Conditional Estimation With The Missing Values Of Wind Velocity Based On VAR Model

The vector autoregression (VAR) model is one of the most successful, flexible, and easy to use models for the analysis of multivariate time series. It is a natural extension of the univariate autoregressive model to dynamic multivariate time series. VAR model in economics was made popular by Litterman,Sargent and Sims in1980s, and it has proven to be especially useful for describing the dynamic behavior of economic and financial time series and for forecasts. If some condition is added in the forecasts process, namely conditional forecasting, the precision can be improved more. In this paper, we will use the conditional forecasting to estimate the missing values of time series, and the result is reasonable.More specific, the primary contents of this paper are as follow:First, we introduce the conception of stationary time series and three basic kinds of stationary time series model. Then we discuss one of the important methods of stationary test-ADF unit root test.Second, we discuss the diversified expression of VAR model, and two methods of parameters estimation: one is maximum likelihood estimation and the other is Bayesian estimation. The Bayesian estimation is mostly used in the model which has more variables or the lagged order, that is to say, it has more parameters to estimate.Third, we discuss the two important applications of the VAR model. They are forecasting and Granger causality test. In this section the topic discussed is the conditional forecasting, which is use to forecast the unknown variables with the condition of the known variables.Fourth, The VAR model is established with the wind velocity data from Su Tong Bridge, Changshu observatory, Haimen observatory and Nantong observatory. Based on the VAR model, the result of Granger causality test indicated that there is Granger causality between most of the variables. Consequently the absent wind velocity data of Su Tong Bridge is forecasted with the condition of the wind velocity data from the three observatories, and the result of the conditional forecasting is comparatively perfect.