Abstract
We investigate the relationship between productivity, size and age of large Australian firms employing more than 100 employees or holding assets in excess of $100 million. In addition, we also investigate the extent of productivity persistence among these firms by looking at transition matrices of productivity distribution and productivity-rank mobility. The empirical study is based on the IBISWorld database used to estimate translog cost function to measure (a residual based) productivity. We find evidence, though somewhat weak, that larger and older firms are on average less productive. Further, we find strong evidence for a high degree of inertia in terms of productivity ranking within an industry.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
We define large firms as those employing more than 100 employees or controlling assets valued in excess of AU$100 million.
Our definition of a large firm is different from ABS’ definition in which the employment threshold is 200 persons and the asset value threshold is AU$200 million (ABS No. 8140.0).
In this paper, we use the terms ‘productivity’ and ‘efficiency’ interchangeably.
It is arguable that the persistence in productivity differences between firms can be explained by differences in management capability, especially for large firms which are subjected to the discipline of “market forces” in the form of various corporate activities such as takeovers, mergers and acquisitions. The results of Baily et al. (1992) seem to support the persistence hypothesis; see also the review by Bartelsman and Doms (2000). We believe there are many reasons why some firms could be persistently underperforming when compared with their peers while escaping the discipline of market forces. Inefficient management could be shielded from market forces because of government ownership or regulations; examples in Australia include firms such as Telstra, Qantas, and Australian Wheat Board. Further, to argue that market forces would discipline inefficient managers presuppose that firms’ valuation, as reflected in their share prices, is linked to their productivity performance. However, we are not aware of any studies that demonstrate such a link. Even if there is a link, the persistence could still be explained, for example, by investment risks—some firms are known in the investment community to have low risks and low returns, their management teams are known to be conservative in their investment decisions, and investors take their low productivity as a matter of course. Whether these explanations are valid or not is an interesting empirical question that we cannot answer in this paper, but the issue deserves further investigation.
Since the input cost shares must sum to one, one of the input cost share equation is excluded from the system. The estimation results are invariant to which cost share equation is excluded.
In the empirical estimation of A it , the fixed effect term λ i captures not only permanent differences in productivity levels between firms, but also other systematic, time-invariant, differences across firms. Naturally, it also captures any systematic bias in measuring firms’ input prices and output volume.
See Appendix A for a more detailed discussion of data sources and variable construction.
Appendix B contains the results of restricting the sample to only manufacturing firms; in addition, three different ways of deflating nominal variables into real values have also been attempted. Annex I (available upon request) contains the results of four further variations which include alternative ways of approximating wage cost, capital cost and deflation of nominal variables.
Results of the sensitivity analysis are summarised in Appendix B and Annex I; the latter is available upon request from the authors.
Recall that we have unbalanced panel data. Thus, if a firm with four observations in the sample, its “usual” quintile rank is defined if it occupies the same quintile rank for at least three of the 4 years.
Results of the sensitivity analysis are available from the authors upon request.
All monetary values are deflated using producer price index, which we obtain from the ABS publication, Producer Price Index, 6427.0.
Some firms are only assigned a two-digit ANZSIC because of the multiple production activities the firms engage in would not permit a finer classification. Table 25 presents the distribution of firms in the finest ANZSIC. The corresponding sector-specific PPI is used in each case for deflating the nominal variables in Model III.
References
Australian Bureau of Statistics (ABS), “Business Operations and Industry Performance, 2000–01”, ABS No. 8140.0, 2000-01.
Baily, M.N., Hulten, C., and Campbell, D., “The distribution of productivity in manufacturing plants,” Brookings Paper on Economic Activity: Microeconomics, Washington, D.C., pp. 187–249, 1992.
Bartelsman, E. and Dhrymes, P., “Productivity dynamics: U.S. manufacturing plants, 1972–1986,” Journal of Productivity Analysis, vol. 9, pp. 5Y34, 1998.
Bartelsman, E.J. and Doms, M., “Understanding productivity: lessons from longitudinal microdata,” Journal of Economic Literature, vol. 38(3), pp. 569Y594, 2000.
Bernard, A.B. and Jones, C.I., “Comparing applies to oranges: Productivity convergence and measurement across industries and countries,” American Economic Review, vol. 86, pp. 1216Y1252, 1996.
Bhattacharyya, A., Bhattacharyya, A., and Krishna, M., “Decomposition of technological change and factor bias in Indian power sector: An unbiased panel data approach,” Journal of Productivity Analysis, vol. 8, pp. 35Y52, 1997.
Bland, S. and Will, L., “Resource Movements and Labour Productivity, an Australian Illustration: 1994–95 to 1997–98”, Productivity Commission Staff, Research Paper, AusInfo, Canberra, 2001.
Brynjolfsson, E. and Hitt, L.M., “Beyond computation: Information technology, organizational transformation and business performance,” Journal of Economic Perspectives, vol. 14(4), pp. 23Y48, 2000.
Dawkins, P., Harris, M., and King, S., “Big business, private performance and public policy: An introduction,” in Dawkins, P., Harris, M., and King, S. (eds.), How Big Business Performs: Private Performance and Public Policy. Australia (Allen and Unwin), 1999.
Ericson, R. and Pakes, A., “Markov perfect industry dynamics: A framework for empirical work,” Review of Economic Studies, vol. 62, pp. 53Y82, 1995.
Gardiner, K. and Hills, J., “Policy implications of new data on income mobility,” Economic Journal, vol. 109, pp. F91YF111, 1999.
Girma, S. and Kneller, R., “Productivity levels and dynamics: UK service sector firms 1988–1998,” Research Paper Series #2002/25, Leverhulme Centre, University of Nottingham, 2002.
Hopenhayn, H., “Entry, exit and firm dynamics in long run equilibrium,” Econometrica, vol. 101, pp. 1127Y1150, 1992.
Jovanovic, B., “Selection and the evolution of industry,” Econometrica, vol. 50, pp. 649Y679, 1982.
Pakes, A. and Ericson, R., “Empirical implications of alternative models of firm dynamics,” Journal of Economic Theory, vol. 79, pp. 1Y45, 1998.
Productivity Commission. Microeconomic Reforms and Australian Productivity: Exploring the Links, Research Paper, Canberra: AusInfo, 1999.
Zaide, A., Rake K., and Falkingham J., “Income Mobility in Later Life,” Working Paper, London School of Economics, 2001.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Appendix A: Data sources and variable construction
The main source of the firm financial data is IBISWorld, a commercial information provider company which offers, by subscription, financial information of the top 2,000 companies in Australia and the general business environment within which these companies operate. The company financial information that IBISWorld provides includes total operating revenue, employment, values of fixed assets and liabilities, and information useful to derive total costs and the cost components. The information on firm age is obtained from the Australian Company Number (ACN) Registration Database provided by the Australian Securities and Investments Commission (ASIC). We use the earlier registration date of each firm as its date of birth. We also use other supplementary data sources including various issues of ABS publications such as the Producer Price Index (Catalogue Number 6427.0) and the Australian Industry (8155.0). Below, we outline the construction of variables used in the analyses based on these data.Footnote 16
-
(1)
Total costs
The financial information concerning costs consists of the total operating costs, i.e., the costs of selling goods and services. Unfortunately this information is only available for the period 2001–2003, there was no such information for earlier years. We therefore derive the total costs from the identity: total sales revenue = total accounting costs of sales + gross profit. Thus, we obtain total accounting costs as the difference between total sales revenue and gross profit. We obtain gross profit figures from the expression: Gross profit = net profit before taxes + depreciation + audit fees + net interest payments. Note that the derived total cost of sales excludes extraordinary items. Note also that since depreciation is added back in the gross profit computation, total accounting costs do not include any capital costs. Thus, we obtain the total costs by adding capital costs to the total accounting costs.
-
(2)
Capital costs
The computation of capital costs is based on a consideration of opportunity costs, in the sense of how much capital income could have been generated if the investment into the firm was not made, but alternatively placed on the capital market. The ABS provides money market data, such as long term interest rates on treasury bills. The interest rate we used is a non weighted yearly average of the monthly long-term (10 years) money market interest rate. Note that the interest rate is assumed to be neither firm- nor industry-specific.
The amount of capital a firm has is taken to be the value of total fixed assets. The amount is obtained by deducting, from the value of total assets, the items cash, current and noncurrent receivables and trade debtors, inventories, other current assets and intangibles assets. The firm’s capital costs are simply the opportunity costs of the fixed assets, i.e., fixed assets × long-term interest rate. The cost share of capital is then simply the proportion of capital costs to total costs.
-
(3)
Labour costs
Although the IBIS database contains information on number of employees, there is no information on total wage bills. As such, we need to approximate the total labour costs of firms by using industry averages. We obtain industry-specific labour cost shares from ABS data (ABS Australian Industry, 8155.0). We multiply this by the firms’ total operating costs to get total labour costs.
The labour costs per employee (average wage rate) are then given as the ratio of total labour costs to the total number of fulltime equivalent employees. The cost share of labour is then simply the proportion of labour cost to total costs.
-
(4)
Material costs
The approximate value of material costs is simply the residue after deducting labour and capital costs from total costs. The cost share of materials is the proportion of material costs to total cost.
Appendix B: Productivity estimation and transition of manufacturing firms data
This Appendix reports the results of a sensitivity analysis that focuses on two issues. First, we repeat the empirical exercise in the text using a sub-sample that consists of only manufacturing firms. The manufacturing-only sample consists of 2,838 observations, approximately one third of the full sample. The aim is to see how the cost function estimates would change if we restrict the sample to only manufacturing firms, which presumably have different types of production technology as compared to non-manufacturing firms. The second issue examined is the deflation of nominal variables into real values. Three different ways of deflating nominal variables have been attempted; they result in three different sets of empirical estimates that we label Models I, II and III. In Model I, nominal variables were deflated in the same way as in the text using an economy-wide PPI. In Model II, nominal variables were not deflated; they were used in the estimation without any adjustment. In Model III, nominal variables were deflated using sector-specific PPI up to four-digit ANZSIC.Footnote 17 Making use of the productivity estimates generated by these models, we also estimate the productivity equations (Eqs. 9 and 10 in the text) and construct the 1 and 3-year productivity transition matrices as in the text.
Comparing Model I with the full model in the text should provide an indication of the robustness of the cost function specification, while comparing across Models I, II and III should reveal potential problems in using different ways of deflation. As explained below, the results indicate that the translog cost function specification is rather robust to restricting the sample to manufacturing firms and to different ways of deflating the nominal variables. In particular, the pattern of productivity persistence has remained evident in all transition matrices that we constructed.
Table 15 presents the coefficient estimates of translog cost function under the three models. When compared with the estimates of the full model (Table 7 in the text), the estimates are broadly similar, except for a few second-order terms. In particular, two estimates—for the terms (ln y ln w 2) and (ln y ln w 3)—had a reversal of sign, although we note that their values are close to zero in both cases. Despite the smaller sample size, all estimates have higher t-statistics, suggesting that the translog function produces a better fit under the restricted sample. Comparing across the three models, we see that the coefficient estimates are almost identical, indicating that the different ways of deflating nominal variables does not appear to result in any material differences.
Tables 16, 17 and 18 present the coefficient estimates of the productivity equations, where the dependent variable is the productivity estimates obtained from the respective models in Table 15. In each case, we estimate the OLS, fixed effects and random effects models. So far as the effects of firm age and size on productivity are concerned, Model I produces results that are consistent with those of the full model under OLS, but not under the fixed effects model. The random effects model produces estimates of the same sign but are not statistically significant. The overall picture is thus a little muddled—consistent results are obtained in some models but not in others. Comparing across Tables 16, 17 and 18, we see that the estimates are almost identical, regardless of the methods of deflation.
Tables 19, 20 and 21 show the 1-year transition tables of Models I, II and III, respectively. Note that the number of firms in any given year is small due to the small sample size. However, comparing the percentage figures in each quintile, we see that the pattern of productivity persistence is very much similar to that described in the text—firms are more likely than not to remain in the same quintile from 1 year to the next. The pattern is similar regardless of the methods of deflation. Similar conclusion also applies when we examine the 3-year transition tables in Tables 22, 23 and 24.
Rights and permissions
About this article
Cite this article
Palangkaraya, A., Stierwald, A. & Yong, J. Is Firm Productivity Related to Size and Age? The Case of Large Australian Firms. J Ind Compet Trade 9, 167–195 (2009). https://doi.org/10.1007/s10842-007-0028-4
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10842-007-0028-4