Heteroscedastic model of regression-time series

A heteroscedastic model of regression-time series is established. Based on the autoregression equation for the standard variation of the residual errors obtained by ordinary regression analysis, the least square estimates and the maximum likelihood estimates for the regression, the autoregression and the moving average coefficients are derived. The model can exploit the particular advantages of regression analysis and time series under the condition of the small sample, and make compensation for the residual errors of regression model to increase its precision. Both the independent and the correlated residual errors are respectively discussed in detail. Calculations show that higher precision can be gained by using the present method than the traditional one in analysis and prediction.