IFN Working Paper No. 1149, 2017 Grind or Gamble? An Experimental Analysis of Effort and Spread Seeking in Contests Ola Andersson, Håkan J. Holm and Erik Wengström Research Institute of Industrial Economics P.O. Box 55665 SE-102 15 Stockholm, Sweden [email protected] www.ifn.se Grind or Gamble? An Experimental Analysis of Effort and Spread Seeking in Contests November 2016 Ola Andersson, Håkan J. Holm and Erik Wengström* Abstract: We conduct a contest experiment where participants can invest in increasing both the mean and the spread of an uncertain performance variable. Subjects are treated with different prize schemes and in accordance with theory we observe substantial investments in spread. We find that both types of investments can be controlled with a three level prize scheme. However, the control is imperfect and behavior is characterized by inertia. The winner-take-all prize scheme has many disadvantages including high spread and heterogeneous behavior. The scheme where only one loser is punished appears superior; it generates high mean, low spread and is most popular. JEL Codes: C7; D8; D02; D03 Keywords: Contest; Risk; Spread; Incentives; Institutional Choice, Experiment *Andersson: Research Institute of Industrial Economics (IFN), Lund University, [email protected]; Holm: Lund University, Department of Economics: [email protected] Wengström: Lund University, Department of Economics and University of Copenhagen, Department of Economics, [email protected]. We thank the Swedish Competition Authority for financial support and Claes Ek for helping us run the experiment. We are grateful to participants at the 8th CNEE Workshop, the 11th Nordic Conference on Behavioral and Experimental Economics in Oslo, the Contest: Theory and Evidence Conference in Norwich, 2016, and the seminar at Department of Banking and Finance at University of Innsbruck for helpful comments. 1. Introduction A core question in economics is to provide people with appropriate incentives. It is therefore hardly surprising that many different incentive systems used in practice have been analyzed both theoretically and empirically. One such scheme entails using fixed payments based on rank, so called rank-order tournaments or contests for short. Such payments are practical in many cases, especially when the performance variable is difficult to translate to a cardinal scale and where ordinal assessments are easy. Lazaer and Rosen (1981) showed that payments based on rank leads to efficient choices under risk-neutrality, and also that participants may under certain circumstances prefer to be paid according to rank. The latter property is important whenever the participants in the contest (e.g., workers and managers) have a say in the decision of what incentive system to implement. Most theoretical studies of contests have either focused solely on effort (see e.g., Nalebuff and Stiglitz 1983, Rosen, 1986 and Moldovanu and Sela 2001; see Konrad 2009 for a review) or solely on risk taking (see e.g., Dekel and Scotchmer, 1999, Tsetlin, Gaba, and Winkler, 2004) affecting the mean and the spread, respectively, of the rank determining performance variable. However, it has been convincingly argued by Hvide (2002) that in many important contest situations the contestants have a possibility to affect both the mean and the spread of the performance variable. For instance, CEOs can enter stable mature markets or unstable emerging markets and fund managers can choose a safe or risky portfolio. If this is the case and contestants at no cost can increase the variance of their performances, Hvide (2002) shows that contestants, facing contests that award the top ranking candidate, will end up in an 1 equilibrium characterized by low mean and high spread.1 Since it is reasonable to assume that the principal, i.e. the “contest organizer”, is positively affected by mean increases in the performance and negatively affected by the spread of the performance, this is a particularly bad equilibrium for the contest organizer. In some of the literature, spread increases are somewhat loosely referred to as increases in risk without being precise of its meaning, in particular concerning who is exposed to the risk. To avoid confusion we interpret investments in mean as (productive) effort since it will increase the expected performance to the benefit of the contest organizer. Furthermore, investments in spread are interpreted as unproductive since it is costly and can be assumed to increase the risk the contest organizer faces without affecting the expected performance.2 The distinction between spread of the rank-determining performance variable and the spread of the contestant’s payoffs is worth noting. Investments in mean increases of performance will normally affect the spread of the contestant’s payoff (and hence his risk) without necessarily affecting the variance of the contest organizer. The observation by Hvide (2002) that the contestants seek spread at the cost of effort may be problematic for society as a whole when there are externalities (as in the banking sector), but as noted before, it will be especially costly for the contest organizer. Gilpatric (2009) shows how such spread seeking can be tamed. In a model with three or more contestants and where it is assumed to be costly to increase the spread, he shows that three payoff levels (a prize to 1 In a somewhat similar model Kräkel and Sliwka (2004) analyzes a two-contestant tournament in which contestants differ in abilities and first choose a high or low-spread strategy and then choose effort (which is assumed to affect the mean). In this setting, they show that diverse equilibria are possible and depend on the magnitude of the ability difference, the shape of the cost function, and the prize spread. 2 Note, investments in spread can also be seen as a form of effort, but since these investments are assumed to be costless in some of the literature (see e.g., Hvide, 2002) we reserve the term for investments affecting the mean. 2 the contestant ranked first, an intermediate prize, and a “loser prize” to the contestant ranked last) are sufficient to induce any combination of effort and spread under certain assumptions. The intuition is straightforward: Increasing spread (symmetrically) raises the probability of ending up first and last. Using standard prize schemes, the latter is not punished which distorts spread-choices upwards. By introducing a ‘’looser prize’’ for ending up last such incentives can be tamed. One important difference in assumptions between Gilpatric and Hvide is that the former assumes strictly positive and convex cost of increasing the spread, whereas the latter assumes that increases in spread are free. We think that both assumptions are possible to defend in different empirical settings. Choosing a stock with a large spread instead of one with a low spread appears costless. However, if one assumes that projects possess a normal level of initial spread, then search theory would suggest that it is costly to find projects with the same mean but that have an unusually large spread. One example is the degree of originality in the design of new products. It is time consuming and costly come up with creative deviations from a standard design and it is not clear that consumers eventually consider these deviations to be improvements. On the other hand, a creative design that deviates from the standard product in many attributes that consumers appreciate will have a large market. Hence, increased originality of a new product is costly and increases the probability for both fiascos and best-sellers, which in turn generates a larger spread of returns. While the predictions from theory by Gilpatric (2009) are promising in the sense that spread and productive effort choices can be carefully tailored, we know from various experiments that actual behavior does not always follow crisp equilibrium predictions. In the case with 3 contest behavior one can think of various disturbing behavioral factors such as risk preferences, positional concerns beyond what is motivated by rank based payments and cognitive factors. We therefore conduct a contest experiment with a varying prize structure to investigate if the theoretical predictions get support. We also address the important but empirically open question regarding which prize structure the contestants prefer. In addition, our new design allows us to explore if different prize schemes generate a different level of behavioral heterogeneity when spread choices also are directly involved. Furthermore, the two-variable design makes it possible to study how the simultaneous effort and spread choices are connected. We find clear evidence of investments in spread, which suggest that the concern raised by Hvide (2002) is motivated even in a setting where increasing spread is costly. Furthermore, both effort and spread seeking can be controlled to a certain extent with a three-level prize scheme as suggested by Gilpatric (2009). However, the observed behavior is characterized by inertia and the theoretically predicted treatment differences are starker than the observed ones. We also present results on how the prize schemes perform in this new environment of both investments in mean increases and spread. The prize scheme where only one loser is punished appears superior to the other schemes since it is associated with relatively high effort, low spread seeking and a low behavioral heterogeneity within the competing groups. Somewhat surprisingly, it is also the most popular scheme among the subjects. On the other side, winner- take-all appears to be the worst performing scheme with high spread, highly heterogeneous behavior and lowest popularity scores at the same time as it does not generate significantly higher efforts. 4 We also explore the results on how effort and spread choices are correlated both “within” the subject over time and “across” subjects. We find some evidence that subjects trade-off effort against spread, but this is happens only in one treatment (with two winners and one loser), which suggest that this effect is contingent on the prize structure. However, when we compare the correlation across subjects, we find robust evidence for a positive correlation. Hence, those who make the greatest productive efforts are also the ones most likely to make large investments in destructive spread. This ought to be an important lesson for any contest organizer who suspects that the contestants can affect the spread of their performance. 2. Related empirical studies To our knowledge our study is the first experiment that manipulates the prize structure in a contest where contestants can choose both the mean and the spread of the performance variable. There is one study by Nieken (2010) where subjects in pairs choose a low or high variance distribution first and then effort. This study contains no treatment manipulation and can be considered a laboratory test of Hvide’s (2002) predictions, which partly get support in that when contestants have chosen high variance they exert less productive effort. At the same time, about 50 percent of the contestants do not end up in choosing the high variance distribution even after 27 rounds as they should in theory. This suggests that there is some heterogeneity in how contestants choose spread, and possibly also that all contestants do not fully understand the strategic upside of the high spread strategy. We take this further by investigating how spread choices can be controlled by treatment manipulations of the prize structure. This ought to be highly relevant from a risk-management perspective. We also elicit 5 individual attributes, like personality attributes and cognitive measures to understand the heterogeneity better. Investments by a contestant typically affect the probability distribution of the ranks the contestant ends up with, but will often also affect the contestant’s expected payoff directly. Both effects are likely to affect the payoff distribution of the contestant and (depending on definitions) thus the risk he faces. Eriksen and Kvaløy (2016) focus on this risk when they study betting behavior in a lottery contest where the prize is partly contingent on the size of the bets (i.e., the investment) and where it is rational to bet zero. They find sizeable irrational risk taking in this lottery contest and that risk taking increases when feedback about the winner’s strategy (in the previous round) is given and when the number of contestants increases. Eriksen and Kvaløy (2016) explain that competition “per se” triggers risk seeking even if it is irrational by referring to psychological mechanisms such as the “the contingency of reinforcement” by Skinner (1969) and the “availability heuristic” by Kahneman and Tversky (1973). Although, our contest and spread-seeking concept are different from the contest and risk-taking concept in Eriksen and Kvaløy (2016) it is not obvious why not the same mechanism should be triggered in our contest.3 Hence, if spread seeking is mainly triggered by irrational psychological mechanisms as suggested by Eriksen and Kvalöy (2016) we think it is legitimate to ask if these mechanisms can be tamed (at least to some extent) with appropriately chosen prize schemes. 3 To be fair there are also other differences in our contests. For instance, in Eriksen and Kvaløy (2016) any level of risk taking is excessive. In contrast, we allow for contests with both too high and too low spread seeking compared to equilibrium predictions. 6 In line with the theoretical studies, most contest experiments use a framework where contestants only can invest in effort (see the review by Dechenaux et al., 2015).4 The first study to examine contests was clearly inspired by the theoretical findings by Lazear and Rosen (1981) and conducted by Bull et al. (1987). They found that tournament and piece-rate pay schemes generated the same mean effort, though the contest pay scheme induced a higher variance in effort. Observations like this one has motivated researchers to more carefully study the impact the prize structure has on effort (see e.g., Orrison et al., 2004; Harbring and Irlenbusch, 2008, Müller and Schotter, 2010, Sheremeta 2011, Shupp et al. 2013, Dutcher et al. 2015 and Andersson et al. 2016b). The results from these studies are somewhat mixed, but it is clear that manipulations of the prize structure matters and that the standard winner-takes- it-all (WTA) prize structure may discourage some contestants to exert effort. Our study can be seen as complementary to these since the choice of prize structure get somewhat more involved but also more realistic in cases when subjects also can affect the spread of their performance variable. One interesting empirical question that can be addressed in our study is if contestants who chose low productive effort also chose low unproductive spread and if this is associated with personal characteristics. It is also the case that prize schemes like WTA will in theory tempt contestants to increase the spread of their performance variable whereas other schemes will not. To learn more about this we need both between treatment manipulations of prize schemes and within-subject analyses, which this paper can contribute with. 4 We have found one study where the contestants only choose risk and not effort. In this study (Nieken and Sliwka 2010) the correlation between the contestants performance variable is experimentally manipulated and found to matter for the contestants spread choices. 7 3. Hypotheses The purpose of the experiment is to investigate if effort and spread choices can be controlled in contests when subjects choose both effort and risk. To accomplish this we investigate four different prize schemes with three contestants each. Each contestant 𝑖 ∈ {1,2,3} can make costly choices to increase the mean, µ ∈ [0,30], (i.e., effort) and spread, 𝜎 ∈ [0,30], of their 𝑖 𝑖 performance variable Y =  where  :U((50),50) and iid.5 The cost of i i i i i i choosing µ and σ are given by 𝐶 (𝑥) = 𝐶 (𝑥) = 𝐶(𝑥) = 𝑥2/20 .6 𝜇 𝜎 In all prize schemes the contestants compete for the same total prize sum (of 360 Danish crowns), but the schemes differ in what theory predicts about effort and spread seeking.7 The first scheme is the winner-take-all scheme (WTA), which gives all money to one single winner. The second scheme is called single-loser (SL) since all but the loser divide the prize sum equally. The third and fourth schemes are called something-for-all since all contestants receive something but where the winner get the most and the contestant in the 2nd place get the average payoff and the contestant ending up last receives the least. The third and fourth prize schemes have either small or large differences in prizes, respectively and are denoted SFAS and SFAL. Each prize scheme represents a specific treatment and their exact prizes in Danish crowns are given in Table 1. 5 I.e. when both µ and σ are set to zero the performance variable is uniformly distributed according to U(-50, 50). 6 See Appendix B for details of the model and the predictions underpinning our hypothesis. 7 The exchange rate between US dollars and Danish crowns was about 0.17 at the time of the experiment, which meant that the sum of prizes was equivalent to 61 US dollars. 8