Extended Switching Regression Models With Time-varying Probabilities For Combining Forecasts In Stock Market Volatility

The volatility is inherent in the basic properties of the stock market. In the mature financial system, moderate volatility in stock prices helped to stabilize the national economy, but the intense and frequent fluctuations may affect the investors to make accurate investment decisions difficulty, also affect the national healthy economic development, and even induce a recession. The volatility forecasting of the stock market in risk management, asset pricing, arbitrage research fields have played an important role, especially with an active warrant market, stock index futures and other derivative financial products, the introduction of volatility derivative financial products, as a basis for pricing indicators, the significance of their research has become increasingly prominent.There are three types about metric models of stock market volatility, generally, Exponential Smoothing, GARCH models and the SV models. Exponential Smoothing volatility, in rates of return, assumed constant fluctuations accordance with the sequence of traversal of the historical volatility of earnings as future forecast volatility on the unbiased estimate. However, in reality, the volatility of returns is variable sequence of fluctuations, while the return series at each time the distribution is not the same, that is, the standard deviation of the distribution and change over time. In GARCH models, conditional variance is defined as the square of the residual pre-entry and pre-conditional variance of the deterministic function of the estimated conditional variance which are directly related to the past observations, although the solution of the time-varying volatility, but when the return series unusual observations when making a sudden change in the estimated conditional variance. Conditional variance, in SV models, is no longer a deterministic function, but rather to join a random item to reflect the random factors on the impact of volatility. In the use of the above three types of models to describe fluctuations, modeling and forecasting process, the model does not predict what kind of effect is absolutely superior to other models.As the modeling mechanisms, assumptions and information from different sources, no single volatility models that incorporate and reflect only fluctuations in the local information, using combination of forecasts, there may be more reasonable to describe and characterize volatility characteristics. In the volatility forecasting, usually combined with linear regression models to predict fluctuations in the model's flaw which the weights are fixed, at different times, under different conditions using a fixed-weighted combination of linear regression models to characterize the volatility is not scientific. The combination of switching regression model assumed a variable state to achieve in different economic times under different conditions in which the model parameters and time-varying weights to switch. Overcoming the combination of a linear regression model weights fixed defects, can more accurately to characterize the time-varying characteristics of volatility. In the application of existing switching regression model there are still some obvious shortcomings and merits further in-depth research directions. First of all, most are based on a combination of state variables, switching regression model empirical research, but also rarely involve the introduction of multiple state variable model of the situation; Secondly, a small number of scholars have put forth a number of state variables to consider an extension of switching regression model, very few researchers of the state variables of time-varying characteristics of research.In this paper, based on the switching regression model, given the number of state variables and state variables of the probability distribution of time-variable conditions, presents a new combination of time-varying extended switching regression prediction method. Combination of stock price index of the specific characteristics of the individual forecasting methods, commonly, used to evaluate the prediction model, according to the scope of the individual selected three kinds of individual forecasting methods, and then applied the individual forecasting model to time-varying extended switching regression combination forecasting method. And its application to the Shanghai Composite Index Forecast for empirical analysis, the method predicted better than other commonly used combinations of individual forecasts and projections, and to better simulate the volatility of stock price indices trend for investors to predict stock research the reference model.