dsti/eas/ind/swp/ah(2001)14 working party on statistics · dsti/eas/ind/swp/ah(2001)14 working...

For Official Use DSTI/EAS/IND/SWP/AH(2001)14

Organisation de Coopération et de Développement EconomiquesOrganisation for Economic Co-operation and Development 20-Nov-2001___________________________________________________________________________________________

English text onlyDIRECTORATE FOR SCIENCE, TECHNOLOGY AND INDUSTRYCOMMITTEE ON INDUSTRY AND BUSINESS ENVIRONMENT

Working Party on Statistics

THE SOURCES OF PRODUCTIVITY GROWTH: MICRO-LEVEL EVIDENCE FOR THE OECD

WORKSHOP ON FIRM-LEVEL STATISTICS, 26-27 NOVEMBER 2001

Session 4: Measuring Productivity at Firm Level

This paper was prepared by Matthew Barnes (Queen Mary, University of London and Office for NationalStatistics), Jonathan Haskell (Queen Mary, University of London, Office for National Statistics and CEPR) andMika Maliranta (The Research Institute of the Finnish Economy, ETLA and Statistics Finland). The paperrepresents the views of the authors and does not necessarily reflect the opinions of the affiliating institutions orthe OECD.

Contact: Dirk PILAT; Tel: +33 1 45 24 87 49; Fax: +33 1 44 30 62 58; E-mail: [email protected]

JT00116958

Document complet disponible sur OLIS dans son format d’origineComplete document available on OLIS in its original format

DST

I/EA

S/IND

/SWP

/AH

(2001)14F

or Official U

se

English text only

DSTI/EAS/IND/SWP/AH(2001)14

2

PRELIMINARY

THE SOURCES OF PRODUCTIVITY GROWTH:MICRO-LEVEL EVIDENCE FOR THE OECD*

Matthew BarnesQueen Mary, University of London and Office for National Statistics

Jonathan HaskellQueen Mary, University of London, Office for National Statistics and CEPR

Mika MalirantaThe Research Institute of the Finnish Economy, ETLA and Statistics Finland

November 2001(Initial version: March 2001)

Paper updated for the OECD Workshop on Firm-Level Statistics, OECD, Paris,November 26-27 2001

Abstract

We document the decomposition of productivity growth in OECD member states using the early data fromthe OECD Firm Level Study (Further analysis will follow using updated data). Our results suggest asubstantial role for within effects in most countries. The entry and exit effects seem to vary withinstitutions. Considering employment protection, relative to the US, other countries typically have exitorswho are more productive than average, lowering the exit contribution to productivity growth. But entrantsare also more productive than average, raising the entry contribution to productivity growth. In addition wedocument current developments of the OECD Firm Level Study.

* Contact addresses:

Matthew BarnesDepartment of EconomicsQueen Mary, University of LondonMile End RoadLondon E1 4NS, UKEmail: [email protected]: www.qmul.ac.uk/~ugte193/

Jonathan HaskelDepartment of EconomicsQueen Mary, University of LondonMile End RoadLondon E1 4NS, UKEmail: [email protected]: www.qmul.ac.uk/~ugte153/

Mika MalirantaThe Research Institute of theFinnish Economy, ETLALönnrotinkatu 4 BFIN-00120 HelsinkiFinlandEmail: [email protected]

For financial support Barnes and Haskel thank the Leverhulme Trust for grant F/07476A. UK participation is part ofthe UK Office for National Statistics Business Data Linking Project. Maliranta thanks the National TechnologyAgency (TEKES) for grant DNRA 1312/401/99, part of the project: “Creative destruction as a source of economicdevelopment”.

We thank Paul Geroski, John Haltiwanger, Jacques Mairesse, Steve Nickell and participants at the OECD TechnicalMeeting on the Firm Level Study, January 2001, Phil Hemmings and Dirk Pilat for useful comments. Thanks also toStefano Scarpetta, Eric Bartelsman and all participants in the OECD Firm Level Study for all their effort in makingthis work possible. The data has been processed in each country using programs written by the authors of this paper,so programming errors and errors and opinions in this paper are our own. Opinions expressed in this paper are notnecessarily those of the Office for National Statistics, HM Government, Statistics Finland or the Organisation forEconomic Cooperation and Development.


3

1. Introduction

1. It is now well acknowledged that productivity can grow in two ways (see e.g. Bartelsman andDoms (2000) and Foster, Haltiwanger and Krizan (1998)). First, productivity can grow due to changeswithin existing enterprises, such as the introduction of new technology and organisational change. Second,productivity can grow due to the process of market selection whereby low productivity establishments exitand are replaced by higher productivity entrants, while higher productivity incumbents gain market share.This note reports some cross-country evidence on these two effects. It summarises the work done in anumber of countries under the auspices of the OECD Firm-Level Study co-ordinated by Eric Bartelsman.The data used here are from an early release and are preliminary, further analysis using new data willfollow. See Appendix 1 for a list of participants. Bartelsman, Scarpetta, and Schivardi (2001) summarisethe demographics and survival analysis components of the project and OECD (2000, 2001a) for moredetails about the project and the OECD’s broader “Sources of Growth” project of the Firm Level Study is apart.

2. Method

2. Write sector-wide productivity in year t, Pt as

P pt it iti= ∑ θ (1)

where θi is the employment share of establishment i and Pt and pit are a productivity measure (which maybe labour or TFP). The decomposition proposed by Foster, Haltiwanger and Krizan (1998) (FHK) 1 is interms of productivity relative to the average:

( )

( ) ( )

t it k it it it k t k it iti S i S i S

it it t k it k it k t ki N i X

P p p P p

p P p P

θ θ θ

θ θ− − −∈ ∈ ∈

− − − −∈ ∈

∆ = ∆ + ∆ − + ∆ ∆

+ − − −∑ ∑ ∑

∑ ∑FHK (2)

where S, N and X denotes the establishments that survive, enter and exit respectively between t and t-k.The first term shows the contribution to ∆P of growth within the surviving establishments; the “within”effect. The second term shows the contribution of changes in shares of the survivors weighted by finalperiod productivity relative to the average (often termed the “between” effect). It is positive when marketshares increase for those survivors with above-average base year productivity. The third term is acovariance term that is positive when market share increases (falls) for establishments with growing(falling) productivity. The entry and exit terms are positive when there is entry (exit) of above- (below-)average productivity establishments.

3. As FHK point out however, problems might occur with measurement error. The decompositiondue to Griliches and Regev (1992) (GR) overcomes this

( )

( ) ( )

t i it it ii S i S

it it it k ii N i X

P p p P

p P p P

θ θ

θ θ∈ ∈

−∈ ∈

∆ = ∆ + ∆ −

+ − − −∑ ∑

∑ ∑GR (3)

1 Bartelsmann and Doms (2000) also use this equation but date the exitors productivity in t; this appears to be a

typo.


4

where the bar indicates a time average over the base and end year. The first term measures the withincontribution of survivors’ productivity growth weighted by time-average market shares. The other termsare all relative to time-average productivity. The advantage of this procedure is that averaging removessome of the measurement error. The disadvantage is that interpretation is more obscure. The within effectwill reflect, to a certain extent, a between restructuring effect since it affects θ.

4. We calculate P in a number of different ways. First, P is real gross output per head (Y/L).2 Wedo not have satisfactory cross-country data on industry hours. L is calculated as employment, where wecannot distinguish between part and full time. Second, where data are available, P is calculated as TFP.TFP is the change in gross output less the share weighted changes in materials, capital and labour inputs.The changes are calculated at the micro level and the shares are the shares of industry costs (industryshares are used to avoid measurement error). The industry cost of capital is calculated as the residual fromlabour and material costs. However due to difficulties in some countries obtaining data for materials, ordeflators for materials, the main TFP measure we use excludes materials.3 Hence the industry cost ofcapital is the residual from the labour share.

5. How would be expect the components of (3) to vary? Over time, FHK (for the US) and Disney,Haskel and Heden (DHH) (2000) for the UK both find evidence that the net entry rises in recessions (as themarginal entrant is a good plant) and falls in booms, with the within effect the reverse.

6. Of particular interest in the current setting is how one might expect (3) to vary across countries.Let us start by supposing that in the absence of institutions there is potentially an equal distribution ofproductivity across every nation (including productivity of potential plants). Suppose then that the EU hashigh wage floors as a result of high minimum wages and high benefits. This is clearly too crude sincethere are substantial differences across the EU, but we maintain this fiction for illustration. Then we wouldexpect the US to have a long tail of unproductive plants which are not observed in Europe.4 Regardingentry, entering plants in Europe would have higher p than entering plants in the in the US. Other thingsequal this would raise the entry effect in Europe. But there would be fewer entrants. If all firms had equalmarket shares, this would lower the entry effect, but if the firms who are more productive have highermarket shares this which attenuate this second effect. Thus overall is it ambiguous, but one might expectthe EU to have a higher entry effect. Looking at (3), for example, ΣN θ(p-P) might even be positive sinceonly very good firms with p>P could enter in Europe. In the US poor firms can enter, and so p<P,rendering ΣN θ(p-P) potentially negative.

7. Turning to the exit effect, exiting plants in Europe would have higher p than exiting plants in thein the US. If we expect p<P for exiting plants then p-P for the US would be a large (in absolute value)negative number as only the very poorest firms exit. In Europe it would be closer to zero, perhaps evenpositive. Thus the exit effect in Europe would be lower. Against this however, there would be moreexitors in Europe which would raise the European exit effect in absolute number. In sum, the prediction isambiguous, but if the second effect is small we would expect Europe to have a lower entry effect i.e. eitherpositive or negative but small in absolute value.

2 If gross output is not available we use sales or value added.3 For the UK and Finland TFP including materials was also used. This is termed MFP for clarity.4 This would suggest that P is higher in Europe, which seems not be true (O’Mahony (1996)). One possibility is

that countries have very different distributions, which makes cross country predictions of the entry and exiteffects very difficult. The other is that countries have the same distribution, but the mean is shifted by country-specific factors (culture etc.). Further study of this awaits more usable data on productivity distributions.


5

8. In the above example a floor is placed on wages with the rest of the distribution unaffected. Theother possible effect of more labour market regulation, unions, taxes etc. is that European wage levels areraised relative to what they might have been. This would cause firms to substitute capital for labour. Thusplants would be more capital intensive in Europe, which would raise labour productivity. The effect onproductivity relative to the average is hard to sign.

3. Data

9. Our objective is to undertake decompositions for as many countries, periods and industries aspossible. The table below shows countries, years and broad industries for which we have usable data andsome brief comments on features of the data, to which we return below.

Summary of the data used for productivity decompositions

Periods Coverage Productivity

Country First Last Mfg Serv LP TFP MFP Unit Method Threshold

Finland 75-80 89-94 � � � � � Estab C Emp>5

France 85-90 90-95 � � � � � Firm RTurnover

��

Germany (W) 92-97 93-98 � � � � � Plant S (PW) Emp>1

Italy 82-87 93-98 � � � � � Firm S Turnover ��

Netherlands 83-88 92-97 � Some � � � Firm S+R (PW)Emp>20,

emp<20→S

Portugal 86-91 93-98 � � � � � Firm R Emp>1

UK 80-85 87-92 � � � � � Estab S+R (PW)Emp > 100,

emp<100→S

USA 87-92 & 92-97 � � � � � Estab C Emp>1Source: OECD Firm Level Study participants, OECD 2001b, see Appendix 2 for more details.Note: Netherlands: Services 87-92 to 91-96 only. Key to Method column: C: Census, R: Register, S: Sample, (PW):Population weighted. Most countries impose some restriction on the size of firm/plant that is included in their data. LPis labour productivity, TFP includes labour and capital inputs, and MFP includes labour, capital and materials.

10. The approach is to carry out FHK and GR decompositions as described in (2) and (3) for 5 yearrolling periods for all periods that are available for a particular country. Thus column 2 shows the first andlast 5 year periods available. The priority is to look at labour productivity5 in all available sectors.Primarily this means manufacturing, which has also been broken down according to the OECD’s STAN2000 classification. For some countries analysis has also been extended to cover services. As columns 7and 8 show, for some countries we have TFP which subtracts off labour and capital inputs and for someMFP which additionally subtracts off materials.

11. For the results to be credible the data and calculations have to be comparable. We comment onthe data below. Regarding the calculations, code was written by the current authors and then circulated toall countries. Thus all countries ran the same code. We tried to standardise the data in a number of waysin the code. First, we ensured that the data was cleaned in a consistent way to deal with missing orspurious observations. Second, observations in the top and bottom percentiles of the output variable (grossoutput or value added) were dropped to ensure any extreme outliers were removed. Third, data wereweighted by each country using sample-based data, where possible, to try to ensure results were

5 Defined as real gross output, real value added or real sales per employee.


6

representative of the population. Where possible information on the population was also used to ensureentries and exits due to sampling were eliminated.

4. Results: Comparison of shares

12. Table 1 presents results for FHK and GR decompositions of manufacturing labour productivityfor the participating countries available at time of writing. The results are for 1987-92 which is the mostrecent 5 year interval that maximises coverage across countries.

13. The table reads as follows. The top row shows the data from Finland. Reading across the table,column 1 in the first row shows that labour productivity growth over this period was 23.7%.6 The next cellshows the fraction of that productivity growth due to growth “within” firms, in this case 61%. The nextcells show that 22% is due to the between effect, -1% due to the cross effect and 18% due to the net entryeffect. Columns 6 and 7 then subdivide the net entry effect into the impact from entry, in this case 2% andthat of exit, here –15%. Note that the net entry effect is column 6 minus column 7. Thus if column 6 (7) isnegative, this signifies that productivity in the entering (exiting) plants is below average. So, if plants whoexit are below average productivity, we expect column 7 to be negative. The next panel shows the resultsfor the GR decomposition. They have the same pattern, except there is no cross term.

14. Panel (a) of Table 1 suggests the following. First, the within effect is substantial acrosscountries, accounting for not less than 60% of productivity growth (Italy) and 95% in France. Second, netentry is about a third of productivity growth in Italy, Portugal, the UK and the US. Third, the effects ofentry are generally positive, suggesting that entering firms achieve higher productivity than the base yearaverage in all countries bar France and the US. The exit effect is generally negative, suggesting that plantsexit with lower productivity than the average in all countries. Finally, the cross term is consistentlynegative, consistent with the FHK findings for the US. Since the weight is employment, this suggests thatmuch of the productivity growth was occurring in plants that were downsizing.

15. Before drawing any strong conclusions however, we must be sure that these results reflectunderlying economic differences and not just data problems. Starting with Finland, the data is based on aCensus and entry and exit is for establishments with at least 5 persons. Thus we expect it to be reasonablyrepresentative of the population. The French data is a register, and is at the firm level where firms have tohave at least 20 employees or more than �� below the official employment data and so the sample is not likely to be representative of the population.We would also expect larger firms to be oversampled which would likely lower the net entry effect andraise the within effect. This appears to be the case. Since the data are at firm level then entry and exit ofplants will be understated if occurring within multi-plant firms, but overstated if a merger is counted as anexit. The Italian data is also a sample requiring �� deleting firms falling below the threshold and adding new firms in. The observation is a firm. We wouldtherefore expect the Italian data to understate true entry and exit and entrants and exitors to have higherproductivity on average. Furthermore, the sampling rules are likely to over-record as exitors firms withlow productivity. The Netherlands data are based on register of firms with over 20 persons plus a samplefor under 20 (which is then weighted). The data is at the firm level. The coverage in Portugal is a register,and the are firm level, although splits and mergers are purged from the data (so that they do not constitute

6 These productivity changes do not all accord well with “official” OECD data, likely due to numerous

methodological differences. In particular the deflation method used and the definition of output (gross outputversus value added). Further differences may occur due to aggregation and population weighting issues. Thisbears further investigation.


7

entry or exit). The UK data is similar to the Netherlands, with partial sampling of smaller establishments,but weighting to try to overcome this. Finally, the US data is a Census at the establishment level.

16. Overall then, we would expect the data for Finland, Netherlands, Portugal, UK and USA to befairly comparable, but that it is somewhat harder to compare these with Italy and France. Thus the veryhigh within effect in France is likely to be due to the substantial overrepresentation of large firms, whilstthe entry and exit results in Italy are not likely to correspond to “true” entry and exit. Comparing Finland,Netherlands, Portugal, UK and USA therefore we see that the Netherlands has the highest within effect,and the UK the lowest.

17. An important feature of the data worth noting are the entry and exit effects. Leaving aside Franceand Italy, who have the data problems set out above, the US is notable in having the only negative entryeffect, suggesting that US establishments who enter become, on average, low productivity7. This is incontrast to the other countries who show a positive entry effect, suggesting that these entrants becomeabove average productivity establishments. This is in line with the argument set out above, namely thatpoor productivity firms can enter in the US but not in the EU. The exit effects are also interesting, with(again ignoring France and Italy) the US having a very high (in absolute terms) exit effect (-44 in thetable). This suggests that the exiting plants in the US are very much less productive than those in the EU.Interestingly then, the net entry effects do not look very different across countries, but the underlying entryand exit patterns do differ.

18. The next question is whether the results are sensitive to decomposition method. The lower panelof Table 1 shows the results using the GR method. The fraction due to the within effect is lower and thatdue to net entry is higher, as would be expected from the averaging procedure implicit in the GR method.Importantly however the ranking between countries of the contribution of each effect is almost unchanged.

19. Following from this, we know that the within component rises in booms whilst the net entryeffect falls in recessions (see FHK for the US, and Maliranta (2001) for Finland on this). Thus the resultsin Table 1 might be due to the fact that different economies are in different cyclical periods. We thereforerecomputed the results for 1985-90 which was, for almost all economies, an upturn. These results are setout in table 2, where due to data availability we are unable to perform a 1985-90 decomposition forPortugal and the US. The numbers are somewhat different with a generally higher within componentconfirming pervious findings. Interestingly however the inter-country rankings are hardly changed (withthe sole exception of the UK). Thus for example the Netherlands, who had the top ranked withincomponent in the 1987-92 decompositions have the second ranked in the 1985-90 decompositions.

20. Table 3 shows the results for a more recent 5 year split, 1992-97. Data for Germany is availablefor this period, but the UK, France and Finland are not available. German data is a sample, so there maybe some doubts about the entry and exit figures, although the data is population-weighted. Germany andthe US show a very high within effect and almost no net entry effect. The US and Netherlands also havevery high within effects. Note again the US has the biggest negative entry and exit effects.

21. Turning to individual industries, Tables 4 and 5 show data for the Textiles and ElectricalEngineering sectors respectively. One has to be more careful here since some sectors can be very small insome countries and hence data error or entry or exit of only a few firms can have a very large impact. Intextiles, leaving again France aside, the US is notable for the highest negative entry and negative exiteffect. In Electrical engineering (Table 5) the US has a positive entry effect however.

7 They could have higher than average productivity but some high market share firms might have lower than

average rendering the overall effect negative; we examine this by looking at each industry below.


8

22. We investigated the entry and exit effects further in a number of ways. First, to explore the USnegative entry and exit effect more, we used the within country decompositions by STAN and calculatedthe percentage of the industries in each country which had a negative entry and exit effect. As Table 6shows this was the highest in the US. Second, we calculated the percentage of 27 industries in 1992 and1997 where the US had the most negative entry and exit effect8. For 1992 the figures were 52% and 52%and for 1997 the figures were 63% and 44%. Thus we can be reasonably confident that the large negativeUS entry and exit figures are indicative of underlying economic behaviour.

23. Finally, we investigate the effects of legislation. We used the cross-country data reported inBlanchard and Wolfers (1999) which is an index of employment protection.9 Since we have USdecompositions for the 5 year intervals 1987-1992 and 1992-1997, Figure 1 plots the FHK share ofproductivity accounted for by the entry effect for the two 5 year intervals 1987-1992 and 1992-1997. Notethat we do not have data for the UK for 1992-97 and we have dropped France due to data worries. Thepoints in italics are the 1992-97 data and in regular font are 1987-92. One has of course to be cautioussince there are not many data points, but the overall relation does seem to be upward sloping, suggestingthat more employment protection leads to a more positive entry share effect. Hence firms entering with ahigher-than-average level of productivity. Figure 2 shows the same pattern for exit. In each case Portugal,Italy and the Netherlands have seen falls in employment protection that should raise entry and exit effectstowards being positive; this is the case in Portugal for exit and Italy for entry.

24. Overall then it would seem that restrictive hiring policies affect the contribution of entry and exitto productivity growth. The exit contribution is lowered as firms with higher productivity leave. Thisretards productivity growth. Against this the entry contribution is raised, as firms with better productivityenter. Of course, such a measure is not an ideal indicator of floors to wages, and indeed if the costs of suchpolicies are incident on the wage, then they may be neutral as far as entry and exit are concerned. Butinterestingly the hiring policies variable is correlated (coefficient 0.73) with the OECD product marketcompetition index (Nicoletti et al, 2000, figure 14). This index is however only available for 1998 which isafter most of our decompositions.

25. Finally, given that employment protection raises the entry effect but lowers the exit effect, figure3 plots the net entry contribution against the employment protection index. Once again with few datapoints one should be cautious, but overall the picture hints that these labour market institutions do not toaffect the net entry effect.

26. Returning to the decompositions, tables 7 and 8 show decompositions for TFP where availablefor 1987-92 and 1985-90. TFP is negative in a number of countries making interpretation difficult. Forthe countries for which one can make a reasonable comparison, the within effects do seem to be smallerand the entry effect larger.

27. Tables 10 to 15 show the results for productivity in services. We have rather few countries hereand data is not generally available for a large number of periods. Regarding all services, Tables 9 and 10shows very low or negative productivity growth in the countries for which we have usable data, suggestingmeasurement problems with service sector productivity. The following Tables therefore try to breakservices down into different sectors, where there may be more chance of better measurement. Tables 11shows data for wholesale, retailing and repairs. Productivity growth is generally positive here, with highentry effects for Germany. Transport etc. (Table 12) and Financial Intermediation (Table 13) show mixed 8 This number includes only those industries for which there are observations for the USA.9 The index is generated by Blanchard and Wolfers and is based on the OECD job protection index which is a

broad index of notice period, severance pay for a senior worker, difficulty of dismissal and availability oftemporary employment.


9

effects whilst Real Estate and Community services (Table 14 and 15) show negative productivity growthwhich suggests measurement problems.

5. Current work

28. At the time of writing a new round of data collection is under way. This round is using newlywritten code for the productivity component of the project that is aware of the numerous issues in the datathat is being used. We have designed this code (with Eric Bartelsman and working closely with all thenational participants) to be easily extendable and to maximise comparability. In addition we have addedBaldwin (2001) decompositions and further indicators to look at productivity distributions and correlates.In the Baldwin variants of GR and FHK, the aggregate productivity measure in the between, the entry andthe exit terms, are replaced by the average productivity of exiting firms in the base year. With thisadjustment, the exit term becomes identically equal to zero. The contribution of an entrant is positive if itis more productive than the average exiting firm it is thought to replace. The contribution of a growingincumbent is positive if the firm is more productive than the average exiting firm. See Appendix 3 forproductivity pseudo-code for the new run. One aim is to build towards a standard approach forundertaking such international collaborative projects using confidential data in the future as outlined inBartelsman and Barnes (2001).

6. Conclusions

29. We document the decomposition of productivity growth in OECD member states using the earlydata from the OECD Firm Level Study. Our results suggest:

a) A substantial role for within effects in most countries.

b) The entry and exit effects seem to vary with institutions. Considering employmentprotection, relative to the US, other countries typically have exitors who are more productivethan average, lowering the exit contribution to productivity growth. But entrants are alsomore productive than average, raising the entry contribution to productivity growth.


10

Table 1. Labour productivity decompositions, Manufacturing, 1987-1992a: FHK Method

1 2 3 4 5 6 7

CountryLabour ProdGrowth (5 yr) Within Between Cross Net Entry Entry Exit

Finland 23.7 61 22 -1 18 2 -15

France 4.2 95 23 -22 3 -37 -40

Italy 20.9 60 21 -14 33 32 -2

Netherlands 11.8 90 19 -25 17 41 24

Portugal 29.5 82 -4 -16 37 7 -30

UK 9.0 64 10 -8 34 21 -13

USA 8.1 85 31 -44 28 -16 -44

b: GR Method1 2 3 4 5 6

CountryLabour ProdGrowth (5 yr) Within Between Net Entry Entry Exit

Finland 23.7 61 21 18 -3 -21

France 4.2 85 11 4 -44 -49

Italy 20.9 53 12 35 17 -17

Netherlands 11.8 77 7 15 23 8

Portugal 29.5 75 -9 34 -7 -41

UK 9.0 60 3 37 9 -28

USA 8.1 63 8 29 -24 -53Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: The productivity growth rate quoted is over a five-year period. Method used is FHK in panel a (see equation (2))and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Manufacturing is defined asthe 2-digit ISIC sectors 15 to 37. All values in columns 2 through 7 (6 in b) are per cent of total change in column 2.Net entry = entry - exit: Figures may not add up due to rounding.


11

Table 2. Labour productivity decompositions, Manufacturing, 1985-1990*a: FHK Method

1 2 3 4 5 6 7 8

Country PeriodLabour ProdGrowth (5 yr) Within Between Cross Net Entry Entry Exit

Finland 1985 - 90 29.9 74 10 -4 19 5 -14

France 1985 - 90 10.4 89 9 -8 10 -13 -23

Italy 1985 - 90 27.8 65 14 -9 30 25 -6

Netherlands 1985 - 90 7.5 103 0 -6 3 48 45

Portugal 1987 - 92 29.5 82 -4 -16 37 7 -30

UK 1985 - 90 8.1 99 -4 -1 6 25 19

USA 1987 - 92 8.1 85 31 -44 28 -16 -44

b: GR Method1 2 3 4 5 6 7

Country PeriodLabour ProdGrowth (5 yr) Within Between Net Entry Entry Exit

Finland 1985 - 90 29.9 73 7 20 0 -20

France 1985 - 90 10.4 85 2 13 -20 -34

Italy 1985 - 90 27.8 60 8 32 11 -21

Netherlands 1985 - 90 7.5 100 -8 8 33 25

Portugal 1987 - 92 29.5 75 -9 34 -7 -41

UK 1985 - 90 8.1 98 -7 9 14 5

USA 1987 - 92 8.1 63 8 29 -24 -53Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: The productivity growth rate quoted is over a five-year period. Method used is FHK in panel a (see equation (2))and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Manufacturing is defined asthe 2-digit ISIC sectors 15 to 37. All values in columns 3 through 8 (7 in b) are per cent of total change in column 2. *Portugal and USA are 1987-92 as in table 1 due to data availability. Net entry = entry - exit: Figures may not add updue to rounding.


12

Table 3. Labour productivity decompositions, Manufacturing, 1992-1997a: FHK Method

1 2 3 4 5 6 7


Italy 25.8 71 15 -15 29 23 -6

Germany (W) 11.1 125 -2 -19 -4 1 -5

Netherlands 20.6 83 -2 -14 32 27 -5

Portugal 26.0 65 -5 3 38 12 -25

USA 15.2 94 15 -25 17 -6 -22

b: GR Method1 2 3 4 5 6


Italy 25.8 63 7 30 9 -21

Germany (W) 11.1 115 -12 -3 -1 3

Netherlands 20.6 76 -11 35 17 -18

Portugal 26.0 66 -7 40 1 -40

USA 15.2 81 1 18 -13 -31Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: The productivity growth rate quoted is over a five-year period. Method used is FHK in panel a (see equation (2))and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Manufacturing is defined asthe 2-digit ISIC sectors 15 to 37. All values in columns 2 through 7 (6 in b) are per cent of total change in column 2.Data for Germany are for the former West Germany only. Net entry = entry - exit: Figures may not add up due torounding.


13

Table 4. Labour productivity decompositions, Textiles, Textile Products, Leather & Footwear, 1987-1992

a: FHK Method

1 2 3 4 5 6 7


Finland 11.5 -7 37 7 63 -2 -65

France 3.1 17 118 -33 -1 -106 -105

Italy 20.1 51 25 -13 37 44 7

Netherlands 11.3 72 53 -37 13 26 14

Portugal 32.4 73 1 -2 29 6 -23

UK 5.8 74 -45 0 72 11 -61

USA 8.7 64 64 -47 19 -61 -81

b: GR Method1 2 3 4 5 6

CountryLabour ProdGrowth (5 yr) Within Between

NetEntry Entry Exit

Finland 11.5 -4 28 75 -5 -81France 3.1 0 98 2 -115 -117Italy 20.1 45 17 38 31 -6Netherlands 11.3 53 34 14 11 -3Portugal 32.4 72 2 26 -10 -36UK 5.8 73 -55 81 2 -79USA 8.7 40 38 22 -72 -94

Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: The productivity growth rate quoted is over a five-year period. Method used is FHK in panel a (see equation (2))and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Textiles, Textile Products,Leather & Footwear is defined as the 2-digit ISIC sectors 17 to 19. All values in columns 2 through 7 (6 in b) are percent of total change in column 2. Net entry = entry - exit: Figures may not add up due to rounding.


14

Table 5. Labour productivity decompositions, Electrical and Optical Equipment, 1987-1992a: FHK Method

1 2 3 4 5 6 7


Finland 44.4 58 -4 22 25 24 -1

France 6.5 82 12 -4 10 -3 -12

Italy 27.7 75 0 -9 34 44 10

Netherlands 23.6 136 -6 -31 2 1 0

Portugal 4.9 588 -76 -417 5 -23 -28

UK 24.7 42 12 4 42 42 1

USA 26.6 82 10 -12 20 11 -9

b: GR Method1 2 3 4 5 6


Finland 44.4 68 7 25 13 -12

France 6.5 80 5 15 -12 -27

Italy 27.7 71 -7 37 28 -8

Netherlands 23.6 120 -16 -4 -12 -8

Portugal 4.9 380 -273 -7 -41 -35

UK 24.7 44 14 42 28 -13

USA 26.6 76 3 21 3 -18Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: The productivity growth rate quoted is over a five-year period. Method used is FHK in panel a (see equation (2))and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Electrical and OpticalEquipment is defined as the 2-digit ISIC sectors 30 to 33. All values in columns 2 through 7 (6 in b) are per cent oftotal change in column 2. Net entry = entry - exit: Figures may not add up due to rounding.

Table 6. Proportion of STAN sectors with negative entry and exit

Country % Entry < 0 % Exit < 0

Finland 30.1 81.3France 81.9 88.9Germany 33.3 33.3Italy 11.7 69.2Netherlands 9.7 53.8Portugal 23.7 73.7UK 14.9 68.9USA 61.1 96.3

Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: STAN sectors includes both aggregated and bottom level industries/groups of industries. Data for all years forwhich results are available for each country.


15

Table 7. Total factor productivity decompositions, Manufacturing, 1987-1992a: FHK Method

1 2 3 4 5 6 7

Country TFP Growth Within Between Cross Net Entry Entry Exit

Finland 5.4 -94 53 65 76 54 -22

France -7.7 132 -21 2 -13 -12 1

Italy 15.5 53 23 -9 33 35 2

Netherlands 2.7 154 91 -97 -48 6 54

UK -4.5 154 23 -54 -23 -5 17

b: GR Method1 2 3 4 5 6

Country TFP Growth Within Between Net Entry Entry Exit

Finland 5.4 -62 85 77 50 -27

France -7.7 133 -21 -12 -19 -8

Italy 15.5 48 17 35 22 -13

Netherlands 2.7 105 44 -49 -11 38

UK -4.5 127 -7 -20 -19 2Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: The productivity growth rate quoted is over a five-year period. Method used is FHK in panel a (see equation (2))and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Manufacturing is defined asthe 2-digit ISIC sectors 15 to 37. All values in columns 2 through 7 (6 in b) are per cent of total change in column 2.Net entry = entry - exit: Figures may not add up due to rounding.


16

Table 8. Total factor productivity decompositions, Manufacturing, 1985-1990a: FHK Method

1 2 3 4 5 6 7

Country TFP Growth Within Between Cross Net Entry Entry Exit

Finland 12.6 41 19 3 37 26 -11

France 4.8 69 25 -18 24 -1 -25

Italy 27.2 53 13 -4 37 29 -9

Netherlands -4.4 36 -11 31 43 20 -23

UK 9.1 -4.65 -3 41 67 42 -25

b: GR Method1 2 3 4 5 6

Country TFP Growth Within Between Net Entry Entry Exit

Finland 12.6 42 19 39 22 -17

France 4.8 60 12 28 -8 -36

Italy 27.2 51 10 39 15 -24

Netherlands -4.4 52 0 48 5 -43

UK 9.1 16 15 70 29 -40Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: The productivity growth rate quoted is over a five-year period. Method used is FHK in panel a (see equation (2))and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Manufacturing is defined asthe 2-digit ISIC sectors 15 to 37. All values in columns 2 through 7 (6 in b) are per cent of total change in column 2.Net entry = entry - exit: Figures may not add up due to rounding.


17

Table 9. Productivity decompositions, Total Services, 1993-1998a: FHK Method

1 2 3 4 5 6 7

Country Prod Growth Within Between Cross Net Entry Entry Exit

Germany (W) 1.1 563 -196 -278 11 -328 -338

Italy -10.9 -91 -79 57 213 170 -43

Italy (TFP) -1.0 -361 -1273 249 1485 687 -798

Portugal -17.3 -89 12 62 114 102 -13

b: GR Method1 2 3 4 5 6

Country Prod Growth Within Between Net Entry Entry Exit

Germany (W) 1.1 424 -334 10 -331 -341

Italy -10.9 -63 -42 205 148 -57

Italy (TFP) -1.0 -236 -1135 1471 663 -808

Portugal -17.3 -58 46 112 85 -27Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: Decompositions are of labour productivity unless stated otherwise. The productivity growth rate quoted is over afive-year period. Method used is FHK in panel a (see equation (2)) and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Total services are defined as the 2-digit ISIC sectors 50 to 99. All values incolumns 2 through 7 (6 in b) are per cent of total change in column 2. Net entry = entry - exit: Figures may not add updue to rounding.


18

Table 10. Productivity decompositions, Business Sector Services, 1993-1998a: FHK Method

1 2 3 4 5 6 7


Italy -11.2 -87 -83 57 214 168 -46

Italy (TFP) -1.3 -287 -1029 201 1215 572 -643

Portugal -18.3 -92 9 62 121 95 -26

b: GR Method1 2 3 4 5 6


Italy -11.2 -59 -47 205 145 -60

Italy (TFP) -1.3 -186 -915 1201 548 -653

Portugal -18.3 -61 43 118 79 -39Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: Decompositions are of labour productivity unless stated otherwise. The productivity growth rate quoted is over afive-year period. Method used is FHK in panel a (see equation (2)) and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Business sector services are defined as the 2-digit ISIC sectors 50 to 74. Allvalues in columns 2 through 7 (6 in b) are per cent of total change in column 2. Net entry = entry - exit: Figures maynot add up due to rounding.


19

Table 11. Productivity decompositions, Wholesale and Retail Trade; Repairs, 1993-1998a: FHK Method

1 2 3 4 5 6 7


Germany (W) 5.4 3 40 6 51 1 -50

Italy 10.5 102 57 -77 18 1 -17

Italy (TFP) 17.6 34 27 -35 73 65 -8

Portugal -9.0 -4 -38 84 58 81 23

b: GR Method1 2 3 4 5 6


Germany (W) 5.4 6 43 51 -3 -53

Italy 10.5 64 19 17 -13 -30

Italy (TFP) 17.6 17 12 71 48 -23

Portugal -9.0 38 4 58 63 5Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: Decompositions are of labour productivity unless stated otherwise. The productivity growth rate quoted is over afive-year period. Method used is FHK in panel a (see equation (2)) and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Wholesale and Retail Trade; Repairs is defined as the 2-digit ISIC sectors 50to 52. All values in columns 2 through 7 (6 in b) are per cent of total change in column 2. Net entry = entry - exit:Figures may not add up due to rounding.


20

Table 12. Productivity decompositions, Transport and Storage and Communication, 1993-1998a: FHK Method

1 2 3 4 5 6 7


Germany (W) 16.3 120 -3 -27 10 39 29

Italy -16.1 -79 -57 37 199 118 -81

Italy (TFP) 9.3 33 67 -10 10 63 53

Portugal 54.7 110 11 -31 10 28 18

b: GR Method1 2 3 4 5 6


Germany (W) 16.3 107 -17 11 33 22

Italy -16.1 -60 -24 184 89 -95

Italy (TFP) 9.3 28 88 -16 34 50

Portugal 54.7 94 0 6 13 7Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: Decompositions are of labour productivity unless stated otherwise. The productivity growth rate quoted is over afive-year period. Method used is FHK in panel a (see equation (2)) and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Transport and storage and communication is defined as the 2-digit ISICsectors 60 to 64. All values in columns 2 through 7 (6 in b) are per cent of total change in column 2. Net entry = entry- exit: Figures may not add up due to rounding.


21

Table 13. Productivity decompositions, Financial Intermediation, 1993-1998a: FHK Method

1 2 3 4 5 6 7


Germany (W) 22.3 121 -13 -9 1 -4 -6

Italy 47.4 20 2 10 67 56 -11

Italy (TFP) 83.5 9 6 7 78 73 -5

Portugal 36.1 89 -7 14 3 0 -3

b: GR Method1 2 3 4 5 6


Germany (W) 22.3 117 -18 1 -5 -6

Italy 47.4 26 -2 77 34 -43

Italy (TFP) 83.5 12 1 87 45 -42

Portugal 36.1 96 -7 11 -4 -15Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: Decompositions are of labour productivity unless stated otherwise. The productivity growth rate quoted is over afive-year period. Method used is FHK in panel a (see equation (2)) and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Financial intermediation is defined as the 2-digit ISIC sectors 65 to 67. Allvalues in columns 2 through 7 (6 in b) are per cent of total change in column 2. Net entry = entry - exit: Figures maynot add up due to rounding.


22

Table 14. Productivity decompositions, Real Estate, Renting and Business Activities,1993-1998

a: FHK Method

1 2 3 4 5 6 7


Germany (W) -1.8 368 -440 260 -88 -145 -57

Italy -31.2 0 2 13 85 63 -22

Italy (TFP) -8.1 38 17 -10 54 -8 -62

Portugal -73.2 20 21 2 57 52 -5

b: GR Method1 2 3 4 5 6


Germany (W) -1.8 498 -306 -92 -153 -62

Italy -31.2 7 16 77 42 -35

Italy (TFP) -8.1 33 23 43 -34 -77

Portugal -73.2 21 33 46 30 -16Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: Decompositions are of labour productivity unless stated otherwise. The productivity growth rate quoted is over afive-year period. Method used is FHK in panel a (see equation (2)) and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Real estate, renting and business activities are defined as the 2-digit ISICsectors 70 to 74. All values in columns 2 through 7 (6 in b) are per cent of total change in column 2. Net entry = entry- exit: Figures may not add up due to rounding.


23

Table 15. Productivity decompositions, Community, Social and Personal Services, 1993-1998a: FHK Method

1 2 3 4 5 6 7


Germany (W) -2.9 0 22 88 -10 91 102

Italy -2.9 -449 2 196 350 540 190

Italy (TFP) 37.5 42 -22 -8 89 81 -7

Portugal -3.6 68 26 110 -105 170 275

b: GR Method1 2 3 4 5 6


Germany (W) -2.9 44 67 -11 89 100

Italy -2.9 -351 112 339 520 181

Italy (TFP) 37.5 38 -15 77 59 -18

Portugal -3.6 124 84 -107 149 257Source: Authors’ calculations using decomposition results from OECD Firm Level Study.Note: Decompositions are of labour productivity unless stated otherwise. The productivity growth rate quoted is over afive-year period. Method used is FHK in panel a (see equation (2)) and GR in panel b (see equation (3)) with a five-year rolling window (see text for details). Community, social and personal services are defined as the 2-digit ISICsectors 75 to 99. All values in columns 2 through 7 (6 in b) are per cent of total change in column 2. Net entry = entry- exit: Figures may not add up due to rounding.


24

Figure 1. Employment protection and share of productivity growth from entry

Em

plo

ymen

t P

rote

ctio

n

% Prod growth from entry (FHK)

-0.20 -0.10 0.00 0.10 0.20 0.30 0.40 0

1

2

3

4

Finland

Italy Italy

Netherla Netherla

Portugal Portugal

UK

USA USA

Source: Blanchard and Wolfers (2000) and results from OECD firm level studyNote: Lower values for employment protection imply less protection. Country names in italics are values for 1992-1997. Other values are for 1987-1992.


25

Figure 2. Employment protection and share of productivity growth from exit

Em

plo

ymen

t P

rote

ctio

n

% Prod growth from exit (FHK)

-0.40 -0.30 -0.20 -0.10 0.00 0.10 0.20 0.30 0

1

2

3

4

Finland

Italy Italy

Netherla Netherla

Portugal Portugal

UK

USA USA



26

Figure 3. Employment protection and share of productivity growth from net entry

Em

plo

ymen

t P

rote

ctio

n

% Prod growth from net entry (FHK)

0.15 0.20 0.25 0.30 0.35 0.40 0

1

2

3

4

Finland

Italy Italy

Netherla Netherla

Portugal Portugal

UK

USA USA



27

REFERENCES

Baldwin, J.R., and Gu, W., (2001a), “Plant Turnover and Productivity Growth in CanadianManufacturing”, mimeo, Statistics Canada

Bartelsman, E. and Barnes, M., (2001), “Comparative Analysis of Firm-level Data: a low marginal costapproach”, Draft Paper

Bartelsman, E., and Doms, M., (2000), “Understanding Productivity: Lessons from LongitudinalMicrodata”, Journal of Economic Literature, 38, 3, 569-594.

Bartelsman, E.J., Scarpetta, S., and Schivardi, F., (2001), “Comparative Analysis of Firm Demographicsand Survival: Evidence for the OECD Countries”, mimeo

Blanchard, O., and Wolfers, J., (2000), “The Role of Shocks and Institutions in the Rise of EuropeanUnemployment: the Aggregate Evidence”, Economic Journal, 110, 462, C1-C33.

Disney, R., Haskel, J., and Heden, Y., (2000), “Restructuring and Productivity Growth in UKManufacturing”, CEPR Discussion Paper.

Foster, L., Haltiwanger J., and Krizan C., (1998), "Aggregate productivity growth: Lessons fromMicroeconomic Evidence", National Bureau of Economic Research Working Paper 6803.

Griliches, Z. and Regev, H., (1992), “Productivity and Firm Turnover in Israeli Industry 1979-88”,National Bureau of Economic Research Working Paper 4059.

Maliranta, M., (2001), “Productivity Growth and Micro-level Restructuring: Finnish experiences duringthe turbulent decades”, Draft Paper, The Research Institute of Finnish Economy, ETLA.

Nicolletti, G., Scarpetta, S., and Boylaud, O., (2000), “Summary indicatiors of product market regulationwith an extension to employment protection legislation”, OECD Economics Department WorkingPaper No. 226, Paris

O’Mahony, M., (1999), “Britain’s Productivity Performance 1950-96: An International Perspective”,National Institute of Economic and Social Research.

OECD (2000), “Recent Growth Trends in OECD Countries”, OECD Economic Outlook, June 2000

OECD (2001a), “Productivity and Firm Dynamics: Evidence from Microdata”, OECD Economic Outlook,June 2001

OECD (2001b), “Productivity and Firm Dynamics: Evidence from Microdata”, OECD EconomicsDepartment, ECO/CPE/WP1(2001)8/ANN2 March 2001 (JT00103733)


28

APPENDIX 1

LIST OF OECD FIRM-LEVEL STUDY PARTICIPANTS

Project Coordinated by: Eric Bartelsman (Project leader) and Stefano Scarpetta

CanadaJohn Baldwin (Statistics Canada)

DenmarkTor Eriksson (Aarhus School of Business)

FinlandSeppo Laaksonen (Statistics Finland), Satu Nurmi (Helsinki School of Economics and Statistics Finland)and Mika Maliranta (Research Institute of the Finnish Economy and Statistics Finland)

FranceBruno Crépon and Richard Duhautois (INSEE, Paris)

GermanyThorsten Schank (University of Mannheim and IAB)

ItalyFabiano Schivardi (Bank of Italy)

NetherlandsEric Bartelsman (Free University Amsterdam and Statistics Netherlands)

PortugalPedro Portugal Dias (Bank of Portugal)

SwedenYlva Hedén (Bank of Sweden)

United KingdomMatthew Barnes and Jonathan Haskel (Queen Mary, University of London and Office for NationalStatistics)

United StatesRon Jarmin and Javier Miranda (Center for Economic Studies, US Census Bureau)

OECDStefano Scarpetta and Phil Hemmings (OECD Economics Department, Paris)


29

APPENDIX 2.

META DATA TABLES(REPRODUCED FROM ECO/CPE/WP1(2001)8/ANN2 – FILENAME JT00103733,

PRODUCED BY OECD ECONOMICS DEPARTMENT)


30

Table A1.2. Description of data used in productivity decompositions

Finland France Germany (West) Italy

Type of data (‘Register’,‘Sample’, or ‘Other’)

Census Register Sample Sample

Name of data source(s) Industrial statistics Fiscal database (‘BRN’ file) withadditional information from theEnterprise survey ('EAE’ file)

IAB Establishment Panel Company Accounts Database

Comment on register orsampling method

For technical reasons not allobservations could be used inconstructing the longitudinal datain the manufacturing sector withthe result that employment figuresin manufacturing implied in thedata fall short of those from othersources.

Sample based on random drawsfrom cells based on 16 sectorsand 10 plant sizes. Total sample(all Germany) approx. 8 000.Sample data are weighted togenerate population-equivalentdata.

Approximately 40 000 firms per year.Sampling method: firms with at least5 million euro of turnover, or withmultiple bank relationships. The totalsample is kept roughly the samesize, adding or deleting firms that arein the proximity of the selectionthreshold.

Unit of observation Plant and enterprise code (thusindustrial plants included)

Firm Plant Firm

Comment on unit ofobservation

Legal entity with a unified balancesheet.

Periodicity and timing Annual (end of year) Annual (end of year) Annual Annual, end of year

First 5-year period 1975-19801988-1993 (services)

1985-1990 1992-1997 1983-1988

Last 5-year period 1993-19981993-1998 (services)

1990-1995 1993-1998 (sales data limitnumber of years that can becovered)

1993-1998

Breaks 1994-1995, change in sizethreshold

No In 1993-1994 there has been achange in the data collectionprocedures. As a result, entry isabnormally high in those 2 years,and similar for exit, 1994-1995.


31

Table A1.2. Description of data used in productivity decompositions (continued)

Finland France Germany (West) Italy

Size threshold All plants with at least 5 persons.Since 1995, all plants owned byfirms having at least 20employees

’BRN’ file covers firms with morethan 3.8 million FFr turnover peryear in manufacturing and1.1 million FFr turnover in theservice sector are covered. EAEfile is a sample of firms with morethan 20 employees

Plants with at least one employee Firms with more than 5 millioneuro turnover per year

Sectoral coverage Manufacturing (except 2observations for services)

Manufacturing Manufacturing and total services All sectors

Issues relating to output data Value added Gross output used in calculations.

Issues relating to labour inputdata

Employees Employment is added on to thebalance sheet data. Despite someconcerns, random checks on theemployment figures suggest theyare reliable. Only the number ofemployees is available.

Issues relating to capital stock No capital stock data available Capital stock is reconstructedfrom balance sheet informationusing a permanent inventorymethod. Initial capital stock isestimated using a measure ofaverage age of capital withappropriate deflation.

Issues relating to price data Value-added price data onlyavailable at the 2-digit level (about15 industries). Producer price andunit value indices available at the3 or 4-digit level.

All price data at the ‘naf 36’ level There are breaks in price databetween 1993 and 1999.

All price data at 2-digit level


32

Table A1.2. Descriptions of data used in productivity decompositions (continued)

Netherlands Portugal United Kingdom United States

Type of data (‘Register’,‘Sample’, or ‘Other’)

Register and sample Register Sample Quinquennial Census ofProduction

Name of data source(s) Production Statistics Survey Quadros do pessoal (administrativeestablishment-based database)

Annual Census of Production.(ACOP) Respondents Database(ARD)

Census of Manufacturing

Comment on register orsampling method

The Production Statistics Surveyincludes all firms with at least 20employed persons, and a randomsample for smaller firms. Sampledata (for smaller firms) are weightedto generate population-equivalentdata.

The self-employed, publicemployees and private services tohouseholds not included.

Sample data are weighted togenerate population-equivalent data.Weights derived from employment onCSO Business Register (ARD non-selected files).

Universe

Unit of observation Firm Firm (plant data also available butnot used in this study)

Establishment (lowest autonomousunit within firm)

Establishment and firm

Comment on unit ofobservation

Change in definition of reporting unitin 1987. Impact not considered to belarge. In 1994: New register, movedto Eurostat enterprise definitions.Almost total break in data series.

Firm level tabulations supplied

Periodicity and timing Annual Annual.March (1983-1993)October (1994-1998)

Annual (timing varies) 5-year

First 5-year period Manufacturing: 1980-1985Business services: 1987-1992

1980-1985 1987-1992

Last 5-year period Manufacturing: 1992-1997Business services: 1991-1996

1987-1992 1992-1997 (no intervening years)


33

Table A1.2. Descriptions of data used in productivity decompositions (continued)

Netherlands Portugal United Kingdom United States

Breaks 1993: change in industryclassification

1995: change in SIC code 1984: significant change inregister (due to inclusion of VATregister). "One-year" categorylarge due to incorrectclassification between theregisters.1987: change in definition ofreporting unit, impact not great.1994: new register,comprehensive linking not yetachieved.

No

Size threshold Firms with at least 20 employees inmanufacturing, firms with at least 5employees in business services.

At least one employee At least one employee (smallerobservations may be older due torestrictions to protect small firms)

At least one employee

Sectoral coverage Manufacturing, business services(computer and related activities,other business activities)

All but public administration Manufacturing only Manufacturing

Issues relating tooutput data

Gross output: value of totalturnover plus change in stocks +margin on trading and otherrevenues.

Gross output Gross output adjusted forinventories and deflated usingGray/Bartelsman/Becker 4-digitSIC deflators

Issues relating tolabour input data

Employees Employees Employees Number employees on March 12

Issues relating tocapital stock

No capital stock available Generated from investmentquestions on ARD using perpetualinventory method. Initial stocksbased on industry dataapportioned using energy usagedata from ARD

Issues relating toprice data

Producer price indices for totalturnover. If available at the 3-digitlevel of ISIC; otherwise at the 2-digit level

2-digit level (from nationalaccounts)

4-digit for output and materials,2/3 digit for capital


34

APPENDIX 3

PSEUDO-CODE “PRODUCTIVITY DECOMPOSITIONS AND CORRELATES” –PRELIMINARY VERSION

This file contains the pseudo-code, as well as descriptions of input and output datasets needed for theproductivity themes.

SECTION A. INPUT DATASETS:

I. production surveys:PSyy for years yy=startyr to endyr

In each file variables a variable containing a unique unit (firm, estab) code,a national industry or sector code, and the relevant variables;Mappings between variable names in your dataset and the ’concepts’ used in the program(unit -unique unit code , nq - sales or gross output, e - employment, etc)are made in the call to macro OECDPROD: e.g. unit=firmid, nq=sales, e=employment, ....

relevant macro parameters:survlib, survfile, startyr, endyr, incr, unit, ind, nv, nq, e, pay, nm, k, andtypestr.

NOTE: In most traditional Stats agencies, the production surveys fordifferent major sectors will be in separate files. Please stack thesetogether in each year, while making sure that your ’unit’ stays unique.If your agencies’ computer system can not handle this much data, buy acurrent vintage laptop that provides ample space.

II. price deflators Prices: dataset has variables for ind - nationalindustry code, year - variable with year, one or more deflatorvariables for piq- gross output, piv- value added, pim - materials

You make your own mappings of these variables in the call to OECDPROD program;

relevant macro parameters:

plib, pfile, piq, piv, pim, year

NOTE: Country teams are requested to return the dataset prices to Coordinator; by default this is &outlib..&ccc.pdefl, with ind, year, and one or more of: piq piv pim


35

III. concordances and hierarchy meta-data

STAN hierarchy, showing aggregation tree of STAN codes used in OECD project.Project coordinator will supply this file, along with all program code.

IND to STAN concordance, is a ’many-to-one’ mapping, such that each uniqueind code occurring in any of the PSyy files has a corresponding STAN code(if possible IND mapped to STAN codes at lowest level in STAN hierarchy).

relevant macro parameters:stanlib, concfile, hierfile

IV. Register files for checking entry/exit

Business register, with external info on existence of unit each yearfor which the regfile dataset exists; Variables unit and weight. Thevariable weight is used re-weight observations in decomposition.Register file must exist for each year yy for which there are Prod Surv files.

macro parameters:reglib, regfile, weight, unit, startyr, endyr, incr

V. Productivity correlates

Correlates dataset, one for each year to be merged into prodsurveydata. Each correlate dataset has unique variable for UNIT, and one ormore categorical variables and/or continuous variables. The year thecorrelate refers to may not be the year of the correlate dataset it isput into. For example, one may want to relate use of automationequipment in 1985 to productivity in 1990. In this case, put the datainto correlates dataset CORR1990.

relevant macro parameters:corrlib, corrfile, corryrs, ccat, ccont, unit

NOTE: Country team needs to provide us with definitions of thevariables in ccat or ccont that are used in each year. What survey are they from,how ’dense’ was the linking to the production survey, how was variable transformed, etc.

SECTION B. OUTPUT DATASETS

Decompositions:ccc.ttt.DC, e.g. NLDTFPDC, or FRALPVDC, where ttt in (LP LPV TFP MFP) andccc in (DNK FIN FRA GER NLD POR GBR USA CAN ITA)

Variables:


36

STAN, &year, and within, between, entry and exit components forFHK, GR, and 2 Baldwin variant decompositionsAlso included are total employment and output for CO, EN, EX, aswell as unweighted mean and standard deviation of appropriate prod measurefor each CO, EN, EX group;

Distributions:

ccc.st distributions of all variables of interestYEAR VNAME QRT MEAN STD;where vname is in xlist and ylist, qrt is ’own’ quartile= 1 2 3 4, total=0;

ccc.st2 distr of correlates by quartile of productivity variableAN VNAME YNAME QRT MEAN STD;where vname is name of variable in xlist, qrt is quartile of y in ylist ;for all vnames in xlist;

ccc.st3 distr of productivity variables by quartile of correlatesAN VNAME XNAME QRT MEAN STD;where vname is name of variable in ylist, qrt is quartile of xname in xlist ;for all xnames in xlist;

Regressions:

A baseline regression of productivity CCC.BASERAll-variable regression CCC.ALLR

This is run on both levels and first-differences. Forfirst-differences it stacks all the possible years with ’SPAN’ yearsof difference. For levels, it uses the data from the last of the spanyears in the above dataset; So if you have data from 1987 through1993, and a 5-year span, we have to annual cross sections of firstdifferences stack, and also only the years 92 and 93 stacked for thelevel regression. This simplification is related to ourdeflation. (Laspeyres, with baseyear ’span’ years earlier.)

dsti/eas/ind/swp/ah(2001)14 working party on statistics · dsti/eas/ind/swp/ah(2001)14 working...

Documents