We construct three measures of overconfidence, or proxies for Δ, based on the personal portfolio decisions of CEOs: Holder 67, Longholder, and Net Buyer. The first two measures, Holder 67 and Longholder, use the timing of option exercises to identify overconfidence. The third measure, Net Buyer, uses the habitual acquisition of company stock. Our measures exploit the underdiversification of CEOs. CEO compensation contracts regularly contain large quantities of stock and option grants. To maximize the incentive effects of these holdings, the options cannot be traded and the sale of stock may be restricted. Further, firms prohibit CEOs from perfectly hedging against the risk by short-selling company stock. Most importantly, CEOs' human capital is invested in their firms, so that a bad outcome in the firm will not only negatively impact their personal portfolios, but will also reduce their outside employment options. All of these effects leave CEOs highly exposed to the idiosyncratic risk of their company.
Unlike perfectly hedged outside investors, then, CEOs must trade off the option value of holding stock options against the costs of underdiversification. Though the optimal schedule for exercise depends on their individual wealth, degree of risk aversion, and diversification (Hall and Murphy (2002)), it is generally true that risk aversion and underdiversification predict early exercise of executive options. Similarly, underdiversified CEOs, in order to divest themselves of idiosyncratic risk, should minimize their holdings of company stock. Overconfidence, however, may lead CEOs to overestimate the future returns of their investment projects. Therefore, they believe that the stock prices of their companies will continue to rise under their leadership more than they objectively should expect. As a result, overconfidence induces them to postpone option exercise or even to buy additional company stock in order to benefit personally from the expected future gains.
These arguments may not hold for CEOs who display overoptimism about exogenous variables or overestimate the precision of their beliefs. Overoptimism about the general economic environment need not induce investment in the CEOs' own companies. Miscalibration reduces the expected volatility of the stock and thus the value of holding options. The key aspects of overconfidence for our results are overestimation of skill and self-attribution by the CEO.
Holder 67. For our first measure, we consider the status of each individual option package in our sample at the end of the vesting period. To maintain comparability across packages with vesting periods of different duration, we examine the first year in which all of the packages in the sample are at least partially exercisable, year 5. We then compute the percentage in-the-money for each package. Risk aversion and underdiversification predict that CEOs should exercise options immediately after the vesting period if the amount in-the-money is beyond a rational benchmark. We use the Hall and Murphy (2002) framework as a theoretical guide in choosing a reasonable benchmark for the percentage in-the-money at or above which CEOs should exercise newly vested options. We also follow their calibrations to proxy for (unobserved) CEO wealth, diversification, and risk aversion. We apply the resulting threshold for early exercise to the full sample of CEOs (rather than attempting to calibrate individually) and check the robustness of our results to a wide range of reasonable parameter values.
To begin, we take 67% in-the-money during the fifth year as our threshold. If an option is more than 67% in-the-money at some point in year 5, the CEO should have exercised at least some portion of the package during or before the fifth year. This threshold corresponds to a risk aversion of three in a constant relative risk-aversion (CRRA) utility specification and to a percentage of wealth in company equity equal to 66. We then construct Holder 67 as follows. We consider the subsample of CEOs who at least twice during the sample period had options that were valued above the threshold during the fifth year. We then identify the first instance (if any) at which the CEO failed to exercise such an option during or before the fifth year. From this point in time onward, we classify the CEO as overconfident if he subsequently exhibits the same behavior at least one more time during his tenure as CEO. As we are interested in a “permanent” rather than “transitory” overconfidence effect, our measure targets CEOs who “habitually” exercise options late. We repeat this exercise starting at 50% in-the-money and incrementing by five up to 150% in-the-money to verify the robustness of our results to variations in the parameters (e.g., 100% corresponds to ρ= 3; 50% of wealth in stock).
Whenever we use these Holder measures in an estimation, we restrict the sample to CEOs who at least twice during the sample period had options that were valued above the threshold during the fifth year. This restriction guarantees that every CEO in the subsample had the opportunity to be classified as overconfident and, thus, limits the degree of unobserved overconfidence in the control group. It also guarantees that we are not overproportionally identifying CEOs as overconfident when the stock of their firm is doing well. On the other hand, our restriction considerably limits the number of observations; for Holder 67, it declines from 3,728 to 1,058.
Longholder. To construct our second measure, Longholder, we focus on the expiration date of option packages rather than the end of the vesting period. We classify a CEO as overconfident (for all of his years in the sample) if he ever holds an option until the last year of its duration. As the typical option in the sample has 10 years' duration and is fully vested (at the latest) by year 5, the CEO chooses to hold, rather than exercise, the option for at least 5 years. Thus, again, our measure targets habitual failure to diversify, or a personality, rather than time-varying, overconfidence effect. Further, over 85% of options that are held until their final year are in-the-money and the median percentage in-the-money for such options is 253%. Thus, the CEO could have profitably exercised these options before their last year. Indeed, failure to exercise these options prior to expiration is difficult to reconcile with any reasonable calibration of the Hall and Murphy framework. Therefore, the Longholder measure alleviates the dependence on calibrated thresholds for rational exercise. It also enables us to circumvent the sample restriction (and potential loss of power) of the Holder 67 measures. It is thus best suited for tests in which we must reduce or split the sample (e.g., Prediction 2).
Net Buyer. To construct our third measure, Net Buyer, we exploit the tendency of some CEOs to purchase additional company stock despite their already high exposure to company risk. Specifically, we consider the subsample of CEOs who keep their position as CEO for at least 10 of the 15 years in our sample. We identify CEOs as overconfident if they were net buyers of company equity during their first five years in our sample,11 that is, if they bought stock on net in more years than they sold on net during their first five sample years.12 Whenever we use this measure in an estimation, we exclude the first five years of the CEOs' tenures. Thus, we use disjoint subsamples of CEO years to establish overconfidence and to measure its potential effects on investment. This approach ameliorates endogeneity concerns, but at the cost of (again) reducing sample size substantially—here to 842.
In the middle and right columns of Table I, we show firm and CEO summary statistics for subsamples of firm-years determined by overconfidence sample restrictions and measured overconfidence. For brevity, we use only Holder 67 in these comparisons. Thus, all firm-years included in either subsample are from firms with CEOs who meet the selection criterion of having at least two options that exceeded the 67% threshold in year 5. The patterns are similar for the other overconfidence measures.13
Note that of the 113 CEOs who meet the selection criterion, 58 display overconfidence in their personal portfolio decisions. By comparison, 85 of 661 CEOs are overconfident using the Longholder measure and 97 of 158 using Net Buyer. Thus, the measure ensuring the largest sample size, Longholder, is also the most conservative measure of overconfidence. There are no significant differences in observable firm or CEO characteristics across the overconfident and control samples. In particular, the distribution of firms across the Fama–French industry groups is virtually identical in the overall sample, the subsample satisfying the selection criterion, and the overconfident subsample. Thus, our overconfidence measure appears orthogonal to firm characteristics, at least as measured at the industry level.
In Table II, we present the pairwise correlations among the three overconfidence measures and CEO and firm characteristics. We consider the overconfidence measures two at a time due to differences in the subsamples on which we apply them. Panel A presents the correlations for Longholder and Holder 67 on the Holder 67 subsample; Panel B presents the correlations for Longholder and Net Buyer on the Net Buyer subsample; and, Panel C presents the correlations for Net Buyer and Holder 67 on the intersection of the Net Buyer and Holder 67 subsamples. The correlation between Longholder and Holder 67 is strong and highly significant (0.2472). The correlation between Net Buyer and Holder 67 is also positive, though weaker (0.0628). There is only a small positive correlation between Net Buyer and Longholder. Overall, these relations suggest that our measures are capturing the same effect. We find few reliable correlations between our overconfidence measures and firm or CEO characteristics. The relations often reverse for different overconfidence measures or for the same measure on different subsamples of data, suggesting that overconfidence is not proxying for firm characteristics or observable CEO characteristics.
Table II. Correlations Sample A: CEOs with options more than 67% in-the-money in the fifth year at least two times. Sample B: CEOs with at least 10 years in the data and no more than one missing observation of ownership in the first five years. Data from the first five years of each CEO's tenure excluded. Sample C: CEOs with options more than 67% in-the-money in the fifth year at least two times and with at least 10 years in the data and no more than one missing observation of ownership in the first five years. Data from the first five years of each CEO's tenure excluded.
|Overconfidence and Firm Characteristics|
| ||Holder 67||Longholder||Net Buyer||Size||Q||CF/k||Stock Ownersh.||Vested Options||Corp. Gov.|
|Holder 67||1.00||n/a||1.00|| |
|Net Buyer||n/a||n/a||0.06||n/a||0.01||n/a||n/a||1.00||1.00|| |
|Stock ownership||−0.10||n/a||−0.18||−0.07||−0.19||n/a||n/a||−0.20||−0.18||−0.20||−0.12||−0.13||−0.01||−0.01||−0.09||0.06||0.11||0.00||1.00||1.00||1.00|| |
|Vested options||0.10||n/a||0.06||0.17||0.30||n/a||n/a||−0.11||−0.15||−0.31||−0.26||−0.37||0.18||0.18||0.20||0.25||0.27||0.34||0.10||−0.07||0.01||1.00||1.00||1.00|| |
|Overconfidence and CEO Characteristics|
| ||Holder 67||Longholder||Net Buyer||Depression Baby||CEO/Pres/Chm.||Finance Career||Techn. Career||Tenure|
|“Depression baby”||−0.06||n/a||−0.15||−0.03||−0.07||n/a||n/a||0.04||0.04||1.00||1.00||1.00|| |
|CEO & President & Chairman||0.04||n/a||0.05||−0.04||−0.06||n/a||n/a||0.13||0.07||−0.01||−0.02||−0.10||1.00||1.00||1.00|| |
|Finance career||−0.09||n/a||−0.16||−0.10||−0.13||n/a||n/a||0.08||0.29||−0.05||0.06||−0.03||−0.14||−0.12||−0.19||1.00||1.00||1.00|| |
|Technical career||−0.01||n/a||0.02||−0.12||−0.05||n/a||n/a||−0.25||−0.26||0.10||0.07||0.15||0.07||0.01||0.04||−0.15||−0.15||−0.07||1.00||1.00||1.00|| |
Of particular interest are the relations among stock ownership, vested options, and our overconfidence measures. Mechanistically, an overconfident CEO who delays option exercise or purchases additional company stock will increase his holdings. However, other factors such as firm size, firm age, corporate governance, and tenure, are substantially more important in determining the level of ownership. Overall, then, it is not surprising that there is no consistent correlation between stock or option ownership and the overconfidence measures. Further, the board typically grants stock and options to confer incentives on the CEO. However, traditional agency theory suggests that the incentive effect of stock and options will reduce investment–cash flow sensitivity—the opposite prediction of our overconfidence model. Thus, simple measures of stock and option ownership are not adequate to capture CEO overconfidence, particularly in this context.
Before turning to the effects of overconfidence on investment, we briefly consider some alternative explanations of the measures.
1. Inside information. A CEO may decide not to decrease exposure to company risk because of private information about future stock prices that makes holding options or buying stock attractive. Inside information also predicts investment–cash flow sensitivity. Since the information has not been incorporated into the market price, the firm's stock is undervalued and investment may be sensitive to cash flow for the usual Myers–Majluf reasons.
One of the key distinctions between overconfidence and information is persistence. Positive information is most naturally viewed as a transitory, rather than fixed effect. Information with enough precision to justify increasing exposure to company risk is likely to be short term, and it is unlikely that the same CEO would repeatedly receive positive draws. Our overconfidence measures, on the other hand, target a habitual tendency to hold too much company risk (or a fixed overconfidence effect). Holder 67 requires that a CEO fail to exercise options that are beyond the threshold at least twice. Longholder requires that a CEO fail to exercise options for at least five years. Net Buyer requires that the factor leading the CEO to purchase additional company stock still affects investment decisions in a disjoint future time period.
Still, Holder 67 places no restriction on how long the CEO must hold the option beyond the fifth year and, thus, could potentially capture short-term delays in option exercise. To distinguish the two stories for this measure, we analyze the exercise behavior of CEOs over time. Under the inside information hypothesis, we would expect a CEO to sometimes hold his options (when he has positive inside information) and to sometimes exercise them early (when he has negative inside information). To test this prediction, we run a random effects probit regression of the probability that a CEO holds an option that is at least 67% in-the-money in the fifth year on the number of times that the CEO has held such an option in the past.14 The sample consists of the 759 firm-years in which a CEO had options beyond the 67% threshold in the fifth year after the grant date. The dependent variable is equal to one whenever the CEO did not exercise any portion of those options prior to that year. Panel A of Table III presents the regression results. In column 1, the coefficient of “past late exercises” is positive (0.2493) and highly significant (z = 4.40). This implies that the typical CEO, rather than varying his exercise behavior over time, either persistently holds options beyond the 67% threshold or persistently exercises early. The results are robust to the inclusion of Q (column 2) and the firm's earnings–price ratio (column 3) as controls. Consistent with Jenter (2002), high values of Q—perhaps, in this context, a proxy for market overvaluation—appear to decrease the probability of late exercise. The earnings–price ratio effect, on the other hand, appears to go the wrong direction for an overvaluation story (though it is statistically insignificant). Panel B of Table III shows the percentage of CEOs who hold an option that is 67% in-the-money divided into categories based on the number of past late exercises. Overall, the results suggest that the number of times a CEO has held a 67%, in-the-money option in the past is considerably more important in determining the CEO's future exercise behavior than any information about current or future stock price performance—an indication of a personal fixed effect on option exercise decisions.
Table III. Persistence of Exercising Behavior In Panel A, the dependent variable is a dummy variable taking the value one if the CEO fails to exercise a five-year-old option that reaches at least 67% in-the-money in the current period. Past late exercises are the number of times that the CEO has exercised such options late in the past. Q is the market value of assets over the book value of assets at the beginning of the year. Earnings/Price ratio is the minimal earnings to price ratio during the fiscal year. Panel B presents statistics on late exercises of stock options partitioned by the number of past late exercises by the CEO in question.
|Panel A. Random Effects Probit Regression|
|Sample: Observations with 67% in-the-Money Options (in Year Five)|
|Past late exercises||0.2493||0.2569||0.2571||0.266|
|Q|| ||−0.1519 (1.79)*|| ||−0.1514 (1.81)*|
|Earnings/price ratio|| ||−0.709 (0.77)||−0.8128 (0.89)|
|Number of CEOs||278||273||272||271|
|Panel B. Percent of “Late Exercisers” Partitioned by Number of Past Late Exercises|
|Sample: Observations with 67% in-the-Money Options (in Year Five)|
|Past Late Exercises||% Who Exercise Late||Number of CEOs|
The second key distinction between overconfidence and information is performance. If positive information is the true reason for not diversifying the personal portfolio, then CEOs who exhibit this behavior should earn positive abnormal returns over a strategy of diversification. We calculate the distribution of returns among all CEOs who had options beyond the 67% threshold, but did not exercise. We compare those returns to the (hypothetical) returns from exercising the options during the fifth year and investing the proceeds in the S&P 500. As we do not know the exact price at which CEOs exercised their options, we calculate the returns under three alternative assumptions. First, we assume that CEOs are able to perfectly time the market in the short run and exercise at the maximum price during the fiscal year of their actual and hypothetical exercise. Then, as alternatives, we consider exercise at the mean or median price during the year. Table IV shows that on average CEOs do not beat the market by holding options beyond the threshold. The return differentials are small with huge standard deviations under any assumption about exercise behavior. Similarly, the average CEO does not consistently beat the market by holding options: The average CEO who holds beyond the 67% threshold beats the S&P 500 only 45.86% of the time. The results are similar for the Longholder measure. We find that CEOs whose options were at least 40% in-the-money in year 9 (i.e., past the relevant Hall and Murphy threshold for that year) and who did not exercise would have been better off on average had they exercised and invested in the S&P 500 (though the mean is not significant). This result also holds when we consider the alternative strategies of exercising in the eighth, seventh, sixth, or fifth years with appropriate adjustments to the threshold (Malmendier and Tate (2003)). Thus, there is no evidence that positive information, on average, motivates CEOs who hold options beyond the theoretically motivated thresholds.
Table IV. Distribution of Returns of “Late Exercisers” (67%, Fifth Year) The table presents data on the returns of late exercising CEOs (Holders 67) by percentiles. The first column presents the percentage in-the-money at the maximum price during the fifth fiscal year from grant date for each option package that is held beyond the 67% threshold. The second, third, and fourth columns present the returns (in %) relative to exercising the options during year 5 and investing instead in the S&P 500, assuming exercise at the maximum, mean, and median stock prices during the fiscal year, respectively. We also present the last percentile for which the return is negative under each price assumption. All returns are annualized. Sample: CEOs who have option packages at least 67% in-the-money in the fifth year after the option grant and who have not exercised the options before the fifth year.
|Percentage in-the-Money in Year 5||Return (in %) Relative to Exercising during Year 5 and Investing in S&P 500|
|Exercise at Fiscal-Year Maximum Price||Exercise at Fiscal-Year Mean Price||Exercise at Fiscal-Year Median Price|
|Percentile||%in the Money||Percentile||Return||Percentile||Return||Percentile||Return|
|40th||366.88||40th 46th||−2.79 −0.66||40th 46th||−2.50 −0.38||40th 49th||−2.56 −0.05|
|Mean||1,275.90|| ||3.60|| ||4.85|| ||3.57|
|Standard deviation||3,336.66|| ||20.23|| ||20.96|| ||21.15|
|Observations||182|| ||182|| ||182|| ||182|
|CEOs||86|| ||86|| ||86|| ||86|
2. Signaling. Another reason why CEOs may want to hold company risk is to convey a (potentially) costly signal to the capital market that their firm's prospects are better than the prospects of similar firms. However, the most natural version of the signaling story would not predict heightened investment–cash flow sensitivity. Signaling should alleviate informational asymmetries and, thus, eliminate investment–cash flow sensitivity among the firms in which CEOs hold their options. Thus, the tests of our investment predictions themselves will help to dispel this alternative explanation.
Moreover, the usefulness of option exercises as a signaling device is doubtful. Financial services firms and the financial press, while following stock purchases and sales of insiders closely, generally discount option exercises as signals of future stock prices. They point to vesting and expiration times as the main determinants of exercise.15 Our only stock-based proxy for overconfidence, Net Buyer, measures overconfidence and investment–cash flow sensitivity for two disjoint time periods to specifically eliminate the influence of signaling.
The same arguments address the hypothesis that CEOs use their insider trades opportunistically to raise the capital market valuation of their firms.
3. Risk tolerance. Alternatively, one might want to interpret our measures of overconfidence as measures of risk attitude. A CEO may hold his options beyond the threshold simply because he is less risk averse and, therefore, less affected by underdiversification. If anything, however, lower risk aversion should predict lower investment–cash flow sensitivity since less-risk-averse managers should be more willing to lever up the firm. Thus, once again, the tests of the empirical predictions of our model will provide evidence against this story. Moreover, while higher risk tolerance induces option holders to exercise their options later, it does not imply that the CEO should habitually buy additional stock of his company. Indeed, to explain the Net Buyer measure, CEOs would need to be risk seeking on average.
4. Tax reasons. An option holder may postpone exercise to delay the payment of taxes on his profits. Personal income tax deferral, however, would not predict higher sensitivity of investment to cash flow among holders, nor does it apply to additional stock purchases.
5. Procrastination. Finally, CEOs might hold options until expiration if they are “inertial” in the sense of O'Donoghue and Rabin (2001): Inertia on their personal account may carry over to the corporate account of the firm in a reluctance to conduct equity issues, or more generally, to a preference for the “quiet life” (Bertrand and Mullainathan (2003)). We find, however, that more than 8% of the CEOs classified as overconfident under the Longholder measure conduct other transactions on their personal portfolios in the two years prior to the year their “longheld” option expires. We also find in Malmendier and Tate (2003) that Longholder CEOs are significantly more likely to conduct acquisitions than their peers and, thus, do not appear to procrastinate on the corporate account. Finally, an inertial CEO should not habitually purchase company equity and, so, should not be classified as overconfident under the Net Buyer measure.16
We take further steps to distinguish overconfidence from alternative explanations (particularly information) in the context of the investment regressions that follow.