Effects of Executive Compensation Complexity on Investor Behaviour in an Experimental Stock Market

^†Preliminary work related to this study appeared in Welker (2012). The authors appreciate the helpful comments and suggestions from Eddy Cardinaels (editor) and an anonymous reviewer.

Abstract

This study experimentally investigates whether shareholders correctly anticipate the incentive effects of increasingly complex compensation packages given to managers, and whether potential biases in individual shareholder beliefs carry over to price and volume effects in a stock market. In the experiment, a manager makes a decision about the stochastic dividend process generating the firm's fundamental value, and shareholders make estimations about fundamental value and trade shares in a stock market. The manager's action is either known or hidden and, nested in the hidden-action condition, compensation is either simple or complex. We hypothesise that hidden action gives rise to perceived behavioural uncertainty, and that this uncertainty has valuation relevance in that it affects both individual estimations of fundamental firm value and stock prices. Furthermore, we hypothesise that perceived behavioural uncertainty increases with the complexity of compensation, and that compensation complexity also affects trading volume. We find supporting evidence for our conjectures. Both estimations and market prices are biased when the manager's action is hidden, and biases are stronger when compensation is complex. We further find that trading volume decreases when compensation is complex, even though the heterogeneity of investors' individual beliefs increases.

Notes

^† Preliminary work related to this study appeared in Welker (2012). The authors appreciate the helpful comments and suggestions from Eddy Cardinaels (editor) and an anonymous reviewer.

¹ To our knowledge, prior research has not explicitly addressed a principal's anticipation of an agent's behaviour after a contract has been signed. There is, however, experimental research that analyses a principal's choices between alternative formal control instruments, which allows indirect inferences with respect to the principals’ expectations. Examples are Fehr and Schmidt (2004) and Falk and Kosfeld (2006).

² Experimental asset market studies, starting with Plott and Sunder (1982), usually incorporate information asymmetries between traders into the analysis of prices and trading volumes; see, e.g. Plott and Sunder (1988), Copeland and Friedman (1987, 1991), Bloomfield (1996), Theissen (2000), and Bloomfield et al. (2009).

³ We did not refer to ambiguity in deriving Hypotheses 1 and 2 for two reasons: First, we expect ambiguity aversion to have no effect on estimations, as the estimation task does not represent calculating a reservation price for a risky asset, but only an expected value of dividends. Second, the effect of ambiguity on market prices is unclear, as those investors who perceive the highest ambiguity are least likely to actually trade. See Easley and O'Hara (2010), Epstein and Schneider (2010), and Bossaerts et al. (2010).

⁴ Expected variable compensation with simple compensation was 900 (700) for bag A (B). With complex compensation, the probabilities of receiving the bonuses were 98.9% (85.0%) and 82.9% (38.3%) for bag A (B), respectively, resulting in an expected variable compensation of 904.35 (658.13) for bag A (B). Parameters were chosen such that managerial wealth given bag A was on an (approximately) identical level for the two treatments.

⁵ Under the scheme described, the expected payoff is maximised by forecasting (10 times) the median of the dividend distribution. The median is equal to the expected value for bag A. For bag B, even though the distribution is skewed, the 0.5-percentile is still at the expected value of (10 times) 7.

⁶ Welker (2012) provides additional information on changes in investors' estimations over time, investigates reactions of estimations and prices to dividend announcements, and analyses whether representative patterns in dividend realisations affect investors' beliefs.

⁷ As in each market 100 shares were in supply, absolute numbers for trading volume are equal to the percentage of the outstanding shares traded in the market.

⁸ The bias in estimations is statistically significant (Mann–Whitney U-test with the average estimation of all investors in a market over all periods as the independent unit of observation; p = 0.042, two-sided).

⁹ Note that a rational Bayesian update will lead to a significant change in the belief of an investor who is uncertain about the manager's action ex ante only if the dividend pattern is extreme (a large number of blue or of red balls). As a consequence, an investor's change in beliefs about the action chosen is unlikely to be large, and the investor is unlikely to become certain over time.

¹⁰ As we had to make sure that the number of investors in a market would be sufficient to allow for heterogeneity in expectations, we followed previous studies and chose relatively large market sizes (10 investors in a market). With our sample size of five markets per treatment condition, we are close to the sample sizes found in comparable studies. For example, Gillette et al. (1999) have eight and five markets in two treatments; Haruvy et al. (2007) use six markets per treatment to study the dynamics of beliefs and prices; Bloomfield et al. (2009) have six markets per treatment in their analysis of noise trading.

¹¹ Preliminary evidence on the effects of hidden action and compensation complexity on investors' beliefs and asset markets appeared in Welker (2012).

¹² The standard deviation of estimations and market depth are positively correlated, but the level is low (0.198).