基于高频数据的时变VaR建模和预测研究

MODELING AND FORECASTING OF DYNAMIC VAR SERIES BASED ON HIGH FREQUENCY DATA

提出了VaR时间序列动态预测的方法.首先以上证综合指数和深证综合指数日内分钟数据为基础,根据不同方法计算出每日VaR值,然后给出了VaR时间序列的统计特征,包括平稳性和长记忆性,最后对VaR序列建立ARMA模型和ARFIMA模型,并比较了两种模型预测效果.我们的结果表明:1)基于德尔塔正态法的VaR序列其ARMA模型预测效果好于历史模拟法和蒙特卡洛模拟法的预测效果;2)尽管VaR序列存在长记忆性,但所有VaR序列的ARMA模型预测效果好于ARFIMA模型的预测效果.

A method is proposed to predict dynamically the VaR time series.Daily VaR value was calculated differently based on daily data in min of Shanghai Composite Index and Shenzhen Composite Index. Statistic features of VaR time series were given,including stability and long-term memory characteristics. ARMA model and ARFIMA model were then built from VaR time series and the two models were compared to find the best predicted result.Analysis indicated that ARMA model of VaR time series based on delta-normal method had better predictive effect than that based on historical simulation method and Monte Carlo simulation method;Although VaR time series was of long memory,ARMA model of VaR time series demonstrated better predictive result than ARFIMA model.