基于数据挖掘的VaR测量算法研究

Research of Algorithm to Measure Value-at-Risk Based on Data Mining

提出了将数据挖掘中分位数图的思想应用于VaR测量的新方法,为风险控制和投资决策探索了一种新思路.从设定时间区间内的证券组合对数收益率序列里,求出给定置信水平下不同时间段的分位数,得到分位数序列,由此建立回归方程,预测组合未来特定一段时间内收益率序列的分位数,然后再根据组合的市值大小,直接计算出VaR值.该算法能够有效地处理组合收益非正态分布和非线性组合的情况,在社保基金投资管理系统中得到应用,在国内证券市场数据上进行实验,并采用失败频率检验法来验证算法的精确性,结果表明,该算法是可靠的,有较好的鲁棒性.

The thought of quantile plot in data mining was promoted novelly to measure VaR for risk control and invest decision. First, quantiles of different time period were estimated from the series of portfolio＇s log return with given confidence level. Secondly, quantile regression equation was constructed from quantile series to forecast portfolio＇s log return in future. Thirdly, according to portfolio＇s market value, VaR was calculated. This novel algorithm can effectively process portfolio return＇s non-normal distribution and non-linear portfolio. The algorithm was adopted by Chinese Social Security Fund invest management system and the experi- ments were based on the data of Chinese securities exchange. Its accuracy was verified with Failure Frequency Method, and the results show that the algorithm is reliable and robust.