Multivariate Volatility Of The RMB Exchange Rate Based On Independent Component Analysis

The volatility is the conditional variance of return on assets. Characterization and theprediction of volatility of return on assets is a very important element of the financial sector,whether in asset pricing, risk management or portfolio. RMB exchange rate as a financial timeseries, research on multiple volatility is still in its initial stage. As free variables of K differentcountries against the RMB exchange rate covariance matrix are K(K+1)/2, Construct a timeseries model for each free variable will result in a large number of model parameters toestimate. So in this context, multivariate volatility models based on independent componentanalysis are widely used.This paper studies a multivariate volatility model based on independent componentanalysis, improving the IC-GARCH model to get IC-GJRGARCH model and IC-IGARCHmodel. Focus on the MAD comparison of RMB exchange rate volatility forecast for thetraditional model, improving the former model and the improved model. The results show that,based on independent component analysis of IC-GARCH better than OGARCH and CCCmodels, but also through the promotion of the IC-IGARCH and IC-GJRGARCH model,prediction is superior to the IC-GARCH.This article has the following main research achievements:1. Research multivariate volatility model IC-GJRGARCH model and IC-IGARCHmodel based on independent component analysis. Firstly conduct the independent componentanalysis for the return of the RMB exchange rate using FASTICA algorithm. Secondly, buildthe GJR GARCH model and the IGARCH model for the conditional independent componentsand do model testing and parameter estimation. The Empirical Analysis and its result of the1194records from many countries against RMB exchange rate is conducted.2. Carry out research on the residual distribution of the improved model IC-IGARCHmodel. Compare the MAD of RMB exchange rate volatility forecast with the residuals in theGaussian distribution, generalized error distribution and t distribution. The results show thatthe forecast of the improved model IC-IGARCH model with residual type of generalized errordistribution and t distribution is better than its prediction with the Gaussian distribution.Finally, a brief discussion of the evaluation criteria of the multivariate volatility modelbased on independent component analysis for dimension reduction techniques and modelprediction, proposed future research directions.