Application of the Trading Strategies

Previous forecast evaluations indicate that the 15-day-ahead forecasting model generally outperforms the other h-day-ahead volatility forecasting. We then propose the option trading strategies based on the 15-day-ahead predicted change in future volatility. The competitive volatility forecasting models are applied to the option trading strategies and the performances of different models are compared. The option strategies are traded based on the predictive ability of sentiment levels of (changes in) the future volatility and the results are shown in Table 22. The long (short) straddle is traded while positive (negative) future volatility changes are predicted. The benchmark trading strategies are the long and short straddle without any decision support. Panel A (Panel B) in Table 22 depicts the monthly rate of return of the long (short) straddle traded 15 days before the options final settlement day based on different volatility models.

Table 22 Performance of Option Trading Strategy

ISP Benchmark Recruiting Sentiment Levels Recruiting Sentiment Changes Without

any

Panel B: Performance of short straddle based on the decision support of volatility forecasting (%)

30 -0.58 -4.04 -1.52 -1.82 -1.80 1.91 -0.58 -0.58 -0.58 -0.58 -1.84 -4.15

60 -4.47 -0.69 -1.85 1.19 -4.47 3.61 0.78 -4.47 -4.47 -4.47 1.45

90 1.21 -6.07 -4.31 1.49 1.35 -0.80 -4.00 1.21 1.21 -0.19 -1.02

120 -3.41 -7.80 -0.68 -5.10 -2.19 -3.63 -3.41 -1.41 -3.41 -0.71 -4.36

Notes: This table presents the performance of the option trading strategy for options traded 15 days before the final settlement day based on different volatility forecasting models. Panel A (Panel B) summarizes the monthly rate of return (%) for a long (short) straddle referring to equation (6.6). Model (1) in Table 6 is the benchmark volatility forecasting model based on multivariate historical volatility measures, realized volatility (RV), the absolute return (|R|) and the high-low range (HL), and is simplified as MHV. Model (2) to Model (6) (Model (7) to Model (11)) in Table 6 are volatility forecasting models‟ recruiting levels of (changes in) investor sentiments. +TVIX represents the volatility forecasting based on the MHV, and the sentiment proxy of TVIX is included as are the other symbols. The in-sample-period (ISP) is set as 30, 60, 90 and 120 days. Values in boldface and italics are long or short strategies which produce the best average monthly rate of return (%) based on MHV recruiting levels of or changes in investor sentiments.

The performance of each model is evaluated based on the monthly rate of return by referring to equation (6.6). For space considerations, the cumulative profit-loss and cost of capital of each model are omitted but are available from the authors upon request. The trading performance of the forecasting model that recruits the sentiment index results in a better average rate of return compared to the benchmark MHV model, especially when the in-sample period is 60 days. Most of the performance of the long straddle strategy that is based on alternative models, including the benchmark MHV model, is superior to the benchmark strategy of the long straddle without any filter, although not all of the strategies traded based on different sentiment integrated models and in-sample periods outperform the benchmark strategy. The performance of the short straddle based on the volatility forecasting, however, does not consistently present a better rate of return than the benchmark strategy for the short straddle without any filter. The trading performance concludes that the short straddle 15 days before the final settlement day based on the +TO model, the forecasting model based on the MHV recruiting level of the turnover ratio, gives rise to a monthly rate of return of 3.61%, which is better than the risk-free rate. The long straddle 15 days before the final settlement day based on +TO (+ΔTO) further produces a monthly rate of return of 28.07% (19.47%) while the levels (changes) are considered. The effective option trading strategy suggests that a long (short) straddle based on the positive (negative) changes of volatility forecasting including the sentiment level of the

„turnover ratio (TO)‟ achieves the average monthly return of 15.84%.

6.5 Sub-Conclusions

The algorithm of option trading strategies based on volatility forecasting is evaluated in this study. The difference between this paper and the previous literature is that we construct a volatility forecasting model that recruits the investor sentiments.

The contribution of this study is that the algorithm of the effective option trading strategy proposed is based on a superior model. We also bridge the gap between investor sentiments and the decision support system from a behavioral finance point of view.

The algorithm is established by means of the following steps. First, possible sentiment proxies for the equity and derivatives markets are collected such as the volatility index which is a proxy for the investors‟ fear gauge, put-call trading volume ratio, put-call open interest ratio, market turnover ratio and the ARMS index. Second, the causal relationship between investor sentiments and future volatility is examined to confirm the predicted ability of sentiment indicators. Third, the multiple-factor forecasting model is built up by including each sentiment indicator based on the benchmark forecasting model (MHV), including absolute daily returns, daily high-low range and daily realized volatility. Fourth, the forecasting ability of competitive models is compared and the forecast evaluation is measured by the regression-based forecast efficiency test and the mean absolute percentage error (MAPE). The parameters used in the option trading strategy, including the in-sample-period and the holding period, are identified in this step. Finally, we simulate the option trading strategies based on the predicted future volatility change.

An effective multiple-factor volatility forecasting model that recruits the sentiment indicators from the stock and derivatives markets is presented.

The causality and the regression based forecast efficiency tests support the view that the sentiment proxies of market turnover and the volatility index include levels and changes that can help predict future volatility. The algorithm for the option trading strategies is supposed to long (short) straddle 15 days before the final settlement days of the option contract based on a 60-day in-sample-period volatility forecasting model. Volatility forecasting that recruits market turnover is the best filter and the average monthly return is about 28.07% (3.61%) for a long (short) straddle, which implies an average monthly return of 15.84% considering the margin based transaction cost. An effective option trading strategy that refers to a predicted positive (negative) change in future volatility that recruits market turnover is suggested in this study.

In conclusion, our empirical findings agree with the noise trader explanation that the causality runs from sentiment to market behavior. The results also support the view that the forecasting models of volatility need to assign a prominent role to investor sentiments. We posit that proxies of investor sentiments support the decision to engage in option trading, and that the trading algorithm based on the volatility forecasting recruiting investor sentiments can be further applied in the electronic trading platforms and other artificial intelligence decision support systems.