- Email: [email protected]
Channels of size adjustment and firm performance
Economics Letters 116 (2012) 202–206
Contents lists available at SciVerse ScienceDirect
Economics Letters journal homepage: www.elsevier.com/locate/ecolet
Channels of size adjustment and firm performance✩ Holger Breinlich a,∗ , Stefan Niemann b , Edna Solomon c a
University of Essex, CEP, LdA and CEPR, United Kingdom
b
University of Essex, United Kingdom
c
University of Greenwich, United Kingdom
article
info
Article history: Received 18 December 2011 Received in revised form 10 February 2012 Accepted 15 February 2012 Available online 22 February 2012
abstract We use business register data for the United Kingdom to document the importance of the different channels that firms use to adjust their size. We show how the choice of adjustment channel impacts upon firm-level variables such as wages or productivity. © 2012 Elsevier B.V. All rights reserved.
JEL classification: E22 G31 G32 G33 G34 L25 Keywords: Adjustment channels Mergers and acquisitions Greenfield investment Investment Employment
1. Introduction Firms constantly adapt to changes in their market environment through changes in the scale and scope of their operations. The magnitude and consequences of the resulting micro-level adjustments have been extensively documented in the literature (e.g. Davis et al., 2006). While attention has also been paid to the different channels through which firm growth and contractions take place (e.g. Davis et al., 1996), there is little systematic analysis of how firms choose between the available channels and what
✩ This work contains statistical data from ONS which is Crown Copyright. The use of the ONS statistical data in this work does not imply the endorsement of the ONS in relation to the interpretation or analysis of the statistical data. This work uses research datasets which may not exactly reproduce National Statistics aggregates. We gratefully acknowledge financial support by the Leverhulme Trust through Research Project Grant F/00 213/P. ∗ Correspondence to: Wivenhoe Park, Colchester CO4 3SQ, United Kingdom. Tel.: +44 0 1206 87 2768; fax: +44 0 1206 87 2724. E-mail address: [email protected] (H. Breinlich).
0165-1765/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2012.02.020
the consequences of these choices are for firm-level performance. Relatedly, most theoretical models of firm dynamics focus on the overall change in employment or turnover of firms, but are silent about how these changes are achieved (e.g. Jovanovic, 1982; Hopenhayn, 1992; Hopenhayn and Rogerson, 1993; Asplund and Nocke, 2006). In practice, there are three principal ways in which firms can expand or contract. First, they can adjust employment or output at existing production facilities while continuing to use them (‘internal adjustment’). Second, contracting firms can shut down establishments or divisions, and expanding firms can decide to open up new ones (‘greenfield investment/disinvestment’). Third, firms can use the market for corporate control to buy or sell parts or the entirety of their operations (‘mergers and acquisitions’, M&As). In this paper, we use unique business register data for the United Kingdom to present a novel set of stylized facts on the relative importance of these three channels, and what the associated changes in a number of key firm-level variables such as wages or productivity are.
H. Breinlich et al. / Economics Letters 116 (2012) 202–206
2. Data and methodology We use two data sources for the United Kingdom for the period between 1997 and 2005.1 The first is the Business Structure Database (BSD) which covers essentially the entire British economy, accounting for 99% of aggregate employment and turnover. The second database we use is the Annual Respondents Database (ARD). The ARD is based on a stratified sample (drawn from the BSD population) of over 40,000 UK private sector companies per year and contains a large number of variables not available in the BSD, such as wages, investment and intermediate inputs. We merge the BSD with the ARD using identifiers common to both datasets. The BSD allows the analysis of demographic events over time by capturing the ownership structure of firms and plants via a system of reference numbers at different levels of aggregation. In this paper we focus on so-called enterprise groups and enterprises, which are best thought of as firms and plants, respectively.2 In our methodology, the most basic event is a change in employment at a continuing plant (‘internal adjustment’). This is easily observed from the entries of two adjacent years for the same plant. If a plant identifier disappears from the data, we code this as plant exit (‘greenfield disinvestment’). Likewise, the appearance of a new identifier is coded as plant entry (‘greenfield investment’). Finally, the combination of plant and firm references allows for the analysis of ownership changes. For example, if firm A buys plant 1 from firm B, the plant reference number of plant 1 remains unchanged but its firm identifier changes from A to B.
203
We also investigate how the choice of adjustment channel correlates with changes in a number of firm characteristics. In analogy to similar decompositions in the productivity literature (e.g. Foster et al., 2006), we decompose the change of a firmlevel variable ywt into the contributions of the different adjustment channels:
1ywt ≈
set −1 1yet +
e∈C +
+
(yet −1 − yw\et −1 )1set
e∈C +
1yet 1set +
e∈C +
+
set −1 1yet
e∈C −
(yet −1 − yw\et −1 )1set +
e∈C −
+
set (yet − yw\et −1 ) −
e∈N
+
1yet 1set
e∈C −
set −1 (yet −1 − yw\et −1 )
e∈X
set (yet − yw\et −1 ) −
e∈A
set −1 (yet −1 − yw\et −1 ) (1)
e∈S
where 1y denotes log-changes in the plant/firm variable of interest between periods t − 1 and t, and set the share of plant e in the total employment of firm w in period t. The first two lines of (1) capture the contribution of internal adjustment (C + and C − stand for continuing and expanding, and continuing and contracting, respectively). The fourth and fifth line of (1) capture the contribution of the greenfield investment, exit, acquisitions, and sales (N, X , A, and S, respectively). This decomposition suggests looking at the following quantities:
• ‘Composition effects’. Changes in ywt due to changes in the 3. Results We start our presentation of stylized facts with an overview of the importance of the different adjustment channels for total firm growth. In line with existing work we focus on continuing firms, i.e., firms which existed in the current period, will not exit in their entirety in the next period, and which change employment between periods. The most basic question we are interested in is how frequently the various channels are used. Panel A of Table 1 displays the fraction of all employment adjustments which take place through each of the three channels.3 It is evident that M&As and greenfield investment are rare events. On average, these two channels were used in only about 3% of employment changes, with the vast majority of both expansions and contractions occurring via internal adjustments. Panel B shows that, when they take place, M&As and greenfield adjustments are major events. The average M&A expansion (contraction) is 15 (20) times bigger than the average internal expansion (contraction). Greenfield investments are smaller than M&A expansions but still around 10 times bigger than the average internal expansion; greenfield disinvestments, in contrast, are of comparable magnitude to M&As. Panel B implies that despite their infrequent occurrence, greenfield investment and M&As still account for a large fraction of overall employment adjustments. The two forms together make up 23% of economy-wide employment expansions and 37% of employment contractions of continuing firms, with M&As being quantitatively more important than greenfield adjustments (see Panel C).
1 For a detailed description of our data and methodology, as well as additional results, see Breinlich et al. (2011). 2 Henceforth, we will uniformly use the terms firm and plant instead of enterprise group and enterprise. 3 We use employment rather than turnover adjustments since the former are much more directly under the control of a firm.
composition of the set, or the relative importance, of the plants making up a firm. What matters for ywt here is the level of variable yet of the plant undertaking an adjustment, relative to all other plants in the same firm (i.e., yet − yw\et −1 and yet −1 − yw\et −1 ).4 • ‘Change effects’. Changes in ywt due to changes in yet for existing and continuing plants. In the above decomposition, these are the changes in the plant-level variable yet associated with internal expansion or contraction (1yet ). 3.1. Composition effects To obtain estimates for yet − yw\et −1 and yet −1 − yw\et −1 , we first calculate the employment-weighted average of y across all the plants of the firm not undergoing the demographic event in question (yw\et −1 ). We then subtract these averages from the level of y of the firm undergoing the demographic event (yet or yet −1 ), using an appropriate lag structure as indicated in (1). This procedure yields one difference for each of the firms and events for which we have sufficient data. Table 2 shows estimates of the mean of these differences across firms, together with the corresponding standard error and the significance level for a twosided tests of whether the mean is different from zero. We present results for a number of firm-level variables commonly analyzed in the literature: wages, labor and total factor productivity, and a measure of firm profitability (operating profits per employee). Given that these variables are not available in the BSD, we use the merged ARD–BSD dataset here and in Section 3.2, with data on the firm characteristics (y) coming from the ARD and the demographic events from the BSD. As seen, plants undergoing one of the four external adjustment forms (birth, exit, acquisition, sale) tend to be less productive
4 The subscript w \ e denotes all plants of firm w excluding plant e, and y w\e is the employment-weighted arithmetic average of variable y across these plants.
204
H. Breinlich et al. / Economics Letters 116 (2012) 202–206
Table 1 Descriptive statistics, adjustment strategies (1997–2005). Source: Office for National Statistics and authors’ calculations. Internal adjustment
M&A
Greenfield
All adjustments
Panel A: usage frequency of adjustment channels (%, total count in last column) Gross expansions Gross contractions
99.11 98.38
1.53 1.71
0.60 1.12
138.19 148.34
86.74 161.29
1,672,516 1,418,009
Panel B: average size of adjustment by channel (number of employees) Gross expansions Gross contractions
9.07 7.45
11.46 11.57
Panel C: aggregate importance of adjustment channels (% of total adjustment, last column indicates total annual average employment change) Gross expansions Gross contractions
77.32 62.65
18.23 21.63
4.52 15.49
2,771,429 2,371,429
Notes: Panel A shows the fraction of all employment adjustments by continuing firms involving internal adjustment, M&As and greenfield investment or disinvestment. Note that firms can use several channels at the same time so that the percentages do not have to add up to 100%. Panel B shows the average transaction size of each of these channels, and Panel C their contribution to overall adjustment. All figures are averages over 1997–2005. Table 2 Composition effects of the choice of adjustment channel (1997–2005). Source: Office for National Statistics and authors’ calculations.
Labor productivity Wages per employee TFP (OLS) Operating profits per employee Observations
Birth
Exit
Acquisition
Sale
Internal expansion
Internal contraction
−0.050
−0.145
−0.021
−0.146
0.009 (0.017) 0.022 (0.010)** 0.013 (0.007)* −0.023 (0.028) 2435–3301
−0.065
(0.052) 0.025 (0.031) −0.053 (0.026)** −0.024 (0.111) 234–379
(0.034)***
−0.034
(0.015)
−0.031
(0.020)*
−0.033
(0.009)***
−0.019
(0.014)**
−0.079
(0.016)***
−0.030
(0.007)***
−0.021
(0.055) 623–951
(0.027)***
−0.090 (0.012)**
−0.198
(0.026) 2336–3236
(0.046)*** 885–1205
(0.016)***
−0.038 (0.010)***
−0.012 (0.007)*
−0.159 (0.026)*** 2799–3842
Notes: Table shows average log differences between plants undergoing an adjustment form and the other plants in the same firm (see text for details). They are derived from simple descriptive OLS regressions. Figures in brackets denote robust standard errors clustered at the enterprise level. * Denote significance at the 10% level. ** Denote significance at the 5% level. *** Denote significance at the 1% level. Table 3 Change effects for existing and continuing plants (1997–2005). Source: Office for National Statistics and authors’ calculations.
Internal expansion Internal contraction Acquisition Sale Observations
Labor productivity
Wages per employee
TFP (OLS)
Operating profits per employee
−0.020
−0.021
−0.002
−0.021
(0.011)* 0.061 (0.011)*** 0.015 (0.015) 0.010 (0.012) 36,206
(0.006)*** 0.047 (0.006)*** 0.014 (0.007)* 0.003 (0.006) 38,945
(0.006) 0.013 (0.006)** 0.016 (0.007)** −0.000 (0.006) 30,243
(0.023) 0.067 (0.023)*** 0.009 (0.029) 0.054 (0.025)** 28,358
Notes: Table shows the effects of the corporate events listed in the first column on existing and continuing plants. Results are estimated using OLS regressions with industry–year fixed effects. Figures in brackets are robust standard errors clustered at the plant level. Each column denotes a separate regression, where the growth rates of the variables of interest are regressed on the event dummies. The omitted group are plants that do neither expand nor contract. * Indicate statistical significance at the 10% level. ** Indicate statistical significance at the 5% level. *** Indicate statistical significance at the 1% level.
and pay lower wages. There are no significant differences in profitability, however, with the exception of plants which are sold off — these report profits of 20% less than the remaining plants in the same firm. Turning to the internal adjustment forms, contracting plants show performance characteristics comparable to those being sold off, such as low productivity and profitability. Expanding plants, on the other hand, tend to be slightly more productive and pay higher wages than the other plants in the same firm.
be obtained via simple dummy variable regressions of 1yet on binary indicators for internal expansion/contraction. We use plants which neither expand nor contract employment between adjacent periods as the omitted category in these regressions, so that all changes are expressed relative to this group. For comparison with the existing literature, we also include plants undergoing an acquisition or sale. Table 3 presents results for the same variables as in Section 3.1, again using the merged ARD–BSD dataset. We control for industry–year fixed effects in all regressions, so that results rely on within-sector-year variation only.5
3.2. Change effects Estimates of the average log-change in the plant-level variable yet associated with internal expansion or contraction (1yet ) can
5 If, after controlling for industry–year fixed effects, plants with unchanged employment represent a suitable control group for the other demographic events,
H. Breinlich et al. / Economics Letters 116 (2012) 202–206
The general pattern in our results is that internal expansions tend to reduce wages and profits, as well as labor and total factor productivity. Internal contractions have almost exactly the opposite pattern. They increase both productivity measures as well as profitability and wages. Patterns are less pronounced for both acquisitions and sales, but some interesting patterns emerge here as well. For example, similar to Maksimovic and Phillips (2001) and Schoar (2002), we find that acquired plants subsequently increase their productivity, although this effect is only significant for the TFP measure. Besides serving as input into the decomposition (1), these results allow for some interesting comparisons with the existing literature, which usually focuses on only a subset of the channels analyzed here. For example, our results show that while asset transfers via M&As are not always associated with significant productivity gains, internal expansions actually reduce productivity, with the difference between the two expansion forms being highly statistically significant. This demonstrates that, when analyzing the consequences of corporate events such as acquisitions, the choice of control group is crucial. In the previous example, both acquisitions and internal expansions serve the purpose of increasing overall firm size, and might imply similar adjustment processes to the operations of a firm. It would thus seem natural to use internally expanding plants as a control group in the estimation of the effects of acquisitions. As our results demonstrate, this choice would let acquisitions appear in a much better light than when compared to, for example, plants not undergoing any change. 3.3. Contribution of adjustment channels to changes in firm-level variables We now combine the results from Sections 3.1 and 3.2 with the decomposition proposed in (1) to shed light on the contribution of our adjustment channels to changes in the firm-level variables d here. Recall from (1) that we need the changes in the variable of interest (y) associated with the different adjustment channels, as well as the employment shares (s) of the plants undergoing a given demographic event. Ideally, we would like to use data on both y and s from the merged ARD–BSD dataset. A problem in this context is that the matching procedure between the ARD and the BSD only results in a small subset of firms with a complete set of plants — which is of course needed to calculate the employment shares in (1). The reason for this is that sampling in the ARD takes place at a scale equivalent to the plant level (not the firm level), and the ARD covers less than 5% of the plants in the BSD in a given year. Consequently, we only have the full set of plants for less than 2% of firms. To make progress, we thus use employment shares from the BSD (where coverage is close to complete, but we only have information on employment and turnover) with our estimates of the average changes in y associated with the different adjustment forms derived in Sections 3.1 and 3.2. That is, we rewrite (1) as
1 ywt =
set −1 1 yet +
e∈C +
+
(yet −1 − yw\et −1 )1set
e∈C +
e∈C +
1 yet 1set +
set −1 1 yet
e∈C −
the following results have a causal interpretation. However, it is possible that there are pre-existing level or trend differences which induce firms to select into a particular adjustment channel. Identifying truly causal effects would have to rely on sources of exogenous variation in the choice of adjustment channels, which is beyond the scope of this short note and the subject of ongoing work.
+
205
(yet −1 − yw\et −1 )1set +
e∈C −
+
1 yet 1set
e∈C −
set (yet − yw\et −1 ) −
e∈N
+
set −1 (yet −1 − yw\et −1 )
e∈X
set (yet − yw\et −1 ) −
e∈A
set −1 (yet −1 − yw\et −1 ) (2)
e∈S
where hats above the variables indicate quantities estimated in the last sections, using data from the merged ARD–BSD sample. For example, we found that exiting plants have a labor productivity which is on average 14.5% lower than the remaining plants of the same firm (see Table 2). We thus set
−
set −1 (yet −1 − yw\et −1 ) = −
e∈X
set −1 × (−0.145)
e∈X
where set −1 , the employment share of the exiting plant, is now calculated from the BSD data. Implicitly, we are thus making the assumption that the sample used for the estimations in the previous sections is representative of the BSD in terms of the effects associated with the different adjustment channels. To make this assumption more plausible, we only keep those firms in the BSD from which at least one plant can be matched to the ARD at some point in our sample. This makes the set of firms used in our decomposition analysis more comparable to the type of firms represented in the ARD.6 For each firm in the BSD, and for each left-hand side variable y, the decomposition in (2) yields a predicted growth rate of y, as well as the contributions of the different adjustment channels. In Table 4, we report an average value for each decomposition term, expressed as a fraction of the sum of all terms (in absolute values). Depending on the variable in question, the external adjustment forms (in particular, exits and sales) account for around 50% of the overall mean changes at the firm level. External adjustment forms are particularly important for growth in total factor productivity, where they account for 70% of the overall mean change. Our results thus indicate that, despite their infrequent occurrence, external adjustments contribute to a large degree to the overall changes in firm-level variables. While there are of course important differences in terms of the underlying data and the decomposition approaches used, this dominance of external adjustment is reminiscent of recent findings in the productivity literature using plant-level data (e.g., Disney et al., 2003; Foster et al., 2006). 4. Conclusions We presented a novel set of stylized facts on forms of firm expansion and contraction, using unique business register data for the United Kingdom. In contrast to contributions in the existing literature, our data enabled us to distinguish between all three principal adjustment channels that firms can use to change the scale and scope of their operations: changes of employment at existing establishments, greenfield investments
6 Note that we thus do not require our estimates to be representative of the BSD as a whole but only of those firms which at some point had at least one of their plants sampled in the ARD. Since ARD plants are the unit of analysis in Tables 2 and 3, our predicted values in (2) should be more representative of this restricted sample. Given that the ARD oversamples larger plants, and that larger plants tend to be part of larger firms, such firms will be overrepresented in the restricted sample. As such, our results are best thought of as applying mainly to larger firms. As is true in the UK as in other countries, however, these firms make up the majority of economic activity. In additional robustness checks (available from the authors upon request) we show that further restricting our sample to the subset of firms with plants which we have actually used in the estimations in Sections 3.1 and 3.2 (rather than all ARD plants, some of which have missing data) does not alter any of the conclusions drawn below.
206
H. Breinlich et al. / Economics Letters 116 (2012) 202–206
Table 4 Firm-level decompositions (1997–2005) — average value for each decomposition term, expressed as a fraction of sum of all terms (in absolute values). Source: Office for National Statistics and authors’ calculations. Component
Labor productivity (%)
Wages per employee (%)
TFP (OLS) (%)
Operating profits per employee (%)
Within effect (internal expansion) Between effect (internal expansion) Cross effect (internal expansion) Within effect (internal contraction) Between effect (internal contraction) Cross effect (internal contraction) Contribution of birth Contribution of exit Contribution of acquisition Contribution of sale Avg. predicted change in decomposed variable Observations
11.53 0.15 0.17 26.54 0.49 0.23 0.28 34.16 4.03 22.41 2.50 445,134
19.67 0.61 0.29 33.33 0.47 0.28 0.23 13.01 9.67 22.44 0.91 445,136
4.87 0.94 0.07 24.28 0.38 0.21 1.25 32.98 15.48 19.53 0.49 445,138
12.56 0.41 0.19 30.20 1.25 0.26 0.14 19.30 4.18 31.52 2.30 445,136
Notes: Table shows the average contribution of each component in the first column to changes in the firm-level variables listed in the top row.
and disinvestment, and M&As. We documented the relative importance of these three channels and showed what the associated changes in a number of firm-level variables were. References Asplund, M., Nocke, V., 2006. Firm turnover in imperfectly competitive markets. Review of Economic Studies 73, 295–327. Breinlich, H., Niemann, S., Solomon, E., 2011. Channels of Size Adjustment and Firm Performance. Discussion Paper No. 703, University of Essex. Davis, S., Faberman, R., Haltiwanger, J., 2006. The flow approach to labor markets: new data sources and micro–macro links. Journal of Economic Perspectives 20 (3), 3–26. Davis, S., Haltiwanger, J., Schuh, S., 1996. Job Creation and Destruction. MIT Press, Cambridge.
Disney, R., Haskel, J., Heden, Y., 2003. Restructuring and productivity growth in UK manufacturing. Economic Journal 113, 666–694. Foster, L., Haltiwanger, J, Krizan, C.J., 2006. Market selection, reallocation, and restructuring in the US retail trade sector in the 1990s. Review of Economics and Statistics 88, 748–758. Hopenhayn, H., 1992. Entry, exit, and firm dynamics in long run equilibrium. Econometrica 60, 1127–1150. Hopenhayn, H., Rogerson, R., 1993. Job turnover and policy evaluation: a general equilibrium analysis. Journal of Political Economy 101, 915–938. Jovanovic, Boyan, 1982. Selection and the evolution of industry. Econometrica 50, 649–670. Maksimovic, V., Phillips, G., 2001. The market for corporate assets: who engages in mergers and asset sales and are there efficiency gains? Journal of Finance 56, 2019–2065. Schoar, A., 2002. Effects of corporate diversification on productivity. Journal of Finance 57, 2379–2403.
Contents lists available at SciVerse ScienceDirect
Economics Letters journal homepage: www.elsevier.com/locate/ecolet
Channels of size adjustment and firm performance✩ Holger Breinlich a,∗ , Stefan Niemann b , Edna Solomon c a
University of Essex, CEP, LdA and CEPR, United Kingdom
b
University of Essex, United Kingdom
c
University of Greenwich, United Kingdom
article
info
Article history: Received 18 December 2011 Received in revised form 10 February 2012 Accepted 15 February 2012 Available online 22 February 2012
abstract We use business register data for the United Kingdom to document the importance of the different channels that firms use to adjust their size. We show how the choice of adjustment channel impacts upon firm-level variables such as wages or productivity. © 2012 Elsevier B.V. All rights reserved.
JEL classification: E22 G31 G32 G33 G34 L25 Keywords: Adjustment channels Mergers and acquisitions Greenfield investment Investment Employment
1. Introduction Firms constantly adapt to changes in their market environment through changes in the scale and scope of their operations. The magnitude and consequences of the resulting micro-level adjustments have been extensively documented in the literature (e.g. Davis et al., 2006). While attention has also been paid to the different channels through which firm growth and contractions take place (e.g. Davis et al., 1996), there is little systematic analysis of how firms choose between the available channels and what
✩ This work contains statistical data from ONS which is Crown Copyright. The use of the ONS statistical data in this work does not imply the endorsement of the ONS in relation to the interpretation or analysis of the statistical data. This work uses research datasets which may not exactly reproduce National Statistics aggregates. We gratefully acknowledge financial support by the Leverhulme Trust through Research Project Grant F/00 213/P. ∗ Correspondence to: Wivenhoe Park, Colchester CO4 3SQ, United Kingdom. Tel.: +44 0 1206 87 2768; fax: +44 0 1206 87 2724. E-mail address: [email protected] (H. Breinlich).
0165-1765/$ – see front matter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2012.02.020
the consequences of these choices are for firm-level performance. Relatedly, most theoretical models of firm dynamics focus on the overall change in employment or turnover of firms, but are silent about how these changes are achieved (e.g. Jovanovic, 1982; Hopenhayn, 1992; Hopenhayn and Rogerson, 1993; Asplund and Nocke, 2006). In practice, there are three principal ways in which firms can expand or contract. First, they can adjust employment or output at existing production facilities while continuing to use them (‘internal adjustment’). Second, contracting firms can shut down establishments or divisions, and expanding firms can decide to open up new ones (‘greenfield investment/disinvestment’). Third, firms can use the market for corporate control to buy or sell parts or the entirety of their operations (‘mergers and acquisitions’, M&As). In this paper, we use unique business register data for the United Kingdom to present a novel set of stylized facts on the relative importance of these three channels, and what the associated changes in a number of key firm-level variables such as wages or productivity are.
H. Breinlich et al. / Economics Letters 116 (2012) 202–206
2. Data and methodology We use two data sources for the United Kingdom for the period between 1997 and 2005.1 The first is the Business Structure Database (BSD) which covers essentially the entire British economy, accounting for 99% of aggregate employment and turnover. The second database we use is the Annual Respondents Database (ARD). The ARD is based on a stratified sample (drawn from the BSD population) of over 40,000 UK private sector companies per year and contains a large number of variables not available in the BSD, such as wages, investment and intermediate inputs. We merge the BSD with the ARD using identifiers common to both datasets. The BSD allows the analysis of demographic events over time by capturing the ownership structure of firms and plants via a system of reference numbers at different levels of aggregation. In this paper we focus on so-called enterprise groups and enterprises, which are best thought of as firms and plants, respectively.2 In our methodology, the most basic event is a change in employment at a continuing plant (‘internal adjustment’). This is easily observed from the entries of two adjacent years for the same plant. If a plant identifier disappears from the data, we code this as plant exit (‘greenfield disinvestment’). Likewise, the appearance of a new identifier is coded as plant entry (‘greenfield investment’). Finally, the combination of plant and firm references allows for the analysis of ownership changes. For example, if firm A buys plant 1 from firm B, the plant reference number of plant 1 remains unchanged but its firm identifier changes from A to B.
203
We also investigate how the choice of adjustment channel correlates with changes in a number of firm characteristics. In analogy to similar decompositions in the productivity literature (e.g. Foster et al., 2006), we decompose the change of a firmlevel variable ywt into the contributions of the different adjustment channels:
1ywt ≈
set −1 1yet +
e∈C +
+
(yet −1 − yw\et −1 )1set
e∈C +
1yet 1set +
e∈C +
+
set −1 1yet
e∈C −
(yet −1 − yw\et −1 )1set +
e∈C −
+
set (yet − yw\et −1 ) −
e∈N
+
1yet 1set
e∈C −
set −1 (yet −1 − yw\et −1 )
e∈X
set (yet − yw\et −1 ) −
e∈A
set −1 (yet −1 − yw\et −1 ) (1)
e∈S
where 1y denotes log-changes in the plant/firm variable of interest between periods t − 1 and t, and set the share of plant e in the total employment of firm w in period t. The first two lines of (1) capture the contribution of internal adjustment (C + and C − stand for continuing and expanding, and continuing and contracting, respectively). The fourth and fifth line of (1) capture the contribution of the greenfield investment, exit, acquisitions, and sales (N, X , A, and S, respectively). This decomposition suggests looking at the following quantities:
• ‘Composition effects’. Changes in ywt due to changes in the 3. Results We start our presentation of stylized facts with an overview of the importance of the different adjustment channels for total firm growth. In line with existing work we focus on continuing firms, i.e., firms which existed in the current period, will not exit in their entirety in the next period, and which change employment between periods. The most basic question we are interested in is how frequently the various channels are used. Panel A of Table 1 displays the fraction of all employment adjustments which take place through each of the three channels.3 It is evident that M&As and greenfield investment are rare events. On average, these two channels were used in only about 3% of employment changes, with the vast majority of both expansions and contractions occurring via internal adjustments. Panel B shows that, when they take place, M&As and greenfield adjustments are major events. The average M&A expansion (contraction) is 15 (20) times bigger than the average internal expansion (contraction). Greenfield investments are smaller than M&A expansions but still around 10 times bigger than the average internal expansion; greenfield disinvestments, in contrast, are of comparable magnitude to M&As. Panel B implies that despite their infrequent occurrence, greenfield investment and M&As still account for a large fraction of overall employment adjustments. The two forms together make up 23% of economy-wide employment expansions and 37% of employment contractions of continuing firms, with M&As being quantitatively more important than greenfield adjustments (see Panel C).
1 For a detailed description of our data and methodology, as well as additional results, see Breinlich et al. (2011). 2 Henceforth, we will uniformly use the terms firm and plant instead of enterprise group and enterprise. 3 We use employment rather than turnover adjustments since the former are much more directly under the control of a firm.
composition of the set, or the relative importance, of the plants making up a firm. What matters for ywt here is the level of variable yet of the plant undertaking an adjustment, relative to all other plants in the same firm (i.e., yet − yw\et −1 and yet −1 − yw\et −1 ).4 • ‘Change effects’. Changes in ywt due to changes in yet for existing and continuing plants. In the above decomposition, these are the changes in the plant-level variable yet associated with internal expansion or contraction (1yet ). 3.1. Composition effects To obtain estimates for yet − yw\et −1 and yet −1 − yw\et −1 , we first calculate the employment-weighted average of y across all the plants of the firm not undergoing the demographic event in question (yw\et −1 ). We then subtract these averages from the level of y of the firm undergoing the demographic event (yet or yet −1 ), using an appropriate lag structure as indicated in (1). This procedure yields one difference for each of the firms and events for which we have sufficient data. Table 2 shows estimates of the mean of these differences across firms, together with the corresponding standard error and the significance level for a twosided tests of whether the mean is different from zero. We present results for a number of firm-level variables commonly analyzed in the literature: wages, labor and total factor productivity, and a measure of firm profitability (operating profits per employee). Given that these variables are not available in the BSD, we use the merged ARD–BSD dataset here and in Section 3.2, with data on the firm characteristics (y) coming from the ARD and the demographic events from the BSD. As seen, plants undergoing one of the four external adjustment forms (birth, exit, acquisition, sale) tend to be less productive
4 The subscript w \ e denotes all plants of firm w excluding plant e, and y w\e is the employment-weighted arithmetic average of variable y across these plants.
204
H. Breinlich et al. / Economics Letters 116 (2012) 202–206
Table 1 Descriptive statistics, adjustment strategies (1997–2005). Source: Office for National Statistics and authors’ calculations. Internal adjustment
M&A
Greenfield
All adjustments
Panel A: usage frequency of adjustment channels (%, total count in last column) Gross expansions Gross contractions
99.11 98.38
1.53 1.71
0.60 1.12
138.19 148.34
86.74 161.29
1,672,516 1,418,009
Panel B: average size of adjustment by channel (number of employees) Gross expansions Gross contractions
9.07 7.45
11.46 11.57
Panel C: aggregate importance of adjustment channels (% of total adjustment, last column indicates total annual average employment change) Gross expansions Gross contractions
77.32 62.65
18.23 21.63
4.52 15.49
2,771,429 2,371,429
Notes: Panel A shows the fraction of all employment adjustments by continuing firms involving internal adjustment, M&As and greenfield investment or disinvestment. Note that firms can use several channels at the same time so that the percentages do not have to add up to 100%. Panel B shows the average transaction size of each of these channels, and Panel C their contribution to overall adjustment. All figures are averages over 1997–2005. Table 2 Composition effects of the choice of adjustment channel (1997–2005). Source: Office for National Statistics and authors’ calculations.
Labor productivity Wages per employee TFP (OLS) Operating profits per employee Observations
Birth
Exit
Acquisition
Sale
Internal expansion
Internal contraction
−0.050
−0.145
−0.021
−0.146
0.009 (0.017) 0.022 (0.010)** 0.013 (0.007)* −0.023 (0.028) 2435–3301
−0.065
(0.052) 0.025 (0.031) −0.053 (0.026)** −0.024 (0.111) 234–379
(0.034)***
−0.034
(0.015)
−0.031
(0.020)*
−0.033
(0.009)***
−0.019
(0.014)**
−0.079
(0.016)***
−0.030
(0.007)***
−0.021
(0.055) 623–951
(0.027)***
−0.090 (0.012)**
−0.198
(0.026) 2336–3236
(0.046)*** 885–1205
(0.016)***
−0.038 (0.010)***
−0.012 (0.007)*
−0.159 (0.026)*** 2799–3842
Notes: Table shows average log differences between plants undergoing an adjustment form and the other plants in the same firm (see text for details). They are derived from simple descriptive OLS regressions. Figures in brackets denote robust standard errors clustered at the enterprise level. * Denote significance at the 10% level. ** Denote significance at the 5% level. *** Denote significance at the 1% level. Table 3 Change effects for existing and continuing plants (1997–2005). Source: Office for National Statistics and authors’ calculations.
Internal expansion Internal contraction Acquisition Sale Observations
Labor productivity
Wages per employee
TFP (OLS)
Operating profits per employee
−0.020
−0.021
−0.002
−0.021
(0.011)* 0.061 (0.011)*** 0.015 (0.015) 0.010 (0.012) 36,206
(0.006)*** 0.047 (0.006)*** 0.014 (0.007)* 0.003 (0.006) 38,945
(0.006) 0.013 (0.006)** 0.016 (0.007)** −0.000 (0.006) 30,243
(0.023) 0.067 (0.023)*** 0.009 (0.029) 0.054 (0.025)** 28,358
Notes: Table shows the effects of the corporate events listed in the first column on existing and continuing plants. Results are estimated using OLS regressions with industry–year fixed effects. Figures in brackets are robust standard errors clustered at the plant level. Each column denotes a separate regression, where the growth rates of the variables of interest are regressed on the event dummies. The omitted group are plants that do neither expand nor contract. * Indicate statistical significance at the 10% level. ** Indicate statistical significance at the 5% level. *** Indicate statistical significance at the 1% level.
and pay lower wages. There are no significant differences in profitability, however, with the exception of plants which are sold off — these report profits of 20% less than the remaining plants in the same firm. Turning to the internal adjustment forms, contracting plants show performance characteristics comparable to those being sold off, such as low productivity and profitability. Expanding plants, on the other hand, tend to be slightly more productive and pay higher wages than the other plants in the same firm.
be obtained via simple dummy variable regressions of 1yet on binary indicators for internal expansion/contraction. We use plants which neither expand nor contract employment between adjacent periods as the omitted category in these regressions, so that all changes are expressed relative to this group. For comparison with the existing literature, we also include plants undergoing an acquisition or sale. Table 3 presents results for the same variables as in Section 3.1, again using the merged ARD–BSD dataset. We control for industry–year fixed effects in all regressions, so that results rely on within-sector-year variation only.5
3.2. Change effects Estimates of the average log-change in the plant-level variable yet associated with internal expansion or contraction (1yet ) can
5 If, after controlling for industry–year fixed effects, plants with unchanged employment represent a suitable control group for the other demographic events,
H. Breinlich et al. / Economics Letters 116 (2012) 202–206
The general pattern in our results is that internal expansions tend to reduce wages and profits, as well as labor and total factor productivity. Internal contractions have almost exactly the opposite pattern. They increase both productivity measures as well as profitability and wages. Patterns are less pronounced for both acquisitions and sales, but some interesting patterns emerge here as well. For example, similar to Maksimovic and Phillips (2001) and Schoar (2002), we find that acquired plants subsequently increase their productivity, although this effect is only significant for the TFP measure. Besides serving as input into the decomposition (1), these results allow for some interesting comparisons with the existing literature, which usually focuses on only a subset of the channels analyzed here. For example, our results show that while asset transfers via M&As are not always associated with significant productivity gains, internal expansions actually reduce productivity, with the difference between the two expansion forms being highly statistically significant. This demonstrates that, when analyzing the consequences of corporate events such as acquisitions, the choice of control group is crucial. In the previous example, both acquisitions and internal expansions serve the purpose of increasing overall firm size, and might imply similar adjustment processes to the operations of a firm. It would thus seem natural to use internally expanding plants as a control group in the estimation of the effects of acquisitions. As our results demonstrate, this choice would let acquisitions appear in a much better light than when compared to, for example, plants not undergoing any change. 3.3. Contribution of adjustment channels to changes in firm-level variables We now combine the results from Sections 3.1 and 3.2 with the decomposition proposed in (1) to shed light on the contribution of our adjustment channels to changes in the firm-level variables d here. Recall from (1) that we need the changes in the variable of interest (y) associated with the different adjustment channels, as well as the employment shares (s) of the plants undergoing a given demographic event. Ideally, we would like to use data on both y and s from the merged ARD–BSD dataset. A problem in this context is that the matching procedure between the ARD and the BSD only results in a small subset of firms with a complete set of plants — which is of course needed to calculate the employment shares in (1). The reason for this is that sampling in the ARD takes place at a scale equivalent to the plant level (not the firm level), and the ARD covers less than 5% of the plants in the BSD in a given year. Consequently, we only have the full set of plants for less than 2% of firms. To make progress, we thus use employment shares from the BSD (where coverage is close to complete, but we only have information on employment and turnover) with our estimates of the average changes in y associated with the different adjustment forms derived in Sections 3.1 and 3.2. That is, we rewrite (1) as
1 ywt =
set −1 1 yet +
e∈C +
+
(yet −1 − yw\et −1 )1set
e∈C +
e∈C +
1 yet 1set +
set −1 1 yet
e∈C −
the following results have a causal interpretation. However, it is possible that there are pre-existing level or trend differences which induce firms to select into a particular adjustment channel. Identifying truly causal effects would have to rely on sources of exogenous variation in the choice of adjustment channels, which is beyond the scope of this short note and the subject of ongoing work.
+
205
(yet −1 − yw\et −1 )1set +
e∈C −
+
1 yet 1set
e∈C −
set (yet − yw\et −1 ) −
e∈N
+
set −1 (yet −1 − yw\et −1 )
e∈X
set (yet − yw\et −1 ) −
e∈A
set −1 (yet −1 − yw\et −1 ) (2)
e∈S
where hats above the variables indicate quantities estimated in the last sections, using data from the merged ARD–BSD sample. For example, we found that exiting plants have a labor productivity which is on average 14.5% lower than the remaining plants of the same firm (see Table 2). We thus set
−
set −1 (yet −1 − yw\et −1 ) = −
e∈X
set −1 × (−0.145)
e∈X
where set −1 , the employment share of the exiting plant, is now calculated from the BSD data. Implicitly, we are thus making the assumption that the sample used for the estimations in the previous sections is representative of the BSD in terms of the effects associated with the different adjustment channels. To make this assumption more plausible, we only keep those firms in the BSD from which at least one plant can be matched to the ARD at some point in our sample. This makes the set of firms used in our decomposition analysis more comparable to the type of firms represented in the ARD.6 For each firm in the BSD, and for each left-hand side variable y, the decomposition in (2) yields a predicted growth rate of y, as well as the contributions of the different adjustment channels. In Table 4, we report an average value for each decomposition term, expressed as a fraction of the sum of all terms (in absolute values). Depending on the variable in question, the external adjustment forms (in particular, exits and sales) account for around 50% of the overall mean changes at the firm level. External adjustment forms are particularly important for growth in total factor productivity, where they account for 70% of the overall mean change. Our results thus indicate that, despite their infrequent occurrence, external adjustments contribute to a large degree to the overall changes in firm-level variables. While there are of course important differences in terms of the underlying data and the decomposition approaches used, this dominance of external adjustment is reminiscent of recent findings in the productivity literature using plant-level data (e.g., Disney et al., 2003; Foster et al., 2006). 4. Conclusions We presented a novel set of stylized facts on forms of firm expansion and contraction, using unique business register data for the United Kingdom. In contrast to contributions in the existing literature, our data enabled us to distinguish between all three principal adjustment channels that firms can use to change the scale and scope of their operations: changes of employment at existing establishments, greenfield investments
6 Note that we thus do not require our estimates to be representative of the BSD as a whole but only of those firms which at some point had at least one of their plants sampled in the ARD. Since ARD plants are the unit of analysis in Tables 2 and 3, our predicted values in (2) should be more representative of this restricted sample. Given that the ARD oversamples larger plants, and that larger plants tend to be part of larger firms, such firms will be overrepresented in the restricted sample. As such, our results are best thought of as applying mainly to larger firms. As is true in the UK as in other countries, however, these firms make up the majority of economic activity. In additional robustness checks (available from the authors upon request) we show that further restricting our sample to the subset of firms with plants which we have actually used in the estimations in Sections 3.1 and 3.2 (rather than all ARD plants, some of which have missing data) does not alter any of the conclusions drawn below.
206
H. Breinlich et al. / Economics Letters 116 (2012) 202–206
Table 4 Firm-level decompositions (1997–2005) — average value for each decomposition term, expressed as a fraction of sum of all terms (in absolute values). Source: Office for National Statistics and authors’ calculations. Component
Labor productivity (%)
Wages per employee (%)
TFP (OLS) (%)
Operating profits per employee (%)
Within effect (internal expansion) Between effect (internal expansion) Cross effect (internal expansion) Within effect (internal contraction) Between effect (internal contraction) Cross effect (internal contraction) Contribution of birth Contribution of exit Contribution of acquisition Contribution of sale Avg. predicted change in decomposed variable Observations
11.53 0.15 0.17 26.54 0.49 0.23 0.28 34.16 4.03 22.41 2.50 445,134
19.67 0.61 0.29 33.33 0.47 0.28 0.23 13.01 9.67 22.44 0.91 445,136
4.87 0.94 0.07 24.28 0.38 0.21 1.25 32.98 15.48 19.53 0.49 445,138
12.56 0.41 0.19 30.20 1.25 0.26 0.14 19.30 4.18 31.52 2.30 445,136
Notes: Table shows the average contribution of each component in the first column to changes in the firm-level variables listed in the top row.
and disinvestment, and M&As. We documented the relative importance of these three channels and showed what the associated changes in a number of firm-level variables were. References Asplund, M., Nocke, V., 2006. Firm turnover in imperfectly competitive markets. Review of Economic Studies 73, 295–327. Breinlich, H., Niemann, S., Solomon, E., 2011. Channels of Size Adjustment and Firm Performance. Discussion Paper No. 703, University of Essex. Davis, S., Faberman, R., Haltiwanger, J., 2006. The flow approach to labor markets: new data sources and micro–macro links. Journal of Economic Perspectives 20 (3), 3–26. Davis, S., Haltiwanger, J., Schuh, S., 1996. Job Creation and Destruction. MIT Press, Cambridge.
Disney, R., Haskel, J., Heden, Y., 2003. Restructuring and productivity growth in UK manufacturing. Economic Journal 113, 666–694. Foster, L., Haltiwanger, J, Krizan, C.J., 2006. Market selection, reallocation, and restructuring in the US retail trade sector in the 1990s. Review of Economics and Statistics 88, 748–758. Hopenhayn, H., 1992. Entry, exit, and firm dynamics in long run equilibrium. Econometrica 60, 1127–1150. Hopenhayn, H., Rogerson, R., 1993. Job turnover and policy evaluation: a general equilibrium analysis. Journal of Political Economy 101, 915–938. Jovanovic, Boyan, 1982. Selection and the evolution of industry. Econometrica 50, 649–670. Maksimovic, V., Phillips, G., 2001. The market for corporate assets: who engages in mergers and asset sales and are there efficiency gains? Journal of Finance 56, 2019–2065. Schoar, A., 2002. Effects of corporate diversification on productivity. Journal of Finance 57, 2379–2403.