73. Creation and Control of Bubbles: Managers Compensation Schemes, Risk Aversion, and Wealth and Short Sale Constraints

The paper was previously published as Ang et al. (2009). The creation and control of speculative bubbles in a laboratory setting. In Lee, A., and Lee, C.F. (Eds.), Handbook of Quantitative Finance and Risk Management (pp. 137–164). Springer, New York.

Stiglitz (1990), in his overview of a symposium on bubbles, defines the existence of bubbles to be: “if the reason that the price is high today is only because investors believe that the selling price will be high tomorrow – when ‘fundamental’ factors do not seem to justify such a price.” Similarly, he defines the breaking of a bubble as marked price declines which occur without any apparent new information.

Other notable examples of bubbles include the Dutch tulip mania in the seventeenth century, the South Sea Islands Company bubbles Voth and Temin (2003), John Law’s Mississippi Company scheme bubbles of the eighteenth century, the US stock market boom of the late 1920s, the Florida land price bubbles of the 1920s, the great bull market of the 1950s and 1960s, the high-tech stock boom of the early 1980s, and the boom and bust of the California and Massachusetts housing markets in recent years. However, due to the difficulties in specifying the fundamentals, there are still disagreements as to whether these cases could be explained by the fundamental, e.g., Garber (1990) versus White (1990).

Outstanding surveys of this literature are provided by Porter and Smith (2003), Camerer (1989), Sunder (1992).

Griffin et al. (2003) examine the extant theoretical literature about bubbles which includes models where naive individuals cause excessive price movements and smart money trades against (and potentially eliminates) a bubble versus models where sophisticated investors follow market prices and help drive a bubble. In considering these competing views over the tech bubble period on Nasdaq, they find evidence which supports the view that institutions contributed more than individuals to the spectacular Nasdaq rise and fall.

Becker and Huselid (1992), Ehrenberg and Bognanno (1990) have documented in field studies that such tournament compensation systems are effective in raising performance in professional golf and auto racing competitions.

It is possible that if there is sufficient number of short horizon portfolio managers herding in the manner described by Froot et al. (1992), a bubble can start on basis of any information. Shleifer and Vishny (1990) also propose that the portfolio managers have short horizon; however, it is the risk of uncertain return from investing in the longer horizon that prevented disequilibrium to be arbitraged away.

In the experiment, a trader has at least the following choices available:

See Smith et al. (1988) for an example of when reinitialization is not used.

It is important to note that there is a difference between a one- and two-period horizon and a shortened horizon. In a one-period model, only a single dividend is valued. In a two-period model, two dividends are valued. In our shortened investment horizon, the trader is induced to operate within a horizon that is different from that of his operating environment. That is, within a two-period operating environment, the trader is given an incentive to operate with a shorter (possibly single)-period horizon. This is quite different from a single-period model. This shortened horizon is a stronger test of market efficiency, in that the pressures are away from rather toward rational equilibrium prices, (as defined in Eq. 73.4, subsequently). The methodology is meant to emulate modern portfolio managers operating in an environment of perpetual horizon stock securities yet receiving tournament incentives to outperform colleagues on a short-term basis.

Francs are the currency used within this study. They have been used successfully by Plott and Sunder (1982), Ang and Schwarz (1985), as well as others. Their primary benefit is to avoid the technical problem of dealing with small dollar amounts.

The compensation schemes depict the different ways portfolio managers are being rewarded: those who are above the average or beaten the market (Schedule Six) and those who are the superstars (Schedule Two).

The authors are aware of the work of Holt and Laury (2002) which was not available at the time of this study. According to Holt and Laury, their experiment shows that increases in the payoff level increase RRA. However, when estimating RRA, Holt and Laury assume that subject’s utilities depend only on payments in the experiment. They fail to account for the wealth subjects have from other sources (see Heinemann 2003).

The Jackson Personality Inventory is scientifically designed questionnaire for the purpose of measuring a variety of traits of interest in the study of personality. It was developed for use on populations of average or above average ability. Jackson states (1976, p. 9), “It is particularly appropriate for use in schools, colleges, and universities as an aid to counseling, for personality research in a variety of settings, and in business and industry.” Of the 16 measurement scales of personality presented, one scale directly measures monetary risk taking using a set of 20 true and false questions. Mean and standard deviation measures for 2,000 male and 2,000 female college students are provided. Jackson et al. (1972) demonstrate four facets of risk taking: physical, monetary, social, and ethical. The authors’ questionnaires are situational in that the respondent is asked to choose the probability that would be necessary to induce the respondent to choose a risky over a certain outcome. Jackson (1977) presents high internal consistency correlation between the risk measurement techniques.

Note that all odd-numbered experiments used the dividend design in Table 73.2. In order to differentiate between (1) learning about a stationary environment and (2) learning efficient valuation within laboratory markets, we created nonstationarity in equilibrium prices across experiments. In particular, for all even-numbered experiments, the dividend payoffs of Table 73.2 were simply cut in half so that rational equilibrium prices were also one-half that of the odd-numbered experiments. When this equilibrium dividend rotation is viewed in conjunction with the previously mentioned rotation of trader types, it becomes apparent that each individual trader was likely to view the environment (at least initially) as nonstationary. Consequently, any results that we show regarding equilibrium pricing and convergence would suggest that learning about valuation methods rather than a stationary environment creates rational valuation. That is, we are concerned about learning which takes place within the trader (how he values) not about the environment (stationary value). We are able to pursue this expanded question due to our debt to earlier authors who have already well established the presence of the latter.

While period A prices exceeded the calculated PFE price of 460, this price is somewhat unknown to traders at this point. Prior trading results had created a history of B period prices averaging 320. Consequently, it was rational for a PFE trader to pay up to 550 (230 for A period plus 320 for B period sales price). The last trade in period 5A of 505 was well below that level. A more detailed presentation of the experimental results further reveals the rationality of these prices and is available from the authors upon request.

Again, period A prices seem to drift upward due to initial excess pricing in period B.

Our design is to eliminate the bubble effect of miscalculation caused by inexperienced traders as suggested by White (1990) and King et al. (1990). It is more useful and realistic to study the formation and control of bubbles in markets of experienced traders.

To the author’s knowledge, this is the first time traders in an experimental market of this type have used their own money to trade and still produced bubbles.

For instance, new strategies were employed at various stages (which perpetuated continuing uncertainty in the markets). At one time, the market actually stood still for an extended period. Then traders began to liquidate at any price rather than to replicate their earlier strategy of waiting until late in the period to sell out at bubble prices. Other traders began to try and scalp the market by driving prices both up and down, thereby generating capital gains in both price directions. Even others began to try and force losses on traders with large inventories and thereby improve their relative ranking. This was accomplished successfully in period 2A by selling at a loss (at a price below market prices) in order to create a low settle price, M (the second to last trade). Other attempts at this strategy followed in all remaining A periods. Nevertheless, bubbles persisted and many traders were frustrated in their inability to arbitrage them away.

Traders completed survey questionnaire at the completion of experiments 4, 6, and 10.

Given that in experiment 6, period A prices averaged around 600, initial trading capital of 3,000 francs would provide buying power of roughly five securities. Consequently, the new buying power and selling power were a priori relatively equal. Even though period A prices turned out to be quite a bit lower in experiments 7–10, this did not create a great advantage to buyers since the supply of securities (5 traders × 12 traders = 60) was relatively large for a 6-min trading period. As such, there was an ample supply of securities relative to buying power in order to drive prices down should traders turn bearish.

We are unable to recruit all 12 traders back for experiments 7–10 due to graduation, taking of jobs, etc. We were, however, able to retain 7 of the original 12 traders. These traders had now participated in six previous experiments. The five replacements were drawn from the original pool of subjects that had completed the risk attribute questionnaires. These new traders were chosen to replace the risk types that had vacated so that in general, we maintained a wide dispersion of risk types within the market. In addition, some of these new traders had sat in as observers to previous experiments. Others viewed videos of the earlier experiments. All were instructed in the past experimental results, and the various strategies previously used were explained. As such, we do not believe that this change is a critical factor in the continuation of our investigation.

An analysis of many of the last trades of period A for experiments 7–10 often shows either a sharp spike up or down. This illustrates that the traders had become very efficient (through learning) in their manipulation of closing prices. Given the large supply of securities available to squelch a price bubble, speculators were no longer singularly (due to large initial endowments of trading capital) able to create capital gains by driving market prices up. With this constraint, they quickly learned that all they needed to accomplish was to purchase the most securities at present prices and then drive the market up on the final few trades. This was often easily accomplished in that 1) only the second to last trade needed to higher in line with the calculation rules of the TPI and 2) as no surprise, there were always many traders who were willing to sell their securities at a price above the current level. The art to this strategy became a matter of timing; do not try to buy the market too early lest you run out of capital, and do not be too late lest you be unable to make the second to the last trade. There did not appear to be too much of a problem for buyers in accomplishing this in experiments 7 and 8; however, starting in experiment 9, some short traders, having become annoyed at bullish traders getting the tournament prize, began jockeying in these last seconds with the long traders in order to drive prices down. The results of such feuds appear in periods 3A, 4A, and 5A of experiment 9 and each A period of experiment 10. The winner of these duels increasingly became the trader who was best able to execute his trade. Eventually, trading activity become so enraged in the last 15 s of trading that the open outcry systems of double-oral auction began to break down.

Experiments 11–14 were conducted at a second university, and therefore, the results provide information about the external validity of our experiments outside the setting of a single university.

A detailed examination of individual trades reveals the speculative group of traders who are found to be more innovative in designing new trading strategies both in the creating and bursting of bubbles. The finding is consistent with the observation made by Benjamin Friedman (1992) in his review of a dozen NBER working papers on asset pricing. He finds these recent research results demonstrate that rational speculative behaviors such as an attempt by investors to learn from other investors, to affect another’s opinion, or to simply engage in protective trading could in some context, such as imperfect information, magnify price fluctuations.

See the classic textbooks by Greene (2012), Wooldridge (2010), or Hayashi (2000) for details on the implementation and interpretation of OLS.

Furthermore, although insignificant, the p-value for I*S is equal to.14, suggesting that the speculative difference may be even greater under a shortened investment horizon.

In addition, we perform further OLS regressions and report the results in Table 73.5.