关键词: 多尺度套期保值, 经验模态分解, DCC-GARCH, 最小方差法, 最小CVaR法

Abstract: Comprehensively integrating market information to estimate the optimal hedging ratio has always been the key for price risk management. In recent years, with the development of mathematics and econometrics, scholars have proposed many hedging models. However, most of the existing studies ignore the impact of different time scales on hedging. In practice, market participants have different hedging horizons, and the time-series data of financial markets has both time and frequency characteristics. The traditional hedging model is only constructed on the perspective of the time domain, and it is difficult to fully extract the multi-scale information of data. This paper uses the Empirical Mode Decomposition (EMD) method to study the multi-scale hedging problem of CSI 300 index futures. It is helpful for investors to comprehensively consider the hedging horizon, choose the appropriate hedging model and risk measurement indicators to estimate the optimal hedging ratio. Meanwhile, the research results are of guiding significance for policy makers and investors to fully learn the hedging function of futures markets.
This work selects the trading data of CSI 300 index spots and futures, which comes from the Wind database. The Empirical Mode Decomposition (EMD) method is used to divide the CSI 300 index spots and futures return into short-term, medium-term, and long-term time scales. The return means of spots and futures at different time scales are similar to that of the original returns. The difference among them is mainly manifested in volatility. The volatility information extracted from the short-term scale is the most important, while the long-term scale represents the long-term trend of the market, and contributes little to the volatility. Furthermore, combined with DCC-GARCH model, the multi-scale hedging problem of CSI 300 index futures under the hedging framework of minimum variance and minimum CVaR respectively are studied. The hedging ratios of the dynamic DCC-GARCH model are estimated. The hedging performance of the dynamic model are compared with that of traditional static models, i.e., simple minimum variance, simple minimum CVaR, ordinary least squares (OLS) and vector autoregression (VAR). The results show that the trends of the optimal hedging ratios at the original and short-term scales are similar. As the time scale increases, the ratios gradually decrease. In terms of hedge performance, the DCC-GARCH model outperforms the static models at the original and the short-term scales, which can significantly reduce portfolio VaR and increase the risk reduction ratio. However, both the DCC-GARCH and static VAR models are not suitable for the medium and long-term scales. For DCC-GARCH, the hedging performance calculated by the minimum CVaR method is better than that calculated by the minimum variance method.
In our paper, it is assumed that the estimated parameters of the hedging ratio of different models are known. In reality, the return of CSI 300 spots and futures are stochastic and uncertain. Both the means and volatilities of returns have estimation risk. Under the condition of uncertain parameters of models, the research on hedging between spots and futures is more in line with the real investment environment and has wider applicability. Further research can introduce Bayesian methods into the hedging problem and construct the Bayesian hedging strategies under uncertain parameters. Another interesting direction is to consider ambiguity aversion psychology of investors from the perspective of behavioral finance, and study the optimal hedging strategies under different degrees of ambiguity aversion.

Key words: multi-scale hedging, EMD, DCC-GARCH, minimum variance method, minimum CVaR method