WPS5278 Policy Research Working Paper 5278 Will the Crisis Affect the Economic Recovery in Eastern European Countries? Evidence from Firm Level Data Paulo Correa Mariana Iootty The World Bank Europe and Central Asia Region Private and Financial Sector Development Sector Unit April 2010 Policy Research Working Paper 5278 Abstract Two sources of growth are firm learning and innovation. European countries. The results show that innovative Using a unique panel data for 1,686 firms in six countries and young firms were significantly more affected by the (Bulgaria, Hungary, Latvia, Lithuania, Romania, crisis than non innovative and older enterprises. The and Turkey), this paper applies panel data estimators authors interpret these results as an indication that the and Juhn-Murphy Pierce decomposition in order to achievement of pre-crisis growth rates in those countries identify the effects of the global economic crisis on sales may be difficult. growth of innovative and young enterprises in Eastern This paper--a product of the Private and Financial Sector Development Sector Unit, Europe and Central Asia Region--is part of a larger effort in the department to study the effects of the financial crisis on the corporate sector. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at pcorrea@ worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team WILL THE CRISIS AFFECT THE ECONOMIC RECOVERY IN EASTERN EUROPEAN COUNTRIES? EVIDENCE FROM FIRM LEVEL DATA1 Paulo Correa Lead Economist, Finance and Private Sector Development, Europe and Central Asia Region, World Bank Mariana Iootty Consultant, Finance and Private Sector Development, Europe and Central Asia Region, World Bank and Federal Rural University of Rio de Janeiro 1 We would like to thank to Gerardo Corrochano (Sector Director, ECSPF), Lalit Raina (Sector Manager, ECSPF) and Sophie Sirtaine (Sector Manager, ECSPF) for their guidance. Donato de Rosa, Joćo Pedro Azevedo, Murat Seker, Peter Harrold, Ulrich Zachau and Willem van Eeghen provided valuable suggestions on earlier draft of this paper. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. 1 Introduction One question of the current international debate particularly relevant to Eastern European countries refers to the potential drivers of the economic recovery. As international trade declines and financial conditions worsen, firm learning and innovation become more relevant sources of growth in the postcrisis period. Conditional on size and survival rate, young firms are expected to grow faster than older firms, among other reasons, due to diminishing returns to learning (Klepper and Thompson (2007); Dunne, Roberts and Samuelson (1989)). Innovative firms are also expected to provide a positive contribution to economic growth, as innovation and R&D tend to positively affect firm productivity and sales performance (Aw, Roberts and Xu (2008)).2 In this sense, understanding how the crisis will affect innovative and young enterprises helps comprehends Eastern European growth prospects. In this paper, we investigate the sales growth of innovative and young firms in the region, before and after the crisis. We apply two methods. First, we use paneldata estimators in order to assess the difference of sales growth performance between innovative (young) and noninnovative (older) companies, over the period, when controlling for firm characteristics. Second, we use the JuhnMurphyPierce technique to decompose the difference in sales growth performance between those groups of firms, before and after the crisis, into: i) differences in the distribution of observed characteristics (the characteristics effect); ii) differences in the "market value" of such characteristics (the returns effect); and iii) differences in unobservable attributes and "market values" (the unexplained effect). This decomposition technique not only allows us to check the econometric findings but also helps us to assess the sources of sales differentials between those firms' cohorts. Panel data analysis shows that sales growth rate of innovative firms have declined more than that of noninnovative companies, a result that is robust regardless the estimator applied, and the criteria used to categorize innovation (introduction of a 2 The micromacro link between firmlevel innovation, theories of aggregate technological change and growth can be found in Klette and Kortum (2004). 2 product/process, or development of R&D activities). Our analysis also points that the decrease of sales growth rate of younger firms was more severe than older companies. The decomposition exercise confirms that the positive difference of sales growth rate in favor of innovative and in favor of young companies, before the crisis, was reversed afterwards. For innovative companies, we found evidence that the reduction in the "premium" for the ability to innovate ­ which we interpret as Schumpeter's "animal spirit" ­ after the crisis contributed to explain the worse performance of innovative firms. Conversely, in the case of young enterprises, results suggest that their "ability to learn" contributed to mitigate the negative impact of the crisis on sales growth performance. Overall, we interpret these results as indication that the achievement of precrisis growth rates in those countries may be difficult. The paper is organized as follows. The next section describes the data used and exposes the empirical methodology applied. Section 3 presents results on the empirical analysis, and Section 4 summarizes the main results. 2 ­ Data and Methodology We use a unique panel data for 1,686 firms in six countries (Bulgaria, Hungary, Latvia, Lithuania, Romania, and Turkey) covering manufacturing, retail and other service sectors.3 This panel results from the matching of two datasets. The first one is the World Bank's Financial Crisis Survey (EFCS) implemented in June/July 2009 and designed to capture the effects of the crisis on key elements of the private economy: sales, employment, finances, and expectations about the future. The second one is the World Bank's Enterprise Survey (ES) carried out in 2008. With an original sample that is stratified by firm size, sector and region and representative of the private nonagricultural formal economy in each country, the ES 2008 provides information on firm characteristics, various performance measures and the business environment, most of them refereed to fiscal year of 2007, the precrisis scenario. The merger between the two datasets was possible since the EFCS covers a subsample of firms drawn from the set of firms interviewed in the ES 2008. As the participation in EFCS was voluntary and the entire original sample of the ES, which covered 3Turkey is the exception, where only the original manufacturing subsample was targeted. 3 2,499 firms, was contacted to determine whether these firms were still in existence or if they had failed and/or became inactive, the sampling weights of the ES 2008 were adjusted, based in the nonresponse firms. These adjusted weights were then used in order to produce the estimations that are, therefore, representative for the nonagricultural private economy within each country. We classify the firms into groups of: i) young and older firms; and ii) innovative or noninnovative companies. To distinguish the age status of the firm, we define a dummy for firms up to five years old in before (2007) the crisis. To distinguish the innovative status of the firm, we used as a criterion the introduction of new product or process in the period 20052007, according to ES 2008.4 As we do not have this information for 2009, we then define a dummy that equals one if the firm has introduced a product/process in 20052007 period, and repeat this value for the postcrisis period.5 Alternatively, we use as a criterion to classify innovative companies the development of R&D activities in 20052007 period, also according to ES 2008.6 Table 1 presents the estimated proportions of innovative and noninnovative companies across countries, following the two criteria, while Table 2 shows the estimated proportions of firms by age. Noninnovative firms as well as older firms make the largest groups in all countries. 4The question used in ES 2008 was "In the last three years, has this establishment introduced new products or services?" 5Therefore, even though we recognize that innovation process is not static over time, we work with an innovation category that is time invariant, as we assume that innovative firms in the precrisis scenario (2004 2007) are potentially innovative after the crisis (2009). 6The question used in ES 2008 was "In the last three years, has this establishment invested in research and development (inhouse or outsourced)?" 4 Table 1 ­ Proportions of innovative/noninnovative companies, by country Noninnovative Innovative Country Did not introduce new product/process Introduce of new product/process N. Obs Turkey 0.60 0.40 513 Romania 0.51 0.49 365 Hungary 0.66 0.34 187 Latvia 0.43 0.57 226 Lithuania 0.34 0.66 238 Bulgaria 0.50 0.50 149 Total 0.52 0.49 1,678 Country Did not invest in R&D Invested in R&D N. Obs Turkey 0.70 0.30 513 Romania 0.77 0.23 359 Hungary 0.90 0.10 187 Latvia 0.84 0.16 225 Lithuania 0.86 0.14 239 Bulgaria 0.63 0.37 150 Total 0.76 0.24 1,673 Source: EFCS 2009ES 2008 Table 2 ­ Proportions of young/older companies (before the crisis), by country Young Older N.obs Country (< = 5 years old) (>= 6 years old) Turkey 0.11 0.89 514 Romania 0.20 0.80 370 Hungary 0.11 0.89 187 Latvia 0.17 0.83 226 Lithuania 0.25 0.75 239 Bulgaria 0.11 0.89 150 Total 0.11 0.89 1,686 Source: EFCS 2009ES 2008 Using the innovation and age groups defined above, we examine a particular measure of firm performance: growth in sales. We calculate the sales growth rates before and after the crisis (Table 3) for these categories of firms. For the precrisis period, we use the sales value of 2007 fiscal year and three years before that, from ES 2008, and compute the annualized growth rate of sales in the 20042007 period.7 For the postcrisis period, we use annual growth rate of sales in the 20082009 period8, according to EFCS.9 7 Sales values for 2004 and 2007 were originally in local currencies and were converted to current US dollars using exchange rate from International Financial Statistics Database. We then controlled for outliers of sales 5 Table 3 Sales Growth Rates (in %) comparisons: before and after the crisis Precrisis Postcrisis Average 95% Conf. Interv. Average 95% Conf. Interv. Innovative (introduce new product/process) 36.38 27.05 45.71 26.48 29.68 23.29 Noninnovative (did not introduce new product/process) 23.39 16.70 30.07 25.34 28.46 22.22 Innovative (invest in R&D) 45.29 27.87 62.71 27.02 31.82 22.21 Noninnovative (did not invest in R&D) 24.61 19.76 29.45 25.55 28.08 23.02 Young 45.52 28.10 62.94 30.13 36.77 23.48 (< = 5 years old) Older 26.90 20.97 32.84 25.42 27.78 23.06 (>= 6 years old) Source: EFCS 2009 ES 2008 Data shows that before the crisis, growth rates for innovative (young) firms showed to be significantly higher than noninnovative (older) companies, at 5% level of significance.10 After the crisis, however, the reduction in growth rate was higher for innovative (young) firms, though not statistically significant. These results seem to suggest that the crisis impact was higher for innovative and young companies. However, besides lacking of statistical significance for the comparisons after the crisis, these results reflect only a partial analysis. One can argue whether the difference of sales growth performance between the refereed groups of firms persist when controlling for other firm characteristics that are likely to affect sales performance. In order to address this, we apply an empirical approach to discuss to what extent innovative (young) firms were more affected by the global downturn once firm idiosyncratic characteristics are taken into account. Two empirical methods are used. values for both 2004 and 2007. The criterion was to consider as an outlier the observations lying outside the interval defined as the average three standard deviations of the refereed variable expressed in logarithm form. 8 June/July 2008 to June/July 2009. As the 2008/2009 growth rate ranges from 100% to 100%, we didn't control for outliers. 9 Some World Bank reports (like the EU10 Regular Economic Report) state that the acute phase of the crisis in these countries was in the second half of 2008 (specifically September 2008), but the severity of the impact was perceived only in 2009 statistics, even though the economic activity has stabilized in the second half of 2009. As we only have data for June/July 2009, we just use this point as a postcrisis reference. 10 For the precrisis comparison between innovative and noninnovative companies, the results are valid regardless the criteria used to classify innovative companies. 6 First, we apply a paneldata analysis in order to assess the difference of sales growth performance between innovative (young) and noninnovative (older) companies, from pre to postcrisis period, when controlling for certain firm characteristics. This panel analysis uses the specification listed in Equation 1. Annual growth rate of sales of firm i at two (t=2) time periods (2004/2007 and 2008/2009) is represented by g .11 (1150 full time employees), (51250) and ( 251) stand as controls for firm size in the two periods. 12 The omitted category is . is another firmlevel control variable; it is a dummy distinguishing firms that generate more than 10% of their sales from exports.13 Sector and country control for the 2digit industry and the country of the firm.14 As already defined, is a variable controlling for firm age in the precrisis period. The variable Inov reflects, as stated earlier, the innovative status of the firm. Skill depicts the use of skilled workforce by the firm; it is a dummy defined as more than 20% of employees with university degree in 2007, the precrisis period. As it happens with the Inov variable, due to lack of data to reflect the use of skilled workforce by the firm in the 2008/2009 period, we repeat the value of Skill in t=2 as the same as in t=1.15 As we are interested in examining how the sales growth performance of innovative/noninnovative firms and of young/older companies have evolved over time, we interact and with a time trend variable (we do the same for ). Equation 1 is then defined as: 1 11As already stated, for the precrisis period, we use the annualized growth rate of sales in the 20042007 period, based on data from ES 2008. For the postcrisis period, we use annual growth rate of sales in the June/July 2008 june/july 2009 period, according to EFCS. 12 Specifically, the dummy size variables for the two periods (t=1 and t=2) are defined based on information of, respectively, 2007, from ES 2008, and 2009, from EFCS 2009. 13 The dummy of export orientation for the two periods (t=1 and t=2) are defined based on information of, respectively, 2007, from ES 2008, and 2009, from EFCS 2009. 14 While controlling for industry fixed effects we can account for any factor that can create heterogeneity across industries, as differences in the level of competition and in technology use. Meanwhile, country fixed effects allows to isolate potential differences in macro policies that may affect the evolution of firms in each country. The omitted categories are "other manufacturing" sector and Bulgaria, respectively. 15 Therefore, the variable is also timeinvariant, and reflects, in the postcrisis period, the potentially skilled intensive firms. 7 where is a vector of control variables representing size, export orientation, country and sector, as previously defined. The error term is defined as: 2 where are random individualspecific effects and is an idiosyncratic error. Following a random effect model, both and the idiosyncratic error are assumed to be uncorrelated with regressors. We use two different estimators for this random effect model: the FGLS (feasible generalized least squares estimator) and the PFGLS (pooled feasible generalized least squares estimator), also called populationaveraged estimator. The difference between them is related to the assumptions about the structure of serial correlation of the model error, . While the FGLS restricts this serial correlation to be the same at all lags (so the errors are called equicorrelated), the PFGLS estimator here applied places no restriction on the error.16 Both of the estimators obtain cluster robust standard errors and use probability weights in order to get results that are representative for all firms in each country. The second empirical method to be used is the JuhnMurphyPierce (JMP) decomposition. This technique was first employed by Juhn, Murphy and Pierce (1993) and applies microsimulation techniques to decompose the differences between two outcomes distributions (from previous OLS fitted results) for two population subgroups.17 Following Juhn, Murphy, and Pierce (1993) the framework is then defined. First, after estimating an OLS model to explain the sales growth rate of two population sub groups separately (let's assume, for instance, the innovative and noninnovative companies) in one given period (for instance, before the crisis), we define two models: 16 Let be the error correlation over time for individual i, the FGLS estimator sets for all so that errors are assumed to be equicorrelated. The populationaveraged estimator (PFGLS) we use places no restriction on . 17 This method is a variant of the classical Oaxaca and Blinder decomposition technique, widely applied in labor economics to explain wage differentials between groups of workers (Oaxaca, 1973 and Blinder,1973) When compared to OaxacaBlinder method, the JMP decomposition technique has the distinct advantage of dealing explicitly with the residuals from the estimation, therefore considering three components (characteristics, return and unexplained effects), while OaxacaBlinder deals with only two (characteristics and return effects). 8 (3) 4 where 1 and 2 set the innovative and noninnovative groups of firms; and are the vectors of the values of the dependent variable sales growth rate in the two firms' groups; and are the data matrices (observed characteristics)18 for the two firms' groups; and are the vectors of estimated coefficients (observable returns ­ "market values" of those characteristics) for the two subgroups; and, and are the vectors of estimated residuals (unobservable, i.e., unmeasured characteristics and returns) for the same subgroups. Let . and . denote the cumulative distribution functions of the residuals for innovative and noninnovative groups, respectively. Therefore, if we take | 5 to be the percentile of an individual firm's (i) residual in the residual distribution of innovative firms, we can write: | (6) where . is the inverse of the cumulative distribution function for the "innovative firms" group. The same procedure can be done to write the individual residual in the residual distribution of noninnovative companies. Next, assume that . to be the reference residual distribution (e.g., the average residual distribution over both samples) and let to be an estimate of benchmark coefficients (e.g, the coefficients from a pooled model over the two firms' groups. Based on 18 Both the theoretical and the empirical literature suggest that attributes like human capital, size, age, and export orientation are likely to affect firm evolution. Klette and Kortum (2004), Aw, Roberts and Xu (2008), Seker (2009) and Bernard et al (2007) are some examples. Therefore, when comparing innovative and noninnovative companies, the set of observable characteristics that influences sales growth includes the following variables: size (dummies of 110 full time employees,1150, 51250 and 251), firm age (a dummy for firms up to five years old), use of skilled labor (a dummy for firms with more than 20% of employees with university degree in 2007), export orientation (a dummy distinguishing firms that generate more than 10% of their sales from exports), country and sector fixed effects. For the comparison young and older firms, the set of characteristics includes: size in the period, innovation status (a dummy distinguishing companies that have introduced a new product/process in the 20042007 period), use of skilled labor in the precrisis period, export orientation in the period, country and sector fixed effects. 9 this framework we can construct hypothetical outcome distributions with any of the three components held fixed. We can then determine: (A) hypothetical outcomes with varying observable characteristics but fixed coefficients and fixed residual distribution as | 7 | 8 (B) hypothetical outcomes with varying observable characteristics and varying coefficients but a fixed residual distribution as | 9 | 10 (C) outcomes with varying observable characteristics, varying coefficients and varying residual distribution as | 11 | 12 These last outcomes are actually equal to the originally observed values: | 13 | 14 Let a capital letter stand for a summary statistic of the distribution of the variable denoted by the corresponding lowercase letter. For instance, Y may be the mean of the distribution of y. So, the differential can be decomposed as: (15) where T is the total difference; C can be attributed to differences in observable characteristics the characteristic (or quantity) effect; R to differences in observable 10 returns ­ the return (or price) effect, and U to differences in unobservable quantities and returns ­ the unexplained (or residual) effect. In this paper, as we are using the JMP method to analyze the difference of sales growth performance distribution between the two groups of firms, the first component characteristics (or quantity) effect captures the part of the difference of sales growth performance between the two groups of firms that is due to differences in observable characteristics (quantities). In other words, it quantifies to what extent innovative (young) firms have a more favorable "endowment" in terms of observable characteristics as compared to noninnovative (older) companies. For example, one firm category might have, on average, larger firms, larger use of skilled workforce and larger export orientation, and these might lead to higher growth sales performance. The second component, the return (or price) effect, reflects the part of the gap of sales growth performance between the two groups of firms that is due differences in returns of those observable characteristics. In other words, it measures to what extent the returns of those observable characteristics on the sales performance of a firm differ between innovative (young) and noninnovative (older) companies. For example, one firm category might get larger returns from the same observable characteristics when compared to other firm group. The third component, the unexplained (or residual) effect, measures the part of the difference of the sales growth performance between the two groups of firms that is due to differences in unobservable characteristics and returns or to measurement error. Assuming that the OLS models that explain sales performance of the two companies' cohorts are satisfactorily specified, any unmeasured factors that affect the performance of these two groups will be capture by this residual effect.19 For the comparison between innovative and noninnovative firms, we interpret this residual effect as a reflection of the difference between the two firm cohorts in terms of some intangible assets, such as the specialized knowledge embodied in researchers, the firms' entrepreneurship ability or its "animal spirits". When comparing young and older firms, this residual effect could measure 19 As already stated, the choice of the variables used in the OLS models that explain the sales growth rate of the two groups of firms draws on the theoretical and empirical literature of firm evolution. 11 those firms' intangible assets related to age, such as the ability to appropriate of learning benefits. 3 ­ Main Results 3.1 ­ Panel Data Analysis The regression results for Equation (1) are given in Table 4. Both the FGLS estimator and the populationaveraged estimator (PFGLS) results are displayed.20 The relevant parameters to be examined in both of the results sets are those related to the following variables: i) time interactions with innovative status of firm ( and with the age status of the firm ( ; and ii) time dummy. The coefficients associated to time interactions variables basically represent the difference in the changes of sales growth rate between the two categories over time. In other words, it provides the predicted sales growth variation of innovative (young) firms, compared to noninnovative (older) companies, over time (i.e, from precrisis to postcrisis period).21 Meanwhile, the dummy variable aims to reflect the aggregate difference in growth sales performance of all firms before and after the crisis when controlling for firm characteristics. This could be assumed as a proxy of the aggregate crisis impact on the enterprise performance in these countries. 20 Before presenting the results, it is worth noting that these estimators have shown to be consistent; the Hausman test presented evidence in favor of the random effect model. The overall statistic of the Hausman test (which assumed the null hypothesis that random effect estimator was fully efficient) was: Prob>Chi2=0.9726. 21 For example, for the innovative/noninnovative comparison, the coefficient associated to the interaction variable ( ) represents the differenceindifference estimate, as defined as: , , , , where , is the growth sales rate of innovative companies in the postcrisis period ­ 2008/2009, and , is the annualized growth sales rate of innovative companies in the precrisis period ­ 2004/2007. For noninnovative companies , and , are the growth sales rate in the postcrisis and precrisis periods, respectively. It is important to distinguish the interpretation of ) coefficient from the interpretation of `s coefficient. While the last coefficient measures the difference between the average sales growth rates of innovative and noninnovative companies over time, the coefficient associated to ( ) measures the difference in the changes of sales growth rate between the two categories over time. For instance, if the coefficient of is positive, this would mean that, on average, the growth rate of innovative firms is larger than noninnovative companies. A negative sign for the coefficient of would not contradict this; it would mean that, over time, the decrease in the sales growth rate of innovative companies has been larger when compared to the decrease in sales growth rate of noninnovative companies. 12 The results in columns 2 and 3 showed that when controlling for firmspecific characteristics, innovative firms have performed significantly worse than noninnovative companies over the refereed period (from precrisis to postcrisis period).22 The pooled FGLS estimator revealed that the sales growth rate of innovative companies has decreased 15.97 percentage points more than the sales growth rate of noninnovative over time, at 5% level of significance. The FGLS estimator confirms this result as it shows that the contraction in the sales growth performance of innovative companies was 16.31 points larger than the growth performance of noninnovative firms, also at 5% level of significance.23 In order to examine if these results are robust to the way innovative firms are classified, we reestimate the sales growth equation using the development of R&D activities in the 20042007 period as an alternative criterion to categorize innovative firms (see columns 4 and 5). The populationaveraged estimator showed that, when comparing the pre and postcrisis periods and controlling for firmspecific characteristics, the sales growth performance of companies with R&D activities has decreased 25.34 percentage points more than firms without R&D activities; this difference was statistically significant at 1% level. The FGLS estimator also reveals that firms with R&D activities performed worse over time, though without statistical significance. 22 Even though innovative companies have presented larger negative changes of sales growth rate over time, they did grow more than noninnovative firms, on average (see result for variable). 23 To check whether possible correlations between innovation and export orientation variables could influence the results, we applied the population average estimator to three different specifications: i) Equation 1 including , ), ( , and variables; ii) Equation 1 including , and ) variables; and iii) Equation 1 including ( and variables. The results, presented in Tables II and III in the Annex, showed that the inclusion of export related variables ( and do not change the fact that innovative companies have performed significantly worse than noninnovative companies over time, which always showed to be statistically significant. Besides, the results of and have never presented statistical significance, regardless the specification used. 13 Table 4 ­ Explaining the sales growth rate over time: a panel data analysis* Innovation measured as introduction Innovation measured by the of product/process performance of R&D pooled FGLS pooled FGLS estimator FGLS estimator estimator FGLS estimator Coefficient (2) (3) (4) (5) time 47.273*** 47.728*** 48.740*** 54.442*** (3.676) (4.601) (3.305) (4.715) innovative 14.229** 15.603** 22.022** 5.200 (6.357) (6.834) (9.618) (7.599) time X innovative 15.975** 16.316** 25.345*** 5.160 (6.717) (6.890) (9.398) (7.694) Young 18.775** 19.490*** 18.453** 18.628*** (9.498) (7.004) (9.071) (7.027) time X young 17.623* 20.679*** 17.446* 19.782*** (10.042) (7.410) (9.675) (7.449) skilledintensive 7.805 13.336** 7.538 11.392** (5.970) (5.343) (5.475) (5.030) Time X skill 12.174* 13.089** 12.431** 10.663** (6.288) (5.540) (5.960) (5.190) Small 6.818** 7.046*** 7.703** 7.726*** (3.337) (2.528) (3.193) (2.503) Medium 11.141*** 11.299*** 11.553*** 12.059*** (4.115) (3.738) (4.244) (4.128) Large 7.243* 11.090*** 7.267* 12.064*** (4.028) (3.339) (4.180) (3.650) Exportoriented 6.502* 4.264 6.859* 3.794 (3.726) (3.754) (3.783) (3.647) _cons 12.069** 16.063*** 12.097*** 21.291*** (4.734) (5.474) (4.639) (4.802) N.obs 2720 2720 2718 2718 Prob >chi2 0.000 0.000 0.000 0.000 Robust standard errors are in parenthesis. *** p<0.01, ** p<0.05, * p<0.1 * Table I in the Annex presents the complete results, including the sector and country controls Results in columns 2 and 3 also indicated that young firms have been significantly more affected than older companies.24 The results of the populationaveraged estimator 24 For the young/older comparison, the coefficient associated to the interaction variable ( ) represents the differenceindifference estimate, as defined as: , , , , where , is the growth sales rate of young companies in the postcrisis period ­ 2008/2009, and , is the annualized growth sales rate of young companies in the precrisis period ­ 2004/2007. Similarly, , and , are the growth sales rate of older firms in the postcrisis and precrisis periods, 14 suggest that the reduction of sales growth performance of young firms was 17.62 percentage points larger than older firms over time (at 10% significance level); the FGLS estimator suggests that this difference was 20.67 percentage points (at 1% significance level). If we consider the results columns 4 and 5, the negative difference for young companies are confirmed.25 Also remarkable is the result of the coefficient of the dummy time. The results for this coefficient showed that, when controlling for firm characteristics, sales growth of all companies have presented, on average, a reduction that varies from 47.27 to 54.42 percentage points. This evidence, which showed to be statistically significant at 1% level, could be interpreted as the crisis aggregate impact on the enterprise performance in these countries. One last feature of Table 3 that is worth mentioning is the result of the coefficient associated to the variable . The results (in columns 2 and 3) showed that sales growth rate of skilled intensive companies (those employing more than 20 percent of workers with university degrees) have increased more than nonskilled firms. From column 2, the populationaveraged estimator showed that the sales growth of skilled companies has augmented 12.17 percentage points more than the sales growth of non skilled firms; from column 3, the population average estimator results points to a difference of 13.08 percentage points in favor of skilled companies over time.26 3.2 JuhnMurphyPierce Decomposition27 In order to understand the sources of the difference in sales growth performance between innovative (young) and noninnovative (older) companies, we apply the JMP decomposition technique in two periods, before and after the crisis. respectively. Since the Youngi variable is defined as a dummy for firms up to five years old before the crisis (in 2007), it is worth mentioning that when we compare the sales growth rate of young and older firms, we are actually comparing the performance of firms that, before the crisis, were young (<=5 years old) and older (>= 6 years old). This comparison is then consistent with the innovative/noninnovative comparison as all these firm categories are considered as timeinvariant in the model. 25 Even though young companies have presented larger negative changes of sales growth rate over time, they did grow more than older firms, on average (see result for variable). 26 It must be stressed that skilled companies grew less than non skilled firms, on average (see result for variable). 27 This decomposition analysis was performed using Stata (version 11). 15 For the innovative/noninnovative comparison, the results of the JMP decomposition are listed in Table 5. As the aim of the analyses is to explain the sales growth performance of innovative companies relative to noninnovative firms, the distinct components that explain the sales performance gap between these firm groups (characteristics, return and unexplained effects) are considered from the point of view of innovative companies. Table 5 shows that, for the precrisis period, the total difference between sales growth performance of innovative and noninnovative companies amounts to 13.41 percentage points (in favor of innovative). The observable characteristics effect is also positive implying that the characteristics of innovative firms were actually more favorable than those of noninnovative companies. Turning to the composition of this characteristic effect into the explanatory variables assumed (size, use of skilled labor, export orientation, age, country and sector), we see that, among the firm characteristics, the major contributions for this positive difference came from size (see Table IV in the Annex). Table 5 ­ Decomposing the total difference of sales growth rate between innovative and noninnovative firms: JMP results before and after the crisis#* Total Difference Characteristics Effect Return Effect Unexplained Effect 13.411 1.038 12.073 0.300 Before (5.750) (0.388) (0.915) (0.152) 100% 7.7% 90.0% 2.2% 1.417 1.185 0.250 0.018 After (0.679) (0.582) (0.103) (0.007) 100% 83.6% 17.7% 1.3% # The results refer to the sample means of the sales growth rate distribution * Bootstrap standard error in parentheses The observable return effect is also positive, revealing that the returns of those characteristics were higher for innovative companies. In other words, it means that the transformation of given "inputs" in results by the innovative firms leaded to higher sales growth rate than the transformation by noninnovative companies, and this accounts for the largest part (90%) of the sales growth gap between the two groups of firms. 16 Finally, the unexplained components showed to be positive, suggesting that, on average, unobservable factors favor the innovative companies. As mentioned before, we interpret this as an indication of the existence of a positive premium for the entrepreneurship ability, or the "animal spirits", of innovative companies. In the aftercrisis period, Table 5 shows a negative gap between the sales growth performance of innovative and noninnovative companies. This result points out a drastic change in the pattern of comparative sales performance of the two groups of firms when contrasting to the precrisis period. What explains this deterioration can be identified when examining the three effects. First, the observable characteristic effect is negative, implying that the characteristics of innovative firms turned to be less favorable than those of noninnovative companies, and this effect accounts for the largest part (83.6%) of the sales growth gap between the two types of firms after the crisis. The observable return effect is negative as well, indicating that the returns of those characteristics turned to be smaller for innovative companies. On the other hand, the unexplained component remains positive, suggesting that unobservable factors still favor the innovative companies, and served to reduce the sales growth gap between the two firms' categories. Nevertheless, it is also remarkable that this residual was significantly reduced after the crisis, which can be seen as evidence that the "premium" for the ability to innovate ("animal spirits") has declined after the economic downturn. Table 6 presents the JMP decomposition technique, before and after the crisis, to examine the gap of sales growth performance between young and older companies. As the aim of the exercise is to explain the sales growth performance of young companies relative to older firms, the three effects that explain the sales performance gap between the two groups are considered from the point of view of young companies. 17 Table 6 ­ Decomposing the total difference of sales growth rate between young and older firms: JMP results before and after the crisis#* Total Difference Characteristics Effect Return Effect Unexplained Effect 19.261 20.985 15.559 17.283 Before (9.164) (7.485) (1.491) (6.569) 100% 109.0% 80.8% 89.7% 10.696 4.434 5.011 1.252 After (3.828) (1.775) (1.005) (0.559) 100% 41.4% 46.8% 11.7% # The results refer to the sample means of the sales growth rate distribution * Bootstrap standard error in parentheses For the precrisis period, Table 6 shows that the total difference between sales growth performance of young and older companies is 19.26 percentage points (in favor of young). The observable characteristics effect and the return effect were also positive implying that the characteristics of young firms were actually more favorable than those of older companies, especially regarding use of skilled workforce and export orientation (see Table V in the Annex). The returns of the observable characteristics were higher for younger companies. However, the residual (or unexplained) effect was negative, which we interpreted as evidence that the ability of young firms to benefit from the learning process was inferior to the ability of older companies. This negative effect acted to reduce the positive gap of sales growth between the two groups of firms before the crisis. In the postcrisis period, the results show a negative gap between the sales growth performance of young and older companies, indicating a reversal of the comparative sales performance of the two groups of firms. The observable characteristic effect have also turned to be negative, which means that, on average, the characteristics of young firms turned to be less favorable than those of older companies after the crisis. If we separate this negative characteristic effect, we see that the major contributions for this result came from size (see Table V in Annex). This could be interpreted as evidence that after the crisis young companies might have faced difficulties evolving into larger firms. The unexplained component was again negative, suggesting that the ability of young firms to capture the learning benefits remained inferior to the ability of older companies. It is worth mentioning, however, that this residual effect turned to be less negative in the 18 postcrisis period, possibly suggesting that young firms were forced to improve their learning capabilities in order to operate in a more competitive environment. Yet this improvement was not enough to compensate the worse sales growth performance of young enterprises after the crisis. 4 Conclusions In this paper we examined the effects of the global downturn on innovative firms and on young firms in order to understand the growth prospects of Eastern European countries using a unique panel data for 1,686 firms in six countries (Bulgaria, Hungary, Latvia, Lithuania, Romania, and Turkey) covering manufacturing, retail and other service sectors. Two empirical methods were used. First, we applied a paneldata analysis in order to assess the difference of sales growth performance between innovative and non innovative firms, and between young and older firms, over the pre and postcrisis periods, when controlling for certain firm characteristics. Second, we used the Juhn MurphyPierce technique to decompose the difference in sales growth performance between innovative (young) noninnovative (older) companies, over time, into three effects: characteristics effect; returns effect and unexplained effect. The results show that innovative firms have suffered more than noninnovative companies, regardless the estimator applied (populationaveraged estimator or FGLS), and the criteria used to categorize innovation (introduction of a product or a process or development of R&D activities). We also showed that the decrease of sales growth rate of younger firms was more severe than of older companies. When comparing innovative and noninnovative companies, the decomposition technique showed three main outcomes. First, the positive difference in sales growth performance in favor of innovative companies has been reversed after the crisis. Second, while before the crisis that positive gap in favor of innovative companies was due to their better characteristics, especially regarding size, and, mainly, to higher returns obtained by innovative firms, after the crisis the gap reversion was due mostly to the deterioration of characteristics of innovative companies. Finally, the results showed that the residual (or unexplained) effect was always positive, before and after the crisis, suggesting that the 19 unobservable factors favor the innovative companies, which could also be interpreted as a premium for the entrepreneurship ability, or "animal spirits", of innovative companies. However, it is also remarkable that this residual was significantly reduced after the crisis, which can be seen as evidence that the "premium" for innovative companies was diminished over time. When contrasting young and older companies, the use of JuhnMurphyPierce decomposition also revealed three main results. First, we found evidence that the positive gap in sales growth performance in favor of young companies has been reversed after the crisis. Second, the results showed that after the crisis there was a deterioration of the characteristics of the young companies, especially regarding firm size. Finally, the results showed that the residual (or unexplained) effect was always negative, before and after the crisis, suggesting that the ability of young firms to capture the learning benefits was always inferior to the ability of older companies. It is worth mentioning, however, that this residual effect turned to be less negative in the postcrisis period, which can suggest that young firms were forced to improve their learning capabilities in order to operate in a more competitive environment. Nevertheless, this improvement was not enough to compensate the worse sales growth performance of young enterprises after the crisis. In sum, our results seem to provide robust evidence that innovative and young companies were more affected by the global downturn. Two obstacles for the achievement of precrisis growth rates in the region related to this result are a premature exit among young firms and a decline on innovation performance. The exit of potentially viable young firms ­ assuming the same firm entry rates of the precrisis period ­ and the consequent premature "aging" of the enterprise sector might soften market selection and therefore reduce productivity gains. The impact on innovative firms may hold back investment in R&D among other reasons because lower sales ­ in a context where diminished access to credit shifted internal funding to finance working capital imply less funding for innovation. For those reasons, we interpret our results as indication that the achievement of precrisis growth rates in those countries may be difficult. 20 References Aw, Bee Y., Mark J. Roberts and Daniel Y. Xu (2008), "R&D Investments, Exporting and the Evolution of Firm Productivity," American Economic Review, Papers and Proceedings, 98:2, pp. 451456. Bernard, A.B, J. Branford Jensen, S.J Redding, and Schott, P.K. (2007), "Firms in International Trade", Journal of Economic Perspectives, 21(3), pp.10530. Blinder, Alan S. (1973). "Wage Discrimination: Reduced Form and Structural Variables." Journal of Human Resources, 8: 436455 Dunne, Timothy, Roberts, Mark. J, and Larry, Samuelson, (1989). "The Growth and Failure of U.S. Manufacturing Plants," The Quarterly Journal of Economics, MIT Press, vol. 104(4): 67198 Geroski, P.A., 2005. Understanding the implications of empirical work on corporate growth rates. Managerial and Decision Economics 26:129­138. Grossman, G. and E. Helpman, 1991. Innovation and Growth in the Global Economy, Cambridge: MIT Press, 1991, Juhn, Chinhui, Kevin M. Murphy, Brooks Pierce (1993). "Wage Inequality and the Rise in Returns to Skill". Journal of Political Economy 101(3): 410442 Klepper, S., & Thompson, P. (2007). Submarkets and the evolution of market structure. RAND Journal of Economics 34(4): 862888 Oaxaca, Ronald (1973). "MaleFemale Wage Differentials in Urban Labor Markets," International Economic Review, 14 (October): 693709 Romer, Paul M. (1986). "Increasing Returns and Long Run Growth", Journal of Political Economy, 94, 100237. Seker, Murat (2009). "Importing, Exporting and Innovation in Developing Countries", World Bank Policy Research Working Paper n.o 5156. World Bank (2009). "EU10 Regular Economic Report: From Stabilization to Recovery October 2009" 21 Annex Table I ­ Explaining the sales growth rate over time: a panel data analysis based on pooled FGLS and FGLS estimators (complete results) # Innovation measured as introduction of Innovation measured by the performance of product/process R&D PFGLS FGLS PFGLS FGLS estimator estimator estimator estimator time 47.273*** 47.728*** 48.740*** 54.442*** (3.676) (4.601) (3.305) (4.715) innovative 14.229** 15.603** 22.022** 5.200 (6.357) (6.834) (9.618) (7.599) time X innovative 15.975** 16.316** 25.345*** 5.160 (6.717) (6.890) (9.398) (7.694) Young 18.775** 19.490*** 18.453** 18.628*** (9.498) (7.004) (9.071) (7.027) time X young 17.623* 20.679*** 17.446* 19.782*** (10.042) (7.410) (9.675) (7.449) skilledintensive 7.805 13.336** 7.538 11.392** (5.970) (5.343) (5.475) (5.030) Time X skill 12.174* 13.089** 12.431** 10.663** (6.288) (5.540) (5.960) (5.190) Small 6.818** 7.046*** 7.703** 7.726*** (3.337) (2.528) (3.193) (2.503) Medium 11.141*** 11.299*** 11.553*** 12.059*** (4.115) (3.738) (4.244) (4.128) Large 7.243* 11.090*** 7.267* 12.064*** (4.028) (3.339) (4.180) (3.650) Exportoriented 6.502* 4.264 6.859* 3.794 (3.726) (3.754) (3.783) (3.647) Turkey 10.065** 9.360* 10.096** 9.422* (4.942) (5.342) (5.040) (5.436) Romania 8.942* 7.230 10.515* 7.451 (5.107) (6.032) (5.465) (6.005) Hungary 0.203 1.849 2.001 1.816 (2.698) (2.412) (3.000) (2.525) Latvia 7.752** 4.732* 5.112 3.412 (3.420) (2.743) (3.340) (2.738) Lithuania 0.695 0.789 2.254 1.245 (3.653) (2.758) (3.764) (2.650) Food 2.668 2.533 3.458 2.942 (4.218) (5.636) (4.179) (5.501) #Robust standard errors in parenthesis. *** p<0.01, ** p<0.05, * p<0.1 22 Table I (cont.) ­ Explaining the sales growth rate over time: a panel data analysis based on pooled FGLS and FGLS estimators (complete results) # Innovation measured as introduction of Innovation measured by the performance of product/process R&D PFGLS FGLS PFGLS FGLS estimator estimator estimator estimator Textiles 1.348 7.811 2.000 8.411 (5.830) (7.034) (5.853) (6.949) Garments 9.075 11.372** 8.922 11.539** (5.594) (5.629) (5.830) (5.554) Chemicals 5.421 7.410 4.854 7.999 (6.616) (15.012) (6.719) (14.841) Plastic & Rubber 20.910 5.129 21.586 5.684 (17.124) (8.095) (17.210) (8.085) Non metallic mineral products 1.372 7.392 0.920 7.294 (5.335) (7.223) (5.815) (7.148) Basic metals 10.436 1.987 10.777 1.359 (10.691) (12.759) (11.436) (12.927) Fabricated metal products 4.404 9.218* 5.177 9.671* (4.930) (5.470) (4.902) (5.448) Machinery and equipments 10.445 6.085 8.265 6.669 (18.134) (20.430) (17.048) (19.837) Electronics 11.297* 7.352 11.024* 6.989 (6.370) (6.144) (6.073) (6.099) Construction 4.699 8.118** 6.487 9.877** (4.480) (4.024) (4.691) (4.227) Services of Motor Vehicles 3.904 3.881 4.135 2.683 (4.900) (5.693) (5.165) (5.671) Wholesale 5.010 3.351 4.576 4.092 (4.133) (4.789) (4.216) (4.728) Retail 8.407* 4.973 8.147* 5.141 (4.638) (5.065) (4.527) (4.963) Hotel and restaurants 5.704 0.361 5.446 0.301 (5.565) (5.888) (5.484) (5.944) Transport 6.144 1.658 4.708 2.881 (6.706) (5.319) (6.434) (5.307) IT 15.981** 5.423 15.371** 5.070 (7.370) (5.151) (6.925) (5.345) Other 7.305 6.343 11.015 6.940 (11.934) (9.586) (13.984) (9.530) _cons 12.069** 16.063*** 12.097*** 21.291*** (4.734) (5.474) (4.639) (4.802) N.obs 2720 2720 2718 2718 #Robust standard errors in parenthesis. *** p<0.01, ** p<0.05, * p<0.1 23 Table II ­ Explaining the sales growth rate over time with and without export orientation variables: a panel data analysis based on pooled FGLS estimator (complete results) # Coefficient Model 1 Model 2 Model 3 time 46.844*** 49.193*** 53.399*** (3.952) (3.848) (4.020) innovative 14.146** 13.723** (6.359) (6.264) time X innovative 15.891** 15.530** (6.726) (6.629) Young 18.775** 19.518** 17.988* (9.497) (9.471) (9.364) time X young 17.618* 18.673* 16.579* (10.041) (9.993) (9.992) skilledintensive 7.773 7.942 5.863 (5.972) (6.034) (5.418) time X skill 12.128* 12.047* 9.597* (6.286) (6.312) (5.825) Small 6.793** 6.518* 7.785** (3.347) (3.395) (3.179) Medium 11.051*** 10.186** 12.392*** (4.163) (4.097) (4.057) Large 7.178* 6.021 8.348** (4.043) (4.027) (3.868) Exportoriented 5.324 3.238 (5.311) (5.074) time x export oriented 1.813 3.572 (5.965) (5.976) Turkey 10.086** 8.996* 9.853** (4.945) (4.821) (4.972) Romania 9.283* 5.685 8.919* (5.394) (3.959) (5.395) Hungary 0.240 0.016 0.534 (2.715) (2.673) (2.672) Latvia 7.712** 8.069** 7.144** (3.422) (3.447) (3.311) Lithuania 0.721 1.016 0.318 (3.655) (3.623) (3.530) Food 2.726 3.345 2.805 (4.259) (4.158) (4.317) Textiles 1.319 0.826 0.369 (5.824) (5.827) (5.975) #Innovation measured as introduction of product/process; robust standard errors in parenthesis.*** p<0.01, ** p<0.05, * p<0.1 24 Table II (cont.)­ Explaining the sales growth rate over time with and without export orientation variables: a panel data analysis based on pooled FGLS estimator (complete results) # Coefficient Model 1 Model 2 Model 3 Garments 9.083 9.952* 9.396* (5.603) (5.647) (5.570) Chemicals 5.478 5.086 5.636 (6.596) (6.742) (6.513) Plastic & Rubber 20.916 20.865 19.480 (17.134) (17.238) (16.670) Non metallic mineral products 1.312 1.119 0.851 (5.383) (5.372) (5.449) Basic metals 10.381 10.544 9.765 (10.701) (10.836) (11.129) Fabricated metal products 4.443 4.510 5.991 (4.900) (4.932) (5.002) Machinery and equipments 10.490 10.782 10.678 (18.146) (17.968) (17.496) Electronics 11.362* 10.847* 12.026** (6.404) (6.470) (5.952) Construction 4.631 4.055 5.748 (4.507) (4.416) (4.533) Services of Motor Vehicles 3.897 4.373 2.846 (4.908) (5.027) (5.050) Wholesale 5.053 4.551 5.428 (4.158) (4.036) (4.076) Retail 8.468* 8.847* 8.192* (4.684) (4.723) (4.644) Hotel and restaurants 5.638 5.201 5.657 (5.593) (5.513) (5.466) Transport 6.162 4.525 4.976 (6.717) (6.573) (6.738) IT 16.051** 16.619** 17.945** (7.393) (7.424) (7.596) Other 7.322 7.417 9.255 (11.884) (11.054) (11.819) _cons 11.805** 12.735*** 17.251*** (4.955) (4.725) (5.451) N.obs 2720 2720 2730 #nnovation measured as introduction of product/process; robust standard errors in parenthesis *** p<0.01, ** p<0.05, * p<0.1 25 Table III ­ Explaining the sales growth rate over time with and without export orientation variables: a panel data analysis based on FGLS estimator (complete results) # Coefficient Model 1 Model 2 Model 3 time 47.924*** 48.510*** 55.528*** (5.114) (4.622) (5.026) innovative 15.625** 15.336** (6.825) (6.728) time X innovative 16.342** 16.109** (6.882) (6.806) Young 19.498*** 19.601*** 18.713*** (7.002) (6.995) (7.028) time X young 20.683*** 20.849*** 19.767*** (7.409) (7.390) (7.456) skilledintensive 13.348** 13.440** 11.214** (5.357) (5.383) (5.032) time X skill 13.103** 13.080** 10.282** (5.551) (5.540) (5.122) Small 7.049*** 6.788*** 7.819*** (2.537) (2.517) (2.490) Medium 11.314*** 10.585*** 12.417*** (3.748) (3.378) (3.900) Large 11.107*** 10.184*** 12.755*** (3.326) (3.320) (3.424) Exportoriented 4.688 3.372 (6.757) (6.640) time x export oriented 0.694 0.375 (6.677) (6.706) Turkey 9.339* 8.973* 9.229* (5.339) (5.302) (5.355) Romania 7.094 5.315 6.941 (6.054) (4.834) (6.113) Hungary 1.859 1.815 2.482 (2.420) (2.402) (2.388) Latvia 4.738* 4.952* 4.002 (2.746) (2.771) (2.657) Lithuania 0.785 0.952 0.886 (2.762) (2.768) (2.657) Food 2.526 3.151 2.819 (5.647) (5.591) (5.642) #Innovation measured as introduction of product/process; robust standard errors in parenthesis. *** p<0.01, ** p<0.05, * p<0.1 26 Table III(cont.) ­ Explaining the sales growth rate over time with and without export orientation variables: a panel data analysis based on FGLS estimator (complete results) # Coefficient Model 1 Model 2 Model 3 Textiles 7.797 8.465 8.799 (7.018) (7.073) (7.143) Garments 11.370** 11.845** 11.730** (5.628) (5.642) (5.628) Chemicals 7.440 7.643 8.089 (15.106) (15.048) (14.955) Plastic & Rubber 5.151 5.066 6.059 (8.116) (8.030) (8.005) Non metallic mineral products 7.388 7.669 7.310 (7.221) (7.228) (7.221) Basic metals 2.022 2.499 0.940 (12.789) (12.683) (12.938) Fabricated metal products 9.206* 9.436* 9.832* (5.455) (5.413) (5.551) Machinery and equipments 6.092 6.024 7.387 (20.429) (20.420) (19.979) Electronics 7.335 7.128 7.548 (6.148) (6.268) (5.910) Construction 8.147** 7.214* 9.662** (4.082) (3.850) (4.267) Services of Motor Vehicles 3.867 4.619 2.648 (5.710) (5.865) (5.803) Wholesale 3.331 3.542 4.202 (4.828) (4.779) (4.792) Retail 4.945 5.820 5.081 (5.115) (5.363) (5.125) Hotel and restaurants 0.398 0.351 0.509 (5.930) (5.907) (6.064) Transport 1.684 2.131 2.855 (5.337) (5.295) (5.375) IT 5.411 6.006 5.917 (5.166) (5.069) (5.088) Other 6.305 5.874 7.071 (9.551) (9.569) (9.596) _cons 16.186*** 15.887*** 22.577*** (5.836) (5.488) (5.651) N.obs 2720 2720 2730 #Innovation measured as introduction of product/process; robust standard errors in parenthesis. *** p<0.01, ** p<0.05, * p<0.1 27 Table IV ­ Decomposing the total difference of sales growth rate between innovative and noninnovative firms: JMP results before and after the crisis specific characteristics effects Total Total Returns Unexplained characteristics Difference skill age size export country sector Effect Effect Effect Before 13.411 1.038 1.518 0.358 0.704 0.441 2.122 0.529 12.073 0.300 After 1.417 1.185 0.641 0.103 0.612 0.002 2.625 0.292 0.250 0.018 Table V Decomposing the total difference of sales growth rate between young and older firms: JMP results before and after the crisis Specific characteristics effects Total Total Returns Unexplained characteristics Difference skill innovation size export country sector Effect Effect Effect Before 19.261 20.985 17.284 0.228 1.197 0.351 4.936 0.160 15.559 17.283 After 10.696 4.434 1.658 0.212 1.707 0.235 4.207 0.625 5.011 1.252 28