Risk-taking in conglomerates : a comparative analysis of risk-taking by divisional managers and CEOs of stand-alone firms

(1)

Risk-Taking in Conglomerates: A Comparative Analysis of Risk-Taking

by Divisional Managers and CEOs of Stand-Alone Firms.

Master Thesis (MSc Business Economics, Finance track)

University of Amsterdam

Name: Milco Breed

Date: July 2016

(2)

2

Statement of Originality

This documents is written by Student Milco Breed who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

This research compares the taking behavior of divisional managers relative to the risk-taking behavior of CEOs of matched stand-alone firms. Divisional managers can have the incentive to take more risk, because their holding/parent company can subsidize losses in separate divisions. Two hypotheses test if divisional managers take more risk relative to CEOs of matched stand-alone firms. The results for the first hypothesis provide no evidence of a coinsurance effect. Taken as a whole, divisions and conglomerates take more risk than matched stand-alone firms. The second hypothesis that tests for managerial moral hazard finds no evidence that misbehavior by divisional managers is causing this additional risk.

(4)

4

Introduction

A conglomerate is a corporation that is made up of a number of different, often unrelated businesses. In a conglomerate, the holding or parent company owns a controlling stake in a number of smaller companies. Each of the conglomerate subsidiary businesses, called division or segment, typically runs independently of the other business divisions. This thesis is focusing on the relationship of the subsidiaries and the risk-taking behavior of the managers of the subsidiaries. The goal is to find out if if a relation exists between conglomerate firms and risk-taking behavior of divisional managers.

The existing literature mainly focuses on efficiency of internal capital markets (Williamson, 1975; Gertner et al., 1994; Stein, 1997), inefficient internal capital markets (Shin & Stulz, 1998; Rajan et al., 2000; Scharfstein & Stein, 2000) and managerial behavior (Denis et al., 1997; Glaser et al, 2013). Conglomerate firms contain several divisions and they operate in diverse industries. An important advantage for the subsidiaries of conglomerates is that their holding/parent firm is more diversified and so they can subsidize any losses in separate divisions. On the other hand, conglomerate firms introduce additional agency problems between firms. These agency problems can destroy value associated with the conglomerate structure (Laeven & Levine, 2007).

While most research is focusing on the value decreasing effect of conglomerates, relatively little research is done about specific agency problems of conglomerate firms. One possible agency problem is the risk-taking behavior of divisional managers in conglomerate firms. Divisional managers in conglomerates could have the incentive to take relatively more risk compared to their peers in stand-alone operating firms since their holding/parent firm would most likely subsidize any losses. This incentive of the divisional managers can lead to inefficiency and potential losses for the conglomerate firms.

The main question of this thesis is: “Do divisional managers in conglomerates take more risk than CEOs in matched stand-alone firms?” To answer this question, first will be examined to the relationship between risk of divisions and risk of matched stand-alone firms. With evidence that divisions take more risk than matched stand-alone peers, a possible reason can be investigated. One explanation is that managers of divisions behave self-interested. This increases the risk of divisions. An alternative explanation for the increase of risk-taking of divisions is that conglomerates attract riskier divisions and they want the divisions to take more risk. In this case a division characteristic is causing the

(6)

6

additional risk of divisions. In this thesis the focus is on the first explanation that divisional managers behave self-interested.

This research contributes to a better understanding of how conglomerates operate. Researchers argue whether conglomerates have a diversification premium (Villalonga, 2004; Hund Monk and Tice, 2012) or a diversification discount (Leaven & Levine, 2007; Maksimovic & Philips, 2013). Which benefits of conglomerates lead to a diversification premium? And on the other hand: which agency problems are causing the conglomerates to be inefficient? A difference of risk-taking behavior between conglomerates and stand-alone firms can be an agency problem. This study can provide additional evidence on a specific agency problem of conglomerates. Moreover, this study will contribute to the bigger question: “Are internal capital markets maintained by conglomerate firms an efficient organizational form that adds value? Or are conglomerate firms inefficient organizations that are driven by agency conflicts between separate divisions?”

To estimate the risk-taking behavior of divisions and stand-alone firms a proxy is used. The proxy for risk-taking behavior is the cash flow volatility. In a first step the cash flow volatility of divisions is analyzed to the cash flow volatility of matched stand-alone firms. The matched stand-alone firms are selected on the same size and industry as the division. The data includes firm characteristics of divisions, firm characteristics of stand-alone firms, personal characteristics of divisional managers and personal characteristics of CEOs of stand-alone firms. From the data of firm characteristics the cash flow volatility is constructed. With the constructed cash flow volatility, two hypotheses are tested.

The first hypothesis is based on the coinsurance effect (Lewellen, 1971). To test if the coinsurance effect exists, the risk-taking behavior of divisions is compared to the risk-taking behavior of matched stand-alone firms. This test is based on two different levels. The first test compares the risk-taking behavior of divisions relative to their matched stand-alone firms. The second test compares the risk-taking behavior of conglomerates relative to their matched stand-alone firms. If the proxy for risk-taking behavior of divisions and conglomerates is significantly higher than the proxy for risk-taking behavior of matched stand-alone firms, then this is taken as evidence that the coinsurance effect does not exist. The second hypothesis tests whether divisional managers behave self-interested. The models include personal characteristics of divisional managers and CEOs of stand-alone

(7)

7

firms. The two hypotheses together can answer whether divisional managers of conglomerates take more risk than CEOs of matched stand-alone firms.

In section 1 the literature review is discussed. This literature describes the efficiency of internal capital markets. Furthermore, the literature discusses whether conglomerates have a diversification premium or a diversification discount. What agency problems are causing the diversification premium of conglomerates? In addition, section 1 examines the characteristics of divisions and the personal characteristics that could influence the behavior towards risk of managers. At the end of this section the hypotheses are constructed and discussed.

Section 2 describes the methodology to test the hypotheses. The difference between a division and a stand-alone firm is based on the person who is in charge. The divisions are managed by divisional managers, while the stand-alone firms are managed by a CEO. In addition, the construction of the cash flow volatility, the proxy for risk-taking behavior, is described. Beside the construction of the cash flow volatility, the matching procedure and the models to test the two hypotheses are reviewed. The first model estimates whether risk-taking of divisions is higher relative to the risk-risk-taking of matched stand-alone firms. The second model estimates whether personal characteristics influence the risk-taking behavior of divisional managers.

Section 3 contains the data collection and the descriptive statistics. The data collection is reviewed. Which variables are included and from which data sources the data is collected? Furthermore, the basic statistics of the data are presented for the whole sample. In addition, the data of subsamples are presented. The subsamples are: stand-alone firms, divisions and conglomerates with at least two divisions.

Section 4 provides the main results of the estimated models, which are described in the methodology. The first two models provide evidence whether the coinsurance effect exists or not. The first model compares risk-taking of divisions relative to matched stand-alone peers. The second model compares risk-taking of conglomerates relative to matched stand-alone peers. From these two models there can be concluded whether the coinsurance effect exists. The third model estimates whether managers behave self-interested. At the end of section 4, the most important findings are summarized.

Section 5 presents additional results and robustness checks. In this part, an alternative explanation (gamble for resurrection) is proposed and tested. Moreover, a

(8)

8

difference matching procedure is used as robustness check. The results of both tests are included in this section.

In section 6 the conclusion and discussion are presented. An answer is given to the question: “Do divisional managers in conglomerates take more risk than CEOs of matched stand-alone firms?” The answer of this question is based on the outcomes of the two hypotheses. Finally, some limitations and suggestions for future research of the study are given.

Section 1. Literature Review

A) Conglomerates and their internal capital structure.

Conglomerates are diversified firms investing in different industries. Their internal capital market is more diversified compared to their stand-alone peers, which could result in better investment opportunities and better firm performance. Williamson (1975) suggests that the internal capital market of diversified firms might allocate capital more efficiently than the external capital market. The top management of a diversified firm is better informed about investment opportunities than external investors, which could increase the efficiency of allocating capital. Conglomerates are diversified firms. Combining this with the theory of Williamson, conglomerates can choose to best investment opportunity to allocate their capital.

To provide evidence in line with the theory of Williamson (1975), Gertner et al. (1994) and Stein (1997) presented models identifying circumstances where internal capital markets lead to more efficient investment decisions. Gertner et al. (1994) analyzed the costs and benefits of internal versus external capital allocations. They argue that internal capital allocations lead to more monitoring compared to bank lending and it makes it easier to efficiently reallocate the assets of projects that are performing poorly. However, a downside is that internal capital markets reduce managers’ entrepreneurial incentives. This effect is not covered in their paper. Basically, they ignored the potential for information or agency problems at the level of the provider of capital.

In addition, Stein (1997) looked at the role of the headquarters in allocating scarce resources to competing projects in an internal capital market. Headquarters are able to shift funds from one project to another. Stein concluded that internal capital markets can be

(9)

9

efficient, but only when there is a high span of control. If corporate managers involve in self-interested behavior, this behavior will not result in the maximized value of the shareholders. Both papers provide evidence that the internal capital markets of conglomerates can add value to the firm. However, information and agency problems can still decrease firm value.

B) Inefficient internal capital markets.

The inefficiency in internal capital markets has been researched as well. Most empirical research about conglomerates is focusing on the disadvantages of conglomerate firms and the inefficiency of internal capital markets. Shin and Stulz (1998) studied internal capital markets of diversified firms and they found evidence that investments in conglomerates are inefficient. Diversified firms enable themselves to fund profitable projects with their internal capital markets, while the external capital markets would not be able to finance them. But the investments by segments of highly diversified firms are less sensitive to their cash flows than the investments of comparable single-segment firms. The investment by a segment of a diversified firm depends significantly less on its own cash flow compared to the cash flows of the firms from other segments. This suggests inefficient internal capital markets in diversified firms.

Furthermore, Rajan et al. (2000) examined the effect of transferring funds from divisions with poor investment opportunities to divisions with good investment opportunities. They expect that the allocation of resources and funds between divisions of a diversified firms will be transferred form the divisions with the poor investment opportunities to the division with the good investment opportunities. However, they found that when diversity in resources and opportunities increases, more cash flows will flow towards the inefficient division. This leads to inefficient investments and suggests that diversity is costly. They argue that internal capital markets are prone to distortions in capital allocation, which results in inefficient investment.

In complement to the findings of Rajan et al. (2000), Scharfstein and Stein (2000) created a model in which they implied a kind of ‘socialism’ in internal capital allocation, whereby weaker divisions get subsidized by stronger ones. Scharfstein and Stein also found distortions in capital allocation resulting in inefficient investment. These distortions in capital allocation are caused by agents in the firm, who allocate resources to another agent to get something in return. In their model the CEO is the only one with an authority to allocate

(10)

10

resources, but in reality more agents in an organization have authority to allocate resources. The results in the paper suggest that the power to make resource-allocation decisions affects efficiency.

The papers of Shin and Stulz (1998), Rajan et al. (2000) and Scharfstein and Stein (2000) found empirical evidence for inefficiency in internal capital markets. Diversified firms do not tend to behave like profit maximizing firms. The allocations of capital and resources do not flow singly to the segments with the best investment opportunities. This failure of the internal capital market results in inefficiency in conglomerates.

C) Diversification Discount or Premium.

If internal capital markets of conglomerates are inefficient, this could lower the value of the conglomerates relative stand-alone firms. Maksimovic and Philips (2013) argue the hypothesis that conglomerates destroy value on average when compared to similar stand-alone firms. Based on existing papers, they found several results consistent with their hypothesis. This suggests that conglomerates destroy value, compared to their stand-alone peers. Furthermore, Laeven and Levine (2007) found results in line with the value-decreasing theory for specific financial conglomerates. The market value of financial conglomerates who engage in multiple activities, are worth less than if those financial conglomerates were broken into financial intermediaries that are specialized in the individual activities. These results are consistent with the theories that agency problems in conglomerates reduce the market value of conglomerates.

On the other hand, benefits of diversification can lead to a diversification premium. Hann, Ogneva and Ozbas (2009) show that diversified firms have a lower cost of capital than portfolios of comparable stand-alone firms and that the reduction is strongly correlated with business units cash flows. This is consistent with a coinsurance effect. The coinsurance effect gives multi-segment firms greater debt capacity than single-line businesses of similar size (Lewellen, 1971). Diversified firms have a lower cost of capital relative to single-segment firms. The benefits of the coinsurance effect can lead to a diversification premium. Villalonga (2004) provide empirical evidence that diversified firms in the United States trade at a large and statistically significant premium relative to specialized firms in the same industry. More diversified firms can allocate their capital to the divisions with the best investment opportunities and engaging in different business should provide a higher return. Villalonga

(11)

11

(2004) uses the Business Information Tracking Series (BITS) database, instead of COMPUSTAT segment database. An important difference of the research of Villalonga with previous research was the definition of a business segment. Also Hund, Munk and Tice (2012) show that conglomerate firms trade at a premium relative to single segment firms on a value-weighted basis. The conglomerate discount is a failure of the matching procedure on size, age and profitability.

Moreover, Matvos and Seru (2012) found evidence that during the recent financial crisis internal capital markets gained in value. The recent financial crisis provides a natural experiment to test how internal capital markets works during that period. Diversified firms with internal capital markets gained in value, relative to firms that did not have internal capital markets.

D) Agency problems.

The reasons for this inefficiency are mainly focused on agency problems. Denis et al. (1997) present evidence on the agency cost explanation for corporate diversification. They found that the level of diversification is negatively related to performance. The existence of agency problems are likely to be responsible for firms value-reducing diversification strategies according Denis et al. (1997). However, they did not found enough evidence that agency problems led managers to engage in those value-destroying strategies. Glaser et al. (2013) analyzed the internal capital markets of a multinational conglomerate. They found that more powerful managers obtain larger allocations and increase investment substantially more than their less connected peers. This causes misallocation of capital. Despite these misallocations, the more powerful managers overinvest in their division. Those overinvestments leads to lower performance. This empirical evidence of Glaser et al. (2013) contributes to the bargaining-power theories, which predicts that capital allocations are partly based on power and connections. Moreover, it contributes to the empire-building theory, which predicts that larger capital allocations does not improve performance.

Furthermore, divisional managers can attract personal benefits. Stein (1997) argues that managers can take private benefits from running particular projects they control. In addition, divisional managers are less entrenched than managers of stand-alone peers and can easier continue bad projects from which they draw private benefits (Miksomovic and Philips, 2013).

(12)

12

To better understand how failures of internal capital markets of conglomerates lead to inefficiency, Ozbas and Scharfstein (2010) analyze the relationship between organizational form and efficiency. The relationship is analyzed by comparing the investment behavior of businesses that function as a part of a conglomerate. Their results present evidence of inefficiencies in internal capital markets. Investments of stand-alone firms are more sensitive to industry Q than the investment of unrelated segments of conglomerate firms. Ozbas and Scharfstein (2010) also discuss scenarios whereby managers have small ownership stakes. They argue that inefficiency of investment behavior of conglomerate firms is caused due agency problems at the top of conglomerates.

An example of misbehavior of managers of diversified firms is given by Hoskisson et al. (1991). Lower level divisional managers tend to negotiate with lower performance targets when they have a greater compensation and when they have more power. Control loss of conglomerates results in risk averse behavior at the decision level of the divisional manager. This declines the financial performance.

Less research is done to the role of the top level management in conglomerates. The role of the headquarters is investigated by Hoang and Ruckes (2014). Hoang and Ruckes found empirical findings that multidivisional firms bias their investment levels in favor of divisions with investments prospects. The headquarters of a multidivisional company has private information about capital productivities of its divisions. They can use this private information to efficiently allocate their capital to the divisions. This means that the current capital allocations serve as a signal to divisional managers about future allocations. This signal can change the attitude toward risk-taking.

E) Division characteristics.

Divisional managers of conglomerates can take more risk than CEOs in matched stand-alone firms, because their holding/parent firm can subsidize losses in separate divisions. In some cases the holding/parent company want their divisions to take more risk. Conglomerates can respond more aggressively when they face competitive threats in products markets. They can use their internal capital markets to maintain the market share when faced with competitors (Faure-Grimaud and Inderst, 2005). On the other side, Swanburg (2014) examined how diversified firms respond to heightened competition following tariff reductions in order to assess the competing theories. Those firms reduce

(13)

13

their investment in the industries with the tariff reductions relative to non-diversified firms. Segments in conglomerates that face heightened competition in one industry are not subsidized by segments that corporate in other industries. This is in contrast to the findings of Faure-Grimaud and Inderst (2005).

Moreover, according to Rajan et al. (2000) divisional managers have the incentive to choose “projects that produce less value, but the revenues can be better defended against redistribution to other divisions” over value maximizing projects when there is a high diversity of investment opportunities. The divisional managers expect that their high division’ profits are more likely to be lost to others. In the model of Rajan et al. (2000) capital is transferred from high-value divisions to low value divisions.

A reason why managers want to engage in conglomeration is that they can easier avoid bankruptcy (Grass, 2012). Managers fear firm failure, because than they will lose their job, economic benefits and status. Another explanation is the spillover effect inside conglomerates (Duchin et al., 2015). A change in industry surplus in one division generates spillovers on managerial payoffs in other divisions of the same firm. This effect tends to be stronger when conglomerates have excess cash and when managers have more influence, but it tends to be weaker in the presence of strong governance. This causes conglomerates to perform worse.

F) Personal characteristics influence risk-taking behavior.

Some difference in risk-taking behavior of managers can be explained based on existing papers about personal characteristics. Byrnes, Miller and Schafer (1999) did research to gender differences in risk-taking. They did tests with male and female participants and found that the male participants took more risk in 14 out of the 16 tests. This is in line with the theory that male take more risk relative to female.

Beside the gender of managers, Pahl and Vroom (1971) found that there was a relationship between age and risk-taking. Pahl and Vroom show a significant negative relationship between age and both risk taking and the value placed upon risk. This means that younger managers tend to take more risk than older managers.

Berger et al. (1997) argue that CEOs with higher cash compensation and longer tenure are more likely to avoid risk. Moreover, they argue that tenure of a CEO is positively correlated with managerial quality or skill. On the other side Guay (1999) argues that CEOs

(14)

14

with higher total cash compensation are better diversified. They have more money invested outside the firm and therefore they are less risk-averse.

Overall, gender, age, tenure and compensation do affect the risk-taking behavior of managers. Expected is that divisional managers behave similar to the managers. Age does has a negative relationship with risk-taking, male managers take more risk relative to female managers, and managers with longer tenure and more cash compensation are more risk-averse.

G) Hypotheses.

The existing literature describes the pros and cons of internal capital markets in conglomerates. The first studies to conglomerates mainly focus on efficiency in internal capital markets. Williamson (1975), Gertner et al. (1994) and Stein (1997) found a positive relation between diversity and firm performance. Their researches were based under strict conditions. Certain information and agency problems were not taken into account.

More recent studies detect failures in the capital allocations, which leads to inefficiency. Shin and Stulz (1998), Rajan et al. (2000) and Scharfstein and Stein (2000) found that firms were not able to optimize the value of the shareholders. This results in inefficient investments in diversified firms. These papers prove that the benefits of conglomerates are costly.

Maksimovic and Philips (2013) and Leaven and Levine (2007) found a diversification discount. On the other hand, Villalonga (2004) and Hund, Munk & Tice (2012) presented empirical evidence for a diversification premium by using another definition for segments. This highlights the fact that in empirical research the definition of a conglomerate is driven by the availability of data and the industry classification. This definition can also affect the results of having a diversification premium or a diversification discount.

Based on the efficiency of internal capital markets, conglomerates can have a diversification premium or a diversification discount. The first hypothesis is based on the coinsurance effect. The coinsurance effect predicts that conglomerates can use the benefits of their internal capital markets. This should result in a diversification premium. The expectations are that the internal capital markets of conglomerates are inefficient and so the coinsurance effect does not exist. This means that conglomerates take relatively more risk than stand-alone firms from the same size and operating in the same industry.

(15)

15

The second hypothesis is the moral hazard hypothesis. According to this hypothesis, the divisional managers behave self-interested. Divisional managers can make decisions in their own interest instead of the interest of the conglomerate. The explanation for the diversification discount is based on the misbehavior of divisional managers. An alternative explanation for the increase of risk-taking of conglomerates is that conglomerates attract riskier divisions and they want the divisions to take more risk. The moral hazard hypothesis distinguishes the two explanations.

Section 2. Methodology

The goal is to get an answer to the question: “Do divisional managers in conglomerates take more risk than CEOs in matched stand-alone firms?” To obtain an answer two hypotheses are tested. The main expectation is that divisional managers in conglomerates tend to behave riskier in decision taking compared to the CEOs of stand-alone firms. To test this expectation, a measurement for risk-taking behavior of divisional managers is constructed and compared to the same measurement of risk-taking behavior of CEOs of stand-alone peers. The proxy used to measure risk-taking behavior is the cash flow volatility.

The main expectation can be test by using the coinsurance effect hypothesis and the moral hazard hypothesis. The coinsurance effect hypothesis analyzes whether divisions and conglomerates take more risk compared to stand-alone firms. The moral hazard hypothesis includes personal characteristics of divisional managers and CEOs of stand-alone firms. This hypothesis investigates whether the risk-taking can be caused by personal characteristics related to risk-taking behavior.

The divisions are managed by a divisional manager, while the stand-alone firms are managed by a CEO. This distinguishes the divisions from the stand-alone firms. In addition, the divisional managers have the ability to choose the desired level of investments. Furthermore, they can use the internal capital market of a conglomerate to increase the cash flows of their division.

The empirical results are based on North-American firms, due the existence of divisional data of North-American firms. Divisional data is matched with firm data based on

(16)

16

size and industry. The size is defined as the total assets. The sic code matches the divisions and conglomerates with stand-alone firms based on the same industry.

A) Instrument risk-taking behavior: Cash flow volatility.

To measure the risk-taking behavior of the divisions and their stand-alone firms’ cash flow volatility is used. John et al. (2008) used the variation in firm-level cash flow over total assets to estimate the “riskiness” of a chosen investment. The cash flow volatility is significantly and negatively associated with firm value. This is consistent with the hypothesis that agency problems are negatively correlated with firm value (Rountree et al, 2008). Cash flow volatility is defined as the time-series standard deviation of the ratio of operating cash flows to assets, calculated using the previous time periods (Ying et al., 2013). A low volatility means that the cash flows are constant over time. In this case, the CEO or the divisional manager takes the same risk every year. A high volatility means the cash flows fluctuate over time and so the CEO or the divisional manager takes more risk in some periods. This increases their risk-taking behavior.

All publicity-listed firms have to report sales, operating profit, depreciation, capital expenditures and total assets at the business segment level, according to SEC regulation S-K (Scharfstein, 1998). This means that the publicity-listed firms do not need to publish all variables to construct a proper proxy for cash flow volatility for all firms at the business segment level.

The cash flow volatility of divisions is based on the following calculations. First the cash flows of the divisions are defined by the operating income after depreciation plus depreciation. Due the fact that data of operating income after depreciation is limited available, an alternative construction of cash flows is by subtracting selling, general & administrative expenses and cost of goods sold from the net sales (Minton and Schrand, 1999). If only data of selling, general & administrative expenses or cost of goods sold is available, only one variable is subtracted from the net sales. This gives a proper estimation for the cash flows of divisions. The number of unique segments with a cash flow measurement is 1163. Some of the segments are part of two or more companies. This holds for 55 segments.

The cash flows of the matched stand-alone firms are defined as sales minus cost of goods sold minus selling, general and administrative expenses minus the change in working

(17)

17

capital for the period. Operating cash flow represents the cash flows available for discretionary investment (Minton and Schrand, 1999). This measurement is available for 26871 stand-alone firms.

The estimations of the cash flows for both divisions and stand-alone firms are divided by the total assets to control for the firm size. This ratio estimates the cash flows as a proportion of the division/firm size. The formula for this equation is shown in formula (1). To alleviate the effect of outliners, the data is winsorized at a five times interquartile range. This means the cash flow volatility is windsorized using as threshold the median cash flow volatility plus/minus five times the interquartile range.

The cash flow volatility is calculated according to formula (2). The estimations of the cash flows over assets are subtracted from the average cash flow over assets of the same division or stand-alone firm. The volatility is based on at least three year consecutive data.

√

∑ ∑

B) Risk in single divisions of conglomerates and their match stand-alone peers.

Agency problems are causing internal capital markets of conglomerates to be inefficient. Risk-taking behavior of divisional managers can cause this failure in the internal capital markets. The behavior towards risk-taking is affected by the fact that their holding/parent firm is more diversified and so they can subsidize any losses in separate divisions. To show the additional risk-taking from the divisions, I analyze the cash flow volatility of segments of conglomerates relative to the cash low volatility of a similar portfolio of single-segment firms. To provide empirical evidence that the coinsurance effect does not exist, it has to satisfy the following two conditions. First, the cash flow volatility of segments is higher than those of matched stand-alone firms. Secondly, the cash flow volatility of conglomerates is higher than the cash flow volatility of the portfolio of stand-alone firms.

(18)

18

I assume that all divisions are part of a conglomerate with at least two divisions. The fact that conglomerates have multiple divisions distinguishes them from the stand-alone firms. Moreover, another assumption is that divisional managers are in charge of the division. They have to take responsibility to the CEO of the conglomerate. The divisional manager is never the CEO of the conglomerate, but is in charge of a single division of the conglomerate. But the divisional manager operates independently.

The procedure of matching the divisions of conglomerates with stand-alone firms is by matching them based on the same industry and size. To classify whether segments and stand-alone firms corporate in the same industry the two digits, three digit or four digit SIC-codes can be used. The two digit SIC-code is the widest approximation for both divisions and stand-alone firms in which industry they corporate. The segments of conglomerates are matched with their stand-alone firms based on the two digit SIC code. A match on the industry level occurs when the stand-alone firm has the same two digits SIC code as a segment.

Besides matching on industry, segments are also matched to stand-alone firms based on firm size. The firm size is defined as the total assets in the first data year. The first data year is chosen, because some divisions have huge differences in assets over years. By using the average total assets, this eliminates a part of the risk-taking of the division. The divisions are matched in size groups. The number of divisions and stand-alone firms per size group is shown in panel A of table 1. The smallest divisions with total assets smaller than one million are matched with the stand-alone firms with assets smaller than one million. The biggest divisions with assets over five billion are matched with the stand-alone firms with assets over five billion as well. All other groups in-between are shown in panel A of table 1.

[Insert table 1 here]

A match based on industry and size occurs when a stand-alone firms is operating in the same industry and has the same firm size as a division. If a segment is matched based on size and industry, the matched portfolio gets a unique value. The total number of matched portfolios is 811 operating in 74 different industries.

Focusing on cash flow volatility can explore empirical evidence for the existence of the coinsurance effect in segments of conglomerates. The first model, as stated below in

(19)

19

equation (3), estimates whether the cash flow volatility of segments is higher than those of the pooled matched stand-alone firms.

This model is based on panel data regressions with time fixed effects, defined a s , and the matched portfolio dummies, defined as . To incorporate fixed effects into the regression, a dummy variable for each matched portfolio is created. Time fixed effects exclude time-dependent omitted variables and allow the intercept to vary over time. Segment dummies are not used, because then the dummy for the segment will drop out of the model and creates multicollinearity.

In this model the cash flow volatility from a chosen segment is the measure for risk-taking behavior. A significant positive coefficient for means divisions have higher cash flow volatility relative to their match stand-alone firms. The included control variables are assets, cash return on assets and capital expenditures. Assets are included to control for the firm size. Cash returns on assets to control for profitability. Cash returns on assets are defined and cash flows in year i divided by the total assets in year i. The last control variable is capital expenditures to control for a regime shift. Changes in firm risk due to changes in the firm policy can attribute to reasons other than managerial risk aversion (Low, 2009).

Because the cash flow volatility is based on several periods, according to Minton and Schrand (1999) the control variables have to be averaged over the same period as the cash flow volatility. Models are presented both with and without the control variables. To ensure that the results are statically robust, the control variables are windsorized using as threshold the median plus/minus five times the interquartile range.

A positive and significant sign for the coefficient of segments would mean that segments take more risk than their matched stand-alone peers. The coefficient is suggesting a positive relationship between being part of a conglomerate and additional risk-taking. In this case the coinsurance effect does not exist based on a divisional level. The coefficient is suggesting a positive relationship between being part of a conglomerate and additional

(20)

risk-20

taking. However, being a segment does not have a causal relationship with the additional risk-taking behavior.

C) Risk in conglomerates and their matched stand-alone peers.

If there is a risk premium (Villalonga,2004) you would expect that the additional risk of divisions pays out and that on average the risk for all the divisions is lower than stand-alone firms. This expectation is based on the fact that conglomerates can use uncorrelated segments to subsidize losses in single divisions. Overall conglomerates would benefit from additional risk on a single segment level. By comparing the measurement for risk-taking behavior of conglomerates to matched stand-alone firms, evidence can be provided whether conglomerates take more risk (on average of all divisions) relative to match stand-alone firms.

The second model estimates whether the cash flow volatility of conglomerates is on average lower than those of a portfolio of stand-alone firms. First, the total assets of segments in conglomerates with at least two segments are summed up. According to panel B of table 1, there are 1799 conglomerates with more than one division. The other 2379 conglomerates have one division. The reason that they have one division in this sample is due the lack of divisional data. Secondly, the conglomerates are matched with stand-alone firms with the same size and the same industry based on two digits SIC code. This matching process is based on industry and size in the same way as the divisions are matched with stand-alone firms. The smallest (below one million) and biggest (above five billion) conglomerates are handled the same as the divisions discussed in the previous part. So they are matched with the smallest and biggest stand-alone firms operating in the same industry. If a conglomerate operates in several divisions, a comparable industry is chosen manually. If both the conglomerate observations and their matched stand-alone are selected the cash flow volatility is calculated as an average of the cash flow volatility of all divisions of the conglomerate. The conglomerates with one division are threaten similar to the single divisions and matched to stand-alone firms in their firm size group and industry. The number of unique matched portfolios for conglomerates is 816. Model (4) estimates whether the cash flow volatility of a conglomerate is higher relative to matched stand-alone firms.

(21)

21

This model includes time fixed effects ( and unique match portfolio dummies . The cash flow volatility is a proxy to measure risk-behavior of conglomerates. If the cash flow volatility of divisions and conglomerate firms are both higher than their matched peers, this would provide that there is no evidence for a coinsurance effect. In this case divisions take more risk than matched stand-alone peers. Moreover, conglomerates (which consist of at least one division) are taking more risk relative to their stand-alone peers. The average cash flow volatility of a conglomerate is higher than the cash flow volatility of stand-alone firms. In this scenario it is likely that conglomerates have a diversification discount.

The control variables in this model are the same as used when comparing divisions with stand-alone peers, namely: assets, cash return on assets and capital expenditures. The control variables are averaged when a conglomerate has multiple divisions.

A positive and significant value for would suggest that conglomerates take more risk than their matched stand-alone firms. If the risk-taking of divisions and the risk-taking of conglomerates is higher than the risk-taking of the matched stand-alone firms, this would mean that there is evidence that the insurance effect does not exist. One condition is that on a divisional level, the risk of divisions is also higher than matched stand-alone firms. Some characteristics of conglomerates are causing that conglomerates take more risk relative to their stand-alone peers. Managers behave self-interested or divisions follow a riskier conglomerate strategy. Both explanations can cause this additional risk-taking behavior.

According to the theory of Stein (1997), internal capital markets can be efficient if there is a high span of control. But, if the span of control lowers, the internal capital markets become more inefficient. To estimate the effect of the span of control, a proxy for the span of control is included. The proxy is the number of divisions of a conglomerate. If the number of divisions becomes higher, the span of control decreases. To estimate this effect, the model of comparing the risk-taking behavior of conglomerates relative to their matched stand-alone peers is slightly changed into model (5).

(22)

22

In this model the stand-alone firms have zero divisions. The conglomerates have one or more segments. The expectation is that if their number of segments increases, the span of control decreases and so the cash flow volatility (as a proxy of risk-taking behavior) does increase. The model has the same control variables as the previous model, which compares the cash flow volatility of conglomerates to their matched stand-alone peers.

D) CEO and divisional manager characteristics on risk-taking behavior.

The second hypothesis is the moral hazard hypothesis. If this hypothesis is correct, we would expect that divisional managers would behave self-interested and would prefer to take excessive risk relative to what would be optimal for the shareholders. The incentives of divisional managers of conglomerates are not in line with the optimization of the shareholders’ value of the conglomerates. Those agency problems lead to inefficient internal capital market and could destroy firm value.

The expectation is that segments take more risk than their stand-alone matched firms. If this theory is true, there are several explanations why this occurs. The first explanation is that managers of divisions behave self-interested. They take more risk relative to CEOs, because their parent/holding company can subsidize any losses. In a situation when the risk pays out they get the benefits of the additional risk they take. On the other hand, in a situation when the risk does not pay out, their holding/parent company can subsidizes the losses. This advantage can only be used by divisional managers and cannot be used by the matched CEOs of stand-alone firms. The second explanation is that the holding/parent company want their divisions to take more risk. Conglomerates have the ability to subsidize losses in separate divisions and conglomerates can attract more risky division. In this case risk taking behavior is part of the conglomerate strategy.

To test this moral hazard hypothesis, a number of observable variables related to CEO characteristics and divisional managers’ characteristics are included into model (6). Model (6) is used to estimate whether characteristics of divisional managers and CEOs influence their risk-taking behavior:

(23)

23

This model has time fixed effect ( and matched portfolio dummies included . Moreover, the term “time fixed effect times conglomerate” is included into the model. Time fixed effect allows the intercept of the model to vary over time. The portfolio dummy controls for comparing firm with almost the size and operating in the same industry. The interaction term of time fixed effect times conglomerate, controls for a difference of the year of the observations between data of stand-alone firms and divisional data. The matched portfolios are the same when comparing single divisions with their matched stand-alone peers.

To estimate if divisional managers behave self-interested, a number of characteristics of the divisional managers and CEOs are included in the model. The variables are: age, gender and tenure. Younger CEOs tend to take more risk relative to older CEOs (Pahl and Vroom, 1971). Beside the age of the CEO, the gender of the CEO also influences their risk-taking behavior. Male directors seem to take more risk than female directors (Byrnes, Miller and Schafer, 1999). Furthermore, CEOs with a longer tenure seem to be more risk-averse (Berget et al, 1997). This suggests a negative relation between tenure and risk-taking behavior. No other variables are included into the model, because of the limited data of divisional managers. The BoardEx database only provides these variables related to risk-taking behavior. The CEO/divisional manager characteristics estimate if age, gender and male influence their risk-taking behavior of both stand-alone firms and conglomerates.

The age variables are based on three different age groups. The first group represents the youngest divisional manager and CEO group. Their age is below the threshold of fifty years old. The middle group is between 51 and 65 years old. The oldest group of divisional managers and CEOs are older than 65 years. In the model the standard group is the old group.

(24)

24

The interaction terms estimate the additional risk-taking of divisional managers based on their characteristics. In case of a divisional manager this additional term predicts whether they take more risk compared to a CEO of a matched stand-alone firm which has the same age, gender and tenure. If the estimations of personal characteristics and the interaction term goes in the same direction (positive or negative) and is significant, there is evidence that personal characteristics affect the risk-taking behavior.

The control variables in this model are the same as used before, namely: assets, cash return of assets and capital expenditures. All control variables are averaged over multiple divisions.

This model tries to find empirical evidence that managers behave self-interested. If the segment coefficient is positive and the personal characteristic variable has the same sign as the interaction variables, this is taken as evidence that the personal characteristic influences the behavior of the divisional manager or the CEO. Only the downside of using the variables age, gender and tenure is that they may not analyze a causal relationship. Divisions can attract younger divisional managers, because the divisions would like to take more risk. In this case there is reversed causality. Compensation data would provide more detailed information about behavior of divisional managers and the CEOs of stand-alone firms. However, there is no available data of compensation from divisional managers. This is a limitation of this thesis.

Section 3. Data and Descriptive statistics

In this section the data collection and the summary statistics are discussed. The summary statistics are also divided in subsamples for more detailed statistics. The subsamples are: single divisions, conglomerates that consist of at least two divisions and stand-alone firms.

A) Data collection.

Divisional data is collected from the COMPUSTAT historical segment database. This data is used to identify divisions of conglomerates. The selected variables of this database are: assets, capital expenditures, cost of goods sold, depreciation, equity in earnings, intersegment eliminations, net income, operating income after depreciation, operating profit, pretax income, total revenues, net sales, income taxes, selling, general and

(25)

25

administrative expense, segment name, company name and the four digits sic code. These observations have a yearly scale. The sample period is 1976-2016.

To match the identified divisions with their matched stand-alone firms, data from those matched stand-alone firms is collected. The following variables from the COMPUSTAT North-America fundamentals database are selected: (total) assets, capital expenditures, cost of goods sold, depreciation and amortization, sales/turnover (net), change in working capital, selling, general and administrative expense, company name and the four digits sic code. This dataset contains North-American firms and has yearly observations from 1976-2016.

Personal characteristics of CEOs of stand-alone firms are collected from the ISS database. The variables included from the database are: gender, age, year of service begins and year of service ends. Personal characteristics of divisional managers are provided from the BoardEx database. The included variables are: the role inside a company to distinguish whether a director is a divisional manager, the role description, gender, age, year service starts and the year service ends.

B) Descriptive statistics of the sample.

Table 2 provides basic summary statistics of the complete dataset. The independent variable is the cash flow volatility. Other firm characteristics included in panel A are: assets, cash flow and sales. Beside firm characteristics, panel A of table 2 includes personal characteristics of the CEOs and the divisional managers.

The number of variables is smaller than the amount of observations of the paper of Villalonga (2004). Only observations with a proxy of cash flow volatility are included in this sample. For this reason the number of observations is smaller than the number of observation in the paper of Villalonga (2004). To get a proxy for cash flow volatility of divisions and stand-alone firms, the data has to satisfy to the following conditions. The first condition is at least three year consecutive data. The second condition is a proxy for the total assets. The third condition is the availability of a constructed proxy for the cash flows of divisions and stand-alone firms.

(26)

26

The total assets of the whole sample are on average five billion. However, this average is probably caused by some large outliers. The 75-percentile value of total assets is 836 million. This is around 16 percent of the value of the average total assets. The big difference between the 75-percentile and the mean can be explained by the huge standard deviation. The biggest company in the sample has assets of 3777 billion. This is unlikely. For this reason the total assets are windsorized for the regression models.

Cash flow has an average of 307 million. This means firms get on average 307 million on cash flow per year. However, this average is affected by some large outliers. The 75-percentile value is 73.45 million. Also remarkable is the minimum and maximum value. This variable is windsorized for the regression models as well.

The variables of cash return on assets, defined as cash flows over assets, are already windsorized. For this variable the mean is lower than the median. None of the companies have more cash flow than the size of the firm or division. This seems reasonable.

The variable of cash flow volatility, the dependent variable in the regression models, is on average 0.14. This means firms deviate with 0.14 from their average cash flows over assets. This coefficient does not describe a magnitude. For future comparisons we take a cash flow volatility of 0.14 as a normal level of deviation.

Beside the firm characteristics, table 2 includes personal characteristics of divisional managers and CEOs of stand-alone firms. The average divisional manager/CEO is 55 years old, male and already works around eight years in their function. Most divisional managers /CEOs are between the 50 years and 60 years old. The youngest divisional manager/CEO is thirty years, while the oldest is 97 years old. That someone of 97 is in charge of a company is unlikely. But maybe he is still symbolic the CEO, because he is the founder of the company.

Most of the people that lead the company or division are male. 97 percent of the companies have a male person as divisional managers or CEO. Their tenure has a large standard deviation. However, the difference between the median value and the mean value is only two years. Also noticeable is that the longest CEO is already on service for 61 years. This is the CEO of 97 year old. Seems unlikely, but is still reasonable.

C) Descriptive statistics of stand-alone firms and single divisions.

Panel B of table 2 shows the descriptive statistics of the variables of both single divisions and the stand-alone firms separately. Beside the variable based on the year of the data, the

(27)

27

selected variables are based on two categories. The firm characteristic variables are: total assets1, cash return on assets2, cash flow volatility3, total capital expenditures4 and number of segments. Stand-alone firms contain of zero divisions in the sample. The CEO or divisional manager characteristic variables are: age, tenure and gender.

The average assets of stand-alone firms are higher relative to divisions. This indicates that the average division is smaller than the average stand-alone firms. This seems reasonable because conglomerate firms can contain more than one division. However, assets of both single divisions and stand-alone firms have a large standard deviation. This indicates high uncertainty. The way to deal with this high uncertainty is by comparing divisions with stand-alone firms on the same size.

On average the cash flows is eight percent of the total assets for the single divisions, while the cash flows relative to assets of stand-alone firms is smaller. If we compare the mean cash flow volatility of single divisions with stand-alone firms, the cash flow volatility of stand-alone firms is higher with 0.01. Assuming 0.14 as a normal level of cash flow volatility, this does not indicates that stand-alone firms take more risk based on the measurement for risk-taking behavior.

The divisional managers seem to be younger than CEOs of stand-alone firms. This can be explained by the theory that agents first need to get experience by leading a division before they can lead a stand-alone firm. In the sample the divisions have more female agents in the leading role compared to stand-alone firms. Moreover, divisional managers tend to stay shorter in their function. This can also be explained by the theory that divisional managers need to get experience, before they growth to a function of being a CEO of a firm.

D) Descriptive statistics of multi-segment conglomerates and stand-alone firms.

Panel B of table 2 also shows the summary statistics of conglomerates that contain at least two divisions. The conglomerates of panel B of table 2 are defined as multi-segment firms with at least two divisions. The included variables are: year, average assets of multiple segments5, average cash return on assets6, average cash flow volatility of conglomerates

1

Total assets is windsorized at 5*interquartile-range

2_{Cash return on assets is windsorized at 5*interquartile-range} 3

Cash flow volatility is calculated based on cash return on assets windsorized at 5*interquartile-range

4

Capital expenditures is windsorized at 5*interquartile-range

5_{Average assets is average of total assets of multiple divisions, total assets is windsorized at}

(28)

28

with multiple divisions 7, average capital expenditures of multiple segments8, number of divisions, age of divisional managers or CEOs, gender of divisional managers or CEOs and the tenure of the divisional managers or the CEOs.

The total assets of conglomerates with multiple divisions are bigger than the total assets of single divisions. This is reasonable, because the sum of the total assets of multiple divisions is likely to be larger than the total assets of a single division. The total assets of conglomerates are also bigger than stand-alone firms. This suggests that conglomerates are on average bigger firms relative to stand-alone firms.

Moreover, panel B of table 2 suggests that the risk-taking behavior is larger for conglomerates compared to stand-alone firms. The cash flow volatility of conglomerates with at least two divisions is 0.34, while the cash flow volatility of stand-alone firms is 0.14. Taking cash flow volatility as a proxy for risk-taking behavior, this suggests that conglomerates take more risk relative to stand-alone peers.

The personal characteristics look similar to the characteristics of stand-alone firms. The average divisional manager is younger than the average CEO. The fraction of male does not change and the tenure of divisional managers is still smaller than the tenure of CEOs of stand-alone firms.

Section 4. Main Results

In this section the models discussed in the methodology are worked out. First the coinsurance effect is tested. This is tested on two different levels. The risk-taking of divisions is compared to their matched stand-alone peers and the risk-taking behavior of conglomerates is tested relative to the matched stand-alone firms of the conglomerate. If both divisions and conglomerates have a higher and significant risk-taking, there is no evidence that the coinsurance effect exists. A possible explanation is that managers of divisions behave self-interested. A model with personal characteristics analyzes this misbehavior.

6

Cash ROA is windsorized at 5*interquartile range

7_{Averaged cash flow volatility of divisions of a conglomerate with at least two divisions} 8

(29)

29

A) Evidence for coinsurance effect based on divisions.

By testing whether the cash flow volatility of segments is lower than those of matched stand-alone firms and the cash flow volatility of conglomerate firms is lower than the cash flow volatility of the matched firms of the matched portfolio, this can provide evidence for the non-existence of the coinsurance effect.

Table 3 shows the results of the regressions in which the cash flow volatility of single divisions is compared to the cash flow volatility of their stand-alone peers. The included control variables are the averaged assets, averaged cash flows and averaged capital expenditures. Multiple regressions are presented with and without control variables. The control variables control for firm characteristics. Regression (1), regression (3), regression (5) and regression (7) are presented without control variables. Regression (2), regression (4), regression (6) and regression (8) till regression (12) have control variables included. Furthermore, multiple regressions with time fixed effects and/or matched portfolio dummies are included. All regressions have matched portfolio dummies. Time fixed effects are included in column 3 till column 8 and column 10 till column 12. Beside regressions with and without control variables and fixed effects, regressions with clustered standard errors are included as well. There are two different kind of clustered standard errors. This first one is based on a divisional level. These clustered standard errors are used in regression (5), regression (6) and regression (11). The second one is based on a company level. These clustered standard errors are used in regression (7), regression (8) and regression (12)

From column 1 and column 2 we can say that the cash flow volatility is higher for segments that are part of a conglomerate if we compare them to their matched stand-alone peers. This comparison takes into account the fixed effects of the firms and divisions in the matched portfolio. The matched portfolio dummy controls for the different in intercepts. An example of a fixed effect in the matched portfolios is industry fixed effects. The cash flow volatility of segments is about 0.02 higher than the cash flow volatility of the matched stand-alone firms. Taking cash flow volatility as a measurement for risk-taking behavior, this means that divisions take more risk than their matched stand-alone peers. The coefficient for being a segment changes slightly when I add control variables. The coefficient stays around the

(30)

30

value of 0.02. The assumed normal level of cash flow volatility is 0.14. This is the average cash flow volatility of stand-alone firms. Relative to this number, the additional risk-taking effect of 0.02 is quite a strong effect.

The control variables are all significant. Total assets have negative coefficient. So if the firm size of a division or a stand-alone firm increases, the risk-taking behavior of a division decreases. In addition, cash return on assets has a negative coefficient. This means that divisions or firms that are more profitable take lower risks. The last included control variable is capital expenditures. This variable controls for a regime shift. Divisions or firms with a more expendable regime take more risk according to column 1 and column 2 of table 3.

In column 3 and column 4, time fixed effects and portfolio dummies are both included. Panel B of table 1 shows that the average data year of the sample of stand-alone firms is on average older relative the sample of divisional data. By adding time fixed effects, the estimations are more reliable. The time fixed effects allow the intercept to change over time. In column 3 and column 4 the results of being a segment are inconsistent. Without control variables the estimations of being a division of a conglomerate is only significant at the five percent level. However, by adding control variables the results get significant at the one percent significant level. Comparing those results to the results of column 1 and column 2, the effect of being a division on cash flow volatility is stronger (coefficient of 0.042) without control variables and weaker (coefficient of 0.014) with control variables. The power of the effect decreases by adding control variables. The effect of the control variables is similar to those of column 1 and column 2.

In column 5 and column 6 clustered standard errors based on a divisional level are added. This allows for cross-correlation of residuals. Residuals are allowed to be correlated across years within a divisions annual observations. One assumption is that different divisions inside one conglomerate operate independent from each other. The effect of adding clustered standard errors based on a divisional level in column 5 and column 6 does change the significance in column 5, but does not change the significance in column 6. The coefficient for being a division is not significant anymore in column 5. The results of column 6 are similar to the results of column 4. The significance of the most important variable in column 6, namely being a segment, is not affected by adding clustered standard errors to the model.

(31)

31

While regression (5) and regression (6) have clustered stand errors based on a divisional level, regression (7) and regression (8) have clustered standard errors based on the company level. In this case the residuals are allowed to be correlated across years within a company’s annual observations. The results in column 7 and column 8 look similar to the results of column 5 and column 6. Changing the clustered standard errors to the company level instead of a divisional level does not affect the results. The significance of the control variables are not affected by clustered standard errors.

According to panel B of table 2 the average year of the data of the divisional data is more recent than the average year of the data of the stand-alone firms. Column 9 to column 12 contain only data of stand-alone firm from 1995 till 2015. The results of regression (9) show that divisions take more risk than matched stand-alone firms. By adding time fixed effects in regression (10) the effect gets slightly weaker, but still remains positive and significant. Also adding clustered standard errors on a divisional level in regression (11) does not change the significant of the coefficient of being a division. Only the significant of the control variable ‘capital expenditures’ drops out. Regression (12) with clustered standard errors on a company level is similar to the regression with clustered standard errors on a divisional level.

Overall table 3 presents that divisions, which are managed by a divisional manager, have a higher cash flow volatility relative to their stand-alone peers, which are managed by a CEO. When taking cash flow volatility as a proxy for risk-taking behavior, it seems that segments take more risk when compared to their stand-alone matched firms.

If the coinsurance effect exists it would lead to a diversification premium. In that case, the expectations are that divisions do not take significantly more risk relative to their stand-alone peers. However, conglomerates can smooth out the risk-taking over multiple divisions. Table 4 compares the cash flow volatility of conglomerates to matched stand-alone firms.

Risk-taking in conglomerates : a comparative analysis of risk-taking by divisional managers and CEOs of stand-alone firms