Labour leverage, equity risk and corporate policy choice

Abstract

This paper investigates the role of labour utilization in assessing equity investment risk and corporate financial policy choices. Several existing models of the firm predict that labour utilization is costly to adjust in the short run. I argue that this leads to a relatively fixed obligation to pay cash to labour, in effect creating an off-balance-sheet intangible liability similar to a lease. The liability creates ‘labour leverage’ risk, analogous to financial leverage risk. Labour leverage is hypothesized to be positively correlated with equity investment risk as measured by characteristics of stock returns. Managers recognize this risk and adjust financial policies including debt financing and dividend policy accordingly. The main empirical results are that labour leverage is positively correlated with equity investment risk, and it plays the predicted role in regressions explaining financial structure and dividend policy. Proxies for labour leverage are simple measures based on existing disclosure. The results are consistent with the conjectures that market participants use labour disclosures to assess risk, and that managers take actions to mitigate this risk. The results are consistent across most sectors of the economy, and consistent over time.

ACKNOWLEDGEMENTS

I thank Chris Anderson, Bill Beedles, George Bittlingmayer, Mark Hirschey, Jevons Lee, Karl Muller, Fred Mittelstaedt, Don Siegel, Tom Noe, Terri Rosett, Ross Watts, Shuang Wu, Jerold Zimmerman, the reviewers and seminar participants at Tulane University, University of Rochester, Claremont McKenna College and University of Kansas for helpful comments. I am responsible for any errors.

Notes

¹The relation between accounting and costs such as hiring and training is examined in a series of papers (Brummet et al., 1968a, 1968b, 1969; Brummet, 1970; and later a book: Flamholtz, 1985), where these authors explore the role of these fixed costs as unrecognized assets that could provide useful information for managerial decision-making and other purposes, and perhaps even be capitalized in financial statements. Dittman et al. (1976) argue that the claims of unrecognized human resource assets are overstated and limited to only a few special cases. Following Oi, there is also an extensive economics literature about limited labour adjustment.

²For example, in his book on labour practices in the US, Paying for Productivity (1990), Blinder estimates that labour accounts for at least 70% of total costs.

³Griffiths et al. (1993: 602) use this term for lagged endogenous variables that are taken as given in the current period but reflect endogenous choices made in previous periods.

⁴Compustat defines item 29, labelled ‘Employees’:

• This item represents the number of company workers as reported [emphasis in the original] to shareholders. This figure is reported by some firms as an average number of employees and by some as the number of employees at year-end. No attempt has been made to differentiate between these bases of reporting. If both are given, the year-end figure is used. … The item includes: 1. All employees of consolidated subsidiaries, both domestic and foreign, 2. All part-time and seasonal employees, 3. Full-time equivalent employees, and 4. Officers. This item excludes: 1. Consultants, 2. Contract workers, 3. Directors, and 4. Employees of unconsolidated subsidiaries.

⁵Compustat defines item 42 ‘Labor and Related Expense’:

• This item represents the costs of employees' wages and benefits allocated to continuing operations. This item includes: 1. Incentive compensation, 2. Other benefit plans, 3. Payroll taxes, 4. Pension costs, 5. Profit sharing, and 6. Salaries and wages. This item excludes: 1. Commissions, and 2. Director costs (if it is not possible to exclude these costs, collect the labor and related amount as reported).

⁶The findings in Ballester et al. (1998), discussed elsewhere, are consistent with this interpretation.

⁷β is calculated by CRSP for calendar years using daily returns and a value-weighted market index. CRSP requires trades on at least 50% of trading days to calculate a β.

⁸The sector dummies use definitions suggested by Barth et al. (1998). These are as follows: Mining & Construction = 1000–1299 and 1400–1999; Food = 2000–2111; Textiles & Printing/Publishing = 2200–2780; Chemicals = 2800–2824 and 2840–2899; Pharmaceuticals = 2830–2836; Extractive Industries = 1300–1399 and 2900–2999; Durable Manufactures = 3000–3569, 3580–3669 and 3680–3999; Computers = 7370–7379, 3570–3579 and 3670–3679; Transportation = 4000–4899; Utilities = 4900–4999; Retail = 5000–5999; Financial Institutions = 6000–6411; Insurance & Real Estate = 6500–6999; Services = 7000–7369 and 7380–8999. Owing to the small number of observations in Insurance & Real Estate (especially early in the sample), I combined these with Financial Institutions. Any remaining SICs are coded as Other.

⁹Results for non-December fiscal year-end firms are generally consistent with results reported for December firms, but not reported due to mismatched timing of variables.

¹⁰Of the 191,851 observations in the full Compustat files including C29 data, only 17% disclose labour expense. In the sample used here, 28% disclose labour expense. This difference appears to reflect the larger size of firms in the sample. In the full Compustat universe, firms disclosing labour cost have median market value of $256 million (January 1998 prices) compared to $64 million for non-disclosers. Hence any biases associated with size in the final sample are magnified in the labour cost leverage sub-sample.

¹¹If β (again from CRSP) is used rather than σ, the pattern across industries is similar but weaker overall, and the decade results show a weakening of the labour leverage effect over time, though still positive and statistically significant (t = 4.81) in the 1990s. It is not clear why there is a difference between the results for σ and β over time, but the results with respect to β are consistent with changes in patterns of labour usage and the labour–capital mix over time. For example, advances in technology, such as robots and the increasing use of computers have made labour easier to replace in some production functions. A second likely candidate for this decline is changes in institutional features of labour markets. In particular, the decline of unions has probably weakened restrictions on substituting capital for labour. A third candidate for the decline is the growing use of equity stakes for employees. ESOPs and other incentive provisions were much less likely to be used both in unionized and non-union firms in the earlier time periods, but have grown in popularity in recent years. In addition, the shift towards new modes of labour relations aimed at increasing labour productivity through involvement (such as TQM), and the dramatic decline in the labour-to-capital ratio over time may have reduced the observed effects of labour on systematic risk from the 1960s to the 1990s. These changes both increase the substitution possibilities between capital and labour, and correlate wages with profits. Hence one interpretation of the pattern for β is that firms have learned to use labour relations practices that mitigate systematic labour risk.

¹²The results reported in Table 5 use standardized responses, so the coefficient estimates are not comparable to those reported in Smith and Watts (1992). Using SW's specification for financing policy with the final sample data used in this paper provides results that are nearly identical in all respects with those reported in SW's Table 1. SW find (with t-statistics in parentheses) E/V = 1.34(17.03) − 0.62(−12.47) × IOS − 0.25(−12.33) × RegulationDummy − 0.03(−3.43) × FirmSize, with adjusted-R ² = 0.79. The results for the final sample used in this paper are (using the original variables rather than standardized) E/V = 1.10(233.87) − 0.47(−104.64) × IOS − 0.22(−96.49) × RegulationDummy − 0.02(−33.31) × FirmSize, with adjusted- R ² = 0.60. In sign and significance the results are quite similar, and the magnitudes of the coefficients appear close.

¹³SW find (t-statistics in parentheses) DividendYield = − 0.05(−5.03) + 0.05(9.19) × IOS + 0.01(6.14) × ! RegulationDummy + 0.01(4.20) × FirmSize, with adjusted-R ² = 0.61. The results for the final sample used in this paper are (using the original variables rather than standardized) DividendYield = − 0.03(−34.06) + 0.03(54.63) × IOS + 0.02(45.68) times RegulationDummy + 0.004(46.01) × FirmSize, with Adjusted-R ² = 0.24.