1. Building a Firm level-based productivity indicators dataset: The experience of CompNet
Assolombarda
Milano
23 June 2017
Filippo di Mauro
National University of Singapore
Business School
Chairman of CompNet

2. Intro
Introduction
Firm level data are becoming increasingly available and are of better quality
An important European projects – CompNet – has created a new data base which is also increasingly accurate for the CROSS country comparison
This makes the new data very relevant for systematic policy analysis in addition to research
Will provide short description of the dataset, its construction, and some main results to illustrate its potential use

3. The rational of firm-level perspective
Firm performance distribution is very disperse and asymmetric
Rather than most firms around an “average” performance, there are lots of firms which have low productivity and only a few which are very productive in the “right-tail” of the distribution (the so called “happy few”)
Evolution of labor productivity distribution in France Manufacturing sector - firms with 20+ employees
Why do economists care about firm heterogeneity?
Because they care that resources (capital and labour) are reallocated from low to high productive firms, in order to increase the economy aggregate performance 

4. Firm performance is heterogeneous and asymmetric
Dispersion of firm productivity within sectors ( )
(productivity ratio of top 10% relative to bottom 10% firms in same 2-digit industry)
Firm productivity distribution in manufacturing ( )
We know this since the early 90s when longitudinal firm-level datasets, above all in the States, started to become available that firms are very different, even if they operate within narrowly defined sectors. You can see that in the left-hand chart showing the productivity level of the top 10% productive firm sin a given sector relative to the bottom 10% firm sin the same sector, and then aggregated to the macro sector and averaged across countries. This ratio is more than 2 across all sectors, and about 5 in new EU countries. In contrast, to give you a reference, manufacturing firms are in average about 30% more productive than service firms. This huge dispersion is a fact in all sectors, countries and periods.
But not only productivity is disperse, it is also very skewed, with a large mass of low productive firms and very few highly productive firms.
The countries in the chart are: old- BE, DK, DE, SP, FR, IT, AT, PT, FI; new- CZ, EE, HR, LV, LT, HU, PO, RO, SI, SK.
Source: ECB staff calculations based on CompNet data.
Notes: Computed at the 2-digit level, aggregated to the macro-sector with VA shares. Unweighted average across 19 countries.
Data refers to the 20E sample.
Sources: ECB staff calculations based on CompNet data, Eurostat data and Statistical office of Germany – AFiD-Panel data for Germany.
Note: Re-scaled so the mean of the distribution equals GDP per capita.
Data refers to the 20E sample.

5. Implications for research and policy
The use of firm-level data provides a more reliable picture of the performance of the economy because:
Aggregate indicators alone (say Labour costs) when interpreted as if they had been generated by the behavior of a representative firm, risk to give partial (if not wrong) messages
In reality firms are very different to each other and it is the specific structure/configuration of such firms that matters for the overall performance of the economy
For instance, is the economy mostly characterised by small firms, by exporting firms, are these firms financially solid, are their labour costs under control and so on?
……Are you being provided with this kind of information?

6. The Micro-founded Competitiveness Research Network (CompNet) dataset
We use existing (no new surveys) firm-level data, mostly from business registers, to construct a wide set of relevant business indicators (productivity, costs, employment..)
ECB computational engine
micro data
country teams expertise
Common codes to aggregate indicators at industry, macro-sector and country level in order to solve confidentiality issues
Common methodology to harmonize the resulting set of indicators across countries in terms of measures definition, treatment of outliers, deflators (based on Eurostat sectorial value added) and PPPs.

7. In addition, however, we have
The Micro-founded Competitiveness Research Network (CompNet) dataset
We have about 20 Country teams, from National Central Banks and Statistical Institutes
Our aim is to have a Minimum Common Set of indicators which are also cross-country comparable
At the Center, we do not see raw firm level data (confidentiality is ensured), but sector averages
In addition, however, we have
 the full distribution for the (more than 70) critical business variables we collect
 Most notably, the database includes more than 300 joint distributions linking different firms’ characteristics

8. Five broad categories of variables are available…

9. Example of joint distributions
Example type of question:
Are low productive firms in a country-sector
characterized by higher credit constraints?

10. Major improvements in the data coverage(1) # of firms: 83%
The 5th wave of the CompNet database
Poland
France
Italy
Spain
Portugal
Germany
Romania
Finland
Estonia
Denmark
Czech
Republic
Slovakia
Malta
Latvia
Croatia
Belgium
Major improvements in the data coverage(1)
# of firms: %
# employees: 85%
Time period: 1995 – 2013
Geographical coverage:
16 EU countries
Many more are working to join:
New Zealand, UK,
Switzerland …
(1): years and countries available in the last round in comparison with Eurostat

11. Some examples of the results

12. …..Most productive firms are not credit constrained
Share of credit contrained firms by deciles of labor productivity ICC index estimated within CompNet
Least productive
Most productive
We match survey results to financial fundamentals. We use the parameters to evaluate whether firms are financially constrained
Firms in stressed countries (Spain, Italy) were more affected by the crisis, particularly the least productive firms

13. Exporters Productivity Premium
How much is important the external dimension?
Exporters Productivity Premium
Being an exporter is associated with higher levels of productivity (about 20%)
Export productivity premium is highly heterogeneous, it varies sharply across countries
During the crisis, the drop in productivity has been more pronounced for exporters
They seem to be more vulnerable to macro-economic shocks

14. Exports are concentrated to most productive firms – and these are few
Exports by productivity decile
(average share of exports as % of total by productivity decile)
Exports concentration ( )
(share of exports of top 5 or 10 exporting firms as % of total)
Since the seminal paper of Melitz, in 2003, it has also been established that only the most productive firms in a given sector are able to pay the fixed and variable costs of exporting as you can see in the left-hand chart where the share of total exports by productivity decile. This explains that exports are extremely concentrated in few firms, the top 5 exporters in a given country account for between 40 and 70% of total exports. The work we have done in this area includes showing these facts, and possible implications, for the European Union, using comparable cross-country data and a consistent framework for analysis. The March issue of the EB will show our main work on this area.
Source: ECB Economic Bulletin, March 2017, based on CompNet data
Note: Average across 15 EU countries: BE, EE, FR, IT, LT, LI, PT, SI, SK, FI, CZ, DK, HR, PO, RO.
Data refer to the 20E sample, adjusted threshold
Source: ECB staff calculations based on CompNet data.
Note: Average across 15 EU countries: BE, EE, FR, IT, LT, LI, PT, SI, SK, FI, CZ, DK, HR, PO, RO.
Data refer to the 20E sample, adjusted threshold 14 14

15. → The impact might change with respect to
Examples of Policy relevant research: VOX EU December 2015
Do exchange rate devaluations work?
→ The impact might change with respect to
Firm size
Firm productivity (TFP)
Important implications on aggregate export performance:
The impact seems to be limited in the short run, as largely determined by the (low) reaction of the largest and most productive firms
Exchange rate devaluation can be very effective in helping more vulnerable exporters to succeed in international markets
Numbers: export elasticities to currency depreciation
Source: Berthou, A. and di Mauro, F. (2015): “Exchange rate devaluations: When they can work and why”, VOX.EU, 24 December.

16. Labour cost and productivity dynamic in Croatia and France
Index (2003 = 1)
ULC
Productivity
Labour cost
Productivity
ULC
L productivity (2003=1)
L cost (2003=1)
ULC (2003=1)
High productive firms:
L productivity (2003=1)
L cost (2003=1)
ULC (2003=1)
Low productive firms:
In the last decade, there has been a growing gap between labour productivity and lab our cost … but for different reasons:
In Croatia, overall labour productivity stagnated or declined after the crisis a
→ see ULC of all firms increasing
In France, costs were diverging, but only for the least productive firms (see hystograms )
a → see ULC of most productive firms remaining subdue

17. Profitability (ROA) and productivity distributions
Most productive
Most productive
Median
Median
Least productive
Least productive
In 2009, there has been a sharp
decline in Returns on Assets (ROA)
Most productive
Low productive firms were the most affected by the crisis
Median
In 2013, all firms recover in their
profitability
Least productive

18. Conclusions
There are well established determinants for higher firm productivity
This includes financial health, labor costs, size of the firm, as well as its export status
Firm level data show that the relative importance of such determinants varies tremendously across countries and sectors
Correctly formulated policies must incorporate such information
CompNet database represents a value-added in terms of coverage and comparability of firm-level data among European countries but also in terms of availability of new indicators jointly related to productivity

19. Thanks for your attention