Productivity Developments In Estonia: Evidence From firm Level Data1
Labor productivity growth has been weak in recent years. Firm-level data can help shed light on where in the economy productivity growth was strong and where it lacked. Moreover, they reveal which firm-level characteristics were critical for productivity performance. It turns out that the bulk of productivity growth in Estonia can be attributed to the more traditional firms, that there was a strong catching-up effect of firms with initially below-average performance, and that the superior performance of younger firms disappeared in the period after the global financial crisis, suggesting reduced dynamism in the economy. Firm characteristics that were associated with strong productivity growth were also associated with weak employment generation, suggesting a pronounced labor rationalization element in productivity growth. There is tentative evidence that this effect might have been stronger in Estonia than elsewhere in Europe.
1. Since 2005, labor productivity trends in Estonia have flattened significantly. Real labor productivity, measured as the ratio of real gross domestic product in 2010 prices over total employment, registered a modest cumulative increase of 14 percent since 2005—an annualized growth rate of only around 1 percent. Value weighted labor productivity, calculated from firm level data was more volatile, but showed a similar trend.2
2. Like in most countries, productivity growth slowed down markedly following the 2008/09 crisis, but the extent is surprising, considering the still very sizeable productivity gap with Western Europe. Measured in real terms (2010 euros), the value added per worker in Estonia is a mere 40 percent of the EU12 average.3 There is however significant variation across sectors—with workers in agriculture only a quarter less productive; while manufacturing productivity in Estonia is less than a third of the EU12 average. On average, productivity gaps are larger in high-technology sectors, both in manufacturing and in services. The gap shrinks when measuring productivity in purchasing power standards, but it remains at a considerable one-third vis-à-vis the EU12 average.
Labor Productivity Trend Growth
Sources: Statistics Estonia; Orbis; IMF staff calculations.
Sources: Eurostat; and IMF staff calculatons.
3. Which segments of the economy drove aggregate productivity developments and what kind of firms were successful in boosting their productivity? This chapter uses firm-level data from Orbis, a worldwide database of primarily private company information, to shed light on these questions. Orbis provides firm-level balance sheet data over the period 2005–14 covering around 50–60 percent of the employed and 40 percent of total value added in the case of Estonia (Table A1).4 Coverage varies across sectors, but even after dropping observations with missing values for key variables such as value added, there is sufficient data from 2005 onward for a meaningful analysis.5 The data can be used to calculate labor productivity, total factor productivity (TFP), employment generation, and value added growth at the firm level. Section A dissects the economy along different dimensions to analyzes which segments are responsible for productivity, employment, and value-added growth. Sections B and C look at which firm characteristics are relevant for productivity and employment growth, respectively. Section D concludes.
A. The Distribution of Productivity, Employment, and Value-added Trends Across the Main Segments of the Economy
4. Dissecting the economy along different dimensions helps identify the locus of productivity growth in the economy and gauge the importance of composition effects on aggregate productivity growth. Five different stratifications of firms operating in Estonia are considered: (i) by economic activity, i.e. agriculture, manufacturing, construction, trade, market services, and basic services; (ii) by level of technological sophistication, i.e. high-tech manufacturing, other manufacturing, high-tech services, and other market services;6 (iii) by firm size, i.e. micro enterprises with less than 10 persons employed, small enterprises with 10–49 persons employed, medium enterprises with 50–249 persons employed, and large enterprises with 250 or more persons employed; and (iv) by degree of involvement in external trade. Did these segments fare differently in terms of productivity, employment, and value-added growth? How much did they contribute to economy-wide trends? Did shifts in the relative importance of these segments materially affect aggregate developments through composition effects?
5. In the dissection by economic activity, agriculture and manufacturing stood out with the largest productivity gains (Figure 1). During 2005–14, labor productivity of an average firm in the agricultural sector nearly doubled and it increased by close to 22½ percent in manufacturing firms. In contrast, labor productivity declined by over 22 percent in market services and 19 percent in basic services, respectively. Much of this divergence likely reflects differences in capital deepening, as changes in TFP, while in the same direction, were considerably less pronounced. Productivity gains pushed up the share of value added in the economy generated by agriculture and mining, and manufacturing with a corresponding decline in other economic activities. Yet, the expansion of value added shares was not sufficient to generate much additional employment. Agriculture and manufacturing saw their shares in employment decline.
Figure 1.Estonia: Trends by Economic Activity
6. In the dissection by economic sophistication, high-technology sectors were not particularly strong drivers of productivity and more traditional manufacturing witnessed impressive gains (Figure 2). Labor productivity and TFP growth was similar for both high-tech and more traditional manufacturing. In services, productivity was flat in high-tech firms, and declined in other services. The value added share of high-technology manufacturing was flat at around 5 percent, while the share of other manufacturing increased 18 percent, from 14 percent of value added to 17 percent. The employment shares of both declined, but the decline was stronger in high-technology manufacturing. In the services sector, high-technology services increased its share in both value added and employment, which was offset by declines in other market services.
Figure 2.Estonia: Trends by Level of Technological Sophistication
7. In the dissection by firm size, productivity increased in all size categories, except for micro enterprises where it declined (Figure 3). While labor productivity decreased by about 14 percent for micro firms, it increased for all other size classes, led by large enterprises where productivity more than doubled. TFP exhibits the same pattern, but variability is smaller and, in particular, the superior performance of large enterprises is much less pronounced. The categories with the strongest and weakest productivity performance expanded their shares in values added—micro firms on account of expanding employment and large firms on account of productivity gains and despite a declining employment share.
Figure 3.Estonia: Trends by Firm Size
8. The dissection by degree of involvement in international trade yields only weak evidence of positive effects on productivity.
On the one hand, labor productivity and TFP increased in the tradeable sector, but declined in the non-tradeable sector, suggesting a positive impact of international trade on productivity growth (Figure 4). The tradeable sector is generally defined as agriculture and allied activities, mining and manufacturing, with the non-tradeable comprising primarily services, along with construction and trade. The tradeable sector also increased its value-added share in the economy, but this expansion was not enough to prevent a fall of its employment share.
Figure 4.Estonia: Trends by Sectors’ Tradability
On the other hand, there is no close link between how much firms sell abroad and how much their productivity grows (Figure 5). Using sectoral data on export shares, which is available at the 4-digit NACE industry level from Statistics Estonia, each firm is placed into one of three buckets according to the export orientation of the industry it belongs to: low, medium, and high share of sales to non-residents in total turnover.7 Average labor productivity declined in all three categories: the decline was 19 percent for firms with low export orientation, 2 percent for firms with medium export orientation, and 1 percent for highly export-oriented firms. This would suggest a boon of export orientation for productivity growth. However, TFP increased for both low and high export oriented firms, but declined for medium export oriented firms, suggesting not systematic link between export orientation and productivity. Value-added and employment shares increased for firms with low-export orientation, fell for those in the medium category, and did not change much in those with high export orientation.
The lack of clear-cut evidence for superior productivity performance of exporters may reflect a variety of factors. It may not show in the exercise based on foreign sales, because it might not be the amount of actual foreign sales, but exposure to foreign competition that disciplines firms into constantly working on productivity improvements. Moreover, it may be exaggerated in the exercise based on belonging to the tradeable or nontradable sector. It might be the nature of the activity of a firm rather than the exposure to stiff foreign competition that is the crucial factor, e.g. manufacturing may generally be more conducive to productivity gains than services. This highlights the more general problem of not controlling for common factors in this analysis.
Figure 5.Estonia: Trends by Export Orientation
9. Composition effects played only a small role in aggregate productivity developments. Had the employment distribution across firms with different degrees of technological sophistication remained the same as in 2005, aggregate labor productivity and TFP would have been around 2.4 percent and 1.1 percent lower in 2014, respectively. A stable employment distribution across firms of different sizes would have raised labor productivity and TFP by 2 percent and 1.1 percent, respectively. Composition effects are even smaller for the dissections according to economic activity and degree of involvement in international trade.
B. Firm Characteristics and Productivity Growth
10. A difference-in-means approach is used to assess the relationship between TFP developments and firm characteristics. Firms are ranked according to their growth in productivity and divided into three buckets. The averages of the top and bottom buckets are then used to explore the differences in firm characteristics for firms belonging to the different groups, i.e. firms that saw the highest increase in TFP vis-à-vis firms that saw the least. The following firm characteristics are examined: initial productivity and performance metrics; firm size and age; and worker skill level, capital intensity, and export orientation. Two periods are considered that roughly correspond to the boom years 2005–09 and the crisis and recovery years 2010–14. The results below are presented in terms of the percentage of the average value (across all firms) of the firm characteristic. The differences are also examined for statistical significance at the 5 percent level.
11. There is evidence of a “catching-up” effect (Figure 6). Firms belonging to the group with higher productivity growth, in both periods, had on average 25 percent lower TFP in the beginning of the period relative to the firms that saw a relatively smaller increase in productivity. The figure was even larger at over 50 percent for labor productivity. The differences are statistically significant. This implies that there was an element of “catching-up,” with firms that had low initial productivity improving their performance relatively more than firms that were already more productive. This “catching-up” hypothesis is also borne out by other performance metrics, such as return on assets and profit margins. The return on assets of firms that showed the greatest increase in productivity was around 70 percent lower in the 2010–14 period, while it was around 40 percent lower in the 2005–09 period. Similar results hold for profit margin—in the 2010–14 period, the profit margin of firms belonging to the group showing higher increases in TFP was only about a fourth, and around half in the 2005–09 period. It should be cautioned however that this analysis relies on firms that stayed in business through the crisis and reported balance sheet data over the entire period under consideration, introducing an unavoidable survivorship bias.
Figure 6.Estonia: Drivers of Productivity
12. Evidence on the relationship between firm size and productivity growth is mixed. Firms that increased productivity most were smaller in value-added terms compared with firms that saw smaller increases in productivity. In the 2010–14 period, firms that did well in terms of TFP growth were around a third smaller, while in the 2005–09 period they were almost half as large. However, the opposite was true when size is measured in terms of employment. Firms that increased productivity most were around 20 percent larger in terms of number of employees in the 2010–14 period. However, the difference was not statistically significant in the 2005-09 sample. Moreover, both groups of firms were smaller than average, i.e. the largest of the firms, in terms of employment, were not part of either the group of firms that had the highest TFP growth nor the group with the lowest TFP growth.
Sources: Orbis; and IMF staff calculations.
13. The role of firm age changed over time and firms with lower average labor costs saw larger productivity gains. The average firm that belonged to the high TFP growth group was older in the 2010–14 period, but was younger in the 2005–09 period. The difference in firm age is statistically significant. While younger firms are typically expected to be more dynamic, innovative, and faster growing, as seen in the 2005–09 sample, it is possible that in the environment of relatively higher uncertainty in the post crisis period, firm maturity and track record were assets that the older firms were able to exploit. This could be interpreted as reduced economic dynamism. Interestingly, average labor costs—a proxy for worker skill level—was lower for firms belonging to the high productivity growth group. It appears that firms that did well relied more on lower-skilled labor.
Sources: Orbis; and IMF staff calculations.
14. Firms that witnessed greater TFP growth were more labor intensive. In both periods, firms that increased productivity most had smaller assets per employee—roughly 2/3—compared with firms that saw the least increase in productivity. This result is reinforced by the employee cost shares, which was higher in firms with higher TFP growth. These, taken together with the earlier result of lower skill level in firms that belonged to the high TFP growth group, suggests that TFP growth was probably primarily driven by the more traditional industries.
Sources: Orbis; and IMF staff calculations.
15. In line with the above findings, export orientation appears to have played a limited role. Export orientation, measured again at the 4-digit NACE industry level, did not play a major role. The difference in export orientation between firms that increased productivity most compared to firms that saw the smallest increase in productivity was statistically insignificant in the 2010–14 period, while in the 2005–09 period, more export-orientated firms saw a smaller increase in productivity.
Sources: Orbis; and IMF staff calculations.
16. Regression analysis generally confirms the above results of the difference-in-means approach (Figure 7). Results continue to hold in univariate regressions, where the drivers of TPF growth are introduced one at a time. The results also go through in a multivariate setting, which controls for other included variables. Initial productivity and performance metrics have the largest impact, followed by average labor cost and capital intensity. Younger firms tend to be more productive, but this effect is weaker in the 2010–14 period. Belonging to the high-tech category and export orientation, albeit the latter with a small magnitude, were positively associated with productivity growth in the multiple regression setting (see Annex for detailed results).8 In the multivariate setting, export orientation has a slightly positive association with productivity growth.
Figure 7.Estonia: Drivers of Productivity: Regression Analysis
C. Firm Characteristics and Employment Growth
17. A difference-in-means analysis is again employed in assessing which firm characteristics are conducive to employment generation. Again, firms are ranked and grouped into three buckets, but this time based on their change in employment. Thus in this section, the average characteristics of firms in the top and bottom buckets in terms of employment generation are compared to identify firm-level drivers of employment trends.
18. More productive firms saw greater increases in employment. In contrast to productivity growth, which was higher in firms that had low initial productivity, firms that increased employment the most were more productive—labor productivity of such firms was around 40 percent higher in the 2010–14 period and they were nearly twice as productive in the 2005–09 period when compared with firms that increased employment the least. The same holds true for other performance metrics, like return on assets, return on equity, and profit margins—firms that generated the most employment were significantly stronger (Figure 8).
Figure 8.Estonia: Drivers of Employment
19. Firms that increased employment more were larger and younger (Figure 9). The size of firms witnessing the greatest increase in employment was larger—around 2.5 times, in terms of value added, and around 70 percent, in terms of employment in the 2010–14 period. Such firms were also younger on average, and the difference in age was statistically significant. Compared with the results for TFP growth, it appears that although the larger and younger firms were increasing employment most, they were nevertheless amongst the group of firms that were at the bottom from regarding TFP growth.
Figure 9.Estonia: Characteristics of Employment Generators
20. Firms that employed higher-skilled labor, were more capital intensive, and more export oriented, increases employment more. Once again in contrast to the TFP growth results, firms that increased employment more had higher average labor costs, implying that they relied on higher-skilled labor. In the 2010–14 period, the average labor cost of the group of firms showing the greatest increase in employment was roughly a quarter higher than those that saw the smallest increase in employment. The difference was even larger, at close to 50 percent, in the 2005–09 period. Firms increasing employment most were also more capital intensive, with significantly higher assets per employee and a lower (close to 25 percent, both before and after the crisis) employee cost share. Finally, firms with larger employment increases were more export oriented.
21. Regression analysis again broadly confirms the findings of the difference-in-means approach (Figure 10). Firms that increased employment, in order of importance, were younger, had higher average labor costs, were more capital intensive and had better performance ratios. Employment growth was also significantly lower in the 2010–14 period for high-tech firms.
Figure 10.Estonia: Drivers of Employment: Regression Analysis
22. The effects of firm characteristics on employment growth and productivity growth were diametrically opposed in most cases. Firm characteristics that were associated with superior productivity performance were also associated with relatively low employment generation. This is true for all firm characteristics, except perhaps for firm age, where the effect on productivity is inconclusive.
|Firm characteristics||Productivity Growth||Employment Growth|
|Average labor costs|
23. Overall, catching-up and improvements in traditional firms were the main drivers of productivity growth. Catching-up, reflected in the important role played by initial productivity and performance metrics, was the main determinant of productivity growth. There is also evidence of strong productivity gains in the more traditional manufacturing sectors that was the backbone of aggregate productivity trends. Furthermore, TFP growth was generally not particularly driven by high-technology firms.
24. There are tentative signs of reduced firm dynamism. Firm age, which can be thought of as a proxy for firm dynamism, became less important for TFP growth in the period since 2010. While younger firms are typically expected to be more dynamic, innovative, and faster growing, as seen in the 2005–09 sample, it is possible that in the environment of relatively higher uncertainty in the post crisis period, firm maturity and track record was an asset that the older firms were able to exploit.
25. Labor rationalization, which seems to be particularly relevant in the case of Estonia, was a main source for productivity growth.
Firms, which increased productivity the most, saw a lower increase in employment via-a-vis firms that increased productivity the least in the period 2010–14.
The labor productivity-to-employment share elasticity, calculated using Eurostat macro-level data for the years 2005 and 2015 for EA12 and Estonia, indicates that for a given percentage change in the employment share, the relative change in labor productivity in the opposite direction is larger in every sector for Estonia. In line with the results above, the sectors displaying the strongest co-movements are the more traditional ones.
Further research is needed to confirm these results and understand better why Estonia stands out.
Sources: Orbis; and IMF staff calculations.
GalPeter N.2013 “Measuring Total Factor Productivity at the Firm Level using OECD-ORBIS” No. 1049 (Paris: Organization of Economic Cooperation and Development).
GilhoolyBob2009 “Firm-level Estimates of Capital Stock and Productivity” Economic and Labor Market Review Vol. 3 No. 5 pp. 36–41.
|Percentage of Employment|
|Percentage of Value Added|
|Number of firms with data on Employees|
|Number of firms with data on Value Added and Labor Productivity|
|Number of firms with data on Total Factor Productivity|
|Average Firm Characteristic||Change in Total Factor Productivity|
|Top third||Bottom third||Difference||t-stat||Top third||Bottom third||Difference||t-stat|
|Total factor productivity||2.54||3.41||−0.87||−37.89||2.64||3.48||−0.83||−32.89|
|Return on equity||−12.48||26.73||−39.20||−23.59||7.53||34.86||−27.34||−18.04|
|Return on assets||−0.31||16.46||−16.77||−36.35||6.32||21.39||−15.07||−30.09|
|Size (Value Added)||1245.56||1955.05||−709.48||−4.46||1715.02||3191.32||−1476.30||−5.90|
|Size (Number of employees)||8.77||7.31||1.45||2.63||13.37||12.92||0.45||0.32|
|Assets per employee||75.82||113.76||−37.94||−2.74||49.12||79.99||−30.87||−2.52|
|Employee cost share||24.31||21.81||2.50||6.93||23.08||18.02||5.06||12.90|
|Average cost of employees||8.18||9.83||−1.65||−7.94||5.63||7.57||−1.94||−11.07|
|Average Firm Characteristic||Change in Employment|
|Total factor productivity||3.12||2.92||0.19||10.36||3.26||2.87||0.39||15.93|
|Return on equity||11.10||0.29||10.81||9.23||27.33||11.46||15.87||11.24|
|Return on assets||9.06||5.31||3.76||11.15||15.30||9.36||5.93||13.55|
|Size (Value Added)||2978.75||1190.75||1788.01||10.69||3428.06||2235.26||1192.80||4.47|
|Size (Number of employees)||13.34||7.85||5.49||6.31||16.50||15.46||1.03||0.83|
|Assets per employee||134.97||111.79||23.18||1.94||142.44||47.09||95.34||7.56|
|Employee cost share||20.32||26.51||−6.19||−21.97||18.27||23.05||−4.78||−13.09|
|Average cost of employees||9.67||7.83||1.84||14.12||7.77||5.25||2.52||19.24|
|TFP Growth||Change in Employment|
|Total factor productivity||−0.3889***||−0.4012***||0.0213***||0.0791***|
|Return on assets||−0.0146***||−0.0166***||0.0018***||0.0027***|
|Return on equity||−0.0030***||−0.0036***||0.0004***||0.0007***|
|Number of employees||0.0002||0.0003*||−0.0002***||−0.0005***|
|Assets per employee||−0.0097***||−0.0074***||0.0032***||0.0099***|
|Labor cost share||0.0033***||0.0081***||−0.0052***||−0.0040***|
|Average cost of employees||−0.0081***||−0.0201***||0.0036***||0.0142***|
|Share of exports||0.0931**||−0.0194||0.1194***||0.0662**|
|Total factor productivity||−0.3506***||−0.3331***||−0.0134***||0.0339***|
|Return on assets||−0.0085***||−0.0099***||0.0020***||0.0015***|
|Average labor cost||0.0070***||0.0012||0.0050***||0.0121***|
|Assets per employee||−0.0089***||−0.0155***||0.0050***||0.0166***|
|Share of exports||0.2166***||0.0605||0.1060***||0.0115|
|Number of obs.||14767||10543||16896||12321|
Prepared by Pragyan Deb with contributions from Andreas Tudyka.
The unweighted average of labor productivity across firms declined over the period.
EU12 refers to the 12 countries that made up the European Union prior to the eastward expansion starting from2004.
Since Orbis data is in nominal terms, real values are obtained using industry level value added and investment deflators available from Eurostat. As a robustness check, the deflators were also de-trended using the Christiano-Fitzgerald time-series filter (at 2 years), which yielded very similar results.
Following Gal (2013) some of the variables are imputed when missing. Specifically, when data on value added is missing, it is imputed using EBITDA and cost of employees. In addition, total asset is used as a proxy when data on (tangible) fixed asset is not available.
Basic services cover public administration, education and health services, and other administrative and support services. Market services include transportation, accommodation, professional, ICT, and financial and real estate services.
When data is not available at the 4-digit NACE level, the 2-digit NACE is used as a fallback. Export orientation, while an improvement over the tradeable and non-tradeable breakdown, is still a crude proxy. A majority of the export orientation data was only available at the two-digit industry level and therefore does not pick up differences in export orientation within a particular (two-digit) industrial sector. Therefore, it is possible that within a particular sector, firms actually involved in exports performed better. In addition, this data does not capture the role of warehousing. A firm selling its products to a domestic warehousing company, which in turn exports the product, will be picked up as a domestic sale in our data. Therefore, some of the firms and sectors may be misclassified in the low export orientation category.
The multivariate specification addresses the question whether being classified as high-tech has an impact on productivity growth compared to all other firms in the sample, while the approach in section B investigates the effect of being high-tech within manufacturing and services.