Combining Forecasts in Various Financial Applications

Abstract

The availability of numerous modeling approaches for volatility forecasting leads to model uncertainty for both researchers and practitioners. Accurate forecasts of volatility are required across most applications in finance such as risk management, portfolio allocation and option pricing. A large number of studies provide evidence in favor of combination methods for forecasting a variety of financial variables, but most of them are implemented on returns’ forecasting. Surprisingly, combinations of volatility forecasts have not received significant attention in the finance literature. This thesis is focused on evaluating the predictive ability of simple and complex combination techniques as well as on developing and investigating innovative methods for combining volatility forecasts with applications in the stock and oil markets. Firstly, combinations of various volatility forecasts based on different combination schemes of S&P500 index are provided. We add to the literature by combining volatility forecasts from models based on daily, intraday and implied volatility data. Moreover, an exhaustive variety of combination methods to forecast volatility ranging from simple techniques to time-varying techniques based on the past performance of the single models and regression techniques is used. The evaluation procedure is based on both statistical and economic loss functions indicating the superior performance of combination techniques. Although combination forecasts based on more complex regression methods perform better than simple combinations and single models, there is no dominant combination technique that outperforms the rest in both statistical and economic terms, implying that different combination schemes are preferable based on the economic application to be used. Secondly, we propose new combination techniques based on portfolio and risk management loss functions to forecast crude oil price volatility. The forecasting performance of three types of volatility forecast combination is evaluated: forecast combinations involving high-frequency models, forecast combinations involving daily models and forecast combinations involving both high-frequency and daily models. By considering combination techniques based on portfolio and risk management loss functions, new evidence may be drawn regarding the combination forecasts techniques. Firstly, the results show that most combination forecasts produce more accurate volatility forecasts in both statistical and economic terms than single volatility models. Secondly, daily data generate higher economic gains when they are combined through portfolio loss functions especially in 1-step and 22-step ahead forecast horizons, while two single models indicate superior forecasting performance for the 5- step ahead forecasts. Thirdly, statistical combination forecasts from high-frequency models are more accurate according to statistical and economic loss functions when they are compared with the economic combinations suggesting that the information contained in these data can adequately predict economic gains even through statistical combinations. Finally, the two information channels lead to higher economic gains when they are combined through portfolio loss functions for the 22-step ahead forecasting horizon.