected loan applications i s larger than the number of applications for a loan. Dummyno loanbecause Dummy that takes the value of 1 if the firm did not apply for a loan complexity because of complexity. 11 Blocks of IC variables Variable name Descriptionof the variable Dummy no loanbecause cost Dummy that takes the value of 1if the firm did not apply for a loan because of its cost. Dummy no loanbecause Dummy that takes the value of 1if the firm did not apply for a loan collateral because of the collateral required. Rejected credit applications Percentage of rejected credit applications. Accepted credit applications Percentage of accepted credit applications. External audit Dummy that takes the value of 1if the firm has its annual financial statements externally audited. 12 Table 2.5: Investment climate (IC) and control (C) variables Blocks of IC variables Variable name Descriptionof the variable Quality, Dummy for quality Dummy taking the value of 1ifthe firm has any kind of quality certification. innovationand certification labor skills Dummy for foreign Dummy taking the value of 1if the plant uses technology licensed from a technology foreign-owned company. Dummy for product Dummy taking the value of 1if the plant has introduced any product innovation innovation inthe last 3 years. Dummy for process Dummytaking the value of 1if the planthas introduced any productionprocess innovation improvement in the last 3 years. Outsourcing Percentage of total annual sales subcontracted. Dummy for R&D Dummy that takes the value of 1if the firm performed R&D activities during the last year. R&Dexpenditures R&Dexpenditures as a percentage of total annual sales. Staff - production workers Percentage of production workers out of all staff. Staff - female workers Percentage of female workers out of all staff. Staff - skilled workers Percentage of skilled production workers out of all staff. Staff -university education Dummy taking the value of 1if the typical production worker has at least one year of university education. Dummy for training Dummy taking the value of 1if the firm provides formal (beyond on-the-job) training to its employees. Trainingto production Percentage of production workers receiving formal (beyond on-the-job) training workers Training to non-production Percentage of non-productionworkers receiving formal (beyond on-the-job) IManager experience I Manager experience in years. Other control Age Age of the firm in 2005. variables Capacity utilization Percentage of total firm productive capacity utilized. Trade union Percentage of workforce unionized Dummy for incorporated Dummy that takes the value of 1ifthe firm i s an incorporated company. company Dummy for limited liability I Dummy that takes the value of 1if the firm i s a limited liability company. company Dummy for FDI Dummy that takes the value of 1if any part of the firm's capital i s foreign. Dummy for public capital Dummy that takes the value of 1if any part of the firm's capital i s public. Exporting experience Number of years of exporting experience. Dummy for local monopoly IIDummy takingthe value one of 1if the firm i s a local monopoly. Dummy 5 or more Dummy taking the value of 1if the plant has 5 or more competitors in the local competitors market. Dummy less than 5 Dummy taking the value of 1if the plant has less than 5 competitors in the local competitors market. Dummy increased sales Dummy taking the value of 1if the plant has increased its sales. Dummy decreased sales Dummy taking the value of 1if the plant has decreased its sales. Dummyfor importer Dummy takingthe value of 1if the firm imports more than 10%of the total purchases of intermediate materials. Share of imports Share of imported inputs over total purchases of intermediate materials and Dummy for exporter Dummy taking the value of 1ifthe firm exports more than 10% of its total annual sales. Share of exports Share o f exports intotal annual sales. Small Dummy taking the value of 1if the firm has less than 20 employees. Medium Dummy taking the value of 1if the firm has 20 or more, but less than 100 13 Table 2.6: Total number of observations before and after cleaning missingvalues and outliers inproduction function (PF) variables Observationsbefore Observationsafter c1eaning c1eaning Missingobservations 61 4 Of which: firms with one PF variable missing 33 0 firms with two PFvariables missing 21 0 firms with three PFvariables missing 3 0 firms with four PFvariables missing 4 4 Outliers 5 6 of which: outliers only in materials (ratio of materials to sales > 1) 5 6 outliers only in labor cost (ratio of labor cost to sales > 1) 0 0 outliers in both materials and labor cost 0 0 Usefulobservations (outliersandmissing excluded) 262 318 The cleaningprocess i s performedinthree steps. I. Those firms with missingvalues in all the PFvariables (sales, materials, labor cost andcapital) are dropped from the sample.For the rest of the missingvalues we applythe proceduredescribedinI1and111. II. We replacethose observationswith ratios of materials to sales or of laborcost to sales greater than one (outliers)following step 111. Ill. We replacethe missingvalues of the PF variablesby their correspondingindustry-region-sizemedians. Ifwe do not haveenoughobservationsinsome cells, we replacethemby the correspondingindustry-sizemedians.Ifwe still do not haveenoughobservationsinthose cells, inthe next step we replacethe missingvaluesby the region-size medians.Ifstill necessary,inthe last step we computethe mediansonly by size andor by industryto replacethose missingvalues. The last row of the table summarizesthe number of useful observationsfor regressionanalysis beforeand after the cleaningprocess. 14 Table 2.7: Representativeness o f production function variables before and after cleaning for missingvalues and outliers By industry and region Industry Guatemala City Rest of the country Total #Obs Perc. #Obs Perc. #Obs Perc. Food 62 24.3 28 38.4 90 27.4 59 24.0 27 37.5 86 27.0 By industry and size I Industry I Medium I Large- I Total I #Obs Perc. #Obs Perc. #Obs Perc. pefore cleaning 38 27.1 27 22.7 25 36.2 90 27.4 I hfter cleaning II36 II26.5 I 261 I 22.2 I 24 I 36.9 I 86 I 27.0 I 34 I 28.6 16 23.2 83 I 25.3 33 I 28.2 16 24.6 82 I 25.8 Other Before cleaning 69 49.3 58 48.7 28 40.6 155 47.3 manufacturing After cleaning 67 49.3 58 49.6 25 38.5 150 47.2 Total Before cleaning 140 100.0 119 100.0 69 100.0 328 100.0 After cleaning 136 100.0 117 100.0 65 100.0 318 100.0 15 Table 2.8: Total number of observations and response rate of I C and C variables in the original sample Blocks of I C 1 I I variables Name of the variable #Observations Responserate Infrastructure Red tape, corruption and crime 16 Blocks of IC variables Name of the variable #Observations Response rate Finance and Largest shareholder --" 376 99 A ,,. . corporate Initial investment: privatebanks governance 324 98.8 Initial investment: public banks 325 99.1 Purchasespaid before delivery 328 100.0 Purchasespaid on delivery 338 innn Quality, Dummy for quality certification 317 96.6 innovationand Dummy for foreign technology labor skills 312 95.1 Dummy for product innovation 312 95.1 Dummy for process innovation 312 95.1 Outsourcing 2737 aa -I Dummy for R&D 312 95.1 R&D expenditures in5 37 n Staff -productionworkers I 312 I 95.1 Staff- female workers 311 94.8 Staff- skilled workers 312 95.1 Staff - university education ?in 94 5 17 Blocks of I C variables Name of the variable #Observations Other control variables 18 Table 2.9: Correlationmatrix among productivity measures 1 TWO steps solow - 1 Single step - Single step - residual Restricted Restricted Unrestricted Dou `Obb- las I Unrestricted Translog `Obb- 1 Doug1as Translog Single step - Cobb-Douglas 0.89 0.92 1 Restricted Translog 0.91 0.93 0.97 1 Single step - Cobb-Douglas 0.08 0.04 -0.22 -0.17 1 Unrestricted ~~~~~l~~ 0.11 0.09 -0.15 -0.11 0.80 1 b) EstimatedProductivityinlogs is obtainedfrom Cobb-Douglas andTranslogproductionfunctions of sales with inputs labor, materials,andcapitalestimatedby OLS under two differentenvironments: (1) Restricted: a singleset of productionfunction coefficientsis obtainedusingdataonplants, for all industries (excludingoutliers). (2) Unrestrictedby industry:a set of productionfunction coefficientsi s obtained for each one of eight industriesusing data on all plants (excludingoutliers). 19 Table 2.11: ICA elasticities and semi-elasticities with respect to productivity, robust White standard errors Blocksof ICA ExplanatoryICA variables governance * significantat 10%; ** significantat 5%; *** significantat 1%; Eachregressionincludesa set of industrydummies and aconstant term. (a) Variables instrumented with the industry-region-sizeaverage. (b) Variables approximated with a proxy (only missingvalues replaced by the industry-region-sizeaverage). Estimation of IC elasticities and semi-elasticities: A) Two steps: In the first step compute the Solow residual with restricted (or unrestricted)input-output elasticities. In the second step compute the IC coefficientsby regressingthe set of IC variables on the Solowresidualby OLS. B) Single step: i)Cobb-Douglas: ComputetheICcoefficientsjointly with therestricted(orunrestricted)input-outputelasticities byOLSinanextended Cobb-Douglas productionfunctionof the form: log Y, = a,log L, t a,w M , a x l o g K , log + + a',cIci+ a;c,+(Y;Jl, + ap+ u, where L, M and K denote the inputs employment, materials and capital respectively,IC and C denote the investmentclimate and other control variables andDj, denotes the industrydummies. ii)Translog: Compute the IC coefficientsjointly with the restricted(or unrestricted) input-outputelasticitiesby OLS in an extended Translogproductionfunction of the form: lQ3Y=qlQ34+4%Y+%1%4 %k4)2t,%kY,'t,%oopgi +~~(logq)oC9Y)+%(4~Q34)+~k~~Q3~)+~~~ 1 1 1 +2 +&++$D, +4+Y Restricted estimation: equal input-outputelasticities for all firms inthe country. Unrestricted by industry estimation: equalinput-outputelasticities for all firms in the same industry. 21 Dummyfor largefirm 0.117 0.106 0.446** 0.462** 0.419* 0.308** Observations 318 318 318 318 318 318 R-squared 0.273 0.257 0.891 0.897 0.912 0.925 Note: We employ a general correction for heteroskedasticity: the robust White standard errors (Table 1.10). This method corrects for any general pattern of heteroskedasticity in the residuals of the productivity regressions. In fact, clustering standard errors i s a particular case of the White standard errors which allow for common variation within clusters. Table 2.11 shows the estimation results with `White' robust standard errors and Table 2.12 shows the results with standard errors clustered by industry, region and size. It i s clear that the significance of the ICA variable coefficients does not change significantly if we cluster the standard errors. 22 Table 2.13: Percentage Contribution of IC and C Variables to the Olley and Pakes Decompositionof the Aggregate Productivity inLogs; Restricted Solow Residual Infrastructure Red tape, corruption and crime Dummy for security (b) 0.36 0.23 0.13 Dummy for crime (b) -0.06 -0.06 -0.01 Manager's time in bureaucratic issues (a) -0.21 -0.21 0.00 Payments to speed up bureaucracy (b) 0.00 -0.01 0.01 corporate Working capital financed by informal sources 0.00 -0.01 0.01 governance ~ T - . . L : ^ - ,,..:*,1 C,"..,,A l.., L,.-lr c^^^^:"1 I^- I I I - Dummyfor credit line 0.12 0.07 0.05 Quality, Dummyfor I S 0 Certification (b) 0.16 0.05 0.11 _ I innovation Percentage o f female workers in staff (b) -0.04 -0.05 0.01 and labor Training to non-production workers (b) I skills 0.08 0.08 0.00 Other control Share o f imported inputs (b) 0.09 0.06 0.03 variables Percentage o f unionized workforce -0.01 0.00 -0.01 Dummyfor FDI 0.08 0.02 0.06 Share of exports 0.30 0.13 0.17 Dummyfor large firm 0.10 0.02 0.08 Industry Apparels and textiles -0.01 -0.02 0.01 dummies Other manufacturing -0.08 -0.07 -0.01 Constant 0.72 0.72 0.00 Residual 0.14 0.00 0.14 Total 2.33 1.65 0.67 Notes: ~~ ~~ * Resultspresentedare ** The relative to aggregate productivity (mixed: share of sales inlevels andproductivity in logs). productivity measure used to construct the tables i s the restricted Solow residualandhence the elasticities and semi-elasticities usedare obtainedfrom the first column of Table 2.11. *** Each term of the Olley and Pakes decomposition of aggregate productivity (PJ can be expressed in terms of the investment climate variables according to the following expression: P. = & ' , I C j + f ~ ' , C ~ + & ' ~ +&, 6+~i Ji + N j d ' , ~ ~ o v ( s j , i , I C j , i ) + N J f L ' c ~ o v ( s j j , C , , ~ ) i N j & ' D ~ ~ o v ( s j , i , D j ) + N j ~ o v ( s j , i , , ~ J , i ) - I ~ 5 IC is the vector of investment climate variables, C the vector of other control variables, Djthe vector of industry dummies, ui and up are the residual and the constant of the productivity regression of the restricted Solow residuals on the IC and C variables (see Escribano et al. (2007) for details). Table 2.13 presents the percentage contributions to aggregate productivity (in logs). 23 Table 2.14: IC impact on the decomposition by inputs of the efficiency term of the Olley and Pakes decomposition inlogs Share of exports 5.66 6.53 1.02 7.16 20.37 Dummy for large firm 2.72 3.99 0.64 1.18 8.52 Industry Apparels and textiles 0.02 0.58 0.15 -0.18 0.57 dummies Other manufacturing 0.55 -0.54 -0.09 0.13 0.06 Residual -3.68 -36.12 -1.04 71.07 30.23 Total 12.71 -13.18 2.70 97.77 100.00 Where I C i s the vector of investment climate variables, C the vector of other control variables, D,the vector of industry dummies, utand are respectively the residual and the constant of the productivity regression on the I C and C variables and using the restricted Solow residual as dependent variable (see Escribano et. al. (2007) for details). Table 2.14 presents the percentage contributions to efficiency which are obtained from the next equation: 24 Table 2.15: Two-stage least squares (2SLS) estimation of employment equation Blocksof ICA % Contribution % variables Explanatory ICA variables Coefficient Contribution Coefficient Productivity -0.377* -27.31 -0.369' -25.50 Real wages Infrastructure Red tape, corruption and crime Financeand corporate governance Quality, innovationand labor skills Other control variables Trade union 10.049** 10.34 I0.049*** 10.46 I Dummy for FDI 0.692** 1.27 pi 0.681** 1.47 Share of exports 0.005 ** 1.90 0.005* 2.01 Instruments I First stage R-sauared 10.29 I evaluation III Hansen Partial R-souared U-value of uartial R-sauared 10.00 0.00 test (U-value) 10.61 1II 10.6 I I I Observations I316 I I316 I I Notes: * significantat 10%; ** significantat 5%; *** significantat 1% (robust standard errors). Each regression includes a set of industry and year dummies and a constant term. (a) Variables instrumented with the industry-region-size average. (b) Variables approximated with a proxy (only missingvalues replacedby the industry-region-size average). Productivity i s endogenous and the list of variables used as excluded instruments is: days to clear customs to export, dummy for gifts in inspections, electricity from generator, shipment losses in exports (Industry-Region-Size average), dummy for conflicts with clients, dummy for crime, working capital financed by informal sources, working capital financed by non-banking financial institutions, dummy for checking or current account, dummy for credit line, staff - female workers, training to nonproduction workers. The percentage contributions of productivity, IC and C variables to average log employment are computed according to the next expression: IC is the vector of investment climate variables, C the vector of other control variables, Dj the vector of industry dummies, vL and pL are respectively the residual and the constant of the real wages 2SLS regression. First stage R-squared from the regression of productivity on both the included and the excluded instruments. The partial R-squared measuresthe squared partial correlation between the excluded instruments and productivity. F-test ofjoint significance of the excluded instruments that corresponds to the partial R-squared. The Hansen test i s a test of overidentifying restrictions. The nullhypothesis i s that the instruments are valid instruments, that is, uncorrelated with the error term, and therefore the excluded instruments are correctly excluded from the estimated equation. 25 Table 2.16: Two stage least squares (2SLS) estimation of realwages equation Blocks of I C A % % variables Explanatory I C A variables Coefficient Contribution Contribution Productivity 0.608** 13.75 0.530* 11.92 Infrastructure Losses due to power outages (b) -0.023* -0.72 -0.023* -0.72 Red tape, corruption Sales reportedto taxes (a) 0.002* 2.49 0.002* 2.05 ~ and crime Paymentsto obtain a contract with the -0.036* -0.89 -0.035 -0.87 evaluation I Hansentest (U-value) 10.93 I 10.93 I I I Observations I316 I I316 I I Notes: * significantat 10%; ** significant at5%; *** significantat 1% (robust standarderrors). Eachregressionincludesa set of industryandyear dummies andaconstant term. (a)Variables instrumentedwith the industry-region-sizeaverage. (b) Variablesapproximatedwith aproxy (only missingvaluesreplacedby the industry-region-sizeaverage). Productivity is endogenous andthe list of variables usedas excluded instruments is: days to clear customsto export, electricity from generator, water from public sources, dummy for security, payments to speed up bureaucracy, working capital financed by non- bankingfinancial institutions. * The percentage contributions of productivity, IC and C variables to average log real wages are computed according to the next expression: IC is the vector of investmentclimate variables, C the vector of other control variables, D, the vector of industrydummies, vw and are respectivelythe residualandthe constantof the real wages 2SLS regression. First stageR-squaredfrom the regressionof productivity onboththe includedandthe excludedinstruments. The partial R-squaredmeasures the squaredpartial correlationbetweenthe excludedinstruments andthe productivity. F-test ofjoint significanceof the excludedinstrumentsthat correspondsto the partial R-squared. The Hansen test i s a test of overidentifying restrictions. The null hypothesis i s that the instruments are valid instruments, that is, uncorrelated with the error term, andtherefore the excludedinstruments are correctly excluded from the estimatedequation. 26 Table 2.17: Two stage least squares (2SLS) estimation of probability of exporting equation Hansen test (p-value) Notes: * significant at 10%; ** significantat 5%; *** significant at 1%(robust standard errors). Each regression includes a set of industry dummies and a constant term. (a) Variables instrumented with the industry-region-size average. (b) Variables approximated with aproxy (only missing values replaced by the industry-region-size average). Productivity i s endogenous and the list of variables used as excluded instruments is: Security expenses, crime losses, initial investment: private banks, dummy for process innovation, staff- skilled workers, trade union. The percentage contributions of productivity, IC and C variables to the probability of exporting are computed according to the next expression: IC is the vector of investment climate variables, C the vector of other control variables, Djthe vector of industry dummies, vExpand 6Expare respectively the residual and the constant of the probability of exporting 2SLS regression. First stageR-squared from the regression of productivityon boththe included and the excluded instruments. The partial R-squared measures the squared partial correlation between the excluded instruments andthe productivity. F-test ofjoint significance of the excluded instruments that corresponds to the partial R-squared. The Hansen test i s a test of overidentifying restrictions. The nullhypothesis i s that the instruments are valid instruments, that is, uncorrelated with the error term, and therefore the excluded instruments are correctly excluded from the estimated equation. 21 Table 2.18: Two stage least squares (2SLS) estimation of probability of exporting I equation Of IC* % % variables ExplanatoryICA variables Coefficient Contribution Coefficient Contribution Productivity 0.106* 357.26 0.11* 367.63 Infrastructure Dummyfor own generator (b) 0.086" 21.43 0.087* 21.60 Wait for a water sumh (b) -0.013* -10.44 -0.013" -10.15 Redtape, corruption andcrime corporate innovation andlabor skills Other control variables Instruments First stage R-squared 0.19 10.18 I Partial R-squared 0.07 I Hansen D-value of partial R-sauared 0.00 HI test (p-value) 10.95 I 10.94 I I Observations I318 I I318 I I Notes: * significantat 10%; ** significantat 5%; *** significantat 1%(robuststandarderrors). Eachregressionincludes a set of industrydummies and a constantterm. (a)Variables instrumentedwith the industry-region-sizeaverage. (b) Variables approximatedwith a proxy (only missingvalues replacedby the industry-region-sizeaverage). Productivityis endogenousandthe list of variables usedas excludedinstruments is: dummy for gifts ininspections, electricity from generator, water from public sources, sales reportedto taxes (I-R-S avg.), working capitalfinanced by informal sources, dummy for I S 0 certification,staff- female workers, training to non-productionworkers, share of importedinputs. * The percentagecontributions of productivity, IC andC variablesto the probability of exportingare computed accordingto the next expression: 3xprespectivelythe IC is the vector of investmentclimatevariables, C the vector of other control variables, DJthe vector of industrydummies, vExpand are residualandthe constant of the probability of exporting2SLS regression. First stageR-squaredfrom the regressionof productivity on boththe includedand the excludedinstruments. The partialR-squaredmeasuresthe squaredpartial correlationbetweenthe excluded instruments andthe productivity. F-testofjoint significance of the excludedinstrumentsthat correspondsto the partialR-squared. The Hansentest i s a test of overidentifyingrestrictions. The nullhypothesis i s that the instruments are valid instruments, that is, uncorrelatedwith the error term, andtherefore the excludedinstruments are correctlyexcluded from the estimatedequation. 28 Figure 2.1: Olley and Pakes DecompositioninLevels by Industry and Region of Aggregate Productivity (Restricted Solow Residual) 18.0 160 140 120 100 8 0 6 0 4.0 2.0 0 0 Figure 2.2: Olley and Pakes Decomposition inLevels by Size and Age ofAggregate Productivity (Restricted Solow Residual) size nse 14.0 120 iao a0 60 40 20 ao Figure 2.3: Mixed Olley and Pakes Decompositionby Industry and Region of Aggregate Productivity (Rt itricted Solow Residual) l n d u a y 3.0 Bn 25 2 0 1 5 1.o 0 5 0 0 29 Figure 2.4: Mixed Olley and Pakes Decompositionby Size andAge ofAggregate Productivity (Restricted Solow Residual) size 25 1 N I ase 20 1.5 1.0 05 00 small Medurn LerBe I -furrs a d Figure 2.5: Relative I C effects by groups of variables on aggregate productivity, average productivity and efficiency (mixed O&P decomposition and simulations of a 20% improvement inI C and C variables) A E A E ari- tothe nixed Ulevand Blest%ccnoCsitimof the Assfsate Raluctivii (WedWclec: RoductivRy measuredin w,shareafsales-wed I inlevds). RSnulatiorrsof aChargeinIC4Vars and%changemAggfsate Rocclciityandmthecbrrponsnts d the aley and we^ -Rim. Irrh-astructures IWtape,comptionandcrime rnFimeandcaporategoLernance 0 Cbality,innmationandlaborskils Iahererddables 30 I I Aggregate productivity Average productivity I IEfficiency term 1 Infrastructures 15.3 25.7 16.7 23.9 22.6 17.4 R e d tape, corruption and crime 45.2 37.3 51.4 45.4 25.1 38.7 Finance and corporate governance 9.1 6.3 9.8 7.8 8.5 5.3 Quality, innovation and labor skills 9.9 6.2 8.7 7.4 12.2 5.9 Other control variables 16.4 24.3 10.1 12.8 26.5 29.7 Total 100 100 100 100 100 100 Notes: A: The percentage contributions of I C and C variables of table C.2a were transformed such that the relative impact of the group `s' to aggregate and average productivity and efficiency may be expressed according to the next expressions (1.A)Wsngg = [il l%LTil+l% cov(s, ,icf,p )I1[jl l+l% cov(si ,icf,pI + ji=ll.). I+1d;r vrts x x 'ricJr -I a ' cov( s! ,ic;,jr I ] -1 (2.A)wsmg= [5p=l ~d&I][ p=1 5 I%`pl$+v!'#$ ? lqrE;,1] C J r l cov(sir .i~f,~ ll[LIIN% C cov(s, ,icf,p )I + ?lNqr Vr#s j,=l C COV(S~,.icS.jr )I ] For all s,r=infrastructures; red tape, corruption and crime; finance and corp. gov.; quality, innov. and labor skills; other control variables, and p=l,..,n are the number o f variables belonging to the group s, andjr=l,..,mr are the variables in the group r, such that r#s. B:The relative impact of each group of IC and C variables is expressed interms of the % change of each term of the O&P decomposition from t=O (before simulation) to t=l (after simulation) (see Escribano et. al. (2007) for details) according to the following expressions (1.B)yngg = [ilI] [il -1 AIPp' AIPp"l+ Vr#s j,=1? Alp; X I] (2.B)wsmg= [5 -1 ?=I AlFil][ ?=I Vrtrj=I 2 AlFil + 1 3AlFil] (3.B)yeff= [~lA~~~~(~:,.Pp",ii~j,Alc~v(~;,i.Pp',i)I ] [ ~ , A ~ C O " ( ~ ~ , . i ~ ~ ~ , t ) ~ I Note that the industry dummies and the constant are excluded from the computations of both cases A and B. 31 Figure 2.6: Relative I C effects on aggregate productivity (Mixed O&P decomposition) -, Infrastructures F&i tape,corrytim&crim ~ i - &corp. p. Clcalbinmatim aher contrd 70 &I& sldlls variables II II x ) =i I 452 0.3 3.4 n 00 00 T.l 11 12 13 14 15 16 T2 2.1 22 2.3 2.4 2.5 2.6 2.7 T.3 3.1 32 3.3 3.4 3.5 T.4 4.1 42 4.3 T.5 5.1 52 5.3 5.4 T.l Total infrastructures 2.3Dummyforconflictsincourts T.4 Total quality, innov. and labor skills I l w t ociedrcustomSt0exp3rt-interaction 2.4DummyforseurityRTmss 4.1DummyforiSOqualityce!tification wthfinstktcb eqmtt 12OEctricityfmmaQma-ator 2.5Dummyforcrime 2.6Mana@stimespentintu.issues 42Staff-femelemrkers 4.3Trainingto mn-pocbctionmrkefs 13Dummyfor'Ejfts'to o h i neleztncitysupply 2.7P a p e mto speeduptureaucmcy 14vIclteroutages T.5 Total other control variables 15vIclterfromplblicsources T.3 Total finance andcorporate governance 5,191areof imp3rtedirputs 16Wp-nentlosses,imptts 3.1Initialinvestment:privatetenks 52Percentageofunionizslmrkforce 32Vlbrkingcapitalfimdbyinfodsources 5,3DummyforFDi T.2 Total redtape corruption and crime 3.3Vlbrkingcapitalfimdbymrrbankingfimial 5,4-of exp3tts PlSelesrep3rldtOtaES institutions 22Dummyforconflictswthclients 3.4Dmmyforcheckingorsa\n'rgaccount 3.4Dummyforcrejt line Figure 2.7: Relative I C effects on average productivity (Mixed O&P decomposition) Infrastructures Fedtape,corrlptim&crim &cap. p. W iimovatim aher contrd &laborsldlls variables 514 242 8 7 Ql a7 62 n 0.1 09 T.l 11 12 13 14 15 16 T2 21 22 2.3 24 25 26 27 T.3 3.1 3.2 3.3 2.4 36 T.4 4.1 42 4.3 T.5 5.1 52 53 5.4 T.l Total infrastructures 2.3Dummyforconflictsincourts 11Daysto clearcustomsto eqmtt - interaction 2.4Dummyforseurityees T.4 Total quality, innov. and labor sklils 4.1DurnmyforiSOqualityce!tification nithfimsthadcbeqmtt 2.5Dummyforcrime 4.2Staff femalemhers - 12Eetricityfromagenemor 4.3Trainingto mrquductionwrkers 13Dummyfor'gifts'to ottainde3ticitysupFiy 2.7Payrentsto speedupburmuracy 2.6M~stimespentinbur.issues 14V&erOuta+ T.5 Total other control variables 15V&erfromplblicsources T.3 Total finance andcorporate governance 5,,Shareof impttdiws 16shipmentlosses,imptts 3.1initialiwestment:privatebanks 32Vlbrkingcapitalf i m dbyinfodsources 5.2Percentawofunionizslmrkforce 5,3DummyforFDI T.2 Total redtape corruption and crime 3,SVlbhingcapitalfimdbymrrbankingfimial 5,4Shareofwtts 2.lSelesrepottdto taxs institutions 22Dummyforconflictsdthclients 3.4DummyforcheMrgorsavingaccount 3.4Dummyforcre3itline 32 Figure 2.8: Relative I C effects on efficiency (Mixed O&P decomposition) 6 Infrastructures F k itap, corrlption&crirre f i m e &cap. q,w. W iinnovation a k r cct-itrd variables &labor skills T.l 11 12 13 14 15 16 T2 21 22 2.3 2.4 2.5 2.6 27 T.3 3.1 3.2 3.3 3.4 3.5 T.4 4.1 42 4.3 T.5 5.1 5.2 5.3 5.4 . T.lTotal infrastructures 2.3Dummyforconflictsinmutts 11Daystoclearcustomstoewrt-interaction 2.4Dummyforsecurityexperses T.4 Total quality, innov. and labor skills Mithfirrnsthatdowtt 25Dummyforcrime 4.lDummyforISOqditycertification 12Eectricityfromagor 2.6Mampfstimespent inbur.issues 4.2Staff-femalenorkern ',fzA"d, 13Dummyfor'gifts'toobtalndectricitysupply 2JPa)nnentstospeedupbureawracy 4.3Tminingto mmprochtionuorkers 14VIBteroutages 15Waberfromplblicsources T.3Total finance and corporate governance variables ~~~~~~ 16Ship~lent losses,imprts 3.1Initialiwestrnent:privateb k s 3pMbrkingcapital f i - e d ~ i ~ osources ~ 5.2Percentapofunionizednorkforce T.2 Total redtape corruption and crime 3.3Mbrkingcapitalfinancedbyrambankingfinancial 53DummyforFDI 5,4Shareofe ~ t t s 2.1Salesreprtedtotax% institdions 22Dumrnyforco~lictsMiithclients 3.4Dumnyforcheckkgorsavingaccount 3.4Dummyforcredit line Figure 2.9: Relative I C effects by groups ofvariables on average productivity (Decomposition inLogs); by size lnfrastuctures Fi?dtape, corruption Financeandcorporate Wity, innovationand Other controlvariable3 andcrime governance laborskills 33 Figure 2.10: ICA PercentageAbsolute Contribution on Economic Performance Variables ._.. - Roductivity Errployrnent Realwages &ports ~- FDI Realwages Redtape, corruption and crime Financeand corporate gowrnance 0 Quality, innovationand laborskils Other control variables I Total I 100 I 100 I 100 I 100 I 100 I Notes: Let the percentage contribution of the IC variable i,to the average value of the dependent variable of equation q be given by (1) where q = logP, lo&, ZogW, EXP, FDI, (see notes on tables D.l-D.4).From (l),relative percentage contribution of the the group 's' of IC variables to the average value of the dependent variable of equation q (e: i s therefore given by ) (2) For all s,r=infrastructures; red tape, corruption and crime; finance and corp. gov.; quality, innov. and labor skills; other control variables, and i=l ...,nare the number of variables belonging to the group s, andjr=l...,mrare the variables inthe group r, such that r#s. * n.s. means that no variables of this group of variables were significant in the corresponding regression. 34 Figure 2.11: Relative I C effects on average log-employment -~ ___- ,o Prod. W. Infrastructures Redtape, corr. & Fin & Quality inn. & labor skills Other ctrlvars crime corp gov. I ''5 O 1'22 I I ii 78 3.5 T.l T.2 1'.3 3.1 3.2 3.3 3.4 3.5 3.6 T.4 4.1 4.2 4.3 T.5 5.1 5.2 T.6 6.1 6.2 6.3 6.4 6.5 T.7 7.1 7.2 7.3 7.4 T.l Productivity T.4 Total red tape corr. and crime 6.3DummyforR&D 4.lWorkforcereportedto taxes 6.4 Staff - skilled workers T.2 Real wages 4.2 Securityexpenses 6.5 Dummyfo rt raining 4.2 Crime losses T.3 Total infrastructures T.7 Total other control variables 3.1Days to clearcustomsto import interaction - T.5 Total finance and corporate 7.1Shareof imported inputs with firms that do import governance 7.2 Trade union 3.2 Dummyforomgenerator 5.3 initiai investment: private banks 7.3DummyforFDI 3.3 Water outages 5.2 Dummyforioan wth collateral 7.4Shareof eworts 3.4 Water from public sources 3.5 Dummyfor internet T.6 Total quality, innovation and 3.6 Dummyfore-maii labor skills 6.lDummyfor IS0 qualitycertification 6.2 Dummyfor process innovation Figure 2.12: Relative IC effects on average log-real wages 1 % Prod Infrast. Redtape, Finance& Quality innovation & Other control variables rate gov labor slulls 70 ptton & crtm cc 60 50 40 30 20 M 5 7 4 2 0 1 T I Ti! 21 T.3 3.1 3.2 T.4 4.1 4.2 T.5 5.1 5.2 T.6 6.1 6.2 T.l Productivity T.4 Total finance and corporate T.6 Total other control variables governance 6.1Exportingexperience T.2 Total infrastructures 4.1Purchases paid beforedelivery 6.2 Dummyfor large firm 2 liossesdueto poweroutages 4.2Working capital financed byinformalsources T.3 Total red tape corruption and T.5 Total quality, innovation and labor crime skills 3.1Sales reportedto taxes 5.1Staff -femaieworkers 3.2Paymentstoobtainacontractwththe 5.2 Staff -productionworkers government 35 Figure 2.13: Relative IC effects on the probability of exporting ______ 6 Rod Infrastructures Redtape, corr, &crime Finance& Quality inn. & labor skills Other ctrl variables corn. aov 5 {, I 0 I 15 13 7 K) 5 0 T.l T 2 2.1 2.2 2.3 2.4 2.5 T.3 3.1 3.2 3.3 3.4 T.4 4.1 4.2 4.3 T.4 4.1 4.2 4.3 4.4 T.5 5.1 5 2 5.: T.l Productivity T.3 Total red tape corr. and crime T.5 Total quality, innov. and labor skills 3.1Workforce reported to taxes 5.1Dummyfor R&D T.2 Total infrastructures 3.2 Dummyfor absenteeism due to crime 5.2 Staff- productionworkers 2.lDays to clear customsto import - 3.3 Number of inspections 5.3 Dummyfor training interaction with firms that do import 3.4 Dummyfor gifts in inspections 5.4 Manager's experience 2.2 Average duration of powroutages 2.3Water from public sources T.4 Total finance and corporate gov. T.6 Total other control variables 2.4 Dummyfor internet 4.1Largest shareholder 6.1Shareof imported inputs 2.5 Shipment losses in exports 4.2Working capitaifinanced byfamilyifriends 6.2Dummymore5competitors 4.3 Dummyfor loan with collateral 6.3 Dummyformediumfirms Figure 2.14: Relative ICA effects on the probability of receiving FDI Rod. Infrastructures RedtaDe. corruDtions &crime Finance& Qualitv innovation Other control variables 3 4 2 5 T 6 6 1 6 2 6 3 T.l Productivity T.4 Total finance and corporate gov. T.6 Total other control variables 4.1Initialinvestment: private banks 6.1Share of exports T.2 Total infrastructures 4.2 Working capital financed bynon banking 6.2 Dummyfor local monopoiy 2.lDummyforowr generator financial institutions 6.3Dummyfor largefirrns 2.2Waitforawatersupply T.5 Total quality, innov. and labor skills T.3 Total red tape corr. and crime 5.1 Durnmyforforeigntechnology 3.lDummyforconflicts with clients 5.2 Staff - skilledworkers 3.2 Dummyforconflicts in courts 3.3Dummyforgifts in inspections 3.4Cost of entry 36 Figure 2.15: Relative I C effects by groups ofvariables on average log-employment; by size 0 I 27.0 17.3 - 5.1 4.2 Productivity Real Wages Infrastructures Red tape, Finance and Quality, Other control corruption and corporate innovation and variables crime governance labor skills I &# Smll firm 0 Mediumfirms 0 Largefirms I I I Figure 2.16: Relative ICA effects by groups of variables on average log-realwages; by size /c '5 70 67.3 35 30 55 50 L5 LO 35 30 27 3 15 10 15 IO 5 1.0 0.7 2 8 3 1 2.9 2.9 4'3 0 Productivity Infrastructures Red tape, Finance and Quality, Other control corruption and corporate innovationand variables crime governance labor skills I ~ Smallfirm Mediumfirm 0 Largefirms Figure 2.17: Relative ICA effects by groups of variables on the probability of exporting; by size zo ~ 25 N P .I 20 115 10 5 i o I Productivity Infrastructures Redtape, Finance and Quality, Other control 1 corruption and corporate innovation and variables crime governance labor skills I 0 Largefirms I Figure 2.18: Relative ICA effects by groups of variables on the probability of receiving FDI;by size k io P 1.5 x 1.0 15 io !5 !O 5 0 5 0 - Productivity Infrastructures Red tape, Finance and Quality, Other control corruption and corporate innovation and variables crime governance labor skills 0 Smallfirms 0 Mediumfirms -_ 0 Large firms 38 Figure 2.19: Managers' perceptions; percentage of firms that considers each one of the following problems as a severe obstacle to firms' economic performance Infrastructures Redtape, co rruption and crime finance Labor skills Total" 11 1.2 1.3 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2 8 2.9 3.1 3.2 3 3 4.1 4.2 5 6 1 Infrastructures. 2 Red Tape, Corruption and Crime. 3. Finance. 1.1Electricity 2.1 Corruption 3.1Access to Land 1.2 Transportation 2.2 Crime.theft and disorder 3.2 Access to Finance 1.3 Customand traderegulatiom 2.3 Anti-conpeitiveor InformalPracticeS 3.3 Macroeconomc uncertainty 2.4 Legal SystenVConflict ReSolUtion 2.5 Regulatory Policy Uncertainty 4. Labour Skills. 2.6TaxRats 4.1 Labor Regulations 2.7Tax Administration 4.2 Skillsand Educationof Avalable Workforce 2.8 Business Licensingand Operating Permits 2.9 Environmentalregulations 5. Total relative weights. &. Average group relative weights. 1E Infrastructure E3 Red tape, corruption and crime Financeand corporate governance Labor skills 1 'Totalsarecomputed astherelative weighof eachgroup of perceptlomoverthesumof all perceptiom'weights 39 Section 3. INTERPRETATIONS OFTHE PRODUCTIVITY RESULTS 1. SECURITY COSTSAND THEIR EFFECT ONPRODUCTIVITY The regression results show a significant positive effect of security costs on productivity compared to declines in productivity associated with speed payments and time dealing with officials. One interpretation i s this as declines in crime improving productivity. However, it seems that a plausible interpretation i s that security costs rise with crime and this positive association with productivity is then puzzling. There is also a concern that the measure might be endogenous. The interpretation of the dummy for security costs i s as follows: those firms having any security costs are on average 37% more productive that the remaining firms. This positive effect on productivity i s independent o f the amount o f money spent on security, it only matters whether the firms spend some money on security or not. The interpretation that 'security costs rise with crime' would be true if the significant variable would have been the total amount spent on security as a percentage of total sales. This variable i s available in the survey and has been incorporated in the selection of significant variables, but finally it was not statistically significant. In fact, in other productivity analyses done for other countries (Africa, Asia, Latin America) where the percentage o f security costs i s significant on TFP, the effect i s negative. That i s in line with the interpretation above, where more crime implies more security expenses and less productivity. We can put more empirical underpinning to this issue by going one step further and computing the interactive effect o f having security expenses (dummy) and losses due to crime in the productivity regressions. This effect i s significant, with the coefficient ranging from -0.015 to -0.019, depending on the productivity specification we are using, see Table 2.19. The interpretation in this case i s straightforward. Let productivity equation be given by: P=bO+bl *durn_secur+b2*(dum-secur*crime-loss)+b3 *IC3+...+bn *ICn+e where P i s productivity, dum-secur i s a dummy for security expenses, crime-loss i s losses due to crime, IC3,...,ICn represent other IC and C variables and e i s a error term. Here b2 measures the interaction effect and b l the direct effect o f the dummy for security on productivity. The effect o f having security expenses i s given by: AP=(bl +b2*crimeJoss) *Adum~secur=(0.518-0.015*crime~loss)*Adum~secur That is: if a firm incurs any security expenses, holding everything else constant, productivity increases on average by 0.518-0.015*crime-loss. This indicates that the positive effect o f security expenses decreases as the intensity o f criminal activity increases (more losses due to crime). For some firms with a large proportion o f sales lost due to crime the effect o f having security expenses may be even negative. 40 So far the effect o f security expenses on crime losses i s still unresolved. The linear correlation between 'dummy for security expenses' and 'losses due to crime' i s negative (- 0.27). Although no causal relationships can be derived from the correlation, it indicates that security may prevent crime, since those firms with security expenses are less likely to suffer criminal attempts. Another plausible interpretation i s that the more losses due to crime (as percentage of sales) the less probability o f having security expenses. The linear regression of crime-loss on dum-secur and a constant i s crime-loss CoefSicient Standard error T-stat dum-segur 1I -6.124 1.333 -4.6 constant 1 12.828 1.173 10.9 This linear regression shows an intuitive negative linear relation between the percentage of sales lost due to crime and dummy for security expenses, so having security expenses reduces losses due to crime. Or, in other words, those firms spending some money preventing crime reduce the losses due to crime. Obviously, this i s a naive linear regression that does not take into account possible endogenous problems and many other econometric issues. Insteadit i s pretty intuitive on the nature o f the relation we are trying to clarify. Table 3.1: IC elasticities and semi-elasticities with respect to productivity TWO steps estimation 1 Single step estimation Water from public sources (b) 0.003*** 0.003*** 0.002"" 0.002"" 0.002"" 0.002** Shipment losses, import (b) -0.009"" -0.009"" -0.009*** -0.008** -0.009** -0.012*** Red tape, Sales reportedto taxes (a) 0.008 0.008 0.013** 0.013"" 0.008 0.005 corruption and Dummy for conflicts with -0.18 -0.185 -0.085 -0.096 -0.112 -0.118 crime clients Dummyfor conflicts incourts 0.126 0.129 0.198 0.127 0.185 0.214 Dummyfor security(b) 0.518*** 0.519*** 0.703*** 0.619*** 0.695*** 0.643*** Interaction (Dummy for -0.015*** -0.015*"* -0.019*** -0.017**" -0.018*** -0.017*** security*Lossesdue to crime) Dummy for crime (b) -0.151 -0.15 -0.093 -0.034 -0.094 0 Manager's time inbureaucratic -0.01 -0.006 -0.021** -0.016" -0.022" -0.013 issues (a) Payments to speed up -0.006 -0.006 -0.007 -0.006 -0.006 -0.004 41 governance instaff(6) Trainingto non-production I 0.001 Io.001 I0.002 I0.002 10.002 Io.001 Finally, the dummy for security expenses may be an endogenous variable. Unfortunately the industry-region-size average o f this variable i s not a good instrument so we had to use the crude plant-level variable inthe regressions. 2. EFFECTSICBLOCKVARIABLESONPRODUCTIVITY OF The effects are accumulated in absolute terms across variables in a block (rather than allowing for some variables to have offsetting effects). Thus the positive effect o f having security costs i s added to the absolute effect o f management time, speed payments and sales reported to authorities to make "red tape, corruption and crime" the block o f I C measures with the biggest impact. The idea i s to add up all the effects in each group in absolute terms and to compute the absolute percentage relative contribution o f each group with respect to the absolute contribution of the investment climate as a whole. This way, we do not offset the positive effects o f some variables with the negative effects of other variables. By using the absolute contributions we measure what the productivity gain i s the investment climate factors improve, and this implies reducing the IC constraints with negative effects and increasingthe IC factors with positive effects. 42 3. GENDER IMPACT The survey questionnaire includes information on the gender o f the principal owner. It would be interesting to note the sector and size distributions of where women are more economically active, whether women report differences in constraints or whether there are differences inthe impact o f objective conditions on performance. The following two tables show the average of the percentage of female workers in staff by industry and size. The sectors where women are more active are garments, textiles, food and chemicals. Table 3.2: Percentageoffemale workersinstaff, averageby industry IIIFoods 34.38 Garments 53.09 Textiles III 34.29III Machinery and equipment 20.00 Chemicals 31.55 Non-metallic min 10.55 Other manufacturing 18.98 By size the distribution of the percentage of women among firm staff is very uniform. Inall categories o f size the percentage o f female workers in staff i s close to 30%. Table 3.3: Percentage of female workers instaff, average by size Small 28.89 Medium 31.33 30.85 We can compute the absolute percentage contribution o f the variable percentage of female workers in staff by industry and size. Table 3.4 shows that the sectors where women are more economically active the relative absolute contribution o f this variable i s larger. Table 3.4: Percentage of female workers instaff, percentage absolute contributions to average log-productivity by industry Foods 12.93 Garments 13.91 Textiles 2.59 Machinery and equipment 1.41 Chemicals 2.19 Non-metallic minerals 1.27 Other manufacturing 1.65 43 Table 3.5 shows the relative contributions by sizes. Inthis case the contributions are uniformly distributed among categories of sizes. Table 3.5: Percentage of female workers instaff, percentage absolute contributions to average log-productivity by size Small I Large Medium 12.02 4. DIFFERENCESSAMPLING (2003VS. 2007DATA) IN A legitimate question is whether differences in perceived constraints between 2003 and 2007 are driven by differences in the sampling. The latest survey includes services as well as manufacturing firms. There may also be differences in the geographic / size / sectoral distribution of respondents. We include the same percentages computed with the sample that we have used to construct the 2003-2007 panel and test changes in the IC coefficients. The panel uses the same sectors, industries and regions in both 2003 and 2007. So the results shouldn't be driven by sampling differences. The percentages are very similar to those of Figure 1.1 (Chapter 1) in almost all the cases, except by some minor changes in the second decimal. So we would conclude that the sampling differences are not affecting the results. Table 3.6: Differences in Sampling (2003 vs. 2007) 2003 2007 Corruption 84.07 62.3 1 Electricity 26.64 51.68 I Macroeconomic instability 62.83 45.99 Crime I 84.07 I 33.64 1 44