Discussion Paper
No. 2012-20 | March 30, 2012
Björn-Christopher Witte
Fund Managers—Why the Best Might be the Worst: On the Evolutionary Vigor of Risk-Seeking Behavior


This article explores the influence of competitive conditions on the evolutionary fitness of risk preferences, using the professional competition between fund managers as a practical example. To explore how different settings of competition parameters, the exclusion rate and the exclusion interval, affect individual investment behavior, an evolutionary model is developed. Using a simple genetic algorithm, two attributes of virtual fund managers evolve: the amount of capital they invest in risky assets and the amount of excessive risk they accept, where a positive value of the latter parameter indicates an inefficient investment portfolio. The simulation experiments illustrate that the influence of competitive conditions on investment behavior and attitudes towards risk is significant. What is alarming is that intense competitive pressure generates risk-seeking behavior and diminishes the predominance of the most skilled. Under these conditions, evolution does not necessarily select managers with efficient portfolios. These results underline the institutional need to create a competitive framework that will not allow risk-taking to constitute an evolutionary advantage per se.

Data Set

JEL Classification:

C73, D81, G11, G24


Cite As

[Please cite the corresponding journal article] Björn-Christopher Witte (2012). Fund Managers—Why the Best Might be the Worst: On the Evolutionary Vigor of Risk-Seeking Behavior. Economics Discussion Papers, No 2012-20, Kiel Institute for the World Economy. http://www.economics-ejournal.org/economics/discussionpapers/2012-20

Comments and Questions

Anonymous - Referee Report 1
May 29, 2012 - 09:35
The paper proposes a model to explain the role of competitive pressure on the evolution of risk preferences. In the model, managers invest in two types of assets: less and more risky assets. They can also undertake additional risk which increases the variance of the portfolio without improving the expected return. Badly performing managers are excluded for some period of time. The paper is well written and the model is quite well explained. However, the main weakness of the paper concerns the fact that the model is not validated: neither its assumptions nor outcomes. The exclusion of badly performing managers constitutes the selection mechanism. However, it is not clear how this mechanism generates outcomes different from selection promoting the best managers, as well as it is not explained whether the author considers exclusion to be a hypothetical mechanism, or whether such a temporary exclusion occurs in real markets. In addition, it is argued in the conclusions that the model reproduces exemplary empirical cases, however, it is not clear to which cases the author refers to. Other remarks: (1) Results are not surprising given models’ assumptions. Investing in additional risk is costless, whereas badly performing agents are excluded from the market. In this case, it is expected that “what survives” is risk-seeking behavior. (2) Some essential terms are not well defined in the paper, for instance, “skilled behaviour” or “efficient portfolio”. (3) The discussion on genetic algorithms is unnecessary since agents are characterized only by two attributes. Instead, recombination and mutation should be better motivated with respect to how they describe behaviors of managers. (4) The first sentence is puzzling: why the author would be surprised by evolutionary outcomes? (5) The distinction between two types of approaches to model trading strategies in the introduction is not clear. The ability to survive market selection (the criterion defining the second type of models) is expected to depend on profits produced by different strategies (the criterion defining the first type of models).

Anonymous - Reply
May 30, 2012 - 10:56
Dear reviewer, thank you very much for your helpful comments. Let me briefly reply to your comments.To (1): What you seem to expect is that results should be suprising instead of confirming intuition. Indeed, the scientific community tends to publish those results which are suprising while retaining those which are not. Unfortunately, this leads to a serious bias in scientific research and a well-known problem. Therefore I don't think that (1) should be an argument.To (2): In the following sentence I give a definition of an efficient portfolio. You may have missed it. "Profile 2 represent an inefficient portfolio, defined as any composition of assets for which there exists an alternative one which offers a better expected payoff without implying a greater payoff variance (here: profile 1). Skill is interpreted here: "On the other hand, if δ_i>0 implies an inefficient portfolio, a positive δ_i can be read as lack of skill" In other words, skill corresponds to the ability to find an efficient portfolio. Maybe I should make the last point more explicit.To (3): I don't see why the existence of "only two attributes" is undermining the application or discussion of genetic algorythms. Furthermore, I explicitely stress, that "The implementation of the genetic algorithm in our model is relatively easy because agents are characterized by two attributes only". To (4): I assume you are refering to the following sentence: "However, we are sometimes surprised about which behavior actually turns out to be “fittest”," I fear I do not understand your point. If we were not surprised sometimes, your critique in (1) would apply to all evolutionary studies.Thanks again for your effort and best regardsB.-C. Witte

Anonymous - Referee Report 2
May 30, 2012 - 09:55
In this paper, fund managers choose how they allocate their portfolio between a risk-free asset and a single risky asset offering a risk premium. They also have the option of adding idiosyncratic risk without changing the expected payoff of their portfolio. The paper considers the evolutionary pressure on portfolio construction as a result of a tournament type setting in which the lower r proportion of the fund managers are dropped from the population at the end of each evaluation interval of v periods. A genetic algorithm is employed to replace excluded managers. The questions are, to what extent do managers allocate towards efficient risk (by investing in the risky asset) and whether they take on inefficient risk (by incorporating idiosyncratic risk into their portfolio). Simulations reveal that the parameter space in (v,r) can be divided into five regions of distinct manager behavior. The general finding is that there is an r dependent threshold in v above which managers hold only the risky asset. There is also a region consisting of high values of r in which the evolutionary pressure forces managers to take on ever increasing idiosyncratic risk. CommentsSubstantial culling combined with a short evaluation period induces managers to take on extra risk in order to increase the likelihood of being in the upper tail of the performance distribution. Longer evaluation periods induce managers to take on efficient risk while discouraging inefficient risk. Short evaluation periods and little culling induce excessively safe behavior where conformity reduces the likelihood of being in the lower tail of the performance distribution. The findings conform to what one would expect from a tournament setting for portfolio management. The author provides a largely ex post intuitive explanation for why each behavior is dominant in its given region. The two regions in which a successful strategy is on the interior of the possible parameter space are the most difficult to explain relying on intuition. These are Regions 2 and 4. Of interest, I think, would be a rigorous mathematical analysis of each of the five regions. I suspect that mathematical derivation of the optimal behavior within each region is possible. Derivation of the parameter threshold I suspect exists at each boundary would also be a welcome addition. The analytical support would, I believe, shed considerable light on the evolutionary outcomes. Is the contribution of the paper potentially significant?At present, this is a decent effort that has potential to be better. Identifying the presence of distortions from a tournament setting is of limited contribution without formal support to explain the distortions that arise. The reason for holding a tournament is to identify quality among heterogeneous agents. With a homogeneous population, the tournament can only do harm. The paper is a foundation upon which more interesting investigations can be performed. Is the analysis correct?Yes, but lacking in formality. The paper would be better served by analytical support.

Anonymous - Reply
May 30, 2012 - 10:58
Dear reviewer, thank you very much for your great effort. I can understand your comments and appreciate them very much. best regards, B.-C. Witte