Abstract
In this study, we show that the option-like structure of equity-based compensation encourages managerial risk-taking and provide new evidence on the way in which CEO’s risk-taking could manifest itself in a multi-segment firm. Our results show that a greater sensitivity of managerial compensation to shareholder wealth—as proxied by CEO’s portfolio vega—leads to greater risk-taking through active capital allocation. We then analyze the impact of risk-taking on shareholder wealth and demonstrate that risk-taking is positively associated with future stock returns. Overall, this article contributes to the literature by providing evidence that equity-based compensation does actually promote the alignment of interests between shareholders and managers.
1. Introduction
It is commonly recognized that managers have discretion in making investment and finance decisions that influence firm risk, and that they are more risk-averse than shareholders since they are undiversified with respect to firm risk. Since they may be exposed to significant risks resulting from human capital being vested in the firm (Amihud and Lev, 1981; Smith and Stulz, 1985) and/or perquisite benefits (Williams, 1987), it is likely that managers become reluctant to take up risk-increasing projects even if they are expected to produce profits, unless they are compensated for bearing such risk. In the 1990s, equity-based compensation was introduced to alleviate such managerial risk-averse behavior by granting managers stocks and stock options. In this study, we analyze the impact on managerial risk-taking of equity-based compensation contracts by looking at CEOs’ capital budgeting decisions within multi-segment firms. As a result, this article makes significant contributions to the literature on both executive compensation and internal capital markets.
First, we provide evidence that equity-based compensation, particularly stock options, improves the alignment of risk-taking interests between shareholders and managers. In particular, CEOs with a greater sensitivity of their equity-related wealth to stock volatility do increase firm-specific risk, which is consistent with the interest of shareholders. To date, the literature has been inconclusive as to whether executive stock options promote managerial risk-taking. Some scholars have argued that stock options do motivate risk-taking since stock options increase in the volatility of the underlying stock while providing at the same time a protection to managers against bad outcomes (Cohen et al., 2000; Coles et al., 2006; Guay, 1999; Low, 2009; Rajgopal and Shevlin, 2002; Smith and Stulz, 1985). Other studies instead have advocated that stock options would create incentives for managers to work more diligently, but the extent of risk-taking ultimately depends on the level of managerial risk-aversion and managers’ wealth at stake in the firm. Accordingly, a stock option may even represent a deterrent to managerial risk-taking (Hirshleifer and Suh, 1992; Ju et al., 2002; Lewellen, 2006; May, 1995).
Second, and more importantly, we document new evidence on how CEOs increase their exposure to risk within multi-segment firms, by looking at the intensity of their active internal capital allocation. We show that the more sensitive the CEOs’ wealth to stock volatility, the more actively they shift available funds across the different segments of the firm. To the best of our knowledge, no previous study has specifically examined the channels through which managerial risk-taking could be implemented as a response to the particular compensation structure selected within a multi-segment firm. In fact, while the area of internal capital markets and its efficiency have been extensively investigated (see, e.g. Billett and Mauer, 2003; Guedj and Scharfstein, 2004; Khanna and Tice, 2001; Maksimovic and Phillips, 2002; Stein, 1997, for analyses on efficient internal capital markets, and Berger and Ofek, 1995; Lamont, 1997; Lang and Stulz, 1994; Rajan et al., 2000; Scharfstein, 1998; Shin and Stulz, 1998; Shleifer and Vishny, 1989, for analyses on inefficient internal capital markets), only a few studies have presented an analysis of the impact of equity-based compensation on CEOs’ capital budgeting decisions (Datta et al., 2009; Yong, 2005), leaving the critical issue of managerial risk-taking still unexplored. 1
Our sample is composed of 551 multi-segment firms in Compustat for which we have available compensation data in Execucomp over the period from 1992 to 2006. We investigate to what extent the convexity of stock options (i.e. the sensitivity of managers’ equity-related wealth to stock volatility) influences managerial risk-taking in a multi-segment firm and the implications of this risk-taking on shareholders’ wealth. Since CEOs in a multi-segment firm are generally responsible for allocating capital across segments to the most promising investment projects, an analysis of multi-segment companies provides an ideal setting for us to explore both the dynamics of managerial risk-taking incentives and its consequent effects on firm valuation. For this purpose, we decided to employ different measures of internal capital allocation to capture the level of “activeness” of CEOs in multi-segment firms.
We begin by first analyzing what are the factors behind such active internal capital allocations and argue that risk-taking incentives—implicit in equity-based compensation—influence active/passive capital allocation. Our hypothesis is that CEOs with more convex equity-based compensation are more likely to engage in active internal capital allocation with the aim to increase firm risk. To test this hypothesis, we employ the semi-parametric fixed-effects censored model developed by Honoré (1992) in order to account for the fact that the activeness measures are censored between 0 and 1, by construction. Our results demonstrate that this is indeed the case: the convexity of managerial compensation (i.e. change in the value of managers’ equity-related portfolio in response to a change in stock-return volatility measured by CEO’s portfolio vega) is positively related to the level of CEOs’ activeness in internal capital allocation. To the extent that well-diversified shareholders prefer more risk-taking to less (this is one of the primary purposes for structuring convex payoffs to managers) ceteris paribus, our results provide convincing evidence that the convexity of equity-based compensation does indeed promote the alignment of interests between shareholders and managers. Importantly, using the exogenous changes in the Delaware takeover regime of the mid-1990s, we provide robust evidence that Delaware-incorporated firms with low equity-based incentives (below-median CEO’s portfolio vega) reduce significantly their active capital allocation.
Next, we show that active internal capital allocation originated from risk-taking incentives is positively associated with an increase in firm risk in the subsequent period. Consistent with prior studies (Low, 2009; Shin and Stulz, 2000), we use equity risk as our proxy of firm risk since an executive stock option is written on a firm’s stock. In particular, we look closely at the idiosyncratic component of stock volatility because it is that part of risk which is entirely specific to a firm’s operations. We also show that firms with equity-based compensation tend to allocate more capital to high-risk segments characterized by greater volatility of stock returns and greater volatility of operating performance.
Finally, we investigate the implications for shareholder value of this risk-taking. Specifically, using pooled ordinary least squares (OLS) regression, we test whether risk-taking through more active internal capital allocation creates value by calculating the future performance of active firms. The results show that activeness driven by risk-taking motives benefits significantly firm shareholders and are consistent with the evidence of Choe et al. (2014) indicating that long-term compensation incentives improve the efficiency of resource allocation and increase firm value.
The remainder of the article is organized as follows. In section 2, we introduce our research questions. In section 3, data and sample summary statistics are provided followed by results in section 4. Section 5 provides the findings of additional robustness tests. Section 6 concludes.
2. Hypotheses
We start by first considering the motivations behind CEOs’ activeness in internal capital allocation. We conjecture that risk-taking incentives induced by convex equity-based compensation may be playing a significant role in creating active internal capital markets. Our hypothesis therefore is that CEOs with more convex equity-based compensation (i.e. a higher sensitivity of equity portfolio value to stock volatility) would engage in more active internal capital (re-)distribution with the aim to affect firm risk. If a positive relationship between the convexity and the degree of activeness was detected, and the active internal capital allocation resulted in an increase in firm risk, it would then be consistent with the theory that high CEO vega promotes a better alignment of interests between shareholders and managers. We summarize the first hypothesis as follows.
H1. The higher compensation sensitivity to stock volatility leads managers to engage in more active internal capital allocation across segments.
Next, we investigate whether active capital budgeting, motivated by greater risk-taking incentives, influences firm risk. If it was true that CEOs attempt to increase firm risk via active capital allocation, then we should observe a positive relationship between the degree of activeness in internal capital allocation and firm risk in the subsequent period(s). As a result, our second hypothesis, H2, can be expressed as follows:
H2. More active internal capital markets cause an increase in firm risk.
If indeed there existed a positive correlation between the lagged degree of activeness and the level of firm risk, then the implication of H2, combined with that of H1, would be that CEOs of diversified firms with more convex equity-based compensation (i.e. higher CEO’s portfolio vega) are more likely to exhibit risk-seeking behavior, and that this behavior is pursued by varying more actively the level of capital expenditures across firm’s segments.
Finally, we examine how active (passive) internal capital allocation contributes to shareholders’ value. Indeed, given the convex structure of compensation, managers could “gamble” on the fluctuations of firm performance by selecting riskier investment projects without considering the profitability of such projects, hoping to generate positive expected returns (even by chance). We argue, however, that the concurrent impact of stock price on both the value of CEOs’ stock options and the wealth of shareholders would prevent unnecessary and value-reducing risk-taking. In this context, managerial risk-taking would be beneficial to both managers and shareholders. Therefore, our third hypothesis can be summarized as follows:
H3. Managerial risk-taking through active internal capital allocation has a positive impact on stock value.
If more active firms were also better performing firms, then the higher compensation sensitivities (i.e. convexity) would be consistent with the alignment of interests between shareholders and managers. In this case, equity-based compensation serves not only as an incentive for managerial risk-taking but also as an incentive for making the best use of managers’ private information. Another possibility is that active internal capital allocation does not have any impact on shareholders’ wealth. Even so, this result would still be consistent with the alignment of interests to the extent that a greater convexity of equity-based compensation alleviates managerial risk-aversion.
3. Sample selection and summary statistics
3.1. Sample selection
The segment data are sourced from the Segment and Industrial Annual file (both active and research) provided by CRSP/COMPUSTAT Merged (CCM) Database. Hence, the sample in this study also includes firms that are subsequently deleted from COMPUSTAT due to mergers and bankruptcies. According to FASB-SFAS No. 14 and SEC regulation S-K, publicly traded US companies are obliged to disclose, for every distinct business segment which constitutes more than 10% of total sales, accounting data such as sales, assets, depreciation, capital expenditures, income, and operating profits since 1978. First, we download the universe of segments reported by the Segment and Industrial Annual file provided by CCM Database over the period from 1991 to 2006. 2 We remove segments which operate in agriculture and financial and services industries (i.e. segment SIC code less than 1000, and greater than or equal to 6000). Segments with incomplete data on capital expenditures, sales, depreciation, operating profits, and SIC codes are dropped. We also eliminate segments with zero depreciation, capital expenditures greater than sales, negative capital expenditures, and negative assets. Firms sometimes report segments which do not have actual economic activities titled as, for example, “corporate adjustment,” “intersegment elimination,” and “reconciling items.” Because the accounting data in these segments represent unallocated amount in sales, assets, and operating profits, we also remove these non-economic segments from our sample (Ahn, 2010). Finally, we exclude firms with sales less than US$20 million. After the filtering process, we have 64,889 segment-firm-year observations. To be included in our sample of diversified firms, a firm needs to have a number of segments equal to or greater than two for at least two consecutive years. After imposing this condition, we have 21,189 segment-firm-year observations for diversified firms, and 29,728 firm-year observations for stand-alone firms over the period from 1992 to 2006.
The executive compensation data are collected from the Standard and Poor’s Execucomp database whose original source is a company’s “annual proxy” (DEF 14A SEC form). The compensation data are available from 1992, and this is the reason why the sample period of this study starts from 1992. Execucomp provides detailed information of the top five (ranked by salary and bonus) executives (including the CEO) in a company on compensation such as salary, bonus, stock and option awards, non-equity incentive plans, and other compensation items. Execucomp mostly provides an indicator for a CEO every year. However, in rare cases of missing a CEO indicator, we regard an executive with the highest sum of salary and bonus as the CEO for that year (Yong, 2005). For the calculation of option sensitivities (described later in the methodology section), we obtain dividend yields and stock price volatilities from Execucomp. For the risk-free rate, we use Treasury yields downloaded from the US Department of Treasury website. 3 Stock prices and company annual fundamental data are downloaded from CRSP Monthly Stock File and COMPUSTAT North America Fundamentals Annual database, respectively. After we merge the compensation and the segment data, we have a sample of 551 diversified firms, or an unbalanced panel of approximately 3612 firm-year observations over the period from 1992 to 2006. 4
3.2. Main variable construction
3.2.1. Measures of activeness
One of the main variables in this study is the measure of activeness in internal capital allocation. Guedj et al. (2009) propose three measures to quantify the activeness level by looking at how much a firm’s capital expenditures in each segment in a particular year deviates from (1) firm’s past capital allocation (labeled as DLC), (2) capital allocation of stand-alone firms in industries to which each segment corresponds (labeled as DIC), and (3) free-cash-flows that each segment generates (labeled as DSF). DLC is calculated as the sum of the absolute fractional changes in capital allocation to each segment over 2 years and can be written as follows
where CAPXi, t is the capital expenditure allocated to a segment i of a firm f in year t, where set F includes segments which are reported for both years, t and t−1. As an example, assume that a firm has segments A and B in common over 2 years, t−1 and t. If the firm allocated 20% of total capital expenditures to segment A and 80% to segment B in year t−1, and equally distributed capital in year t, then the DLCt would be [|0.2−0.5|+|0.8−0.5|]/2 (the sum of fractional capital expenditure changes over the 2 years in segment A and in segment B).
The second, DIC, measures the degree of activeness by quantifying the deviation of the capital expenditure allocated to a particular segment from the corresponding industry’s average capital expenditures over the same time horizon and can be written as follows
In calculating the growth factor, gi, t , we sum the capital expenditures of stand-alone firms (s) in the industry (I) where segment i belongs in year t. Then, we subtract the “synthetic capital expenditures (i.e. the product of the growth factor and segment capital expenditures in the previous year),” which a segment would have received if the firm followed industry trends, from the current year’s capital allocation. The last measure, DSF, takes the segment’s free-cash-flow as a benchmark and is defined as follows
where SFi,t−1 is segment i’s free-cash-flow (defined as the sum of operating profits and depreciation) in year t−1. 5 The DLC, DIC, and DSF are all positive as they are expressed as absolute values. Their values range between 0 (i.e. no change in segment capital allocation relative to their benchmark) and 1 (i.e. 100% change in segment capital allocation), and hence, the higher the values of DLC, DIC, and DSF, the more active the diversified firm’s internal capital allocation activities.
The average vesting period of a typical stock option plan in the United States is approximately 30 months or 2.5 years (see, e.g. Kato et al., 2005; Kole, 1997). Given this practice, we argue that risk-taking incentives triggered by the convexity of equity-based compensation would manifest themselves via active internal capital allocation over a long-time horizon. To be consistent with the average vesting period of stock options, in this study we use a 3-year moving-average capital-activeness measures.
3.2.2. Measures of CEO’s risk-taking incentives
Following Core and Guay’s (2002) approach, we use the vega of CEO’s option portfolio and the delta of CEO’s stock and stock option portfolio to measure CEO’s risk-taking incentives. Guay (1999) demonstrates that option vega is substantially higher than stock vega, and hence, the option vega is a sufficient measure to represent both stock and option vega (Coles et al., 2006). An option is priced using the modified Black and Scholes’ (1973) option pricing model which takes into account dividends (Merton, 1973). We calculate an option vega (delta) as a first-order partial derivative of option value with respect to stock volatility (stock price). In particular, option vega (delta) is the dollar change in the value of stock option given 1% change in stock volatility (stock price). CEO’s portfolio vega is the sum of the vegas of all options held by the CEO. CEO’s portfolio delta is calculated as the sum of his or her option delta and stock delta, where stock delta is defined as a dollar change in CEO’s portfolio consisting of stocks in response to a 1% change in stock price.
For pricing new option grants, Execucomp provides the necessary parameters of the Black–Scholes model. Following the method of option valuation employed by Execucomp, we assume that options are granted on grant year’s fiscal year end since the exact grant date is not available. The time-to-maturity of an option is reduced by about 30% to an amount of 70% of the actual time-to-maturity (Core et al., 2003). This is because executives often do not wait until the expiration date to exercise their options (Hemmer et al., 1996; Huddart and Lang, 1996). We use volatility and dividend yield which are directly obtained from Execucomp, and Treasury bond yield matching the remaining time-to-maturity of the option is used as a risk-free rate.
For previous option grants, variables such as exercise price and time-to-maturity are not provided by Execucomp, and hence, they need to be approximated using Core and Guay’s (2002) 1-year approximation method. 6 For both unexercisable and exercisable options, we compute an average exercise price by subtracting realizable value scaled by number of options from year-end price, where a realizable value is defined as stock price less exercise price. The realizable values are disclosed in the proxy statement and can be accessed through Execucomp. The approximated time-to-maturity for unexercisable option is equal to the time-to-maturity of most recent year’s option grant less 1 year. For exercisable options, we set the time-to-maturity equal to the most recent year’s unexercisable options. If there was no new option granted, the time-to-maturity for unexercisable (exercisable) options is set to 9 (6) years. After approximating these parameters, we then compute the vega of previously granted options using the Black–Scholes option pricing formula.
Several studies demonstrate that CEO’s portfolio vega, rather than delta, is the one that induces managerial risk-taking (e.g. Knopf et al., 2002; Kuang and Qin, 2013; Low, 2009). Since this study is concerned with the direct impact of managerial risk-taking incentives on internal capital allocation decisions, we focus on CEO’s portfolio vega as the measure of risk-taking incentives and include CEO’s portfolio delta as a control variable.
3.3. Summary statistics
In Panel A of Table 1, we report the characteristics of our sample firms. All continuous variables are winsorized at the 1st and the 99th percentiles. Diversified firms in our study have, on average, total assets of US$7028 million, and comprise approximately three segments. We define the book-to-market ratio as the total book value of common equity divided by firm’s market capitalization (computed as the product of share price and number of common shares outstanding as of the fiscal year end). The average book-to-market ratio in our sample is 0.5321, indicating that the majority of our sample is composed of growth firms. The average accounting profitability (ROA) of our sample firms measured the net income divided by total assets is 4.46%. On average, the ratio of total capital expenditures to total assets is 0.06 and the growth rate of capital expenditures computed over 3-year periods is about 15% per year. The average Herfindahl index is 0.57. On average, the leverage ratio is 0.27, where leverage is computed as the sum of both long-term and short-term debt divided by total assets. Firms in our sample have and average Altman’s (1968) Z-score of 3.33, indicating that these firms have generally favorable financial prospects. The average stock return is about 16%. Panel B of Table 1 reports the summary statistics of the the 3-year moving averages of DLC, DIC, and DSF previously defined in section 3.2.1. DLC has a mean of 0.10, and a standard deviation close to 0.07. Firms in our sample show relatively lower DLC values than those of Guedj et al. (2009), where the mean value of DLC is equal 0.12. This difference is most likely ascribable to the fact that our sample consists of diversified firms that also have compensation data available in Execucomp which includes mostly large firms—S&P500 firms from 1992 to 1994 and S&P 1500 companies. Given the evidence of Guedj et al. (2009) that smaller firms are more active, it is not surprising that DLC is lower for our sample firms. In Panel C of Table 1, we report the correlation coefficients between the activeness measures and several other firm characteristics. These correlations range (in absolute value) from a minimum of 0 to a maximum of 0.42. Note that the correlation between DLC and DIC is as high as 0.94, while DSF has a relatively low correlation with both DLC and DIC. Guedj et al. (2009) argue that DSF inherits, by construction, the volatile nature of cash flows and hence it captures very different aspects of the capital allocation process compared to DLC and DIC. It is thus not surprising for DSF to exhibit such differences.
Summary statistics of sample firms.
This table reports the summary statistics of diversified firms included in this study. The segment data are sourced from the Segment and Industrial Annual file (both active and research) provided by CRSP/COMPUSTAT Merged Database (CCM), and firm financial data are sourced from COMPUSTAT North America Fundamentals Annual. Stock market data are sourced from CRSP Monthly Stock File. DLC (DIC) [DSF] is a 3-year moving-average activeness measure, which measures the deviation of a firm’s capital expenditures in each segment in a particular year from lagged capital allocation (industry capital allocation) [segments’ free-cash-flows] as described in section 3.2.1. Panel A reports sample firm characteristics. N(seg) is the number of segments reported by a company each year. AT is the book value of total asset (Compustat code, AT), and the book-to-market ratio (B/M) is defined as the total book value of common equity (CEQ) scaled by market capitalization, computed as the product of share price (PRCC_F) and number of common shares outstanding (CSHO) at fiscal year end. The return on asset (ROA) is defined as net income (NI) divided by total asset, and CAPX/AT is the ratio of total capital expenditures (CAPX) to total assets. H(CAPX) is the Herfindahl index of segment’s capital expenditures, and it is calculated as the sum of the squared proportional capital expenditures for each segment. CAPX growth is an annual growth rate of capital expenditures averaged over 3 years. Leverage is the sum of long-term and short-term debt (DLTT and DLC, respectively) divided by total assets, and Z-score (Altman, 1968) measures the probability of bankruptcy described in equation (4) in section 3.3. Stock return is the annualized stock returns during a fiscal year. Panel B summarizes the capital allocation activeness measures, DLC, DIC, and DSF which are defined in section 3.2.1., while Panel C reports the correlations between various firm characteristics and the capital allocation activeness measures.
Panel A: Sample characteristics.
In Table 2, we provide the summary statistics of CEO compensation variables. Total direct compensation includes salary, bonus, total value of stocks granted, total value of stock options granted, long-term incentive payouts, and other annual compensation variables. Cash compensation is the sum of salary and cash bonus. In our sample, CEOs on average received a total compensation of approximately US$3.6 million of which approximately one-third is paid in cash. As previously documented in the literature, the distribution of CEO’s compensation is positively skewed. In Panel B of Table 2, we provide the annual averages of CEOs’ total/cash compensation, portfolio vegas, and deltas. Dollar values are adjusted for inflation and expressed in 2006 dollars. To better understand yearly trends in compensation, we plot the values documented in Panel B of Table 2. Graph A of Figure 1 exhibits the annual average CEO compensation and the average sensitivities of CEO’s portfolios for our sample firms over the period 1992 to 2006. While total cash compensation remains relatively constant, CEO’s total direct compensation increases significantly over this period. This supports the well-known evidence of the unprecedented use of equity-based compensation since the 1990s. As illustrated in Graph B of Figure 1, CEO’s portfolio vega generally peaks in 2002 and decreases thereafter. By contrast, the pattern in CEO’s portfolio delta is less smooth (refer to Graph C of Figure 1), with a peak of almost US$1.4 million in 2004.
Summary statistics of CEO compensation.
Panel A of Table 2 reports the summary statistics of CEO compensation of sample firms. Total direct compensation includes salary, bonus, total value of restricted stock granted, total value of stock options granted, long-term incentive payouts, and other annual compensation (Execucomp variable, TDC1). Cash compensation is the sum of salary and cash bonus. Value of all options granted in the current year is computed based on the modified Black–Scholes option pricing model (Black and Scholes, 1973), and options that are previously granted are valued using Core and Guay’s (2002) approximation method. CEO’s portfolio delta (Delta) is a dollar change in CEO’s portfolio composed of all stock options and stock holdings in response to 1% change in stock price. CEO’s portfolio vega (Vega) is a dollar change in CEO’s portfolio composed of all stock options as a result of 0.01 change in stock volatility. Panel B provides CPI adjusted-annual averages of CEO compensation. The values are expressed in dollars in 2006.
Panel A: Compensation characteristics of diversified firms.
Time-series trend of CEO compensation in diversified firms over the period 1992–2006.
3.4. Non-parametric analysis
We start our analysis by first providing some descriptive statistics of firm-level and CEO compensation characteristics across quintile portfolios of sorted DLC in Panel A of Table 3. Active firms tend to be small in size and value firms compared to other firms in the sample based on sales and book-to-market ratio. Active firms have also more segments than passive firms, lower asset-adjusted capital spending, and lower profitability as measured by ROA, and a less concentrated capital as proxied by the Herfindahl index of segment capital expenditures. Rajan et al. (2000) assert that the discrepancy in investment opportunities among segments affects managerial decision to allocate resources. For this reason, we use the standard deviation of segments’ Tobin’s Qs to proxy for the dispersion in investment opportunities across segments. Active and passive firms show a similar level of dispersion in segments’ investment opportunities. Also, no particular difference between active and passive firms arises with respect to leverage (the sum of long-term and short-term debt divided by total asset) and financial distress as proxied by Altman’s (1968) Z-score. Although the differences between active and passive firms in the growth rate of asset and capital expenditures are significant, the mean values of these variables in each portfolio seem quite volatile. The characteristics of portfolios sorted on DIC (i.e. deviation from industry capital expenditures) in Panel B of Table 3 and DSF (i.e. deviation from segment’s free cash flows) in Panel C of Table 3 are very similar to those documented in Panel A of Table 3.
Summary statistics on firm characteristics of quintile portfolios sorted on activeness.
This table reports the summary statistics of quintile portfolios sorted by three different activeness measures: DLC, DIC, and DSF. These measures are computed as a 3-year moving-average activeness measure, where activeness is computed as the deviation of a firm’s capital expenditures in each segment in a particular year from: lagged capital allocation (DLC), industry capital allocation (DIC), and segment’s free-cash-flows (DSF), as described in section 3.2.1. Panel A reports average company characteristics of quintile portfolios sorted on DLC. N(seg) is the number of segments reported by a company each year. AT is the book value of total asset (Compustat code, AT), and the book-to-market ratio (B/M) is defined as the total book value of common equity (CEQ) scaled by market capitalization, which is computed as the product of share price (PRCC_F) and number of common shares outstanding (CSHO) at fiscal year end. The return on asset (ROA) is defined as net income (NI) divided by total assets, and CAPX/AT is the ratio of total capital expenditures (CAPX) to total assets. H(CAPX) is the Herfindahl index of segment’s capital expenditures, and it is calculated as the sum of the squared proportional capital expenditures for each segment. CAPX growth is an annual growth rate of capital expenditures averaged over 3 years. σ(SegQ) proxies for the discrepancy in investment opportunities among segments as measured by the standard deviation of segments’ Tobin’s Qs. We first calculate Tobin’s Q of all single-segment firms where Tobin’s Q is defined as total assets minus book value of equity (CEQ) plus market value of equity (PRCC_F*CSHO) scaled by total assets. Each segment is then assigned to the beginning of year’s asset-weighted average of Tobin’s Q of single-segment firms operating in the same Fama–French 48 industry as the segment considered. Leverage is the sum of long-term and short-term debt (DLTT and DLC, respectively) divided by total assets. Z-score (Altman, 1968) measures the probability of bankruptcy. The Altman’s Z-score is calculated using the following formula (Altman, 1968): Z = 1.2X1 + 1.4X2 + 3.3X3 + 0.6X4 + 0.999X5, where X1 = working capital/total asset, X2 = retained earnings/total assets, X3 = earnings before interest and taxes/total assets, X4 = market value of equity/book value of debt, and X5 = sales/total assets. Firms with Z-score above 3.0 are considered to be financially stable while those with Z-score below 1.8 are considered to have a very high risk of bankruptcy. Panel B (Panel C) reports the characteristics of quintile portfolios sorted on DIC (DSF). Our proxy of systematic risk is the estimated coefficient on the Rm-Rf factor of the following models: the CAPM (BETACAPM), the Fama–French three-factor model (BETAFF3), and the Fama–French–Carhart four-factor model (BETAFF4), where each model is estimated using a 36-month rolling window. Our proxy of idiosyncratic risk is calculated as the annualized standard deviation of the residuals over the previous 36-month returns as of the end of June every year based on CAPM (IDVOLCAPM), Fama–French three-factor model (IDVOLFF3), and the Fama–French–Carhart four-factor model (IDVOLFF4). Statistical significance at the 1%, 5%, and 10% is denoted by ***, **, and *, respectively.
Panel A: Company characteristics of quintile portfolios sorted on DLC.
Next, we examine risk characteristics of quintile portfolios sorted on activeness. As a measure of firm risk, we calculate the systematic and unsystematic volatilities based on the following factor models: the Capital Asset Pricing Model (CAPM), the Fama–French three-factor model (1993), and the Fama–French–Carhart four-factor model (1997). Each model is estimated using a 36-month rolling window. Firm’s systematic volatility (BETA) is estimated as the loading on the market factor, while firm’s unsystematic volatility (IDVOL) is the annualized standard deviation of the residuals of a factor model. Panel C of Table 3 shows that both systematic and unsystematic volatilities increase almost monotonically in the degree of activeness as measured by DLC. The differences in risk levels between active and passive portfolios are always statistically significant at the 1% level.
4. Results
4.1. Impact of risk-taking incentives on active internal capital allocation
We utilize the panel structure of the data to account for possible unobserved firm-specific effects that may affect CEO’s decisions on internal capital allocation and test the impact of risk-taking incentives of equity-based compensation on internal capital allocation (H1). In particular, we use the Honoré’s (1992) fixed-effects censored regressions since our dependent variable, DLC, is bounded between 0 and 1 by construction. 7
Our regression specification can be expressed as follows
The dependent variable of this regression, Active, is one of our measures of activeness. The main independent variable of interest is CEO’s portfolio Vega. Additional control variables include: cash compensation (Cash Comp) to capture the level of CEO’s risk-aversion (see Guay, 1999); the delta of CEO’s portfolio of options and stocks (Delta) as Coles et al. (2006) show that CEO’s option delta influences managerial decision on firm’s risk profile; the number of segments to proxy for the degree of diversity (N(seg)), the Herfindahl index of segment-based capital expenditures (H(CAPX)); the dispersion in segments’ investment opportunities (σ(SegQ)); the Altman’s (1968) Z-score as a proxy for the probability of bankruptcy; firm size measured by total sales (Size), book-to-market ratio (B/M), previous 12-month stock returns (Stock Return), and annualized capital expenditures growth rate calculated over the previous 3 years (CAPX growth). All regressions include firm fixed effects. In model specifications (2), (4), and (6), we also control for time-fixed effects to account for market-wide fluctuations.
In Panel A of Table 4, we report the results based on the DLC proxy. Our conclusion does not change when we consider the other two proxies of capital allocation, namely, DIC in Panel B of Table 4 and DSF in Panel C of Table 4. The findings of Panel A show a statistically significant positive loading of the dependent variables on CEO’s option vega (Vega) for our sample of diversified firms. 8 Coles et al. (2006) argue that CEOs’ effort to boost stock prices driven by equity-based compensation may influence managerial decision about firm’s risk profile. To rule out the possibility that our findings might be driven by an omitted indirect effect of option delta on option vega, we also include CEO’s portfolio delta, Delta, as an additional control variable. The results, reported in columns (3) and (4) of each panel, indicate that even after controlling for Delta our conclusion remains unchanged. In addition, Delta does not exert any significant influence on segment-level investment decisions of diversified firms (refer to columns (5) and (6)). This is consistent with Low (2009) who provides causal evidence that CEO’s portfolio vega, not delta, affects the level of CEOs’ (idiosyncratic) risk-taking. Overall, the results of Table 4 validate our hypothesis H1.
Impact of risk-taking incentives on active internal capital allocation.
This table reports the results from Honoré’s (1992) fixed-effects censored regressions. The dependent variable is one of our active capital allocation proxies of DLC (in Panel A), DIC (in Panel B), and DSF (in Panel C). These measures are computed as a 3-year moving-average activeness measure, where activeness is computed as the deviation of a firm’s capital expenditures in each segment in a particular year from: lagged capital allocation (DLC), industry capital allocation (DIC), and segment’s free-cash-flows (DSF), as described in section 3.2.1. The value of all options granted in the current year is computed based on the modified Black–Scholes option pricing model (Black and Scholes, 1973), and options that are previously granted are valued using Core and Guay’s (2002) approximation method. CEO’s portfolio delta (Delta) is the dollar change in CEO’s portfolio composed of all stock options and stock holdings in response to a 1% change in stock price. CEO’s portfolio vega (Vega) is the dollar change in CEO’s portfolio composed of all stock options as a result of a 0.01 change in stock volatility. Cash compensation is the sum of salary and cash bonus, and is expressed in millions of dollars. Size represents firm size measured as the logarithm of total sales, and the book-to-market ratio (B/M) is defined as the total book value of common equity (Compustat code, CEQ) scaled by market capitalization, which is computed as the product of share price (PRCC_F) and number of common shares outstanding (CSHO) at fiscal year end. N(seg) is the number of segments reported by a company each year. H(CAPX) is the Herfindahl index of segment’s capital expenditures, and it is calculated as the sum of the squared proportional capital expenditures for each segment. σ(SegQ) proxies for the discrepancy in investment opportunities among segments as measured by the standard deviation of segments’ Tobin’s Qs. We first calculate Tobin’s Q of all single-segment firms where Tobin’s Q is defined as total assets minus book value of equity (CEQ) plus market value of equity (PRCC_F*CSHO) scaled by total assets. Each segment is then assigned to the beginning of year’s asset-weighted average of Tobin’s Q of single-segment firms which operate in the same Fama–French 48 industry as the segment considered. Z-score (Altman, 1968) measures the probability of bankruptcy described in equation (4) in section 3.3. Stock return is the annualized stock return during the fiscal year. Standard errors are in parentheses, and statistical significance at the 1%, 5%, and 10% is denoted by ***, **, and *, respectively.
Panel A: Impact of risk-taking incentives on active internal capital allocation (DLC).
The coefficients on the other control variables are consistent with our expectations. Specifically, firm size (Size) is negatively related to the probability of being active; value firms (high B/M) and firms with more segments (N(seg)) are more likely to be involved in active internal capital allocation across segments. A higher H(CAPX) implies a higher concentration of capital expenditures in particular segments. The coefficient on H(CAPX) is significantly negative, suggesting that diversified firms with less concentrated capital in particular segments are likely to be more active in their internal capital allocation. We also find evidence that the greater the σ(SegQ), the more likely the CEOs’ participation in active internal capital allocation. However, the effect of this dispersion in segment-based investment opportunities on the active internal capital allocation disappears after controlling for year fixed effects in columns (2), (4), and (6).
4.2. Robustness tests: endogeneity issue and industry-level effects
Coles et al. (2006) point out that failing to disentangle how compensation affects investment policy and risk from how the policy and the corresponding risk profile of a firm affect the compensation scheme of a risk-averse manager could result in a significant simultaneity bias. Our first hypothesis test (H1) is also not free from the issue of endogeneity bias if a firm intentionally increases the sensitivities of CEO’s portfolio to stock price and volatility to affect managerial decision regarding the degree of risk-taking in internal capital allocation.
In addition to using fixed effects in our regression specifications to account for both observed and unobserved variables, we decided to use the exogenous change in the Delaware takeover regime of the mid-1990s. Following Low (2009), we constructed the dummy variable DELAFTLO Vega which is equal to 1 for Delaware-incorporated firms (DEL) if the period is from 1996 to 2006 (AFT) and when CEO’s Vega is below the median value (LO Vega). Similarly, the variable DELAFTLO Delta in Table 5 is a dummy variable which is equal to one for Delaware-incorporated firms (DEL) from 1996 to 2006 (AFT) and when CEO’s Delta is below the median value (LO Delta). We then estimated whether managers with below-median equity-based compensation incentives are less likely to change their active capital allocation across segments following the shift in the Delaware takeover regime. Moreover, since firm’s activeness may vary depending on the nature of the industry a particular segment belongs to, we also include industry fixed effects in all regression models.
Exogenous shock: Change in the Delaware takeover regime (DELAFT).
The dependent variable in this table is one of our active capital allocation proxies of DLC, DIC, and DSF. These measures are computed as a 3-year moving-average activeness measure, where activeness is computed as the deviation of a firm’s capital expenditures in each segment in a particular year from: lagged capital allocation (DLC), industry capital allocation (DIC), and segment’s free-cash-flows (DSF), as described in section 3.2.1. AFTLO Vega (AFTLO Delta) takes the value of one if the year is 1996 and onward, and CEO’s portfolio vega (delta) is below the median value (LO Vega or LO Delta). The variable DELAFTLO Vega (DELAFTLO Delta) takes the value of one for Delaware-incorporated firms (DEL) when CEO’s portfolio vega (delta) is below the median value (LO Vega or LO Delta) and if the year is 1996 and onward. The value of all options granted in the current year is computed based on the modified Black–Scholes option pricing model (Black and Scholes, 1973), and options that are previously granted are valued using Core and Guay’s (2002) approximation method. CEO’s portfolio delta is computed as the dollar change in CEO’s portfolio composed of all stock options and stock holdings in response to a 1% change in stock price. CEO’s portfolio vega is computed as the dollar change in CEO’s portfolio composed of all stock options as a result of a 0.01 change in stock volatility. Cash compensation is the sum of salary and cash bonus, and is expressed in millions of dollars. Percentage industry-adjusted accounting profitability, Industry-adj. ROA, is computed as a diversified firm’s ROA minus ROA of its mimicking firm, where ROA is defined as net income divided by total asset. To compute the ROA of the mimicking firm, we first assign, to each segment, a median ROA of all single-segment firms that operate in the same Fama–French 48 industries as the segment considered. The mimicking firm’s ROA is then calculated as each segment’s asset-weighted average of assigned ROA of each segment. All other control variables for firm characteristics were previously defined in Table 4. Standard errors are in parentheses, and statistical significance at the 1%, 5%, and 10% is denoted by ***, **, and *, respectively.
The findings of these tests are reported in Table 5. The negative loading of the dependent variable DLC on the dummy variable DELAFTLO Vega in column (1) indicates that capital allocation becomes significantly more passive (−0.0322) among Delaware-incorporated firms after 1996 when the CEO is granted a below-median portfolio vega. This result is consistent with Low (2009) who find that following the shift in the Delaware takeover regime, Delaware-incorporated firms decreased significantly total risk, systematic risk, and idiosyncratic risk if their CEO’s portfolio vega was smaller than the median CEO vega value. Furthermore, consistent with the conclusion of Low (2009) and our previous findings in Table 4, CEO’s portfolio Delta does not influence active capital allocation decisions as indicated by the negative but (economically and statistically) insignificant loading of DLC in column (2) on the variable DELAFTLO Delta. Our results remain qualitatively unchanged when we consider the other two active allocation proxies of DIC in columns (3) to (4), and DSF in columns (5) to (6).
4.3. Risk-taking and active internal capital allocation
In this section, we test the hypothesis that being more active translates into greater risk-taking (i.e. H2). For this purpose, we use the idiosyncratic component of firm equity risk as previous studies suggest that managers would only need to deal with the part of risk that cannot be eliminated by diversification. For instance, Low (2009) finds that managerial risk-reduction is mostly implemented through a lower exposure to idiosyncratic volatility.
In all models, we utilize pooled OLS regressions with year fixed effects. The dependent variable is the logarithm of various proxies for idiosyncratic volatility measured as the annualized standard deviation of the 36-month residuals (as of the end of June in each year) from different factor models: CAPM (IDVOLCAPM), Fama–French (1993) three-factor model (IDVOLFF3), and Fama–French–Carhart (1997) four-factor model (IDVOLFF4). The main independent variable of interest is one of our proxies of active capital allocation decisions, namely, DLC, DIC, or DSF. Other lagged control variables include firm characteristics described previously in section 4.1. The results of the following regression model are reported in Table 6
Risk-taking via active internal capital allocation.
This table reports the results of pooled OLS regressions with time-fixed effects. The dependent variable is the logarithm of various proxies of idiosyncratic volatility, calculated as an annualized standard deviation of the residuals from the following factor models estimated over the previous 36 months: CAPM (IDVOLCAPM), Fama–French three-factor model (IDVOLFF3), and the Fama–French–Carhart four-factor model (IDVOLFF4). The main independent variable of interest is the logarithm of one of our three active capital allocation proxies of DLC, DIC, and DSF. DELAFT is a dummy variable which is equal to 1 for Delaware-incorporated firms for the years 1996 to 2006, and 0 otherwise (see Low (2009)). The variable R&D is the level of research and development expenditures of the firm and is computed as COMPUSTAT Data 46 (Research and Development Expense) divided by COMPUSTAT item 6 (Total Assets), with missing values being coded zero. All other control variables for firm characteristics were previously defined in Table 4. Standard errors are in parentheses and statistical significance at the 1%, 5%, and 10% is denoted by ***, **, and *, respectively.
Overall, the findings of Table 6 indicate the existence of a significantly positive relationship between the lagged measure of activeness and idiosyncratic volatilities proxies, which confirms H2. For instance, based on the result in column (1) of Table 6, a one standard deviation increase in the level of activeness (DLC) leads to an increase in firm risk (IDVOLCAPM) by about 0.58%. Furthermore, the signs of the coefficients across all models are consistent with our expectations. Specifically, the size of a firm (Size) is negatively correlated with next-period idiosyncratic risk; less diversified firms with higher H(CAPX) seem to be associated with higher idiosyncratic risk. We also control for a firm’s state of incorporation using the dummy variable DELAFT which is equal to 1 for Delaware-incorporated firms for the years 1996 to 2006, and 0 otherwise. Consistent with Low (2009), Delaware-incorporated firms are less likely to take idiosyncratic risk. Similarly, firms with above-average research and development expenditures (R&D), or greater long-term and short-term debt (Leverage), are also likely to take more risk.
Although there is no formal model of the relationship between active capital allocation decisions across segments and firm risk, the evidence of a positive relationship between CEO’s risk-taking tendency and internal capital markets is not new in the literature. For instance, Yong (2005) finds that there is a positive correlation between new stock option grants and CEO’s risk-taking in internal capital markets. In particular, firms granting executive stock options tend to invest more (less) in segments with high (low) risk, and these high (low) risk segments tend to be added (eliminated) when restructuring.
If firms with more convex equity-based compensation allocate actively more capital to segments characterized by greater risk, we should then observe a positive association between capital allocation decisions and segment-level risk, on average. For this purpose, we estimate two proxies of segment-level risk. Our first proxy is the segment-level financial risk, SEGFINRISK, computed as the standard deviation of the asset-weighted monthly average stock returns for all single-segment firms in the industry for the 24 months prior to the firm’s current fiscal year. Our second proxy is the segment-level operating risk, SEGOPERISK, computed as the standard deviation of the asset-weighted average return on assets (ROA) for all single-segment firms in the industry for the 3 years prior to the firm’s current fiscal year. Following Yong (2005), we first compute the segment-level size-normalized investment, SEGCAPX, in segment i by firm j in year t as follows
To capture active segment-level size-normalized capital allocation, we then calculate the sum of the absolute changes in the size-normalized investment in segment i by firm j from year t−1 to year t, ∆
The relationship between segment-level capital allocation and segment-level risk.
This table shows the coefficients from regressions of the size-adjusted ratio of segment capital allocation against a number of CEO, segment, and firm characteristics. The dependent variable, ΔSEGCAPX, is the sum of the absolute changes in the size-normalized investment in segment i by firm j from year t-1 to year t.
4.4. Implication of risk-taking via active internal capital allocation on shareholder wealth
An important question at this stage is whether this risk-taking via active internal capital allocation ultimately benefits firm’s shareholders. In this section, we consider whether “being active” is detrimental to shareholders’ wealth (i.e. H3). For this purpose, we estimate the following model using pooled OLS regression with year fixed effect
The dependent variable is the industry-adjusted stock return, where Rf, τ + 1 is the annual stock return of stock f, wi is the weight assigned to segment i based on its asset, Ri, τ + 1 is the value-weighted average annual return of single-segment firms in segment i’s industry based on the Fama–French 48 industry classification. Our main independent variable is, Active, as proxied by one of our three active capital allocation measures of DLC, DIC, and DSF. Lagged control variables include (refer to section 4.1 for a description of these variables): N(seg); H(CAPX); Size; B/M; Leverage; Z-score; and σ(SegQ). We also control for the growth rates in assets (AT growth); the growth rate in capital expenditures (CAPX growth), the previous 12-month stock return momentum (Momentum); and industry-adjusted accounting profitability (Industry-adj ROA). The results from pooled OLS regression are reported in Table 8.
Active internal capital allocation and industry-adjusted stock returns.
This table reports the results of pooled OLS regressions with time-fixed effects. The dependent variable is the industry-adjusted stock returns as defined in equation (6). The main independent variable of interest, Active, is represented by one of our active capital allocation proxies of DLC (in Panel A), DIC (in Panel B), and DSF (in Panel C). Percentage industry-adjusted accounting profitability ratio, Industry-adj. ROA, is computed as a diversified firm’s ROA minus ROA of its mimicking firm, where ROA is defined as net income divided by total assets. To compute the ROA of the mimicking firm, we first assign, to each segment, a median ROA of all single-segment firms that operate in the same Fama–French 48 industries as the segment considered. The mimicking firm’s ROA is then calculated as the asset-weighted average of ROA across all firm’s segments. Momentum is computed on the previous 12-month stock returns. All other firm control variables were previously defined in Table 4. Standard errors are in parentheses and statistical significance at the 1%, 5%, and 10% is denoted by ***, **, and *, respectively.
Panel A of Table 8 illustrates the findings of the regression based on DLC. For robustness, in Panel B (Panel C) of Table 8, we document the results obtained when using DIC (DSF) as our main independent variable. The signs on the control variables are consistent with previous literature. Importantly, in all regression specifications active internal capital allocation seems to benefit shareholders as indicated by the positive and significant loadings of industry-adjusted stock returns on the lagged logarithm of DLC, DIC, and DSF. These findings suggest that activeness driven by risk-taking incentives increases shareholder value hence confirming the validity of H3. Thus, equity-based compensation constitutes an effective route to promote interest alignment.
These results of Table 8 conflict with the evidence of Guedj et al. (2009) who show that the degree of activeness in internal capital allocation of diversified firms appears to be negatively correlated with future stock returns. 9 They argue that such inefficiency may derive from managerial incompetence at identifying promising investment opportunities and internal power struggles among divisional managers which further intensify the agency conflicts within the firm. However, the negative relationship documented in their paper casts serious doubts on the plausibility of these explanations as the authors fail to justify why shareholders of firms with more active internal capital markets—which are also characterized by higher default risk and higher idiosyncratic volatility—should require lower expected returns. Their result also implies major failures in internal and external corporate governance mechanisms in disciplining managerial behavior. Indeed, if active capital allocation destroys firm value, CEOs would not dare make bold active decisions that are value-reducing in the first place. Second, once CEOs realize that active capital allocation leads to poorer performance, they would immediately become more “jaded” by passively allocating capital in the future. The result is that there would not be any active capital allocation within firms in the long-run.
5. Additional robustness tests
5.1. Alternative activeness measure
In this section, we aim to show that our results still hold even after taking into account the segments’ size when measuring the degree of managers’ activeness. As such, we introduce an alternative measure of activeness, namely, DIR (i.e. Deviation of Investment Ratio) which captures the level of capital allocation activeness controlling for segments’ assets. The mathematical expression of dir is given as follows, and DIR denotes its 3-year moving average
Assets (AT) and weights (w) of segments are measured at the beginning of the year, while capital expenditures (CAPX) are measured at the end of the year.
Using an alternative activeness measure, DIR (Deviation of Investment Ratios).
This table reports the results obtained for the new activeness measure, DIR, which is intended to capture the level of capital allocation activeness controlling for segments’ assets. Panel A reports statistics of DIR. Panel B shows average activeness measures (DLC, DIC, and DSF) of five portfolios sorted on DIR. Panel C (next page) reports the results of Honoré’s (1992) fixed-effects censored regressions which test the impact of risk-taking incentives on active internal capital allocation. The dependent variable in Panel C is the activeness measure, DIR. Value of all options granted in the current year is computed based on the modified Black–Scholes (Black and Scholes, 1973) option pricing model, and options that are previously granted are valued using Core and Guay’s (2002) approximation method. CEO’s portfolio delta (Delta) is the dollar change in CEO’s portfolio composed of all stock options and stock holdings in response to a 1% change in stock price. CEO’s portfolio vega (Vega) is the dollar change in CEO’s portfolio composed of all stock options as a result of a 0.01 change in stock volatility. Cash compensation is the sum of salary and cash bonus and is expressed in millions of dollars. Size represents firm size measured as the logarithm of total sales, and the book-to-market ratio (B/M) is defined as the total book value of common equity (Compustat code, CEQ) scaled by market capitalization (computed as the product of share price (PRCC_F) and number of common shares outstanding (CSHO) at the fiscal year end). N(seg) is the number of segments reported by the company each year. H(CAPX) is the Herfindahl index of segment’s capital expenditures, and it is calculated as the sum of the squared proportional capital expenditures for each segment. σ(SegQ) proxies for the discrepancy in investment opportunities among segments as measured by a standard deviation of segments’ Tobin’s Qs. We first calculate Tobin’s Q of all single-segment firms. Tobin’s Q is defined as total assets minus book value of equity (CEQ) plus market value of equity (PRCC_F*CSHO) scaled by total assets. Each segment is then assigned to the beginning of year’s asset-weighted average of Tobin’s Q of single-segment firms operating in the same Fama–French 48 industry as the segment considered. Z-score (Altman, 1968) measures the probability of bankruptcy described in equation (4) of section 3.3. Stock return is the annualized stock return during a fiscal year. Standard errors are in parentheses and statistical significance at the 1%, 5%, and 10% is denoted by ***, ** and *, respectively. Panel D reports the estimated coefficients of the regressions of firm risk on active internal capital allocation using pooled OLS regressions with time-fixed effects. The dependent variable is the logarithm of various proxies of idiosyncratic volatility, calculated as the annualized standard deviation of the residuals from the following factor models estimated over the previous 36 months: CAPM (IDVOLCAPM), Fama–French three-factor model (IDVOLFF3), and the Fama–French–Carhart four-factor model (IDVOLFF4). The main independent variable of interest is the logarithm of the activeness proxy DIR. Size represents firm size measured as the logarithm of total sales. H(CAPX) is the Herfindahl index of segment’s capital expenditures, and it is calculated as the sum of the squared proportional capital expenditures for each segment. Leverage is the sum of long-term and short-term debt (Compustat code, DLTT, and DLC, respectively) divided by total asset (AT). Statistical significance at the 1%, 5%, and 10% are denoted by ***, **, and *, respectively. Panel E reports the results of pooled OLS regressions with time-fixed effects. The dependent variable is the industry-adjusted stock returns. The main independent variable is the activeness proxy computed as the logarithm of DIR. Size is the logarithm of sales. Industry-adjusted accounting profitability is computed as a diversified firm’s ROA minus ROA of its mimicking firm, where ROA is defined as net income divided by total asset. To compute the ROA of the mimicking firm, we first assign, to each segment, a median ROA of all single-segment firms that operate in the same Fama–French 48 industry as the segment considered. The mimicking firm’s ROA is then calculated as each segment’s asset-weighted average of assigned ROA of each segment. Momentum is computed using the previous 12-month stock returns. Please refer to Table 4 for a description of all the other control variables. Statistical significance at the 1%, 5%, and 10% are denoted by ***, **, and *, respectively.
Panel A: Descriptive statistics of the alternative activeness measure DIR.
Panel A of Table 9 summarizes the descriptive statistics of the investment ratios of all segments in firms with more than one segment for at least 1 year over the period 1991 to 2006, where the investment ratio is defined as the segment capital expenditures scaled by the segment assets. As shown in Panel A of Table 9, the average DIR, the asset-weighted average change in investment ratio of a segment over two consecutive years, is 0.03 with the maximum value of 0.36 (untabulated), and is positively skewed. We test whether DIR is a qualitatively similar measure to the activeness measures used so far in our analysis (DLC, DIC, and DSF). For this purpose, we construct quintile portfolios sorted on our new activeness measure, DIR. As shown in Panel B of Table 9, the activeness measures, DLC, DIC, and DSF monotonically increase as DIR increases. Since firm’s investment ratio is defined as the proportion of segment capital expenditures relative to the segment assets, and it is usually the case that segment capital expenditures are significantly smaller than segment asset (all of segment investment ratios are less than 0.4), it is not surprising that the values of DIR are quite small.
In Panel C of Table 9, we confirm that a positive relationship exists between CEO’s option vega and the alternative activeness proxy of DIR. Also, managerial decision to be more active in segment capital allocation is more likely to lead to an increase in the level of firm risk in the subsequent period (Panel D). For example, a one standard deviation increase in the level of activeness as measured by DIR is associated with 0.23% increase in firm risk calculated using the CAPM. Furthermore, consistent with our previous findings, activeness continues to be positively related to performance in the next period (refer to Panel E of Table 9). In the whole, the results of Table 9 demonstrate that our conclusions are robust to alternative proxies for active capital allocation.
5.2. Controlling for corporate restructuring
The activeness measures used in this study are based on capital expenditures of segments which are common over two consecutive years but do not control for changes in capital expenditures of added or eliminated segments. Since firms experiencing segment restructuring could have different capital budgeting styles, it is important to control for firm restructuring when testing the relationship between CEO’s equity incentives and CEO’s risk-taking through active capital allocation (H1). We construct an indicator variable, which takes the value of one in year t, if more than one segment is added or dropped in year t relative to year t−1. As shown in Panel A of Table 10, approximately 9% of total firm-year observations experienced segment restructuring at some point during the sample period from 1992 to 2006. The difference in the degree of activeness between the restructuring sample and the non-restructuring sample is statistically significant, highlighting the importance of controlling for segment restructuring. The evidence of Panel B of Table 10 shows that our findings remain unchanged after the Restructuring dummy variable is included in our models. In all models, the coefficients on CEO’s option Vega remain significantly positive. In an unreported test we re-estimated the results of Panel B by excluding the sub-sample of the restructuring firms, and reach qualitatively similar conclusions.
Controlling for corporate restructuring.
Panel A reports the percentage of firm-year observations and the average activeness measures of the restructuring and non-restructuring samples. Panel B illustrates the estimated coefficients of Honoré’s (1992) fixed-effects censored regressions, which test the impact of risk-taking incentives on active internal capital allocation. A detailed description of the dependent and independent variables is reported in Table 4. Restructuring is an indicator variable which takes the value of one, if more than or equal to one segment are added or eliminated by firm j in year t.
Panel A: Activeness measures for restructuring and non-restructuring firms.
6. Conclusion
In this study, we investigate for the first time the impact of managerial risk-taking incentives provided by equity-based compensation on capital budgeting decisions made by CEOs of diversified firms. We provide convincing evidence that CEOs with a more convex equity-based compensation do increase firm-specific risk, which is consistent with prior studies (Cohen et al., 2000; Coles et al., 2006; Guay, 1999; Low, 2009; Rajgopal and Shevlin, 2002; Smith and Stulz, 1985). Second, and more importantly, we show that CEOs engage in risk-taking by actively allocating capital across various lines of business within a firm. The contribution of this study is two-fold. First, it complements prior studies by showing that the convex equity-based compensation helps reduce managerial risk-aversion. Second, to the best of our knowledge, this study is the first to investigate ways in which managerial risk-taking could be implemented in response to the particular compensation structure granted by diversified companies.
Footnotes
Acknowledgements
The authors are grateful for valuable comments from Terry Walter, Susan Thorp, and Balasingham Balachandran. All errors are their own.
Final transcript accepted 30 April 2016 by Karen Benson (AE Finance).
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
1.
2.
The Securities and Exchange Commission (SEC) changed the reporting requirements for executive compensation for fiscal years ending after 15 December 2006. Specifically, the new disclosure rules redefine several compensation components. For instance, some bonuses have been reclassified as non-equity incentive compensation. As this change introduces significant noise in the definition of our compensation variables, we decided to end our sample period in the financial year 2006.
4.
As a comparison, 
who study executive compensation and internal capital market efficiency have a sample size of 1311 diversified firm-years over the period 1992–2003 after imposing few additional filtering rules. The number of observations, 3612 is a total number of firm-year observations of all firms for which we were able to calculate both an activeness measure, DLC in particular, and CEO’s portfolio vega. However, due to unavailability of other firm characteristic data, the total number of observations used in regressions analyses is different.
5.
6.
The authors thank Core and Guay for sharing their codes.
7.
8.
Statistical interpretation of the coefficient estimates from censored regression models requires caution as these show the relationship between independent variables and the “latent dependent variable,” which is transformed to take into account censoring. Consequently, the partial effect of x on y conditional on censoring not just depends on β, and this begs a separate computation of marginal effects. However, Honoré’s model has a limitation in that it removes the unobserved fixed effects which determine the marginal effects, and computes the coefficients using only time variations in the independent variables.
9.
Our sample comprising 3612 firm-year observations (or 551 diversified firms) is significantly smaller than that of as we merge the CRSP/COMPUSTAT database with Execucomp. Although the authors do not report the exact number of observations used to estimate the relationship between performance and activeness, their statistics identify an initial sample of 23,620 firm-year observations.