| The diagnosis and prediction of variable structure points of the time-varying correlation coefficient have been paid increasing attention in recent years. We present a different method from the traditional one for diagnosing whether a local variable structure point exists based on the Copula theory. To avoid the superposition of peaks and troughs in the tested sample interval which is too large, we narrow the sample interval from both sides of each point to be tested. An illustration example shows that our method is more accurate to diagnose the existing of variable structure points of the time-varying correlation coefficient. This paper also incorporates a Grey Markov SCGM(1,1)c model-a model with the advantages of both grey prediction model and Markov model for prediction. To obtain more accurate prediction results, the original sequence is firstly predicted by the SCGM(1,1)c model, and then the prediction results are optimized by using the Markov method. The latter method is more convenient and accurate for sequences with large fluctuations and strong randomness.The empirical research focuses on the Nuclear Power index and the Coal index of the Chinese stock market from September 2010 to March 2015. We estimate the marginal distribution of each index yield by utilizing the nonp arametric kernel estimation method and get the time-varying bivariate normal copula model. The constant and time-varying correlation coefficients of the both sequences are obtained. A remarkable property from the empirical results shows that our diagnostic method for the local variable structure points is more sensitive than the traditional one. Compared to the traditional forecasting models, the prediction results of the Grey Markov SCGM(1,1)c model are also more accurate. The study is significant for investors to respond to the national macroeconomic regulation and control policies and establish dynamic portfolio strategies. |
PDF Full Text Request