Value At Risk Of Convertible Bonds And Extreme Value Theory

Convertible bond has already become an important investment tool in domestic capital market. VaR (Value-at-Risk) is a very popular risk management concept. As a common measuremem of market risk VaR is widely used for the risk exposure of financial portfolios. We estimate and assess tail-related risk on convertible bonds issued using VaR.Though VaR has many compute methods, they have limitations. Because almost all of traditional methods estimating tail-related risk VaR focus on central observations, or, in other words, on returns under normal market conditions. However, VaR is a risk measure that relates solely to the tails of the distribution. The extreme values which lies in the tail are some rarely happened events that have significant influence. Extreme Value Theory (EVT) is the statistical model to study the behavior of extreme values. This paper introduces the basic knowledge of EVT and estimates VaR using EVT.Two separate analyses were done to describe the logarithmic return series of S&P CITIC Convertible Bond index based on application methods of EVT. Because the block maximum method is more suitable to use in those data with seasonal variations situation, this paper studies the peaks over threshold method to estimate VaR. In this paper, we propose the novel method to quantify the volatility clustering effect in the financial time series, and gain the declustering method computing VaR using the peaks over threshold method.Based on the S&P CITIC Convertible Bond index, this paper makes an empirical analysis of VaR estimation using the peaks over threshold method of EVT. Empirical findings conclude that the peaks over threshold method can well approximate the tail of financial return distribution. The Risk Metrics method, EGARCH models, GJR models and the Monte Carlo method were used for the prediction estimation of VaR compare with the EVT method. Kupiec LR Test and Dynamic Quantile Test indicate that the peaks over threshold method can estimate and predict VaR with greater accuracy compare with parametric technique and the Monte Carlo method.