Combining methods for volatility forecasting on stock returns and exchange rates

The research on volatility forecast is aimed at reducing financial investment risks. However, it is not easy to obtain high-accuracy and reliable forecast models as there is no agreement on which method is the best method for forecasting. Discarding the competitive forecast models may lead to the loss of precious information embedded in the model. This study was designed to examine the volatility forecast ability of a newly proposed Smooth Transition (ST) combining volatility forecasts method by applying the high-frequency data and trading volume in the forecast models, as well as benchmarking with the established Mixed Data Sampling (MIDAS) model. Specifically, this study was divided into three sub-studies. In this first sub-study, the proposed ST combining method was evaluated based on seven stock return series: Amsterdam (AEX), Frankfurt (DAX), Hong Kong (Hang Seng), New York (S&P 500), Paris (CAC 40), Tokyo Nikkei225 (Nik), and Shanghai (SSE). Prior to combining forecasts, traditional volatility forecast models (traditional GARCH, Ad Hoc, and STES models) were developed. The results suggested that the ST combining forecast method was able to outperform other traditional forecast models and the trading volume was a robust variable in this finding, substantiating the past literature findings on the correlation of trading volume and stock returns. The second sub-study evaluated the performance of the ST combining approach by applying the high-frequency data of foreign currency series. A total of five core currencies against USD namely, AUD, CAD, EUR, GBP, and JPY were adopted in this study. The combinations of various data frequencies were developed and compared with the other individual models. Besides that, different frequencies of realized variance (RV) as proxies of actual volatility were also constructed in this study to evaluate the influence of the RV proxies in the forecast performance. The results confirmed the superiority role of high-frequency data, in particular the 5-min RV in forecasts. Additionally, it was found that proxy and evaluation criteria (loss function) influence the ranking of the forecast model. The third sub-study was to benchmark the proposed ST combining model with respect to the mixed data sampling (MIDAS) approach in longer lead periods, such as one week ahead for the stock market series and one day ahead for the exchange rate series. The findings indicated that the traditional GARCH model did not warrant longer lead periods. Similarly, there was inadequate evidence to substantiate the robustness of the proposed ST combining forecast method as compared to the MIDAS methods in longer periods. The performance of RV incorporated in STES and GARCH models was more superior in a shorter time horizon. The empirical results of this research demonstrated the performance of the proposed ST combining forecasts model and validated the role of trading volume as well as the RV in the financial markets. Moreover, it also showed that the choice of proxy as actual volatility, evaluation criteria, and time horizons influenced the ranking of forecasts models. The findings provide important insights to practitioners and industry players on the cautions in referencing the volatility indicators in portfolio selection or investment.

Text SPE 2024 50 - Full Text.pdf Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (4MB) Text SPE 2024 50.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (5MB)

Actions (login required)

View Item