Wealth Effects on CEO Compensation: Causal Evidence from the Real Estate Crash of 2006-2010* Beau Page** August 2014

Abstract This paper examined the effect of changes in CEO wealth on CEO equity ownership and compensation using a novel dataset on changes in CEO house prices during the real estate crash of 2006-2010 as a proxy for CEO wealth. Consistent with the predictions of agency theory, a large decrease in house price leads to a decrease in equity ownership for CEOs whose house value is a significant portion of their personal wealth. These CEOs sell more shares after their house loses value. Board grant more equity to affected CEOs, but not enough to offset the CEOs’ equity sales. Total compensation increases, while cash compensation is unaffected.

Keywords: CEO Compensation, Housing Wealth Effects, Real Estate Crash

* I wish to thank my thesis advisor Toni Whited for her guidance and encouragement. Additional thanks go to the member of my committee, Michael Raith and Boris Nikolov. Special thanks to Chet Reeder for his feedback and help in collecting the house price data. ** C. T. Bauer College of Business, University of Houston, e-mail: [email protected]

A firm’s board of directors is responsible for providing the CEO with the proper incentives. One part of this responsibility is determining the CEO’s exposure to the firm’s equity. The optimal amount of managerial ownership depends on a number of firm, industry, and individual specific factors. This paper examines how a CEO’s personal wealth affects his equity incentives, as well as his overall compensation. Specifically, I am interested in how a shock to a manager’s personal wealth causes his compensation and equity exposure to change. As wealth is unobservable, this paper uses a novel dataset of house price changes during the 2006-2010 real estate crash as a proxy for wealth shocks. There are two important aspects to the relation between CEO wealth and equity ownership: incentives and diversification. CEOs are given equity as a work incentive. The idea is to align the CEO’s choices with the goal of maximizing shareholder value, which equity ownership accomplishes by making the CEO’s wealth, and hence his utility, depend on firm value. Assuming a CEO experiences decreasing marginal utility from wealth, as his wealth increases he receives less incentive from a given level of equity ownership. His utility becomes less sensitive to fluctuations in firm value, requiring a greater amount of equity to provide the same level of incentive. While increased equity ownership gives a higher level of incentives, it comes at the cost of decreasing the CEO’s level of investment diversification. Given the typical assumptions about risk aversion, the utility a person receives from a risky payoff depends on her wealth. Specific to CEOs, Lambert, Larcker, and Verrecchia (1991) show that a risk averse manager values risky compensation (such as stock or options) less at lower levels of personal wealth. The more exposed a CEO is to his firm’s stock, as a percentage of his own wealth, the less value he places on it because he cannot diversify away his firm’s idiosyncratic risk. In compensating the CEO, the board’s goal is to balance the positive effect of increased incentive alignment and the negative effect of underdiversification. A CEO who does not have enough incentive may choose projects or make business decisions that provide him with some gain at the expense of the firm’s shareholders. A CEO who has too much ownership in

1

the firm may choose to forgo certain risky positive net present value projects because he is unable to diversify away his personal risk. If a board of directors is actively thinking about these two effects, then, as much as a wealth affects a CEO’s incentive and diversification levels, it will react to substantial changes in the CEO’s wealth. Measuring the effect of wealth on CEOs is difficult because of data limitations. While the SEC requires firms to report detailed data on CEO compensation and equity ownership, there is no requirement to provide data on managers’ personal holdings. In place of CEO wealth, I use data on CEO home values, which has two advantages: houses are an important component of the wealth of U.S. households, and the cross-sectional variance of real estate losses in the 2006 - 2010 period is high. According to the U.S. Census Bureau (2004), in 2004 (the most recent year for which they provide data) the primary residence made up 54.7% of the average U.S. household’s wealth. For the wealthiest Americans, those with a net worth greater than $500,000, this was 28.1%. CEOs are likely to be in this wealthier group. Although they do not hold as large a portion of their wealth in their homes as the average household, even the wealthiest still have a significant percentage of their wealth tied up in their houses. As to the differential effects of the housing crash, in 2008, the worst year for home prices, the Case-Shiller Index recorded a decrease in home prices in Las Vegas of almost 33%, while Denver experienced a decrease in home prices of only 4%. The overall market fell the same amount for a CEO who lives in Denver as for a CEO who lives in Las Vegas, but their house prices fell at very different rates. As a house is a consumption good as well as an investment good, some may feel that“housing wealth isn’t wealth.”1 There is a large literature that discusses the existence and magnitude of the housing wealth effect. Calomiris, Longhofer, and Miles (2012) provide a useful review of the more recent findings. In their paper, Calomiris, Longhofer, and Miles (2012) find a housing wealth effect, and show that it varies based on age and the importance of housing wealth to total wealth. As CEOs are generally wealthy, the size of the wealth effect for them 1

Mervy King, Bank of England Governor, as cited by Buiter (2008).

2

is likely to be smaller than that of the average U.S. household. The size of the housing wealth effect for CEOs is important only inasmuch as it is the mechanism through which a decrease in house prices matters to the CEO’s compensation. If, as this paper argues, changes in housing wealth affect CEO compensation, then larger wealth shocks, which are generally unobservable to researchers, should have even larger effects. I use variation in changes in house prices to identify the causal impact of a decrease in this component of wealth on CEO pay and equity ownership. I find evidence that, for CEOs whose houses are an important component of their wealth, equity ownership decreases when the value of the CEO’s house decreases. An affected CEO sells more shares than a similar CEO who did not experience the decrease in home value, and the board of directors does not fully replace the sold shares with new equity grants. However, equity grants are slightly larger for an affected CEO than an unaffected one. I further separate out the sample by firm size. The wealth effect is seen most strongly for large firm CEOs. Plus, for CEOs of large firms I find evidence that total compensation increases, while cash compensation decreases. The effects are not found in the small firm CEO subsample, which I argue is caused by the fact that small firms are more likely to be dependent on hard to measure local economic effects, confounding the estimation. If this is true, then the large firm CEO subsample is a cleaner test of the wealth effect. This paper fits in the recent stream of literature that attempts to show the effects of outside wealth on CEO pay and firm performance. Baker and Hall (2004) predict and give empirical evidence that a CEO’s overall wealth, proxied for using firm size, affects his compensation and risk aversion. Dittman and Maug (2007) proxy for wealth using prior compensation before calibrating an agency model to predict the proper structure of CEO compensation. Becker (2006) uses data on the actual outside wealth of a sample of Swedish CEOs to show that wealthier CEOs have higher levels of incentives. Liu and Yermack (2007) find evidence that firm performance suffers when a CEO purchases an expensive house. Neyland (2012) studies the effect of CEO divorce on CEO compensation, finding

3

that compensation increases after a divorce. In the next section I develop the paper’s hypotheses. In section 3 I describe the data I use in section 4 to test the hypotheses. Section 5 concludes the paper.

1

Hypothesis development

Agency theory predicts that a CEO’s wealth is an important factor in the level of equity incentives he receives. For a given amount of equity, the wealthier a CEO is, the less sensitive his utility is to a change in firm value. As a result, the level of incentives should be increasing in the level of CEO wealth, which Becker (2006) shows to be true for a sample of Swedish CEOs. A similar prediction holds for changes in the level of a CEO’s wealth. Holding all else equal, if a CEO becomes wealthier, then his utility becomes correspondingly less sensitive to firm value, which leads the board of directors to increase his equity incentives. The logic behind a positive correlation between changes in CEO wealth and changes in equity ownership is clear for an increase in CEO wealth. As a decrease in a CEO’s wealth increases his exposure to the firm, it is not immediately obvious that the correlation between wealth and ownership changes should be positive for decreases in wealth. The question is whether increased exposure is good for shareholders. The important assumption is that before the adverse wealth shock, the CEO’s incentives were set in equilibrium. The board of directors had previously determined that, at the higher wealth level, a certain amount of equity ownership balanced the incentive and diversification effects. A decrease in wealth means that, on the margin, incentives increase and diversification decreases. The previous balance is undone, and if the CEO’s utility becomes too sensitive to firm value, and he is risk averse, then he is likely to look for ways to decrease the volatility of the firm’s returns, even if that means forgoing positive NPV projects. This negative outcome of too much sensitivity, combined with the fact that the CEO’s equity ownership before the change in his wealth was set in equilibrium, predicts that the optimal level of CEO ownership decreases when the

4

CEO’s wealth decreases. There is little data on CEO personal holdings, which makes it difficult to test predictions about changes in CEO wealth. In light of this difficulty, I look at changes in the value of the CEO’s primary residence. The change in a CEO’s house value is of interest for a number of reasons. First, as noted in the introduction, primary residences make up a significant portion of the net worth of U.S. households. Second, the crash in the real estate market in 2007 led to large, unexpected changes in house values. Third, the crash was not nationally uniform, but instead shows a large amount of cross-sectional variation, which is useful in identifying its effect. So, given the importance of primary residences to wealth, and the large magnitude of the 2007 shock, changes in the value of CEO houses should have an effect on CEO equity ownership. Similar to a shock to the CEO’s overall wealth, there should be a positive correlation between changes in house prices and changes in equity ownership. The mechanism through which CEO equity ownership increases is obvious: the board of directors grants new equity to the CEO. Once equity is given, however, the board of directors is usually unable to take it away.2 When a CEO’s wealth decreases, increasing his sensitivity to firm value, he will sell part of his ownership of the firm to decreases the sensitivity, as well as to provide funds to increase his overall diversification. Bettis, Coles, and Lemmon (2000) show that the vast majority of firms have policies in place that restrict a CEO’s ability to sell shares, so the board of directors knows the CEO is making these trades, and does not prevent them. If the board wants the level of CEO ownership to decrease, then it will not replace the shares the CEO has sold. So, two further hypotheses are that CEOs sell more shares when their house prices decrease, and that new equity grants to CEOs are not large enough to replace the number of shares the CEO has sold. Changes in compensation do not only have to work through the equity channel. If the 2

There is an increasing use of clawback provisions in CEO contracts, which would allow the board of directors to take back equity given to the CEO. This is not important here because clawbacks are generally triggered by bad behavior on the part of the CEO (earnings restatements, fraud, misconduct, etc.), not by general economic performance. Also, despite their existence, few clawbacks have actually been performed. See Babenko, Bennett, Bizjak, and Coles (2012).

5

issue is that the CEO’s wealth is too dependent on firm value, and he would prefer to increase his diversification, then the board could grant him more cash instead. He would then use that cash to diversify his personal portfolio. In equilibrium the board of directors should adjust ownership and cash compensation until the marginal benefit of both of these actions is equal. In the case of a decrease in house price, we should then expect to see an increase in the amount of cash compensation given to the CEO. In summary, my primary hypothesis is that a decrease in a CEO’s house value will lead to a corresponding decrease in his level of equity ownership. The mechanism through which this occurs is that CEOs sell more shares after their house decreases in value, and the board does not fully replace the shares the CEO sells. The second hypothesis is that cash compensation increases with a decrease in house value, to allow the CEO to increase his diversification. Finally, I expect that all of these effects should be more prominent for CEOs whose house makes up a more substantial portion of their wealth.

2

Data

My sample is made up of CEOs included in the Standard & Poor’s Execucomp database during the years 2006 to 2009. Execucomp provides data on compensation (total and broken out into categories such as salary and stock option grants) for the top five executives at firms in the S&P 1500, including firms which were in the index historically but no longer are. I identify a firm’s CEO by the CEO indicator in the database. There are situations in which the dates indicated for a CEO’s tenure do not match with the CEO indicator; in those cases I use the dates given for when the CEO was made CEO for identification. After identifying the CEO for each firm, and cleaning the sample (as described below), I drop every firm that does not have the same CEO for the entire sample period, so as to have a balanced panel for my tests. From this reduced sample I randomly select 500 CEOs. I search for the home address of each of the CEOs in the sample. I do this using various

6

online searches. The two primary sources of CEO addresses are WhitePages.com, and the Individual Contributions search section of fec.gov, the Federal Election Commission’s website. WhitePages.com is a directory service, providing contact information for most of the adults living in the U.S. Not every CEO’s home address can be found with WhitePages.com, as people can ask to be removed from the directory, many people have the same name or go by a nickname not given in Execucomp, and there can be confusion when a person changes addresses. To improve the accuracy of identification, I use fec.gov. When U.S. citizens make political contributions, they are required to provide some basic information about themselves, including their address. A large number of CEOs make political contributions, so I am able to find a number of addresses in this way. Of the 500 CEOs I sample, I find home addresses for 440. Once I have the CEOs’ addresses, I use Zillow.com, an online real estate website, to determine the value of their homes during the sample period. Zillow provides an estimate of median house values for most U.S. cities and zip codes. Zillow computes its estimate of house value, called a Zestimate, using a proprietary formula based on the selling prices of houses in the area and house characteristics.3 The benefits of using Zestimates over a more traditional index such as the Case-Schiller index is that Zestimates are available for more cities, and at the zip code level. Case-Schiller indices are only available for the nation, two separate composites of cities, and 20 metropolitan areas. Even with Zestimates there are some data issues. Some cities, such as Houston, Texas, only have data back to 2008. Zestimates are not available in all areas. I drop all observations for which the Zestimate is not available for the CEO’s home zip code. I collect compensation and firm data for the CEOs for whom I have an address. From Execucomp I extract the number of shares held by the CEO at year end, new and existing options holdings, cash compensation (calculated as all compensation not in the form of 3 For more information, Zillow provides a discussion of the Zestimate as well as its accuracy at www.zillow.com/wikipages/What-is-a-Zestimate and www.zillow.com/howto/DataCoverageZestimateAccuracy.htm.

7

equity:

total curr, noneq incent, othcomp, and allothtot ), and total compensation. From

Compustat I extract the value of firm assets and market value. From CRSP I pull the firm’s stock return, as well as the overall market return for each year. To clean the data, I remove firms with missing values, negative or zero assets or market values, and negative equity ownership. I adjust firm size for inflation using the CPI (dollar amounts are in year 2000 dollars). To remove extra noise from my tests, I also drop any CEO whose company has a non-December fiscal year end. This leaves a final sample of 321 CEOs. To determine the level of CEO ownership, I combine shares held with the share equivalence of stock options. Stock option share equivalence is computed as the Black-Scholes delta, or the dollar change of the CEO’s wealth for a one percent increase in share price. New equity grants for a CEO are the sum of stock grants and the delta of option grants during the year. I calculate the amount a CEO sells in a year using the method of Core and Guay (2010). CEO sales in a given year equal the value of end-of-year equity holdings minus the value of new equity grants minus what beginning-of-year equity holdings would be worth at end-of-the-year prices if the CEO’s ownership did not change. This value can be positive (CEO buys more equity with his own money) or negative (CEO sells equity). Panel A of Table 1 presents descriptive statistics on CEO compensation. Mean (median) ownership for my sample of CEOs is $51.00 million ($15.12 million), mean (median) cash compensation is $2.33 million ($1.46 million), the mean (median) value of the equity grant is $2.68 million ($1.37 million), and mean (median) total compensation is $5.01 million ($3.12 million). Ownership is smaller in my sample than it is for all of Execucomp, which has mean (median) ownership of $255.76 million ($17.44 million). This difference is primarily the result of my sample not including the largest firms in the Execucomp database, which means that the average firm in my sample is significantly smaller than the average Execucomp firm, although the difference in median firm sizes between the two samples is small ($2,265.03 million vs. $2,210.84 million). The compensation variables do not differ appreciably from the means and medians for Execucomp as a whole. Table 1 also presents ownership as a

8

fraction of firm value, which has a mean (median) of 2.68% (0.87%). Table 1 reports mean (median) purchases of equity by the CEO of -$7.45 million (-$1.79 million); on average CEOs sold equity during the sample period. For my test I use the first difference in CEO ownership as a percentage of market value, equity grants as a percentage of firm value, shares bought or sold as a percentage of firm value, the first difference of total compensation as a fraction of the firm’s book asset value, and the first difference in cash compensation as a fraction of total compensation. I winsorize each of the difference variables at the 1% level except for the change in cash compensation, which does not appear to have significant outliers. I also winsorize equity grants and CEO sales at the 1% level. Panel B of Table 1 presents statistics related to CEO house values. The mean (median) value of a CEO’s house in the sample is $1.84 million ($1.30 million). The mean (median) ratio of house value to the CEO’s equity ownership (Home Value) is 0.82 (0.082). This variable is a measure of how significant a portion the CEO’s house is of his wealth. It is highly skewed, as some CEOs have houses worth many times their ownership in their firms, while most CEOs have houses worth a smaller fraction of their ownership. Before running tests, I winsorize Home Value at the 1% level to remove the effects of a few large outliers. The mean (median) change in house values in the top third of a CEO’s zip code (ZipTT) is -4.24% (-3.40%), reflecting the housing downturn. Note that there is significant variation in the change in house values; the largest year over year increase (decrease) in house values during the sample period is 16.83% (-34.33%). I use zip code level data instead of house level data in my tests for two reasons: first is that house level data is measured with more error than zip code data because of sample size, and second is that a board of directors is unlikely to know the change in value of a CEO’s house, but it is reasonable to believe they would be aware of general trends in real estate prices. I use the top third of the housing market in each zip code to reflect the fact that CEO houses are generally more expensive than average, and likely to be in the top third for any given zip code.

9

Panel C of Table 1 presents statistics on variables I include in my tests as control variables. As control variables I use the firm’s return, the market return, firm size, the market-to-book ratio, and local unemployment. Firm size is defined as the real book assets of the firm. The market-to-book ratio is equity market value plus the book value of debt divided by firm assets. Local unemployment is the unemployment rate for the county the firm’s headquarters is located, as reported in the Local Area Unemployment table provided by the Bureau of Labor Statistics.4

3

Results

In this section I examine the effect of changing house prices on CEO compensation and equity incentives. As stated in Section 1, because of the effect of changing house prices on CEO wealth, a decrease in the value of a CEO’s house should lead to a decrease in the level of his equity ownership. As firms are unable to take back equity already granted to the CEO, this decrease must be the result of CEOs selling more shares when their house prices decrease. The board of directors must then not replace the shares sold by the CEO, leaving the overall level of equity ownership lower. To help CEOs in increasing diversification, boards of directors would also increase the cash compensation paid to the CEO. I use multivariate regression to test all but one of these hypotheses. As an equity grant cannot be negative, I use a tobit regression when it is the dependent variable. I regress changes in the variables on the percentage changes in house prices in the CEO’s home zip code, the ratio of home value to equity ownership, and an interaction between these two variables. To control for local factors affecting firm choices, such as weak local markets, I include the level of unemployment in the firm’s zip code at the end of the fiscal year. As large firms have more national, or international, markets, unemployment is more likely to control for factors specifically affecting smaller firms. I include the lagged level of equity ownership, and for regressions involving compensation variables, the lagged dependent variable. To 4

http://www.bls.gov/lau

10

control for other factors that affect compensation and incentives, I include the previous year’s level of equity ownership, the natural logarithm of firm size, the market-to-book ratio, the firm’s return, and the market return. Because of the small sample size I do not include industry fixed effects. With the short length of the sample, the market return captures much of the same information as year fixed effects, and so I do not include them either.

3.1

Full sample

Table 2 presents the results for the full sample. While the coefficient on the ZipTT is negative, not positive as predicted, it is statistically insignificant. Interestingly, there is weak evidence that CEOs whose houses are an important portion of their wealth received a slight increase in ownership if their house price did not change, as the coefficient on Home Value is positive and significant at the 10% level. The parameter estimate for the interaction term, which estimates the effect of changing prices as house value increases in importance to the CEO, is positive and statistically significant. So, for a CEO whose house is an important component of his wealth, a decrease in the value of his house is correlated with a decrease in his equity stake in the firm, which is the hypothesized effect. The economic magnitude of this result depends on the importance of the house to the CEO’s wealth. Using the parameter estimates for Home Value and the interaction of Home Value and ZipTT (ignore the coefficient on ZipTT, since it is insignificant), a CEO with the mean Home Value ratio who receives a one standard deviation from the mean negative shock to house value sees his ownership stake decrease by 0.08%. If he has the mean (median) amount of firm ownership, the decrease is worth $40,820 ($12,104). A CEO with one standard deviation above the mean Home Value who receives a one standard deviation below the mean shock to house value sees his ownership stake decrease by 0.56%, worth $284,407 ($84,336) at the mean (median) ownership level. The next two columns of Table 2 show how the CEO’s equity level decreases. In the regression for CEO purchases/sales of stock, the coefficient on the interaction term is positive 11

and significant. For a CEO whose house is a large portion of his equity wealth, a large decrease in the value of his house leads to a larger sale of stock. In the tobit regression of equity grants, a negative and significant coefficient on the interaction term indicates CEOs receive larger grants when their house value decreases and house value is a significant portion of their wealth. As the overall change in equity ownership is negative, the new equity grants do not fully replace the CEO’s sale of stock. The right-most two columns of Table 2 look at whether there is an effect on total or cash compensation when house values decrease. For the full sample there is no effect on either total compensation or on cash compensation. That neither of these two variables are affected suggests the increase in equity grants found above is small. If it were large, then either total compensation would increase and cash compensation stay the same, or cash pay would increase and total pay decrease. The main effect is that, when house prices decrease, a CEO with a significant investment of wealth in his house sells his shares, decreasing his exposure to the firm’s returns.

3.2

Subsamples

In this section I split the sample along two dimensions. First I split the sample into the largest and smallest firms. Second I split the sample into the group with the smallest ratio of house value to equity ownership and the largest. As compensation is generally higher at large firms, it seems likely that small firm CEOs are, on average, less wealthy than CEOs of large firms, and that their houses make up a larger portion of their personal wealth. If true, then the predicted effects of a decrease in house value would be larger for small firm CEOs. Table 3 repeats the analysis of Table 2 for the one-third smallest firms in the samlpe. 4 shows the results for the one-third largest firms in the sample. The large firm sample shows the same effects seen in the full sample, but with but with much larger statistical and economic significance, while the small firm sample is barely affected by changing house prices. 12

The large firm sample provides more evidence of what happens when house prices change. As stated above, the large firm sample shows the same behavior as the full sample, where when a CEO with substantial housing wealth sees a decline in property value his ownership of the firm decreases through equity sales. These CEOs receive larger equity grants than unaffected CEOs, but these grants do not offset the amount of stock sold. At this point for the full sample nothing more could be determined. For the large sample we see that total compensation has a negative and significant coefficient on the interaction term, and cash compensation has a positive and significant interaction term. So affected CEOs see their total compensation increase in reaction to a decrease in their house price. This increase in overall compensation comes entirely from the increase in equity grant, as cash compensation decreases. So here the hypothesis that cash compensation increases when house prices fall is unsupported, even though changes in overall compensation is higher for CEOs affected by the real estate crash. It is surprising that small firm CEOs appear to be unaffected by the housing market. Two possible explanations for this null result are the relation between wealth and firm size, and omitted variables. First, even though we expect CEOs of small firms to be less wealthy than CEOs of large firms, leading to their houses being a larger portion of their wealth, the correlation between Home Value and firm size is statistically insignificant (correlation of -0.056 with a p-value of 0.166). Second, small firms are more likely to be affected by harder to measure local economic conditions. My tests attempt to control for this by the inclusion of local area unemployment as a control variable. If a firm relies on its local market for a significant portion of its business, then local area unemployment captures changes in the local economy that affect demand for the firm’s services. If a firm does not rely on its local market, then local area unemployment should have no predictive power on my regressions. Unemployment has a significant coefficient in two of the small firm regressions and none in the large firm regressions, so it does appear to capture some of the state of the local economy, although it may not capture enough of the local variation to make for clean inference.

13

Table 5 reports the regression results for a sample of the smallest third Home Value ratio observations, and Table 6 reports the results for the largest third Home Value observations. The results here match our expectations, as the low Home Value subsample shows no significant effects from house price movements, while the high Home Value subsample has results similar to the large firms subsample. The main difference between the results for the high Home Value subsample and the large firm subsample is that the increase in total compensation is less statistically significant in the high Home Value subsample, and the change in cash compensation is not significant. Also, the magnitudes of the effects are larger for the large firm subsample.

4

Conclusion

This paper examines the effect of the housing crash of 2007-2010 on the level of CEO equity ownership and on CEO compensation. I provide evidence that CEO equity ownership is affected by shocks to the CEO’s wealth, as predicted by agency theory. As CEO wealth is not available for CEOs of U.S. firms, I provide evidence of a causal relation by showing how exogenouse shocks to the value of CEO primary residences affect CEO compensation. For CEOs whose primary residence is an important part of their wealth, a decrease in house value during the housing crash caused his ownership stake in the firm to decrease by more than a similar CEO who was unaffected by the crash (either because his house did not lose value, or his primary residence was not an important part of his wealth). The mechanism through which this decrease occurs is that the affected CEO sells more shares, which the board does not fully replace through new equity grants. The effect of changes in house value are larger for CEOs of large firms. In addition, the evidence from looking solely at large firms is that cash compensation decreases for CEOs most affected by the housing crash, even though total compensation increases.

14

Table 1: Descriptive Statistics This table displays the descriptive statistics of the variables used in tests of the paper’s hypotheses. The sample is based on a merge of Compustat, Execucomp, and hand-collected house price data for the years 2007 to 2009 at an annual frequency. Ownership ($) is the dollar value, in millions, of CEO equity ownership. Here equity ownership includes both straight equity and equity equivalence of executive stock options. Ownership (%) is the percentage of firm value owned by the CEO. Equity grant is the value, in millions, of new equity grants given to the CEO. Cash compensation is the value, in millions, of all cash payments to the CEO during the year. Total compensation is the sum of equity grants and cash compensation. Equity bought/sold is the value of stock sold by the CEO during the year. House price is the Zestimate, in millions, of the CEO’s house, while Home Value is the ratio of his house price to his equity ownership. ZipTT is the percentage change in value of the median home in the CEO’s home zip code. Local unemployment is the level of unemployment in the firm’s headquarters county.

Panel A. CEO Compensation Ownership ($) Ownership (%) Equity Grant ($) Cash Compensation ($) Total Compensation ($) Equity Bought/Sold ($)

Mean

S. D.

25%

50%

75%

N

50.999 2.676 2.684 2.327 5.011 -7.451

122.917 6.063 4.016 3.962 6.824 43.818

5.158 0.368 0.240 0.926 1.517 -6.661

15.123 0.870 1.366 1.459 3.119 -1.786

40.221 2.257 3.723 2.703 6.625 0.495

618 618 618 618 618 618

2.007 4.907 6.992

0.722 0.027 -8.161

1.300 0.082 -3.396

2.200 0.224 0.449

618 618 618

77,979.160 1.165 169.807 2.393

760.201 1.037 -32.453 4.805

2,210.840 1.301 -1.287 6.043

7,057.970 1.767 33.041 8.331

618 618 618 618

Panel B. CEO House Values House Price ($) 1.835 Home Value 0.822 ZipTT (%) -4.235

Panel C. Control Variables Book Value of Assets 12,463.190 Market-to-Book Ratio 1.633 Firm’s Return (%) 21.765 Local Unemployment (%) 6.610

15

Table 2: Multivariate Regression: Full Sample This table reports the results of estimating OLS (and, in one case, tobit) regressions of the variables of interest on changes in CEO house prices and controls. The different columns report the result for different dependent variables, listed at the top of each column. Each of these involves the change in the variable from 2007 to 2009. CEO Own is the change in the percent of the firm’s outstanding shares owned by the CEO. CEO Sales is the total proportion of the firm’s shares sold by the CEO. Equity Grants are the proportion of the firm’s shares granted to the CEO. Cash Pay is the change in cash compensation, as a percentage of total compensation. Total Pay is the change in total compensation (equity and cash). ZipTT is the percentage change in house prices for the top third of the CEO’s home zip code. Home Value is the ratio of the CEO’s house value to the value of his stock ownership at the beginning of the year. Unemployment is the percent of people unemployed in the firm headquarter’s county, firm return is the return for the year as given by CRSP, market return is the value-weighted market return reported in CRSP, and size is the natural logarithm of firm book value as reported in Compustat. Each column is an OLS regression with heteroscedasticity-robust standard errors except for Equity Grants, which is a tobit regression with a lower bound of 0. Statistical significance is indicated by ∗ , ∗∗ , and ∗∗∗ , which indicate significance levels of 10%, 5%, and 1%, respectively.

ZipTT Home Value ZipTT × HomeValue Lagged Ownership

CEO Own -0.269 (0.691) 0.019∗ (0.012) 0.194∗∗ (0.090) -0.063∗ (0.035)

CEO Sales -0.755 (0.875) 0.050∗∗ (0.020) 1.139∗∗∗ (0.31) -0.060∗ (0.035)

Equity Grants 0.385 (0.393) -0.016∗ (0.009) -0.505∗∗∗ (0.107) -0.012∗∗ (0.005)

-3.934 (3.779) -0.057 (0.060) 0.096∗∗∗ (0.026) 0.488∗∗∗ (0.187) -0.005 (0.030) 0.341 (0.470) 618 0.1024

-6.843∗ (4.088) -0.011 (0.060) 0.087∗∗∗ (0.032) 1.023∗∗∗ (0.238) 0.055 (0.033) -0.226 (0.49) 618 0.1335

2.831∗∗ (1.309) -0.037 (0.025) 0.007 (0.017) -0.535∗∗∗ (0.107) -0.056∗∗∗ (0.017) 0.475∗∗∗ (0.180) 618

Lagged Dependent Unemployment Market-to-Book Ratio Firm Return Market Return Size Intercept N Adj. R2

16

Cash Pay 8.561 (15.064) 0.336 (0.406) 2.925 (3.589) 0.482∗∗ (0.191) -0.623∗∗∗ (0.044) 26.678 (49.444) -2.593∗∗∗ (0.829) 0.503 (0.815) 11.398∗∗∗ (4.053) -2.233∗∗∗ (0.657) 55.47∗∗∗ (7.913) 618 0.3387

Total Pay 0.119 (0.284) -0.003 (0.003) -0.047 (0.031) -0.002 (0.002) -0.434∗∗∗ (0.062) 1.795∗∗ (0.728) 0.079∗ (0.047) 0.009 (0.009) -0.198∗∗∗ (0.058) -0.061∗∗∗ (0.011) 0.349∗∗ (0.159) 616 0.4149

Table 3: Subsample Regression: Small Firms This table reports the results of estimating OLS (and, in one case, tobit) regressions of the variables of interest on changes in CEO house prices and controls for the one-third smallest firms in the sample. The different columns report the result for different dependent variables, listed at the top of each column. Each of these involves the change in the variable from 2007 to 2009. CEO Own is the change in the percent of the firm’s outstanding shares owned by the CEO. CEO Sales is the total proportion of the firm’s shares sold by the CEO. Equity Grants are the proportion of the firm’s shares granted to the CEO. Cash Pay is the change in cash compensation, as a percentage of total compensation. Total Pay is the change in total compensation (equity and cash). ZipTT is the percentage change in house prices for the top third of the CEO’s home zip code. Home Value is the ratio of the CEO’s house value to the value of his stock ownership at the beginning of the year. Unemployment is the percent of people unemployed in the firm headquarter’s county, firm return is the return for the year as given by CRSP, market return is the value-weighted market return reported in CRSP, and size is the natural logarithm of firm book value as reported in Compustat. Each column is an OLS regression with heteroscedasticity-robust standard errors except for Equity Grants, which is a tobit regression with a lower bound of 0. Statistical significance is indicated by ∗ , ∗∗ , and ∗∗∗ , which indicate significance levels of 10%, 5%, and 1%, respectively.

ZipTT Home Value ZipTT × HomeValue Lagged Ownership

CEO Own 1.166 (2.3) 0.029∗ (0.015) -0.343 (0.385) -0.059 (0.039)

CEO Sales 1.519 (2.36) 0.032∗∗ (0.016) -0.561 (0.401) -0.054 (0.039)

Equity Grants -0.461 (0.693) 0.001 (0.011) 0.308 (0.282) -0.013∗∗∗ (0.005)

-11.817 (11.504) -0.093 (0.101) 0.132∗∗ (0.054) 0.427 (0.556) 0.112 (0.189) 0.216 (1.829) 206 0.086

-17.062 (11.455) -0.066 (0.098) 0.148∗∗ (0.062) 0.94 (0.593) 0.199 (0.193) -0.309 (1.836) 206 0.101

6.769∗∗∗ (2.145) -0.014 (0.027) -0.019 (0.019) -0.628∗∗∗ (0.175) -0.066 (0.051) 0.222 (0.374) 206

Lagged Dependent Unemployment Market-to-Book Ratio Firm Return Market Return Size Intercept N Adj. R2

17

Cash Pay 51.25∗∗ (23.724) -0.081 (0.132) -16.676∗∗∗ (3.857) 0.438∗∗ (0.208) -0.63∗∗∗ (0.067) -109.75 (95.264) -2.527∗∗ (1.071) 1.731∗∗∗ (0.517) 8.379 (7.208) -3.298 (2.043) 71.605∗∗∗ (17.545) 206 0.37

Total Pay -0.147 (1.098) -0.009∗ (0.005) -0.21 (0.19) -0.004 (0.004) -0.49∗∗∗ (0.12) 5.71∗∗ (2.307) 0.106 (0.108) 0.011 (0.009) -0.418∗∗ (0.193) -0.237∗∗∗ (0.085) 1.123∗ (0.579) 206 0.481

Table 4: Subsample Regression: Large Firms This table reports the results of estimating OLS (and, in one case, tobit) regressions of the variables of interest on changes in CEO house prices and controls for the one-third largest firms in the sample. The different columns report the result for different dependent variables, listed at the top of each column. Each of these involves the change in the variable from 2007 to 2009. CEO Own is the change in the percent of the firm’s outstanding shares owned by the CEO. CEO Sales is the total proportion of the firm’s shares sold by the CEO. Equity Grants are the proportion of the firm’s shares granted to the CEO. Cash Pay is the change in cash compensation, as a percentage of total compensation. Total Pay is the change in total compensation (equity and cash). ZipTT is the percentage change in house prices for the top third of the CEO’s home zip code. Home Value is the ratio of the CEO’s house value to the value of his stock ownership at the beginning of the year. Unemployment is the percent of people unemployed in the firm headquarter’s county, firm return is the return for the year as given by CRSP, market return is the value-weighted market return reported in CRSP, and size is the natural logarithm of firm book value as reported in Compustat. Each column is an OLS regression with heteroscedasticity-robust standard errors except for Equity Grants, which is a tobit regression with a lower bound of 0. Statistical significance is indicated by ∗ , ∗∗ , and ∗∗∗ , which indicate significance levels of 10%, 5%, and 1%, respectively.

ZipTT Home Value ZipTT × HomeValue Lagged Ownership

CEO Own -0.736 (0.8) 0.052∗∗∗ (0.017) 0.691∗∗∗ (0.172) -0.081 (0.065)

CEO Sales -1.329 (0.839) 0.185∗∗∗ (0.03) 2.745∗∗∗ (0.253) -0.079 (0.066)

Equity Grants 0.54∗ (0.308) -0.099∗∗∗ (0.015) -1.341∗∗∗ (0.134) -0.008 (0.01)

-1.677 (1.85) 0.026 (0.042) 0.319∗∗∗ (0.075) 0.257 (0.183) -0.005 (0.033) 0.044 (0.367) 206 0.134

-1.928 (1.904) 0.075 (0.05) 0.419∗∗∗ (0.089) 0.381∗ (0.195) 0.043 (0.038) -0.61 (0.404) 206 0.43

0.577 (0.965) -0.042 (0.034) -0.079∗ (0.041) -0.179∗∗ (0.087) -0.048∗∗ (0.02) 0.6∗∗∗ (0.222) 206

Lagged Dependent Unemployment Market-to-Book Ratio Firm Return Market Return Size Intercept N Adj. R2

18

Cash Pay -0.868 (24.091) 2.41∗∗∗ (0.409) 24.19∗∗∗ (6.259) 1.154 (0.708) -0.716∗∗∗ (0.088) -0.823 (75.123) -5.03∗∗ (2.362) 3.973 (3.068) 8.782 (7.365) -2.83∗ (1.655) 69.529∗∗∗ (20.076) 206 0.385

Total Pay 0.081 (0.058) -0.007∗∗∗ (0.002) -0.095∗∗∗ (0.018) 0.002 (0.001) -0.815∗∗∗ (0.087) 0.015 (0.155) 0.034∗∗∗ (0.007) 0.005 (0.007) -0.033∗∗ (0.015) -0.021∗∗∗ (0.004) 0.213∗∗∗ (0.044) 205 0.633

Table 5: Subsample Regression: Low House Value to CEO Ownership This table reports the results of estimating regressions of the variables of interest on changes in CEO house prices and controls for the CEOs in the bottom one-third of the Home Value to CEO ownership distribution. The different columns report the result for different dependent variables, listed at the top of each column. Each of these involves the change in the variable from 2007 to 2009. CEO Own is the change in the percent of the firm’s outstanding shares owned by the CEO. CEO Sales is the total proportion of the firm’s shares sold by the CEO. Equity Grants are the proportion of the firm’s shares granted to the CEO. Cash Pay is the change in cash compensation, as a percentage of total compensation. Total Pay is the change in total compensation (equity and cash). ZipTT is the percentage change in house prices for the top third of the CEO’s home zip code. Home Value is the ratio of the CEO’s house value to the value of his stock ownership at the beginning of the year. Unemployment is the percent of people unemployed in the firm headquarter’s county, firm return is the return for the year as given by CRSP, market return is the value-weighted market return reported in CRSP, and size is the natural logarithm of firm book value as reported in Compustat. Each column is an OLS regression with heteroscedasticity-robust standard errors except for Equity Grants, which is a tobit regression with a lower bound of 0. Statistical significance is indicated by ∗ , ∗∗ , and ∗∗∗ , which indicate significance levels of 10%, 5%, and 1%, respectively.

ZipTT Home Value ZipTT × HomeValue Lagged Ownership

CEO Own -1.628 (3.904) 4.103 (10.999) 41.92 (138.426) -0.032 (0.038)

CEO Sales -1.686 (3.893) 2.318 (10.983) 45.473 (139.584) -0.03 (0.038)

Equity Grants 0.082 (0.326) 2.503∗ (1.284) -3.959 (14.19) -0.005∗∗∗ (0.002)

-16.989 (11.601) 0.047 (0.202) 0.352 (0.705) -0.072 (0.563) 0.067 (0.074) -0.03 (1.256) 206 0.03

-17.105 (11.576) 0.065 (0.201) 0.443 (0.699) -0.055 (0.564) 0.093 (0.075) -0.351 (1.257) 206 0.031

0.354 (0.596) -0.006 (0.014) -0.123∗∗∗ (0.036) -0.018 (0.054) -0.024∗∗∗ (0.008) 0.249∗∗ (0.104) 206

Lagged Dependent Unemployment Market-to-Book Ratio Firm Return Market Return Size Intercept N Adj. R2

19

Cash Pay 13.845 (47.134) -386.137∗∗∗ (163.204) 246.318 (1805.71) 0.325 (0.255) -0.603∗∗∗ (0.081) 44.112 (89.217) -5.017∗∗ (2.261) 9.747∗∗ (4.542) -1.141 (6.87) -3.277∗∗∗ (1.106) 72.734∗∗∗ (17.327) 206 0.334

Total Pay -0.062 (0.297) 0.358 (1.195) 1.476 (11.521) -0.002 (0.003) -0.546∗∗∗ (0.142) 0.987 (1.03) 0.041 (0.026) -0.072 (0.053) -0.013 (0.069) -0.05∗∗∗ (0.017) 0.395∗ (0.22) 206 0.285

Table 6: Subsample Regression: High Home Value to CEO Ownership This table reports the results of estimating regressions of the variables of interest on changes in CEO house prices and controls for the CEOs in the top one-third of the Home Value to CEO ownership distribution. The different columns report the result for different dependent variables, listed at the top of each column. Each of these involves the change in the variable from 2007 to 2009. CEO Own is the change in the percent of the firm’s outstanding shares owned by the CEO. CEO Sales is the total proportion of the firm’s shares sold by the CEO. Equity Grants are the proportion of the firm’s shares granted to the CEO. Cash Pay is the change in cash compensation, as a percentage of total compensation. Total Pay is the change in total compensation (equity and cash). ZipTT is the percentage change in house prices for the top third of the CEO’s home zip code. Home Value is the ratio of the CEO’s house value to the value of his stock ownership at the beginning of the year. Unemployment is the percent of people unemployed in the firm headquarter’s county, firm return is the return for the year as given by CRSP, market return is the value-weighted market return reported in CRSP, and size is the natural logarithm of firm book value as reported in Compustat. Each column is an OLS regression with heteroscedasticity-robust standard errors except for Equity Grants, which is a tobit regression with a lower bound of 0. Statistical significance is indicated by ∗ , ∗∗ , and ∗∗∗ , which indicate significance levels of 10%, 5%, and 1%, respectively.

ZipTT Home Value ZipTT × HomeValue Lagged Ownership

CEO Own -0.754 (0.680) 0.015 (0.010) 0.189∗∗ (0.083) -0.145∗ (0.087)

CEO Sales -0.249 (1.171) 0.041∗∗ (0.019) 1.035∗∗∗ (0.312) -0.205∗∗ (0.091)

Equity Grants -0.81 (0.713) -0.011 (0.009) -0.404∗∗∗ (0.112) 0.047 (0.038)

-0.748 (2.486) -0.013 (0.028) 0.081∗∗∗ (0.019) 0.529∗∗∗ (0.188) -0.019 (0.038) 0.304 (0.363) 206 0.183

-7.658 (5.034) -0.018 (0.036) 0.071∗∗∗ (0.027) 1.65∗∗∗ (0.441) -0.006 (0.05) 0.468 (0.551) 206 0.299

6.145∗∗∗ (2.23) 0.008 (0.04) 0.012 (0.018) -1.024∗∗∗ (0.185) -0.009 (0.036) -0.237 (0.353) 206

Lagged Dependent Unemployment Market-to-Book Ratio Firm Return Market Return Size Intercept N Adj. R2

20

Cash Pay 46.929∗ (25.363) 0.408 (0.405) 2.999 (4.191) -0.31 (1.396) -0.815∗∗∗ (0.083) -8.529 (89.618) -3.173∗∗∗ (0.929) -0.104 (0.896) 16.121∗∗ (6.609) -2.923∗∗ (1.468) 77.872∗∗∗ (16.538) 206 0.43

Total Pay 0.081 (0.343) -0.003 (0.004) -0.059∗ (0.035) 0.036 (0.038) -0.427∗∗∗ (0.145) 1.871 (1.396) 0.181∗∗∗ (0.066) 0.005 (0.008) -0.222∗∗ (0.104) -0.057∗∗ (0.023) 0.138 (0.274) 205 0.351

References Babenko, I., B. Bennett, J. M. Bizjak, J. L. Coles, 2012. Clawback provisions. Working paper, Available at SSRN: http://ssrn.com/abstract=2023292. Baker, G. P., B. J. Hall, 2004. CEO incentives and firm size. Journal of Labor Economics 22(4), 767–798. Becker, B., 2006. Wealth and executive compensation. Journal of Finance 61(1), 379–397. Bettis, J. C., J. L. Coles, M. L. Lemmon, 2000. Corporate policies restricting trading by insiders. Journal of Financial Economics 57(2), 191–220. Buiter, W. H., 2008. Lessons from the North Atlantic financial crisis. Center for Economic Policy Research No. 18. Bureau, U. C., 2004. Net Worth and Asset Ownership of Households: 2004. http://www. census.gov/hhes/www/wealth/wealth.html. Calomiris, C. W., S. D. Longhofer, W. Miles, 2012. The housing wealth effect: The crucial roles of demographics, wealth distribution and wealth shares. NBER working paper 17740. Core, J. E., W. R. Guay, 2010. Is CEO pay too high and are incentives too low: A wealthbased contracting framework. Academy of Management Perspectives 24(1), 5–19. Dittman, I., E. Maug, 2007. Lower salaries and no options? On the optimal structure of executive pay. Journal of Finance 62(1), 303–343. Lambert, R. A., D. F. Larcker, R. E. Verrecchia, 1991. Portfolio considerations in valuing executive compensation. Journal of Accounting Research 29(1), 129–149. Liu, C., D. Yermack, 2007. Where are the shareholders’ mansions? CEOs’ home purchases, stock sales, and subsequent company performance. NYU Working Paper No. FIN-07-042. Available at SSRN: http://ssrn.com/abstract=1300781. 21

Neyland, J., 2012. Wealth shocks and executive compensation: Evidence from CEO divorce. Working paper, Available at SSRN: http://ssrn.com/abstract=2140668.

22