5. CONCLUSIONES Y RECOMENDACIONES
6.6. Desarrollo de la propuesta
Previous experimental studies o f the convergence o f asset prices and allocations to theoretically predicted equilibrium values, have incorporated trading period durations that remain constant for a given experiment. Participants in such experiments have known the trading period duration they face. For example, Copeland and Friedm an’s (1987) experimental design incorporates a constant period duration o f 5 minutes, as does Sunder (1992), whereas studies by Forsythe, Palfrey and Plott (1984) and Banks (1985) invoke period lengths o f 7 minutes. In many o f the studies the time remaining in a trading period has been information readily available to participants via computer screens. In addition, in some studies participants have been warned when the trading period nears completion. In Ang and Schwarz (1985) the experimental design includes a known period duration o f 6 minutes, with warning bells that sound after 5 and 5'A minutes have elapsed since the current period began. It is possible that this knowledge has an important influence on the efficiency o f outcomes reported in these and other studies.
In the theoretical literature, Friedman (1984, p.71) concludes that, amongst other features, the knowledge o f the preset time at which trade will end is an institutional feature o f experimental DA markets that enhances the informational and competitive efficiency o f observed outcomes. As the end o f trade approaches the opportunities to gain from holding back on transacting decrease, thus traders are more likely to accept or, more importantly, better (offer a higher price/ask for a lower price) existing bids and asks. The closer the end o f a trading period becomes the more myopic a trader’s behaviour must become if they wish to realise any remaining gains from exchange. Indeed, many o f the previous studies o f experimental DA markets indicate an increase in the rate o f trade in the closing stages o f trading periods. For a graphic example see Plott and Sunder’s (1982, p.680) Figure 4 which shows the time series o f transaction prices in their Market 3.
Friedman, Harrison and Salmon (1984) and Friedman (1991) develop models o f DA markets that predict bid, ask and acceptance behaviour. Both studies adopt the assumption that trade is permissible over a finite interval time [0, T]. Friedm an’s (1991, p.52) BGAN3 assumption allows for the random generation o f new bid and ask prices at random times. The exact distribution o f such prices is unknown, however, it is assumed that traders have sufficient knowledge o f the distribution of
times to compute the expected number o f new prices available to them before the end o f the trading period. To achieve this traders require information regarding the time at which trade ceases in the period, thus there is implied knowledge o f the trading period duration. It seems that this is a fundamental assumption if the model is to predict convergence o f prices and allocations to almost Pareto optimal outcomes (see Proposition 3, op cit p.58).
Thus there is limited theoretical and indirect experimental evidence that the certain knowledge o f when a trading period will end actually influences a market s trading behaviour, with respect to overall efficiency and the timing o f transactions. Participants may delay transacting because they are waiting to see if additional information concerning the prevailing state of nature is forthcoming. Such behaviour would consequently impact upon the speed at which transaction prices converge to theoretically predicted equilibrium values. Thus the results reported in previous experimental studies may in fact have been biased towards, or away from, the rapid attainment o f a FRE equilibrium. Given the high level o f informational and allocational efficiency reported in the literature it is more likely to be the former.
The pilot paper by Duxbury (1997) is the only study o f the impact o f trading period duration on market performance, to date. Drawing extensively on Duxbury (1997), the following section develops a number o f testable hypotheses. The subsequent section provides a detailed review of the results reported in Duxbury (1997).
5.3 Development of Hypotheses
The trading performance reported in previous experimental asset markets may potentially be a result o f the artificial trading period deadline. In an attempt to overcome this potential bias, period duration is introduced into the experimental design as a treatm ent variable. Markets are conducted with variable and known (VK) durations and variable and unknown (VU) durations. In the former markets individuals still know when the trading period will end, whilst in the latter markets individuals’ knowledge o f exactly when a trading period will end is removed. I he outcomes observed from a market under the regime of constant and known (CK) period duration form a control against which the trading behaviour witnessed in VK and VU markets can be compared. Comparison o f market behaviour observed in VK and VU markets permits the impact o f knowledge o f period duration to be isolated
and investigated.
The issue raised is o f substantive theoretical and practical interest. The few theoretical models o f bid, ask and transaction price behaviour in experimental DA markets developed to date rely on assumptions concerning the time remaining until the end o f trade. Indication o f the importance o f these assumptions for the convergence behaviour advocated by these theoretical models is necessary for future theoretical development. This study can be viewed as a ‘boundary’ experiment designed to determine the importance o f the period duration on the efficacy o f theoretical models and so provide guidance with respect to those areas or variables requiring theoretical refinement. The results may also shed light on the disparate results reported by the empirical (low level o f efficiency) and experimental (high level o f efficiency) literature investigating the informational efficiency o f financial markets and associated trading institutions. The experimental results to date may be an artefact, due to the known trading period duration.
The remainder o f this section details the development o f a set o f null (Ho) and alternative (Ha) hypotheses, concerning the impact o f period duration on the volume o f trade, informational efficiency and allocative efficiency observed in the experimental markets. Due to the lack o f theoretical literature concerning the DA market institution the hypotheses offered are exploratory in nature and are intended to motivate the development o f relevant theoretical models.
Early trading volume: The uncertain period duration in VU markets may well result in a reduction in the length o f time taken over bartering and thus an increase in the volume o f trade early in the period as traders attempt to realise potential gains from trade before the trading period closes. Alternatively there may be no impact (see 1H0 below). A priori, it is not clear in which direction VK period duration will impact on the volume o f trade. Therefore, for reasons o f consistency, the alternative hypothesis will be two-sided for all comparisons.
Informational efficiency: Following Friedman s (1984) line o f reasoning, the introduction o f uncertain period duration (VU) may result in a decrease in the informational efficiency o f the experimental markets with respect to the convergence o f transaction prices to theoretically predicted equilibrium values. Non-convergence o f prices could plausibly occur if the predicted higher volume o f trade early in a period meant that, in their haste to trade, individuals imperfectly infer the underlying
state o f nature from the publicly available information. Alternatively, a priori, the predicted higher volume o f trade may result in an increase in the speed of convergence to equilibrium values if the bids, asks and transaction prices correctly reveal the underlying state o f nature. For the VK markets there is no a priori predicted directional change in volume and so consequently it is difficult to conjecture in what direction varying the period duration will impact on the informational efficiency o f such markets.
Allocative efficiency: Similarly, Friedman’s (1984) line o f reasoning, suggests that the introduction o f uncertain period duration (VU) may result in a decrease in the allocative efficiency o f the experimental financial asset market with respect to the convergence o f asset allocations to theoretically predicted equilibrium allocations. The result would be a reduction in the percentage o f potential gains from exchange to be exploited. Once more, non-convergence could plausibly occur if the predicted higher volume o f trade early in a period meant that, in their haste to trade, individuals imperfectly infer the underlying state o f nature. Alternatively the percentage o f potential gains from exchange to be exploited may be increased if, a priori, the predicted higher volume o f trade resulted in the underlying state o f nature being revealed earlier in a period. Thus, more time is available for traders to position their portfolios efficiently. Once again, for the VK markets there is no a priori predicted directional change in volume and so consequently it is difficult to conjecture in what direction varying the period duration will impact on the allocational efficiency ol such markets.
Prem ised on the above line of reasoning, a set o f null and alternative hypotheses can be tentatively formulated. The hypotheses are worded generally so as to be equally valid for both VK and VU markets. All alternative hypotheses are two sided and are not explicitly stated.
1H0: The treatment period duration does not impact on the volum e of trade early in a given period, as compared to that observed under a control market. 2.1 H 0: The mean deviation o f transaction prices from theoretically predicted values in the treatment period duration market does not differ from those observed in a control market.
2.2H0 : There is no correlation between the difference in the volum e of trade,