6.2 Strategy Returns Performance Analysis
6.2.3 Discussion
The results obtained from the analysis of the returns of the different strategies strongly suggest that the the agents using an Option trading strategy obtained higher returns than those not using such strategies in the majority of the experimented scenarios. The specific scenarios where agents trading Option contracts did not obtain higher returns than those not trading Options are in few cases when using the Random A price series and when the Random B price series is used as the price of the underlying asset. For the experiments where the asset price was not the Random B series, the OTMinR strategy performed better than the other strategies. The OTMaxW strategy performed better than the other strategies in all the test cases where the Random B price series was used in the SM Anexperiments.
Relevant issues from the results of the experiments using α–Perfect forecasting are that not all the strategies using Options have a better performance than the strategies not us- ing Options; notably, the OTMaxW Option trading strategy performed worse than both the ATSpecand the ATNoise asset trading strategies in almost every case, excepting the experi- ments using the Random B price series as the price of the asset. Given the definition of the OTMaxWstrategy, an agent using that strategy will select the action for which there is more probability of obtaining positive returns. Therefore, the agent always adopt a speculator– like strategy according to its model of the risk in the market and its forecasting. From this result then, it is possible to conclude that using Options does not guarantee a better out- come than not using Options in all possible scenarios. Moreover, it shows that a bad Option trading strategy can cause a reduced performance.
The OTMix strategy, which is a strategy that combines the OTMinR and OTMaxW strate- gies has a good performance compared to the other strategies (excepting the OTMinR strat- egy). While the returns of the OTMix strategy are not as good as the returns of using only the OTMinR strategy, it still outperforms the asset only trading strategies in the experiments where uncertainty is high and medium (α = 0, α = 0.2, α = 0.4 and α = 0.6). More- over, the OTMix strategy has a better performance than OTMinR in the scenarios where the Random Bprice series is used as the asset price. This, due to the use of the OTMaxW-like strategy.
that the OTMinR strategy also performs remarkably better than the other strategies in the experiments where the asset price is one of the stock market price series. However, for both of the randomly generated price series, the OTMinR strategy is outperformed by other strategies.
The fact that the OTMinR strategy performs bad in all the cases when Random B price series is used and in both set of experiments (when using α–Perfect forecasting or SM An
forecasting) may indicate that this strategy is not good for scenarios where the variance of the price is as high as in the Random B price series. However, in these type of scenarios it is still possible for Option-trading agents to perform better than agents not trading Options as can be seen from the performance of the OTMaxW strategy, which outperforms the other strategies in the Random B cases.
Recapitulating the hypothesis established in Chapter 1, this analysis has shown that agents can indeed benefit from trading Options in the market; the analysis also shows that agents benefit most from trading Options in the cases where the asset prices are based on realprice series. However, the performance of the Option-trading agents using the OTMinR strategy seems to degenerate in markets where the volatility is high (such as the markets represented in the two randomly generated price series). In this cases, the use of other Option trading strategies can still outperform asset-only traders.
Given that the forecasting model of the Option trading agents assumes Normally dis- tributed price series (See section 3.2.2.1), it could be argued that the performance of the Option trading strategies should be expected to be higher than the non-Option trading strate- gies in the randomly generated price series. This because the Option-trading strategies base their valuation of the utility of the actions as a function of the Normal distribution. As the generated price series are more Normal (as shown in Section 5.3) than the stock market price series, it is possible to argue that the Option-trading agents’ forecast should be more accurate. However, the difference in the performance in this cases is not as large as in the cases where the stock market prices are used.
The reason why the performance of the Option-trading agents is better in the stock market based price series may be caused by several factors. One of these factors is that in the cases of the Microsoft and IBM price series, there are some single steps where there is a high decrement in the price. This is demonstrated by the skewness of the price series (shown in Section 5.3) which is −6.18 for the Microsoft price series and −0.50 in the case of the IBM price series. Negative skewness indicates that there are more steps in time where the price is decreased compared to what is expected in a Normal distribution. In the case of the Microsoft price series, the Option trading agents may be protected from such high
6.3. CORRELATION WITH PRICE ANALYSIS 117