Boundaries of the tournament pricing effect in asset markets: evidence from experimental markets.

AuthorIsaac, R. Mark
  1. Introduction

    When traders in experimental asset markets are compensated according to a tournament contract (a "salary-plus-bonus" scheme in which a trader's bonus is proportional to the amount by which that trader exceeds the average performance in that market), we observe violation of convergence results obtained in experiments that did not use such compensation schemes. This effect of tournament incentives has been documented by James and Isaac (2000). In this paper, we search for what Smith and Williams (1990) have termed "the boundaries of falsification" for the proposition that tournament contracts affect asset market prices. Specifically, we examine the robustness of the tournament pricing effect on market prices on several fronts. First, we ask whether the pricing effect survives when not all traders are being compensated according to the beat-the-market contract. Second, we ask whether the tournament pricing effect is robust with respect to perturbations in the trader's compensation contract. Third, we ask whether the pricing effect is observed in two-period asset-life markets (hereafter 2 p.a.1. markets) as it is in 15-period asset-life markets (hereafter 15 p.a.l. markets). (1)

    Section 2 presents the background of the current experiments, the research conjectures, and the experimental design. The experimental results are contained in section 3. Section 4 discusses a phenomenon that we had not encountered until these experiments. Section 5 offers the conclusions and thoughts for further reflection.

  2. Conjectures and Experimental Design

    Background

    James and Isaac (2000) was the first paper to assess the effect on asset market prices of tournament contracts imposed at the individual level. This is distinct from an earlier paper by Brown, Harlow, and Starks (1996) that examines how individual portfolio managers' allocation decisions change in response to such contracts but that does not address the issue of any effects on prices of the assets the portfolio managers are trading. Brown, Harlow, and Starks is itself a specific application of a wider literature on using tournament contracts to address principal-agent conflicts (see also Nalebuff and Stiglitz 1983; Bull, Schotter, and Weigelt 1987; and Ehrenburg and Bognanno 1990). The issue of the effect on market prices is arguably the most important one surrounding the imposition of tournament contracts on portfolio managers because it is these capital market prices that help allocate savings to different forms of investment. Naturally, misleading capital market prices can potentially lead to catastrophica lly misdirected capital spending. (2)

    This key aspect of James and Isaac--that it addresses the effect at the market level of using tournament incentives at the individual level--is also present in this paper. The research objective of this paper is to document the eventual market level ramifications of varying the use of tournament contracts. The nature of those variations will be discussed here.

    The Effect of Tournament Incentives on Asset Market Prices

    Prior to this paper, it has been established that the way in which traders in experimental asset markets are compensated on an individual basis can significantly change the prices that eventuate at the market level. In a seminal paper on experimental asset markets, Smith, Suchanek, and Williams (1988; hereafter SSW) find that as groups of traders were called back for repeat sessions of a 15 p.a.l. market (which can be viewed as an experimental implementation of the theoretical market analyzed by Tirole 1982), the market price of the asset bubbled less in each successive 15-period asset market. This suggests that the no-trading, common expectations equilibrium derived by Tirole might be reached eventually but requires experience on the part of the pool of traders. Of note, in SSW, the traders are compensated such that an experimental dollar corresponds to a U.S. dollar, that is, the traders' respective experimental dollar holdings, deriving variously from cash endowments, dividend payments, and capital gains, are translated one for one into U.S. dollars at the end of the experiment.

    SSW introduced a very workable experimental design; indeed, for the sake of comparability with existing results, James and Isaac (2000) followed it in all but one major detail. That one difference was the manner in which traders are compensated. This aspect was varied in order to gauge the effects on market prices of tournament contracts. An obvious motivation for James and Isaac was the evidence that many real-world traders of assets are compensated by such schemes.

    James and Isaac's experimental design consisted of (i) running the same group of nine subjects through six successive 15 p.a.l. markets, (ii) sequencing those six markets as BBTTBT (i.e., two baseline markets, followed by two treatment markets, followed by one more of each), and (iii) using SSW design parameters in the baseline markets while implementing the treatment markets by replacing SSW's one-to-one conversion from experimental dollars to U.S. dollars with the following tournament compensation:

    U.S. dollar earnings for subject i = 5.00 if [E.sub.i]

    = 5.00 + 2([E.sub.i] - [E.sup.*]) if [E.sub.i] > [E.sup.*],

    where [E.sub.i] represents trader i's experimental dollar earnings at the end of the experiment and [E.sup.*] represents the average experimental dollar earnings at the end of the experiment. (3)

    The results of the James and Isaac experiments showed that replacing one-to-one conversion with tournament contracts drove the asset price away from intrinsic value. Graphically, Figure 1 charts the mean of price deviations from those six experimental sessions.

    The results stand out statistically in part because the BBTTBT sequence of the design exploited SSW's finding of convergence over time in asset markets. Specifically, on the basis of SSW, one is entitled to expect monotonic convergence to intrinsic value, if not cessation of trading, across repeated market sessions in the particular experimental environment they introduce and we adopt. (4) We thus place the treatment markets asymmetrically in the latter part of the sequence in order to capture any deconvergence of an otherwise converging market.

    The BBTTBT design also allows detection of price effects within a single subject group. This is important because across-group variation in price behavior in asset markets can be large, rendering comparisons between baseline-only groups and treatment-only groups potentially misleading in small samples and potentially very expensive in large samples.

    Establishing the Boundaries of Falsification

    The James and Isaac results themselves prompt further questions. For instance, are the results replicable with the original design? If the treatment contract is varied, are similar results still obtained? If there is a mix of traders in the asset market--some compensated one to one, others according to the treatment contract--are similar results still obtained? If the 2 p.a.l. market (used in earlier asset pricing experiments, such as Forsythe, Palfrey, and Plott 1982; hereafter FPP) is used in place of the SSW 15 p.a.l. design, does that change or eliminate the tournament pricing effect? The experimental design in which we attempt to address such questions in this paper can be outlined in table form, presented as Table 1. (5)

    The data generated by our experiments fill in the cells in Table 1. For instance, comparisons between experiments 1 and 3 on the one hand and experiments 2 and 4 on the other can shed light on whether the proportion of traders having tournament compensation contracts during the treatment periods affects market prices. Specifically, given that (i) we know (from SSW) that when zero of nine traders have tournament contracts, prices converge across sessions to intrinsic value, and (ii) we know (from James and Isaac) that when nine of nine traders have tournament contracts, prices deconverge across sessions from intrinsic value, can we further pin down what proportion of tournament contracts is associated with nonintrinsic value pricing as a long-run phenomenon? Will eight out of nine traders having tournament incentives be enough to produce the effect? That is, is the result obtained with nine of nine traders so fragile that it disappears with the substitution of merely one trader with linear compensation? If tou rnament effects do not survive with eight of nine, we can a fortiori rule out that pricing effects would be observed with less than eight out of nine traders receiving the treatment. But suppose that tournament effects are present with eight of nine traders. Will four of nine traders having tournament incentives be enough to produce tournament effects in the market? If not, then that would of course suggest that any "threshold" proportion for observing pricing effects would have to lie between four of nine and eight of nine. (6)

    Similarly, comparisons between experiments 1 and 3 on the one hand and experiments 7 and 8 on the other can shed light on the potential role of perturbations to the treatment contract in mitigating pricing effects. (Experiment 8 has an additional design feature that will be discussed in its own section but that we do not feel hinders comparisons between it and experiments 1 and 3.) The specific perturbation to the compensation contract is in making it a two-step salary component. That is, the modified contract used in experiments 7 and 8 can be thought of as incorporating both an employed salary level and the (lower) opportunity wage that exists if the employee's performance is so bad that he is terminated. Using the same notation as previously,

    U.S. dollar earnings for subject i = 5.00 if [E.sub.b.i], [E.sup.*]

    The difference between these two contract shapes can be seen in Figure 2. The top panel represents the "original" tournament contract. The bottom panel represents the "modified" contract with a...

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT