Based The Gmdh The Theoretical Prediction And Early Warning Models And Empirical Analysis

The forecast and early-warning of import & export trade has become an important research topic in economics. In regard to the forecast, three limitations exist in past research home and abroad: First, the forecast of import & export trade requires large scale data, thus forecast results will be inaccurate with a small sample. However, a detailed statistic of the import & export data is available for several years. Besides, the small sample of the statistic also makes small sample modeling more difficult. Second, the forecast accuracy will be further lowered in long term forecast, which is also a problem of most forecasting models. Third, there is not an appropriate approach to deal with the non-stationary of small sample data. In order to solve these issues adopted an improved GMDH (Group Methods of Data Handling) in small sample data modeling.First of all, this paper analyzed and summed up the import & export forecasting and the status quo of the research. In accordance with the small sample of the import & export of trade in Sichuan, an effective method - GMDH prediction modeling was introduced to avoid the defects of mall samples data which the traditional econometrics modeling cannot be solved and improve the forecast results.Secondly, although the introduction of GMDH prediction model solved the problem of small sample of modeling, forecast accuracy will be lowered with the forecast term become longer for example 12-month long term forecast. Therefore, the GMDH two-level prediction model was used as a reference to improve the accuracy. The predictive scope of the downer-level can be expanded to the upper-level in the two-level model, which means that the spectrum of short-term forecast of average data per-month can be extended to the scope of the average data per-quarter. In this way, the inaccuracy of medium and long-term forecast was solved the GMDH, further improving the forecast accuracy of import & export trade.Finally, great deviations will be occurred when applying GMDH two-level forecast model to non-stationary data. In order to reduce deviations and improve accuracy, the paper introduced the method of automatic search of multi-structure-mutation-point to adjust the non-stationarity. Therefore, a two-level forecast model based on automatic search was established.In order to maintain the sustainable development of foreign trade, early- warning analysis of major export products was done on the basis of forecast. Logit model which is used most frequently currently was improved, with the introduction of GMDH group and algorithms and the use of computer automatically investigating the relationship among variables. Thus, the Logit model based on GMDH was structured. Then, the early-warning analysis of major export products were applied one by one in accordance with the characteristics of Sichuan Province. Reference was provided for the decision-making of government departments as well as import & export enterprises.