Managerial compensation in multi-division firms Shashwat Alok and Radhakrishnan Gopalan March 26, 2012 The authors thank seminar participants at the Olin Business School for valuable comments. Any remaining errors are our own. PhD Student in Finance, Olin Business School,Olin Business School, Washington University in St. Louis. Corresponding author. e-mail: shashwatalok@wustl.edu. Assistant Professor of Finance, Olin Business School, Washington University in St. Louis. Corresponding author. e-mail: gopalan@wustl.edu.
Abstract While literature highlights how agency problems between division managers (DM) and the CEO in multi-division firms can distort capital allocation, it has largely ignored the role of DM incentive contacts in resolving these problems. For the first time in literature, we use hand collected data to study the pay structure of division managers (DM). Using data on 4,080 DM-year pay contracts, we relate DM pay to the performance of both her division and other divisions in the firm. Consistent with the idea that the actions of a DM can impose externalities on the rest of the firm, we find that DM pay loads positively on both her division s and the other divisions performance. Consistent with differences in their spheres of influence we find that the pay for other divisions performance is significantly greater for the firm s CEO. While CEO pay is equally sensitive to the performance of all divisions in the firm, DM pay is more sensitive to her division s performance as compared to the performance of other divisions. Further, pay for division performance is lower in industries with less informative accounting earnings. This highlights an important cost of conglomeration. The sensitivity of DM pay to other divisions performance is especially greater for related divisions and for divisions with fewer growth opportunities as measured by industry Tobin s q and past division sales growth. Focusing on economic conditions, we find that the sensitivity of DM pay to both the performance of her division and other divisions is lower during economic downturns and periods of economic distress. Overall, our evidence suggests that DM compensation is structured to solve agency problems in multi-division firms and reflects the constraints in the contracting environment. JEL Classification:G30, J31 2
Introduction The costs and benefits of housing multiple divisions within the same firm ( conglomeration ) has been of significant research interest (see Stein (2003) and Maksimovic and Phillips (2007) for surveys). While single segment firms depend on the external financial markets for capital, divisions within a conglomerate look to the headquarters or the firm s CEO for capital. Agency problems between the conglomerate s CEO and the division manager (DM from now) can distort capital allocation within conglomerates. Such problems can arise due to asymmetric information (Stein (1997)), differences in risk attitudes (Holmstrom (1979)), career concerns (Holmstrom and Costa (1986)) and private benefits of consumption (Stein (2002)). While a properly designed incentive contract for the DM can mitigate such problems, the lack of stock price for individual divisions can impede the design of such contracts. The large literature on conglomerates has focussed on how the agency problems between the DM and the firm s CEO can distort capital allocation (e.g. Stein (2003), Rajan et al. (2000a)). For the most part this literature has ignored the role of DM incentive contracts in mitigating the problems. There is also very little empirical evidence on how multi-division firms design DM incentive contracts. The objective of this paper is to fill this gap. We analyze data on DM pay for over 4,000 division-years to document the structure of DM pay and the extent to which it varies across firms. The unique aspect of our paper is our ability to combine DM pay data with the performance of both her division and the other divisions in the firm. This allows us, for the first time in literature, to measure the extent of pay-for-performance for DMs. We obtain our data by hand matching two commonly used databases, ExecuComp and Compustat business segment files. ExecuComp provides compensation data for the top five executives of S&P 1500 firms. Along with their compensation, ExecuComp also provides the executive s designation, which reveal some of them to be a high ranking official in a division. For example, the designation of Mr. Arun Sobti in ADC Telecommunications Inc for the year 2000 is President-Broadband Access & Transport Group. To obtain our data, we hand match all such designations in ExecuComp with names of divisions from Compu- 1
stat business segment files. This provides us with a matched sample of DM compensation (from ExecuComp) and division performance (from Compustat business segment files). Our sample spans the period 1992-2009 and includes over 4,080 DM-year pay contracts. When required, we complement our sample with compensation data from the firm s proxy statements. As is clear, our sample is not a random sample of DMs. In Section 3.0.1 we discuss the possible biases due to the non-random nature of the sample and the measures we take to mitigate them. In our empirical analysis, we relate DM pay to the performance of both her division and the other divisions in the firm. We identify three hypothesis that have predictions relevant for our setting. The Risk hypothesis as first formalized in Holmstrom (1979) predicts that an agent s pay should only be related to performance measures directly under her control. To the extent the DM controls the performance of her division, she should obtain pay only for her division s performance. If the performance of other divisions informs the CEO about common risk factors, optimal risk sharing would call for relative performance evaluation and hence a negative relationship between DM pay and the performance of other divisions. The Externality hypothesis on the other hand highlights the links between the divisions of a conglomerate. These links can be real as in when divisions have a customer-supplier relationship, or financial as in when divisions share a common pool of capital. In such situations, DMs can take actions that benefit their division possibly at the expense of firm value. This can be avoided by paying DMs for the performance of both her division and the other divisions in the firm. The Externality hypothesis also predicts how the pay for other division s performance will vary with the extent of relatedness of the divisions and their level of investment opportunities. The need for conglomerates to share capital may vary with economic conditions (e.g., Gopalan and Xie (2008), Dimitrov and Tice (2006)). This would imply time-variation in DM pay-for-performance. We group these predictions under the Time series hypothesis and discuss them along with our results. We begin our empirical analysis by relating DM pay to the performance of her division and that of the other divisions in the firm. We use the return on assets (ROA), which we calculate as the ratio of operating profits over book value of total assets as our measure of 2
performance. We find evidence for significant pay-for-performance for DMs. For the average DM in our sample, a 1% increase in divisional ROA is associated with a 0.426% increase in DM total pay. Given the mean division size and DM pay in our sample, this translates into a $1.14 average increase in DM pay for every $1000 increase in annual divisional profits. In comparison, for the same set of firms, CEO pay on average increases by $2.98 for every $1000 increase in annual divisional profits. Thus DM pay-for-performance sensitivity is almost 40% of that of the firm s CEO. Consistent with the Externality hypothesis, we find that DM pay is positively related to the performance of the other divisions in the firm, especially when we employ firm fixed effects. Further, consistent with the DM having a greater influence over her division s performance, DM-pay is more sensitive to her division s performance as compared to the performance of the other divisions. Specifically a 1% increase in the ROA of the other divisions is associated with only a.22% increase in DM pay. Interestingly, CEO pay is equally sensitive to the performance of all the divisions in the firm. This is consistent with the CEO s actions being perceived to have similar effects across the divisions of the firm. The differences in DM and CEO pay contracts that we document is consistent with the evidence in Aggarwal and Samwick (2003), who show that managerial compensation varies with the level of authority. The extent to which divisional ROA is informative about the true performance of the division can vary across divisions. In our next set of tests we employ two industry-level proxies for accounting informativeness to estimate its effect on DM pay-for-performance. The first is the volatility of accounting profits of all stand alone firms in the industry, with a higher volatility indicating less informative accounting performance (Lambert and Larcker (1987) and Bushman et al. (1996), Ball et al. (2000) and Bushman et al. (2004)). Our second proxy is the extent to which accounting profits are related to contemporaneous stock returns. We measure this by regressing stock returns on accounting profits for all stand alone firms within an industry (Kothari (2001)) and classifying industries with above median regression coefficient as having more informative accounting profits. Consistent with the Risk hypothesis, we find that DM pay for division s performance in industries with more 3
informative accounting profits is twice that in industries with less informative accounting profits. The lower pay-for-performance in divisions with less informative accounting profits may be an important cost of conglomeration. Consistent with the Externality hypothesis we find that DMs of divisions that are related to the rest of the firm obtain greater pay for other division s performance. In these tests we classify divisions in firms with another division in the same three-digit SIC code as being related to the rest of the firm. Rajan et al. (2000a) argue that distortions due to agency conflicts between the DM and the firm s CEO are likely to be especially severe in conglomerates with divisions with very different investment opportunities. If DM pay is designed to mitigate such distortions, then one would expect greater pay for other division s performance in firms with high diversity. Our results actually show greater pay for other division s performance in firms with lower levels of diversity. Prior theoretical literature highlights that capital and pay can act as substitutes in motivating DMs (Scharfstein and Stein (2000)). To specifically test this, we estimate the effect of divisional capital expenditure on DM pay-for-performance and find some weak evidence for higher pay for divisional performance among DMs of divisions with low capital expenditure (see also Wulf (2002)). To test the effect of division growth opportunities on DM pay-for-performance, we use industry Tobin s Q and the past divisional sales growth as our measures of growth opportunities. We find that DMs of divisions with low growth opportunities obtain higher pay for other division s performance. We do not find significant differences in the pay for own division s performance across divisions with different levels of growth opportunities. To the extent conglomerates operate an internal capital market (ICM) and redistribute capital, it would be optimal to take capital from the low growth division and divert it to the high growth division. The low growth division s DM will truthfully report her profits and co-operate with such transfers if her pay is linked to the performance of the high growth division. When we jointly estimate the effect of growth opportunities and the extent of relatedness on the DM s pay for performance, we find greater pay for other division s performance among related divisions with more growth opportunities. This is consistent with the predictions in 4
Anctil and Dutta (1999) who argue that DMs of divisions that are investing and expanding fast may have a greater ability to affect the performance of other divisions, especially if they are related. When we differentiate between economic upturns and downturns, we find that DM-pay is more (less) sensitive to the performance of both her division and the other divisions in the firm during economic upturns/industry boom (economics downturns/industry distress). This evidence is consistent with performance measures being more noisy during downturns and is also consistent with the asymmetric benchmarking for CEO pay as shown in Garvey and Milbourn (2003). To control for sample selection, we repeat all our tests using the Heckman two-step procedure. We use the number of divisions reported by a firm in the Compustat business segment files, Number of divisions and the relative size of the division, Relative size the ratio of the book value of total assets of the division to the book value of total assets of the firm as our instruments for sample selection. The identifying assumption in our Heckman procedure is that while Number of divisions and Relative size should be related to whether or not a DM s pay is included in our sample, conditional on the control variables employed, they should not be correlated with DM total compensation. We believe these two requirements are satisfied in our setting. A DM will be included in our sample if she is among the top five highest paid executives in the firm. Thus DMs from firms with fewer divisions and those from larger divisions are more likely to be included in our sample. Consistent with this conjecture, both Number of divisions and Relative size are significantly related to the likelihood of a DM being included in our sample. We also do not think that conditional on the control variables employed, Number of divisions and Relative size will be correlated with DM total compensation. The specific control variables that we employ include the division s size, firm size, division s past sales growth, capital expenditure, industry earnings volatility and the extent of firm diversity. We find the results discussed above to be robust to repeating all our tests using the Heckman two-step procedure. Overall our evidence is consistent with DM compensation contracts being constrained optimal. Consistent with the Risk hypothesis, we find that DMs have less pay-for-performance 5
in situations where the performance measures are noisy. The latter result highlights an important cost of conglomeration. Consistent with the Externality hypothesis we find that DMs get pay for other division s performance in situations where their actions can affect the performance of the other divisions. Finally, consistent with the Time series hypothesis we find pay for other division s performance to be greater during economic upturns, when the performance measures are likely to be less noisy. The papers that are most related to our paper are Cichello et al. (2008) and Wulf (2002). Cichello et al. (2008) study turnover and promotions of DMs and find divisional ROA to be most closely related to job allocation decisions. While they find evidence of relative performance evaluation (RPE) in promotion decisions they do not find a similar effect for turnover decisions. In contrast, we focus on the structure of DM compensation contracts. Wulf (2002) raises a question similar to ours, although there are important differences between the two papers. Wulf (2002) uses data on 131 multi-division firms for the year 1993 to understand the effect of DM pay on divisional investment. Since she does not have pay for individual DMs, she uses the average DM level pay to represent the pay of the largest division s manager. Focussing on how incentives and capital budgeting may act as substitutes in overcoming agency problems, Wulf (2002) shows that the sensitivity of divisional investment to divisions performance is lower (higher) when DMs have higher (lower) pay for firm performance. Unlike Wulf (2002), we employ a large panel of DM pay contracts from 708 firms to understand its structure. Our data allows us to understand how DM pay varies in the cross-section and through time and thus test a number of different hypothesis. The rest of the paper is organized as follows. In the next section we develop the hypotheses. In Section 2 we describe our empirical specification, key variables and discuss sample selection issues. Section 3 describes the data and provides the summary statistics. In Section 4 we discuss our empirical results and Section 5 concludes. 6
1 Hypotheses In this section we outline the hypothesis that have predictions relevant for our setting. We group the hypothesis into three: Risk hypothesis, Externality hypothesis and Time series hypothesis. Holmstrom (1979) predicts that an agent s pay should only depend on performance measures directly under her control and those that help learn about her actions with least noise. We refer to this as the Risk hypothesis. To the extent a DM only controls the performance of her division, the Risk hypothesis predicts that DM pay should only be related to her division s performance. To the extent the accounting based measures of divisional performance are noisy, the Risk hypothesis also predicts lower pay-for-divisional performance in industries with less informative accounting profits. A division s performance may be affected by both DM effort and exogenous factors outside her control. If a conglomerate s divisions are related in that their performance is subject to similar exogenous risk factors, the performance of other divisions will reveal information about the exogenous factors. Holmstrom (1979) predicts that in such cases, it is optimal to reward the DM for divisional performance relative to the performance of the other divisions. Summarizing, the predictions from the Risk hypothesis are: Prediction 1: DMs should only obtain pay for divisional performance. Prediction 2: Pay for divisional performance should be lower in industries with less informative accounting profits. Prediction 3: In conglomerates with related divisions, DMs should be paid for divisional performance relative to performance of other divisions. Divisions within a conglomerate are often linked. The links can be real as in when divisions have a customer-supplier relationship, or be financial as in when divisions share a common pool of capital. For example, one of the firms in our sample is Alcan Inc for which we have pay and performance information for two DMs; the President and CEO of the Primary metals group and the President and CEO of the Engineered products group. 7
From the names of these divisions, it is evident that some of the products of the primary metals group could be bought by the engineered products group. Thus the actions of the two DMs can affect the performance of both their divisions. We refer to the hypothesis that highlights such links as the Externality hypothesis. With such linked divisions, paying DM solely for divisional performance may induce actions that benefit the division possibly at the expense of overall firm value. To avoid this, firms may link DM pay to the performance of both her division and the other divisions in the firm. To the extent the DM has less influence over the performance of other divisions, DM pay for other division s performance should be less than the pay for her division s performance. DM pay for other division s performance should be greater in conglomerates with related divisions as such divisions are more likely to share resources (Bernardo et al. (2004)). Rajan et al. (2000a) model conflicts in conglomerates with divisions that vary in the extent of investment opportunities and highlight how such conflicts may be resolved by distorting the capital allocation. One can resolve such conflicts through incentive contracts as well. If conflicts are greater in firms with more diverse divisions, then one would expect greater pay for other division s performance in such firms. Conglomerate CEOs can provide incentives to the DM through pay and capital allocation. To the extent the DM has an incentive to manage larger divisions, CEOs can make capital allocation contingent on prior performance and get the DM to take the right actions. To the extent pay and capital allocation are substitute incentive mechanisms, we should expect lower pay-for-division performance in divisions with more capital expenditure. Focussing on capital allocation, conglomerates with divisions that vary in the extent of investment opportunities may find it optimal to transfer capital from the low-growth division to the high-growth division. The DM of the low growth division can be made to go along with such transfers by linking her pay to the performance of the other divisions. This would predict greater pay for other division s performance for DMs of divisions with less investment opportunities. On the other hand, Anctil and Dutta (1999) argue that DMs of divisions that are investing and expanding fast may have greater ability to affect the performance of other divisions. Thus Anctil and Dutta (1999) predict greater pay for other 8
division s performance for DMs of high growth divisions especially if they are related to the rest of the firm. Summarizing, the Externality hypothesis predicts: Prediction 4: DMs should be paid for performance of both her division and the other divisions in the firm. Pay for other division s performance should be greater for DMs in firms that are related to the rest of the firm. Prediction 5: Pay for other division s performance will be greater for DMs in firms with greater diversity. Prediction 6: Pay for division s performance will be lower for managers of divisions with greater capital expenditure. Prediction 7: Pay for other division s performance will be greater for managers of divisions with less investment opportunities. Prediction 7 - Alternative: Pay for other division s performance will be greater for managers of divisions with more investment opportunities. This is especially so if the division is related to the rest of the firm. We now turn to the hypothesis that has predictions on how DM pay-for-performance will vary through time. Gopalan and Xie (2008) show that conglomerates help divisions overcome financial constraints during periods of industry distress. Dimitrov and Tice (2006) highlight similar behavior among conglomerates during periods of economic downturns. To facilitate such a transfer of resources across divisions during downturns, it is important for DMs to care about overall firm performance. This can be achieved by linking their pay to the performance of the other divisions. Note that downturns may also be periods of greater aggregate uncertainty which in turn may increase the noise in observable measures of performance. This would imply a lower overall pay-for-performance sensitivity during such times. In contrast, Bernardo et al. (2001) show that when divisions are related, DMs need to 9
be given high powered incentives to exert effort towards the success of the projects during good times. This implies greater pay for other division s performance in firms with related divisions during economic upturns. Summarizing the predictions from the Time series hypothesis, we have: Prediction 8: Pay for other division s performance will be greater during periods of industry distress and economic downturns. Prediction 8 - Alternative: Overall pay-for-performance will be greater during periods of industry booms and economic upturns. We now discuss our empirical approach and construction of key variables. 2 Empirical design, key variables and sample selection 2.1 Empirical design and key variables We are interested in understanding how DM pay is related to the performance of her division and that of the other divisions in the firm. To achieve this, we estimate variants of the following model. Total compensation ijt = α + β 1 Division ROA it + β 2 Other division ROA it + γ Z ijt + µ t, (1) where subscript i refers to the firm, subscript j refers to the division and subscript t refers to time in years. The dependent variable Total compensation ijt is the sum of annual salary, bonus, other annual compensation, long-term incentive payouts, and other cash payouts. Division ROA (Other division ROA) is the return on assets of the division (other divisions). We calculate ROA as the ratio of operating profits over total assets. In the case of other divisions, we aggregate their operating profits and their total assets. To test how DM pay-for-performance varies with the informativeness of accounting earnings, we employ two industry-level measures of accounting informativeness. Our first 10
measure is the historic volatility of accounting earnings of firms within an industry (Lambert and Larcker (1987) and Bushman et al. (1996), Ball et al. (2000) and Bushman et al. (2004)). Our second measure is the value relevance of earnings, which we measure as the extent to which accounting earnings are related to contemporaneous stock returns (Kothari (2001)). We obtain this measure by regressing annual stock returns of all firms in an industry on their annual earnings per share. The coefficient estimate is referred to as the value relevance of accounting earnings. To test our prediction we divide our sample into divisions with high and low levels of accounting informativeness and estimate (1) in the subsamples and compare our estimates of β 1 and β 2 across the two subsamples. Note that this is equivalent to estimating (1) with a full set of interaction terms between, High information, a dummy variable that identifies industries with more informative accounting profits and all the independent variables and testing for significance of the coefficients on High information Division ROA and High information Other division ROA. We employ a similar procedure to test all our cross-sectional predictions. To test how DM pay-for-performance varies with the extent to which a division is related to the rest of the firm, we identify divisions in firms with another division in the same three digit SIC code industry as related to the rest of the firm. To test Prediction 5, we follow Rajan et al. (2000a) and construct a diversity index for each conglomerate in our sample. We measure diversity as the standard deviation of divisional asset-weighted industry Tobin s Q divided by the equally weighted average industry Tobin s Q of the divisions in the firm. We split our sample into firms with above and below median diversity and estimate (1) in the two subsamples. To test if actual capital allocation and DM pay are substitutes, we employ two proxies based on the actual investment expenditures in the division. Our first proxy is the extent of capital expenditure in the division which we measure as the ratio of division capital expenditure over division total assets. Our second proxy is based on the extent to which the division receives cash inflow from the other divisions in the firm ( subsidy ). Our measure is based on Billett and Mauer (2003) and estimates the divisional subsidy as the difference between the division s capital expenditure and the estimated after tax cash flow of the 11
division. To test Prediction 7 we use two alternative measures of divisional investment opportunities. Our first measure is the industry Tobin s Q which we calculate as the median Tobin s Q of all stand alone firms in the industry during the year. We classify industries with above median Tobin s Q as having more investment opportunities than industries with below median Tobin s Q. 2 Our second measure of division investment opportunities is the past sales growth of the division. We classify divisions with above median annual sales growth during the previous year as having more investment opportunities. To identify periods of industry distress, we use a dummy variable Distress, that takes the value one for divisions in industries in distress. We classify an industry at the threedigit SIC code level to be in distress if the median two-year stock return of all single segment firms is less than -30% (see Opler and Titman (1994) and Gopalan and Xie (2008)). Along similar lines, we classify an industry to be in boom if the median two-year stock return is greater than 30% and construct a dummy variable, Boom to identify divisions in such industries. Our final variable, Recession, identifies recession years. We construct this variable based on recession data from the NBER website to identify the years 1998, 2001-2002 and 2008-2009. In all our regressions we control for firm size using, Log(Total assets), division size using Log(Segment assets), and time fixed effects. The standard errors in all our regressions are robust to heteroscedasticity and are clustered at the individual firm level. 2.2 Sample selection We have pay data only for DMs who are among the top five highest paid executives in the firm and whose designation is descriptive enough to enable us to match them with the division name from Compustat business segment files. Thus our sample is likely to be 2 Finance literature often uses median Tobin s Q of stand alone firms in the industry to proxy for investment opportunities of a division. The justification is based on Montgomery and Wernerfelt (1988) who find that most of the cross-sectional variation in Tobin s Q is explained by industry fixed effects. However, Whited (1999) notes that the industry median is likely to overestimate the division s true Q. Any noise in the measurement of investment opportunity will bias our estimates downward. 12
a non-random subsample of the population of divisions in publicly listed conglomerates. Specifically, the DMs in our sample are likely to control larger divisions so as to be among the top five highest paid executives in the firm than a random DM. This can potentially bias our conclusions. For example, the DMs in our sample may be more influential and receive greater pay for the performance of other divisions because their actions are more likely to have externalities. To control for potential sample selection bias we use the two-step procedure outlined in Heckman (1979). We use the number of divisions in the firm, Number of divisions and the relative size of the division, Relative size the ratio of the book value of total assets of the division to the book value of total assets of the firm as our instruments for sample selection. Our identifying assumption is that while Number of divisions and Relative size should be related to whether or not a DM is included in our sample, conditional on the control variables employed, they should not be correlated with DM total compensation. We believe these two requirements are satisfied in our setting. Number of divisions and Relative size will affect the inclusion of the DM in our sample because DMs from firms with fewer divisions and DMs of the larger divisions are more likely to be among the top five highest paid executives in the firm. Further, there is no reason to expect the Number of divisions or Relative size to affect DM total compensation especially after we control for both the size of the division and firm size. To ensure the second requirement, in addition to the usual set of control variables, we also include Division capital expenditure, Division sales growth, Industry Tobin s Q, Earnings volatility and Herfhindal index-3 digit as additional controls in our two-step procedure. Detailed definition of all the variables we use in our analysis is provided in Appendix A. In estimating the two-step procedure, we use boot strapped standard errors in the second stage because the Inverse Mills ratio that we include in the second stage is a generated regressor. Standard errors are also robust to heteroscedasticity and clustered at the firm level. 13
3 Data and summary statistics 3.1 Data The data for our paper is obtained from four standard sources. Data on stock returns and firm financial data are obtained from CRSP-Compustat merged database. Financial data for individual business divisions are from Compustat business segment files. The data on division manager compensation and designation is collected from ExecuComp and Def 14A proxy statement files. Data on CEO compensation is obtained from ExecuComp. Our sample period extends from 1991-2009. ExecuComp provides the annual compensation data of the top five executives for all S&P 1500 firms. Along with their compensation, ExecuComp also provides the executive s designation, which sometimes reveal them to be a high ranking official in a division. For example the designation of Mr. Arun Sobti in ADC Telecommunications Inc for the year 2000 is President-Broadband Access & Transport Group. We hand match the division name as indicated in the executive s designation with names of divisions from Compustat business segment files to obtain a matched sample of DM compensation and division performance. Using this procedure we are able to identify over 6,747 individual division-year observations. From this overall sample, we retain divisions that have positive and nonmissing values for division sales and are in firms with a minimum of two divisions. This leaves us with 4,080 division-year observations. The drop in the number of observations from 6,747 to 4,080 is mainly because of the requirement of positive sales. All the service divisions such as Maintenance have zero or missing values for division sales. In Section 3.2, we compare the average compensation characteristics of all divisions managers that we identify (the sample of 6,747) to the subsample with non-missing division sales. 3.2 Summary statistics In Panel A of Table 1, we provide the summary pay characteristics of DMs and CEOs in our sample. In the first panel we provide the pay characteristics of the DMs of divisions 14
with positive sales while in the second panel we provide the pay characteristics of DMs with missing sales. Comparing the first panel to the second panel we find that managers of divisions with positive sales have slightly lower compensation as compared to managers of divisions with missing sales and this lower compensation is reflected in all four components of pay. We also find that the DMs of divisions with positive sales are slightly older than DMs of divisions with missing sales. In the third panel we provide the average pay characteristics of the CEOs in our sample. The CEOs are for the divisions with positive sales. We find that total compensation of CEOs is significantly higher than that of division managers ($4,747.31 thousand as compared to $1677.32 thousand). The higher pay is reflected in all the four components of pay. The average CEO is of similar age as the average DM. In Panel B of Table 1 we provide the summary financial characteristics of the divisions in our sample. The mean Log(Division assets) of our sample is 6.44, which translates into a book value of division assets of $627.7 million. Thus the divisions in our sample are fairly large. The median value of Log(Division assets) is comparable to the mean value. In comparison, the mean value of Log(Other division assets) the natural logarithm of the book value of total assets of all the other divisions in the firm is 6.88. Thus the size of the division for which we have DM pay is comparable to the size of the rest of the firm. Since our sample firms have three divisions on average, we are more likely to have pay characteristics of the DM of the largest division. This is not surprising because the DM of the largest division is more likely to be among the top five highest paid executives in the firm. Our sample comprises of the larger firms in Compustat. The mean value of Log(Total assets) of our sample is 7.73 as compared to 5.66 for all firms in Compustat. We find that the divisions in our sample are on average profitable as seen from the mean value of Division ROA of 0.14. Our divisions also tend to be more profitable than the rest of the firm as can be seen by comparing the mean value of Division ROA to that of Other division ROA. The divisions for which we have pay data are also growing fast (mean value of Division sales growth of 0.137), but the rest of the firm appears to be growing at a faster rate (mean value of Other division sales growth of 0.201). The divisions in our sample 15
have growth opportunities as seen from the mean value of Industry Tobin s Q of 1.62. The average Diversity index of the conglomerates in our sample is.26. We classify about 22.4% of the division-years in our sample as distressed as can be seen from the mean value of Distress and classify about 21.5% of the division-years in our sample as being in a boom period as can be seen from the mean value of Boom. Almost one-third of the division-years in our sample are in recession, which is mainly because of the recent financial crisis. We now present the results of the univariate tests of our predictions. Panel C reports the correlation between the variables we use in our analysis. Consistent with externality hypothesis, we find that DM pay is positively correlated with the performance of both her division and the other divisions in the firm. Not surprisingly, DM compensation is higher in larger firms and for DMs of larger divisions. We find that DM compensation is higher for divisions in industries with more volatile performance. Consistent with conglomerates using pay and capital allocation as substitutes, we find that DM pay is lower in divisions with higher capital expenditure and is higher in divisions with high past sales growth. From the table we find that the correlation between Division ROA and Other division s ROA is modest at 26.3%. This ensures our ability to separately identify DM pay for both division performance and other division s performance. In Section 4.1 we discuss the tests we do to ensure that multi-collinearity is not a serious problem for us. Not surprisingly, the more profitable divisions in our sample tend to be larger, receive greater capital (positive correlation with capital expenditure) and have greater growth opportunities (positive correlation with industry Tobin s Q). The correlation between firm size and division size in our sample is very high (0.827), mainly due to the inclusion of a disproportionate number of divisions that are the largest in the firm. Larger divisions appear to have fewer growth opportunities (as seen from the negative correlation between Log(Division assets and Industry Tobin s Q & Division sales growth) and they also do less capital expenditure as a proportion of division size. We now proceed to multivariate tests of our predictions. 16
4 Empirical results 4.1 DM pay-for-performance We begin our empirical analysis with tests of Prediction 1 and present the results in Table 2. The sample in Column (1) includes all executives who manage divisions with positive sales. The dependent variable is Log(Total compensation). Focussing on Column 1, the positive and significant coefficient on Division ROA indicates that DMs obtain pay for division performance. We also find that the coefficient on Other division ROA is also positive but not significant. This is consistent with the Risk hypothesis. From the coefficients on the control variables we find that managers of larger divisions in bigger firms obtain higher total compensation. The coefficient on Division ROA indicates that a 1% increase in divisional ROA is associated with a.426% increase in DM pay. Given that the average value of division total assets in our sample is $627.7 million and the mean total compensation for a DM is $1.677 million, our estimate in Column 1 translates into a $1.14 increase in DM pay for every $1000 increase in annual divisional profits. In Column (2) we repeat our test after including firm fixed effects and find that the coefficient on both Division ROA and Other division ROA is positive and significant. Comparing Column (2) to Column (1) we find that the coefficient on Other division ROA is higher and its standard error is lower in Column (2). Thus once we control for firm fixed effects, our results are actually consistent with the Externality hypothesis. Consistent with DMs having a greater influence on division performance, we find that the pay for division performance is greater than the pay for other division performance. The coefficient on Other division ROA in Column (2) indicates that a 1% increase in other division s ROA is associated with a.22% increase in DM pay. In dollar terms a $1000 increase in the profits of the other divisions is associated with a $0.39 increase in DM pay. Given that we have pay characteristics of the manager of the largest division, one concern would be the extent to which Division ROA and Other division ROA are correlated and hence our ability to independently estimate pay for division and other division perfor- 17
mance. As mentioned before, in the full sample, the correlation between Division ROA and Other division ROA is moderate at 0.263, Further to ensure that multi-collinearity between Division ROA and Other division ROA is not clouding our inferences, we estimate the variance inflation factor (VIF) which measures the increase in variance of estimated coefficient due to collinearity. The VIF in our sample is 1.07 which is considered small. According to the standard rule of thumb (Obrien (2007)) a VIF greater than ten is considered a cause for concern. Further in our sample, the lowest Eigen value is 0.422 and the condition number based on Eigen values is 2.074. An Eigen value close to zero or a very high condition number, say greater than thirty, implies a high degree of linear-dependency between the two variables. These tests help us allay any concerns of multi-collinearity. In Column (3) & (4) we present the results from the Heckman two-step procedure. In Column (3) we present the results of the first-stage regression. The sample for this estimation includes all divisions with non-missing values of sales and total assets in the Compustat business segment files and which belong to the S&P 1500 index (the Execu- Comp sample). The results indicate that more profitable divisions (positive coefficient on Division ROA), larger divisions (positive coefficient on Log(Division assets)), from smaller firms (negative coefficient on Log(Total assets)), from firms with few divisions (negative coefficient on Number of divisions), and divisions whose size is a smaller proportion of firm size (negative coefficient on Relative size), are more likely to be included in our sample. Note that the coefficient on Relative size is opposite to what one would expect mainly because of the inclusion of division size and firm size in our specification. Our results indicate that conditional on firm and division size, divisions that are a smaller proportion of firm size are more likely to be included in our sample. We further find that the divisions in our sample have lower past sales growth, higher industry Tobin s Q, are from industries with higher earnings volatility and from firms with lower Herfindahl index. In Column (4) we present the results of the second stage estimation with Log(Total compensation) as the dependent variable and include the inverse Mill s ratio estimated from the first stage as an additional regressor. Note that we include additional control variables in the Heckman twostep procedure to ensure that Number of divisions and Relative size are truly exogenous. 18
From Column (4) we see that the estimate of DM pay-for-performance after we control for sample selection is similar to what we obtain when we do not control for sample selection (Column (1)). This indicates that sample selection is not biasing our coefficient estimates. From the control variables we find that DMs of divisions that have high past sales growth, those from industries with high Tobin s Q have higher total compensation. The number of observations in Column (4) is less than that in Column (1) because of missing values of division capital expenditure and sales growth. A firm s division manager and its CEO are likely to have different spheres of influence. While the DM s actions are likely to have a direct effect on her division s performance, her actions are likely to affect other division s performance only in an indirect manner. In contrast, a CEO s actions are likely to affect the performance of all the divisions in the firm. If firms design incentive contracts taking these differences into account, then the pay-for-performance sensitivity of CEOs should be different from that of DMs. To see if this is the case, in Column (5), we repeat our estimation with CEO compensation for the firms in our sample as the dependent variable. Consistent with our conjecture, we find that pay-for-performance sensitivity of the CEO is different from that of the DM. Unlike DM pay, we find CEO pay to be equally sensitive to both Division ROA and Other division ROA. The estimates from Column (5) indicate that a 1% increase in divisional ROA is associated with a.394% increase in CEO pay. This translates into an average increase in CEO compensation of $2.98 for every $1000 increase in annual divisional profits. Thus although the coefficients on Division ROA is similar across Column (1) and Column (5), due to differences in the level of DM and CEO pay, the pay-for-performance sensitivity for the DM and the CEO are very different in terms of dollar amounts. In unreported tests we find that the loading on Other division ROA is significantly different between Column (1) and Column (5). In Column (6) we repeat our estimates from Column (5) after including firm fixed effects and in Column (7) we present the results of the second stage regression after controlling for sample selection using the Heckman two-step procedure. Note that the first stage for this estimation is the same as in Column (3). We find the results in both Columns (6) & 19
(7) to be similar to the ones in Column (5). In unreported tests, we repeat our analysis with Bonus and Short-term pay (salary + bonus) as measures of compensation and obtain qualitatively similar results. 4.2 Pay for division and firm performance and accounting informativeness In Table 3, we test Prediction 2. We use the volatility of earnings of firms in an industry as our first measure of accounting informativeness with a higher volatility indicating an industry with less informative profits. We divide our sample into divisions in industries with above and below median earnings volatility and estimate (1) in the two subsamples and present the results in Column (1) and Column (2) respectively. The results indicate that consistent with Prediction 2, DM pay for division performance is indeed lower (higher) for divisions in industries with high (low) earnings volatility. The DM pay for division performance in industries with more informative accounting profits is approximately twice that in industries with less informative accounting profits. Interestingly, we do not find any significant difference in pay for other division s performance across the two subsamples. From the row titled Division ROA we find that the coefficients on Division ROA are significantly different across Columns (1) and (2). Note that our tests in the row titled Division ROA is equivalent to estimating (1) with a full set of interaction terms between High information, a dummy variable that identifies industries with more informative accounting profits and all the independent variables and testing for significance of the coefficient on High information Division ROA. In Columns (3) & (4) we repeat our tests after controlling for the Inverse Mills ratio and obtain similar results. In these tests, due to smaller sample size and the greater noise in our estimates, we find that the coefficient on Division ROA is not significantly different between the two columns although the magnitude of the difference is large. In the next two columns we use the value relevance of accounting earnings as our measure of earnings informativeness. As before, we divide our sample into divisions in industries with 20
above and below median earnings value relevance and estimate (1) within the two subsamples. Our results indicate that DM pay for divisional performance is greater for divisions in industries with more informative earnings. Here again the coefficient in Column (5) is more than twice as large as that in Column (6), and the difference is significant at less than five percent level. In Columns (7) and (8) we repeat the tests after controlling for the Inverse Mills ratio and obtain qualitatively similar results. These results offer significant support for Prediction 2 and highlight one important cost of conglomeration. Conglomerates with divisions in industries with less informative accounting profits offer lower pay-for-performance to the DM. To the extent pay-for-performance is useful in aligning the interests of the DM with that of the firm s shareholders, this will prove costly. 4.2.1 DM pay-for-performance and relatedness of divisions In Table 4 we estimate how DM pay-for-performance varies with the degree of relatedness of the divisions in the firm. According to the Externality hypothesis pay for other division s performance should be higher for divisions that are related to the rest of the firm as such divisions are more likely to share resources with the other parts of the firm. On the other hand, according to the Risk hypothesis, relative performance evaluation (RPE) should call for a negative relationship between DM pay and the performance of the other divisions when the divisions in the firm are related. To test these contrasting predictions, we classify divisions in firms with another division in the same 3-digit SIC code as being related to the rest of the firm and repeat our tests in the subsamples of related and un-related divisions. The evidence in Columns (1) and (2) indicate that consistent with Prediction 4 and inconsistent with RPE, DM pay is positively related to the other division s performance in divisions that are related to the rest of the firm. The loading on Other division ROA in Column 1 is twice that in Column 2 and the coefficient in Column 2 is not statistically different from zero. In unreported tests, we repeat these test with relatedness defined at 2-digit SIC code industry level and obtain qualitatively similar results. In Columns (3) and (4) we repeat our tests after controlling for sample selection using the Heckman two-step procedure and obtain results similar to those in Columns (1) and (2). 21