A Tale of Two Cycles: Closure, Downsizing and Productivity Growth in UK. Manufacturing, Nicholas OULTON

Transcription

1 A Tale of Two Cycles: Closure, Downsizg and Productivity Growth UK Manufacturg, Nicholas OULTON National Institute of Economic and Social Research August 1998 NIESR Discussion Paper No. 140 Address for correspondence National Institute of Economic and Social Research, 2 Dean Trench Street, LONDON SW1P 3HE Tel: (0) ; Fax: (0) n.oulton@niesr.ac.uk

2 CONTENTS Page Abstract 2 1. Introduction 3 2. The ARD 5 3. Structural change UK manufacturg, Decomposg productivity growth 8 5. Decomposg productivity growth practice, The role of closures Productivity growth amongst survivors: downsizg versus upsizg Productivity growth amongst survivors: the role of establishment size Disaggregatg to the sector level A comparison with Conclusions 16 References 17 Appendix 19 1

3 ABSTRACT This paper uses the ARD, the new longitudal database of the Census of Production, to analyse productivity at the establishment level the two cycles of and Contrary to a commonly held view, closures did not play a major role accountg for productivity growth Establishments which exited had lower productivity than survivors but the exits were replaced by entrants which also had low productivity. Most of productivity growth was due to growth with survivors. The greatest gas occurred the 36 establishments employg 7,500 or more 1979; these accounted for a third of productivity growth amongst survivors. Most productivity growth occurred establishments which downsized employment. But despite an overall fall of a quarter employment, 16% of productivity growth occurred establishments which expanded employment. The ma difference between and was the productivity growth rate amongst survivors. In , it was negative overall and over half of employment was itially establishments where productivity fell. Keywords JEL codes Productivity, exit and entry, downsizg, manufacturg D24, J24, L60 2

4 1. Introduction 1 The upswg labour productivity UK manufacturg which began the 1980s occasioned much comment and a number of rival explanations were put forward (Muellbauer 1986; Oulton 1995). One of the most popular hypotheses was that under the gun of the strong pound and the recession manufacturers simply closed their least productive plants or scrapped their least productive assets. So higher productivity was purchased at the cost of lower output and capacity. But two pieces of evidence militate agast this view. First, the 1980s there was a remarkable change the (previously static) employment size distribution of manufacturg establishments: the proportion of employment large plants shrank dramatically. Large plants tend to have higher productivity than smaller ones, so if closure or scrappg were concentrated large plants as the changg size distribution plus anecdotal evidence seem to suggest, then this would have reduced productivity growth, not enhanced it (Oulton 1987). Second, accordg to the popular vtage capital model, under which older, less productive capital is scrapped first, the dustries which output had decled the most should have enjoyed the largest productivity ga. But this is not what happened (Oulton 1989). An alternative explanation is that the trade union reforms of the 1980s removed important obstacles to productivity growth. The productivity improvement was greatest dustries which had previously been the most heavily unionised (Oulton 1990). Usg part a dataset developed Oulton and O Mahony (1994), Bean and Crafts (1995) found that the productivity improvement had been largest dustries where multi-unionism had previously been the most prevalent. 2 A basic difficulty with these studies is that they all used dustry level data. No direct measures of scrappg or closures were available. This is because researchers could 1 This research has been supported by the Department of Trade and Industry, to whom I owe thanks. I am grateful to Peter Hart for helpful comments and to the staff of the ONS at Newport for assistance usg the ARD. The usual disclaimer applies. 2 For further evidence on the role of trade unions, see Denny and Nickell (1992) and Gregg et al. (1993). A pessimistic assessment of the effect of the Thatcher period reforms is Blanchflower and Freeman (1994). Increasg competition may also have played a role (Haskel 1991; Reddg and Proudman 1998). 3

5 generally use only the published data from the Annual Census of Production. 3 By law the published results could not reveal the data relatg to any dividual manufacturg establishment. There is no way of knowg from the published data whether the establishments which were closed had lower or higher productivity than those which survived. Nor can one tell whether plants which were large 1979, assumg they survived, enjoyed higher or lower productivity growth over the next decade than smaller ones. All this has now changed with the advent of the ARD. The ARD is an electronic database of the Annual Census of Production (ACOP). 4 Though it is still an offence to publish the data relatg to an ACOP respondent, the law now allows researchers (with the permission of the ONS) to study the dividual returns to the Census. Because establishments have a unique reference number the history of dividual establishments can be traced over time. The present paper compares productivity growth the two cycles of and , though more attention is paid to the latter. 1973, 1979 and 1989 were all cyclical peaks (or close to beg so). The period contas the major recession of and The next section describes the ARD. Then section 3 looks at structural change UK manufacturg over the period Section 4 lays out a method of decomposg the contribution of different groups with manufacturg to productivity growth. Section 5 then applies this decomposition to a series of different groupgs of analytical terest, e.g. survivors versus non-survivors and a breakdown by itial employment size. This section also presents some results disaggregated to the sectoral level. Section 6 then compares experience with the previous cycle, Section 7 concludes. 3 See however Lansbury and Mayes (1996) who discuss ter alia the relative productivity of entrants and exits manufacturg. However, as their work predated the ARD, they were not able to trace the history of dividual establishments. 4 ARD stands for ABI Respondents Database, and turn ABI stands for Annual Busess Inquiry. Annual Busess Inquiry (Production) is the new name for ACOP. 4

6 2. The ARD The ARD, or ABI Respondents Database to give it its full name, is an electronic database of the Annual Census of Production (ACOP). In prciple, it cludes all the data collected under ACOP from 1970 to the present. It covers the whole of the production sector, manufacturg plus mg and quarryg and, for recent years, construction (Oulton 1997). In the present paper I use only the results for manufacturg. Labour productivity is output per unit of labour. Three measures of output are available from the ARD. In descendg order of size these are: gross output, net output and gross value added. I use gross value added (GVA), maly because it is additive across establishments and dustries. Hours worked are not measured the Census so I use the total number employed as the denomator of the productivity ratio. GVA which is reported current prices was converted to 1990 prices usg producer price dexes for each Class of the 1980 SIC (of which there are 22 with manufacturg: see the Appendix). 5 For the period studied here, dustry is recorded the ARD under the 1980 SIC. The most basic unit the ARD is the local unit, defed as a plant or office at a sgle location. Above that is the establishment, which is the reportg unit. An establishment consists of at least one local unit (itself) and may consist of more. Most of the data the ARD (e.g. GVA) relate to the establishment as a whole. The establishment may or not be a company the legal sense. In 1987, company-based reportg was troduced to the Census and the reportg units are now referred to as busesses. But the larger companies contued to be split up to smaller units. 6 5 For Classes 21, 36 and 44 no PPI exists so the PPI for manufacturg as a whole was used. These deflators were obtaed from Datastream. 6 It is not clear without further research how much difference this change made. For clarity, and because most of the data used here were collected before the change to company-based reportg, I contue to use the term establishment. Note that American usage an establishment is a plant and this should be borne md when comparg the present results with US ones, e.g. Baily et al. (1996). 5

7 3. Structural change UK manufacturg, Table 1 compares performance UK manufacturg over the last two complete cycles, from peak to peak (see also charts 1-3). 7 In the first cycle of 6 years, output and employment both fell and productivity rose a meagre 2.8%. In the second cycle of 10 years, output rose a modest amount, employment fell by a quarter and productivity rose by an impressive 48%. We would like to know how the experience of dividual establishments differed these two periods. The period will be the one studied itially. In 1979, 17,965 manufacturg establishments were selected as respondents to the Census, out of some 108,000 classified to manufacturg. In 1979 as 1989, establishments employg fewer than 20 were nearly all excluded and for those employg the samplg fraction was 50%. 8 The sum of the gross value added reported by respondents is 82% of the figure published the ACOP Summary Volume, which grosses up the total of selected respondents to allow for the non-selected ones. The total of selected employment is 80% of the published figure. For 1989, the correspondg percentages are 83% and 79%. These figures suggest that the Census achieved much the same level of coverage both years, hence no adjustment was made on this account. Establishments were however excluded from the analysis if (a) they were classified to manufacturg 1979 but not 1989 or the reverse (4 establishments) or (b) they had negative value added either year (166 establishments 1979 and , with an aggregate negative value added of around 400m). This resulted 17,733 establishments 1979 and 18, beg eventually cluded the analysis. We distguish between (1) survivors, which are those establishments which are present both years; (2) exits, establishments present only 1979; and (3) entrants, establishments present only Output 1979 was actually a bit lower than 1978 but 1979 is a better year from the pot of view of the availability of variables the ARD. It seems quite likely that manufacturg is currently at another cyclical peak, but if so some time will have to elapse before it can be analysed: the most recent year available the ARD is The samplg scheme was actually more complicated than this; see Oulton (1997). 9 This defition overstates the magnitude of entry and exit somewhat because of samplg the Census. An establishment may be existence both years but only sampled one. In both 1979 and 1989, half of establishments with employees were sampled. Similarly, an establishment may be 6

8 CHART 1 Manufacturg output, (log scale) Output Year 4.3 Source Datastream. CHART 2 Manufacturg employment, (log scale) Employment Year 4.4 Source Datastream. excluded altogether from the Census one year on the grounds of size (normally because it has fewer than 20 employees) though it is large enough to clude another. 7

9 CHART 3 Manufacturg productivity, (log scale) Productivity Year Source Datastream. Table 1 Manufacturg two busess cycles Period Change output, % Change employment, % Change productivity, % Source Datastream. The ONS codes are DUDM for output, DMWB for employment and DMOB for productivity (output per person employed). Table 2 gives some basic statistics for these three categories of establishments. We note that only 46% of establishments present 1979 survived to Conversely, 57% of establishments present 1989 were absent 1979 and so are termed entrants. On this measure alone there was clearly considerable structural change over this period. However, because survivors tended to be substantially larger than exits or 8

10 entrants, they accounted for around two thirds of employment both years. The mean employment size of survivors was and fell by a quarter to 315 by The mean size of exits was half that of survivors But the entrants which replaced the exits were only about a third of the size of survivors Because the size distribution of employment is positively skew, lookg only at mean size can be misleadg. But median establishment size, which is substantially below the mean all categories, also fell. Overall median size fell from 97 to 76, or by 22%. But terestgly the fall was much smaller amongst survivors, only 6%. Another very revealg statistic is the median of the first moment distribution of employment (le 6, Table 2). 10 This tells us the size of establishment such that half the workers work establishments which are larger and half establishments which are smaller (the median of the basic size distribution tells us the establishment size such that half the establishments are larger and half smaller). On this measure, the typical worker was employed 1979 an establishment with 953 employees, 1989 an establishment of 519 employees, a fall of 46%. Amongst survivors, the typical worker was employed an establishment which was 41% smaller 1989 than The explanation for why, amongst survivors, the median size fell by only 6% while the first moment median fell by 46% is that there was a huge decle employment large plants. We return to this topic below when the size distribution is discussed more detail. 10 If f(x) is the density function of a random variable x which takes only positive values, then the density function of the first moment distribution is defed by gx ( ) = xf( x)/ uf( udu ) : see Hart (1975) for analysis and discussion. 0 9

11 Table 2 Survivors, exits and entrants compared: ARD respondents manufacturg, 1979 and 1989 Survivors Exits Entrants Overall Variable No. of establishments 8,074 8,074 9,659 10,640 17,733 18, Share of employment (%) 3. Share of GVA (%) Mean no. of employees Median no. of employees 6. Median of 1 st moment distribution of employment (no.) 7. Mean productivity level (1990 per person employed) 8. Median productivity level (1990 per person employed) 9. S.D. of log productivity th percentile of productivity as % of median th percentile of productivity as % of median 12. Mean crease productivity, (%) 13. Median crease productivity, (%) , ,589 20,790 13,727 19,474 14,575 20,042 12,969 17,245 11,014 16,345 11,968 16, Source ARD. 10

12 As mentioned the troduction, a persistent notion about the 1980s is that it saw a substantial reduction the number of low productivity plants. This impression is not borne out by the evidence of Table 2. The dispersion of productivity across establishments, as measured by the standard deviation of the log of productivity (le 9), actually widened both overall and amongst survivors. Relative to the median productivity level, the bottom 5% of establishments actually did worse 1989 than ten years earlier. Once aga this was true both overall and for survivors. As for the upper tail, this rose relative to the median the case of survivors though not overall. Charts 4-7 show kernel density estimates of productivity levels for survivors, exits and entrants, and overall. They confirm the very wide dispersion which exists at any one time and they certaly show the existence of a very long tail of low productivity establishments both years. 11 The domant impression is of a rightward shift of the whole distribution, with little overall change shape, between the two dates. The distribution of productivity growth amongst survivors appears Chart 8. To make the graph more readable, the 97 establishments who apparently experienced productivity growth exceedg 500% have been excluded. Even so, the distribution has pronounced positive skew: the mean growth is twice the median (with all observations cluded: les 12 and 13, Table 2). A remarkable feature revealed by Chart 8 is the substantial number of survivors with negative productivity growth. In fact, 2,191 survivors, more than a quarter of the total, fell to this category. Their mean size was 285 employees 1979 and , while their mean productivity growth was -27.6%. 11 It is far from clear however that the UK is exceptional this respect (Industry Economics and Statistics Directorate 1997). 11

13 CHART 4 Kernel density of log productivity, 1979: Survivors.15.1 Density ln(prod.) 79 Source ARD. CHART 5 Kernel density of log productivity, 1989: Survivors.15.1 Density ln(prod.) 89 Source ARD. 12

14 CHART 6 Kernel density of log productivity, 1979: Exits.15.1 Density ln(prod.) 79 Source ARD. CHART 7 Kernel density of log productivity, 1989 Entrants.15.1 nsity ln(prod.) 89 Source ARD. 13

15 CHART 8 Kernel density of productivity growth, : Survivors, %.1 Density Source ARD. Note Establishments with productivity growth exceedg 500% excluded. 4. Decomposg productivity growth Suppose we divide manufacturg establishments up to a number of groups. We wish to know what is the contribution of each group to overall productivity growth over a given period. A decomposition close to the one used by Baily et al. (1996) will be employed. Let Y it be output and E it be employment group i at time t. Then labour productivity group i is defed as qit Yit / Eit Labour productivity the aggregate is: q Y E = s q t i it i it i it it 14

16 where sit = Eit / i Eit, the employment share, and s i it = 1. The proportional growth of aggregate labour productivity between time 0 and time T is ( q q )/ q = ( s q s q )/ q T 0 0 i it it i0 i0 0 which can be decomposed as follows: ( q q ) s ( q q ) ( s s )( q q ) ( s s )[( q q ) ( q q )] T i it i i it i i i it i it i i T q0 q0 q0 q0 (1) The first summation on the right hand side is the with group effect, the growth of productivity each group weighted by its base period employment share. The second and third summations are reallocation effects. Aggregate productivity growth is raised if there is a rise the labour share of groups with a higher than average productivity level the base period (the second term). It is also raised if there is a rise the share of groups with higher than average productivity growth (the third term). 12 In words, Aggregate productivity growth = With group effect + Reallocation levels effect + Reallocation growth effect The contribution of the ith group to aggregate productivity growth is therefore measured as: { s ( q q ) + ( s s )( q q ) + ( s s )[( q q ) ( q q )]} / q (2) i0 it i0 it i0 i0 0 it i0 it i0 T 0 0 We may also note that sce growth is defed here discrete terms, the relationship between productivity growth and the growth of output (Y) and employment (E) per period is: 12 The third term differs from the correspondg one Baily et al. (1996) by the clusion of aggregate productivity growth, (q T -q 0 )/q 0. This changes the dividual reallocation growth effects but not the total of such effects. 15

17 1 qit 1 YiT 1 EiT 1/ T = T qi0 T Yi 0 T Ei0 1/ T + ( 1/ T)( EiT / Ei0) (3) The percentage growth over an terval of length T is normalised by dividg by T to facilitate comparisons of growth over tervals of different length. This convention is adopted so that our decompositions will contue to add up. Note that growth per period so defed is not the same as the annual average growth rate. 5. Decomposg productivity growth practice, The role of closures In aggregate, the output of ARD respondents rose by 11%, their employment fell by 29% and their productivity rose by 57% over the ten years (Table 3). 13 Exits accounted for 37% of employment 1979 and productivity these establishments was 17% lower than survivors. So the disappearance of these establishments would certaly have raised productivity. But if nothg else had changed except for exits, i.e. no new entrants and no productivity growth survivors, overall productivity would only have risen by 7%, whereas reality it rose by 57% (5.66% per year, see Table 3) The official figures for aggregate manufacturg are reasonably close to these: see Table 1. It must be remembered that real output is measured here as sgle deflated value added while the official series is a value added weighted average of real gross output dicators. Moreover the official employment series comes from the Census of Employment, not the Census of Production, which used a different register of busesses over this period. 14 In previous tables, statistics have been presented for productivity defed as Y it follows, the productivity of a group is defed as iy it i E it terpretg the tables to follow. / E it. In what / and this should be borne md when 16

18 Table 3 Productivity and employment, and : survivors versus exits and entrants Share of employment 1979 Share of employment last year Productivity 1979 Annualised growth a of value employment productivity added % % 1990 % % % Exits/entrants , Survivors , Total , Exits/entrants , Survivors , Total , a. The growth rate of X from 0 to T is measured as (100/T)*(X T -X 0 )/X 0. Note that this is not the same as the average annual rate of growth. Note The relationship between the growth rates of productivity, value added and employment is given equation (3). Apart from exits, several other changes were gog on at the same time. First, productivity did rise survivors, by 51%. Second, the exits were replaced by new entrants which accounted for a third of 1989 employment. These new establishments had lower productivity than the survivors had attaed by 1989, but the percentage gap was roughly half that between exits and survivors 1979: 9% versus 17%. So overall productivity growth was raised not so much by the exit of low productivity establishments as by their replacement by entrants whose productivity was comparatively high, though still low absolute terms. A similar picture emerges if we do the same calculation for the peak to trough period of , except that survivors play a larger role. Survivors over this 3 year period accounted for 84% of employment 1979 and 87% Exits had productivity which was 22% lower than survivors They were replaced by entrants who had productivity 10% lower. 17

19 We now apply the decomposition of Baily et al. to analysg the role of closures. Table 4 shows the decomposition of productivity growth between survivors on the one hand and exits/entrants on the other. Survivors account for 61% of overall growth. Virtually all their contribution is due to ternal growth, the reallocation effects beg small. The contribution of entrants and exits is also nearly all due to ternal growth, not reallocation effects. Table 4 Decomposition of productivity growth a, and : survivors versus exits and entrants Total contribution Realloc Realloc Total With -ation -ation contribution effect effect: effect: level growth With effect Realloc -ation effect: level Realloc -ation effect: growth Group % % % % % % % % Exits/entrants Survivors Total a. Growth rates are annualised, e.g. growth of X from 0 to T is (100/T)*(X T -X 0 )/X 0. Note Each group s contribution is given by equation (2). For further explanation see text. 5.2 Productivity growth amongst survivors: downsizg versus upsizg Baily et al. (1996) have suggested a fourfold classification terms of success or failure raisg productivity combed with whether or not employment rose (upsizers) or fell (downsizers). This scheme is illustrated the followg table, where q is productivity and E is employment: 18

20 Quadrant 1 Successful upsizers 2 Successful downsizers 3 Unsuccessful downsizers 4 Unsuccessful upsizers Description q>0, E>0 q>0, E<0 q<0, E<0 q<0, E>0 Thus successful upsizers, quadrant 1, are establishments where both productivity and employment rose. 16% of employment 1979 and 31% 1989 fell to this quadrant. Much the largest share of employment 1979, two thirds, fell to quadrant 2, successful downsizers, and their share fell to just under half The rest of employment, around a fifth, was establishments with fallg productivity (Table 5). The fastest rate of productivity growth was found quadrant 2, at an annualised rate of 9.21%. Applyg the decomposition of Baily et al. we fd not surprisgly that quadrant 2 accounts for the lion s share of productivity growth (Table 6). Nevertheless it is noteworthy that the successful upsizers account for 16% of overall growth. However, their positive contribution was almost exactly outweighed by the negative contribution of the other two quadrants (-15%). Baily et al. (1996) found a much larger role for quadrant 1 the US over the period But it must be remembered that the US employment fell by only 4.5% over , compared with 25.4% the UK our period. 19

21 Table 5 Productivity and employment amongst survivors, : by quadrant Quadrant Number of establishments Share of employment 1979 Share of employment 1989 Productivity 1979 Annualised growth a of value employment productivity added No. % % 1990 % % % 1 ( q>0, E>0) 2, , ( q>0, E<0) 3, , ( q<0, E<0) 1, , ( q<0, E>0) , Total 8, , a. The growth rate of X from 0 to T is measured as (100/T)*(X T -X 0 )/X 0. Note that this is not the same as the average annual rate of growth. Note The relationship between the growth rates of productivity, value added and employment is given equation (3). Table 6 Decomposition of productivity growth a amongst survivors, : by quadrant Quadrant With effect Total contribution Reallocation effect: level Reallocation effect: growth % % % % 1 ( q>0, E>0) ( q>0, E<0) ( q<0, E<0) ( q<0, E>0) Total a. Growth rates are annualised, e.g. growth of X from 0 to T is (100/T)*(X T -X 0 )/X 0. Note Each group s contribution is given by equation (2). For further explanation see text. 20

22 5.3 Productivity growth amongst survivors: the role of establishment size We look now more detail at the role of size. Table 7 shows the size distribution amongst respondents (note that there are only a handful of cases with employment less than 20 sce such establishments are normally excluded from the Census). The extreme skewness of the distribution is apparent. The 36 establishments with 7,500 or more employees 1979 accounted for 17% of total employment. Between 1979 and 1989 employment rose absolute terms establishments which had l99 or fewer employees In relative terms, establishments employg up to 999 people 1979 creased their share of total employment. In general, the productivity level 1979 rose with size although the maximum is reached the band 4,000-4,999 employees. Productivity growth also rises with size with exceptionally rapid growth the top size class. However this size class was also the fastest shedder of labour. In the termology of the previous sub-section, the lowest four size classes were successful upsizers, while the remader were successful downsizers. Table 8 shows the contribution of each size class to productivity growth. The top size class, a mere 36 establishments, accounts for almost a third of aggregate productivity growth. By contrast, establishments employg fewer than 500, though 29% of 1979 employment, account for only 15% of productivity growth. Reallocation effects turn out to be quite important when the decomposition is by size. Overall, they reduce the growth which would otherwise have occurred by an annualised 0.72%. Most of this effect is due to the fact that the establishments which grew fastest were also the biggest downsizers. It is apparent that downsizg has played an important role the productivity growth story. But what form has this downsizg taken? Sce an establishment may consist of more than one local unit, a company could downsize by closg local units or by slimmg down employment with each local unit. The ARD records the number of local units of which each establishment is composed so these two possibilities can be quantified. Though this formation is available for 1989 it is missg for 1979, so I use 1980 stead. In 1989, a bit under half of employment our sample was 21

23 Table 7 Productivity and employment amongst survivors, : by 1979 employment size group Employees 1979 (range) Number of establishments Share of employment 1979 Share of employment 1989 Productivity 1979 Annualised growth a of value employment productivity added No. % % 1990 % % % 49 1, , , , , , , , , , ,000-1, , ,500-1, , ,000-2, , ,000-3, , ,000-4, , ,000-7, , , , Total 8, , a. Growth rates are annualised, e.g. growth of X from 0 to T is (100/T)*(X T -X 0 )/X 0. Note that this is not the same as the average annual rate of growth. Note The relationship between the growth rates of productivity, value added and employment is given equation (3). 22

24 Table 8 Decomposition of productivity growth a amongst survivors, : by 1979 employment size group Employees 1979 (range) With effect Total contribution Reallocation effect: level Reallocation effect: growth % % % % ,000-1, ,500-1, ,000-2, ,000-3, ,000-4, ,000-7, , Total a. The growth rate of productivity q from 0 to T is measured as (100/T)*(q T -q 0 )/q 0. Note that this is not the same as the average annual rate of growth. Note Each group s contribution is given by equation (2). For further explanation see text. establishments consistg of only one local unit. More than three quarters of employment was establishments of 4 or fewer local units and over 90% was establishments with 10 or fewer (Table 9). There was little change the distribution between 1980 and If we consider only the 7,451 establishments which are present both 1980 and 1989, we fd that 64.6% of them had no change the number of local units, 17.2% reduced the number and 18.2% creased it. There was fact a small crease the mean number of local units per establishment amongst these survivors and the mean number of employees per local unit such establishments fell by 19% between 1980 and It seems then that downsizg took the form maly of reducg employment contug plants rather than closg plants. 23

25 Table 9 Distribution of employment by number of local units per establishment, 1980 and 1989 No. of local units per establishment Proportion of employment Cumulative proportion of employment % % % % 1 a a. 1 local unit means that the establishment and the local unit cocide. Note 1980: 14,909 establishments with total employment of million. 1989: 18,714 establishments with total employment of million. 24

26 5.4 Disaggregatg to the sector level So far the results have been at the level of all manufacturg. But it is of terest to see whether any differences emerge at lower levels of aggregation. Table 10 shows the breakdown by quadrant for 20 Classes with SIC80. (Two Classes, 21 and 26, are omitted to avoid disclosure due to the small number of establishments; see the Appendix for the names of the Classes). Successful downsizers (quadrant 2) predomate at the sectoral as well as the aggregate level, accountg for the bulk of productivity growth every Class. Productivity growth rates quadrant 2 range from a low of 4.01% Class 23, Extraction of merals n.e.s., to a high of 71.85% (annualised rates) Class 33, Office machery. In the latter case, the productivity growth rate is heavily fluenced by the Producer Price Index for this sector which fell by 41% between 1979 and Successful upsizers (quadrant 1) account for a mority of employment every sector. Their share varies between a low of 3.6% Class 35, Motor vehicles, and a high of 31.5% Class 41, Food. The largest share for quadrant 3, where both employment and productivity are fallg, is found Class 37, Instrument engeerg (33.7%). The reasons for this apparently poor performance would clearly merit further vestigation. 25

27 Table 10 Productivity amongst survivors, : by SIC80 Class SIC80 Class Quadrant Number of establishments Employment share 1979 Employment share 1989 Productivity 1979 Annualised growth a of value employment productivity added % % 1990 % % % , , , , , , , N/A N/A N/A N/A N/A , , , , , , , , , , , , , , , , , , N/A N/A N/A N/A N/A N/A N/A N/A 26

28 Table 10, contued. SIC80 Class Quadrant Number of establishments Employment share 1979 Employment share 1989 Productivity 1979 Annualised growth a of value employment productivity added , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

29 Table 10, contued. SIC80 Class Quadrant Number of establishments Employment share 1979 Employment share 1989 Productivity 1979 Annualised growth a of value employment productivity added , , , , , , , , , , , , , , , , , , , , a. The growth rate of X from 0 to T is measured as (100/T)*(X T -X 0 )/X 0. Note that this is not the same as the average annual rate of growth. Note Classes 21 and 26 and Quadrant 4 of Class 23 are omitted to avoid disclosure, sce these had only a handful of survivg establishments. See the Appendix for the names of the Classes. 28

30 6. A comparison with The period 1973 to 1979 was one which, usg aggregate statistics, manufacturg employment fell by 7.0%, output fell by 4.2% and productivity rose by 2.8% (see Table 1 and Charts 1-3) was a cyclical peak and were recession years, associated with the first oil shock. The period was marked by extensive government tervention, cludg comes policies and the bailg out of lame ducks, and is also notorious for bad labour relations. It therefore provides a natural basis for comparison with Tables do the same analysis for as has just been done for Gettg on for 10,000 establishments, accountg for a quarter of employment, exited between 1973 and This was a lower proportion than durg the subsequent cycle. However, the latter was longer, 11 years peak to peak as opposed to 7. The mean productivity level of survivors fell by 6% over this period. Changes establishment size were fairly small. In 1973 a typical worker worked an establishment employg 991 people; his 1979 counterpart worked one employg 956 workers. There was little change the dispersion of productivity over this period (Table 11). As , the establishments which exited had lower productivity than the survivors, by 12%. Interestgly, the entrants had higher productivity than the survivors Tables 12 and 13, to be compared with Tables 3 and 4, show the arithmetic. Productivity fell amongst survivors, but this was outweighed by the entry and exit process, so the period as a whole shows a small rise productivity. 29

31 Table 11 Survivors, exits and entrants compared: ARD respondents manufacturg, 1973 and 1979 Survivors Exits Entrants Overall Variable No. of establishments 11,625 11,625 9,664 6,138 21,289 17, Share of employment (%) 3. Share of GVA (%) Mean no. of employees Median no. of employees 6. Median of 1 st moment distribution of employment (no.) 7. Mean productivity level (1990 per person employed) 8. Median productivity level (1990 per person employed) 9. S.D. of log productivity th percentile of productivity as % of median th percentile of productivity as % of median 12. Mean crease productivity, (%) 13. Median crease productivity, (%) ,221 1, ,144 14,248 13,374 15,188 14,341 14,573 12,462 11,840 11,145 12,267 11,851 11, Source ARD. 30

32 Table 12 Productivity and employment, : survivors versus exits and entrants Share of employment 1973 Share of employment 1979 Productivity 1973 Annualised growth a of value employment productivity added % % 1990 % % % Exits/entrants , Survivors , Total , a. The growth rate of X from 0 to T is measured as (100/T)*(X T -X 0 )/X 0. Note that this is not the same as the average annual rate of growth. Note The relationship between the growth rates of productivity, value added and employment is given equation (3). Table 13 Decomposition of productivity growth a, : survivors versus exits and entrants Total contribution With Realloc effect -ation effect: level Realloc -ation effect: growth Group % % % % Exits/entrants Survivors Total a. Growth rates are annualised, e.g. growth of X from 0 to T is (100/T)*(X T -X 0 )/X 0. Note Each group s contribution is given by equation (2). For further explanation see text. When we do the analysis by quadrants (Tables 14 and 15, compare Tables 5 and 6), we fd the big difference between the two cycles lies the employment share of successful downsizers (quadrant 2). These accounted for under a third of employment 1973 compared to two thirds The counterpart was much large shares for unsuccessful downsizers and upsizers (quadrants 3 and 4). These two quadrants, which by defition productivity fell, accounted for 54% of itial employment; the correspondg percentage for was 18%. Quadrant 1, the successful upsizers, 31

33 had a similar employment share both cycles. Indeed it made a larger contribution to overall productivity growth amongst survivors the first than the second cycle. But its contribution was more than outweighed by poor performance the other quadrants. Table 14 Productivity and employment amongst survivors, : by quadrant Quadrant Number of establishments Share of employment 1973 Share of employment 1979 Productivity 1973 Annualised growth a of value employment productivity added No. % % 1990 % % % 1 ( q>0, E>0) 2, , ( q>0, E<0) 2, , ( q<0, E<0) 3, , ( q<0, E>0) 3, , Total 11, , a. The growth rate of X from 0 to T is measured as (100/T)*(X T -X 0 )/X 0. Note that this is not the same as the average annual rate of growth. Note The relationship between the growth rates of productivity, value added and employment is given equation (3). Table 15 Decomposition of productivity growth a amongst survivors, : by quadrant Quadrant With effect Total contribution Reallocation effect: level Reallocation effect: growth % % % % 1 ( q>0, E>0) ( q>0, E<0) ( q<0, E<0) ( q<0, E>0) Total a. Growth rates are annualised, e.g. growth of X from 0 to T is (100/T)*(X T -X 0 )/X 0. Note Each group s contribution is given by equation (2). For further explanation see text. 32

34 For successful downsizers (quadrant 2), productivity growth was quite rapid at an annualised 6.29% compared with 9.21% the later cycle. But because of its smaller share, the contribution of this quadrant was about the same as that of quadrant 1. The contributions of the other two quadrants were necessarily negative. Ignorg sign, all four quadrants made contributions of roughly equal size. Table 16 Productivity and employment amongst survivors, : by 1973 employment size group Employees 1973 (range) Number of establishments Share of employment 1973 Share of employment 1979 Productivity 1973 Annualised growth a of value employment productivity added No. % % 1990 % % % 49 2, , , , , , , , , , , , ,000-1, , ,500-1, , ,000-2, , ,000-3, , ,000-4, , ,000-7, , , , Total 11, , a. Growth rates are annualised, e.g. growth of X from 0 to T is (100/T)*(X T -X 0 )/X 0. Note that this is not the same as the average annual rate of growth. Note The relationship between the growth rates of productivity, value added and employment is given equation (3). 33

35 Tables 16 and 17 analyse productivity growth survivors by employment size 1973 (compare Tables 7 and 8). Employment rose moderately establishment with under 500 employees and fell larger establishments (except for the top size class). But whereas productivity rose strongly all size classes, and exceptionally rapidly the largest, productivity fell 10 out of 14 size classes; the other 4 classes, growth was very modest. In consequence, each size class made a small, mostly negative, contribution to overall productivity growth. Table 17 Decomposition of productivity growth a amongst survivors, : by 1973 employment size group Employees 1973 (range) With effect Total contribution Reallocation effect: level Reallocation effect: growth % % % % ,000-1, ,500-1, ,000-2, ,000-3, ,000-4, ,000-7, , Total a. The growth rate of productivity q from 0 to T is measured as (100/T)*(q T -q 0 )/q 0. Note that this is not the same as the average annual rate of growth. Note Each group s contribution is given by equation (2). For further explanation see text. 34

36 7. Conclusions The ma conclusions are as follows: 1. Over , the dispersion of productivity across establishments actually creased. The so-called long tail of under-performers lengthened. There was huge variation the growth of productivity. Amongst survivg establishments, a quarter experienced negative productivity growth and these accounted for 18% of employment 1979 and 21% Over , closures are not able to expla the upsurge productivity which occurred. True, establishments which disappeared had lower productivity than those which survived. But this by itself would have had only a small effect on overall productivity. Moreover the establishments which disappeared were replaced by new entrants which also had comparatively low productivity. 3. The bulk of productivity growth can be ascribed to establishments where employment fell. This is true both aggregate and at the sectoral level. Nevertheless, establishments which employment grew accounted for 16% of productivity growth, despite the 25% fall employment manufacturg as a whole. 4. Productivity growth was highest the largest establishments. In fact, the top 36 establishments, each with 7,500 or more workers and employg 17% of the 1979 workforce, accounted for almost a third of aggregate productivity growth. 5. In the earlier cycle of , productivity growth was quite rapid those establishments where it rose at all. The problem was that amongst survivors, over half of employment was establishments which productivity fell. The purpose of this paper has been maly descriptive. Much further work is clearly needed to exame the determants of entry and exit and the causes of productivity growth In a companion paper I exame the role of physical vestment explag productivity growth (Oulton 1998). 35

37 References Baily, M.N., Bartelsman, E.J. and Haltiwanger, J. (1996). Downsizg and Productivity Growth: Myth or Reality?. In Sources of Productivity Growth, edited by D.G. Mayes. Cambridge: Cambridge University Press. Bean, C. and Crafts, N.F.R. (1995). British Economic Growth Sce 1945: Relative Economic Decle... and Renaissance?. In Economic Growth Europe Sce 1945, edited by N. Crafts and G. Toniolo. Cambridge: Cambridge University Press. Blanchflower, D. and Freeman, R.B. (1994). Did the Thatcher Reforms Change British Labour Market Performance?. In R. Barrell (ed.), The UK Labour Market, Cambridge: Cambridge University Press. Denny, K. and Nickell, S.J. (1992). Unions and Investment British Industry. Economic Journal, 102, pp Gregg, P., Mach, S. and Metcalf, D. (1993). Signals and Cycles: Productivity Growth and Changes Union Status UK Companies, Economic Journal, 103, pp Hart, P.E. (1975). Moment Distributions Economics: An Exposition. Journal of the Royal Statistical Society, Series A, 138, Part 3, Haskel, J., (1991). Imperfect Competition, Work Practices and Productivity Growth. Oxford Bullet of Economics and Statistics, 53, No. 3 (August), pp Industry Economics and Statistics Directorate (1997). Differences companies performance: British dustry s under-performg tail. Department of Trade and Industry. London. Lansbury, M. and Mayes, D.G. (1996). Entry, Exit, Ownership and the Growth of Productivity. In Sources of Productivity Growth, edited by D.G. Mayes. Cambridge: Cambridge University Press. Muellbauer, J. (1986). The Assessment: Productivity and Competitiveness British Manufacturg. Oxford Review of Economic Policy, 2, pp. i-xxv. Oulton, N. (1987). Plant Closures and the Productivity Miracle Manufacturg, National Institute Economic Review, No. 121 (August), pp Oulton, N. (1989). Productivity Growth Manufacturg, : the Roles of New Investment and Scrappg, National Institute Economic Review, No. 127 (February), pp