The Application Of The Combination Forecasting Method In China’ CPI Forecast

In recent years, CPI has been the focus of attention. Especially after the secondhalf of2010, CPI continued to rises,and brought about adverse effects on China’seconomy, income distribution, foreign economic relations. Timely and accuratelyforecasting the CPI is the premise to ensure the national to make effective policies.In allusion to the uncertainty of the economic forecast,basing on summarizingthe previous’ study about combination forecasting methods and forecasting methodsof CPI，this thesis will introduce the combination forecasting method into theapplication of China’s CPI forecast,expecting to improve the accuracy of the CPIforecast.From the point of view of inflation, this thesis determines the leading indicatorand the leading time of CPI, useing VAR model, Granger causality test, variancedecomposition, impulse response function and other econometric methods, and avoidsthe subjectivity in selecting the leading indicators and the leading of CPI. Throughempirical analysis,this thesis obtains seven leading indicators: social retailgoods,public expectations, current money, narrow money supply, broad money supply,exports,imports. In addition, from the analysis of the variance decomposition results,the roles of narrow money supply, broad money supply and imports are particularlysignificant.On that basis, the thesis establishs three kinds of CPI forecasting model,namelyARIMA seasonal forecasting model,ARDL model and VEC model,and has forecastedthe CPI for January to October of2001,obtaining better predictions. The highershort-term forecasting accuracy of ARIMA seasonal forecasting model,thecointegration of ARDL model and VEC model,ensure the accuracy and stability ofthe combination forecasting method.On the base of determining the weight of the combination forecastingmethod.,the thesis establishs the equal weight combination forecasting model and therciprocal of error sum squares combination forecasting model.Comparing theevaluation Index of the prediction between the combination forecasting model and theindividual forecasting model, the results show that combination forecasting modelssignificantly improveds prediction accuracy.Combining with the two factors,that theforecasting model has different predict effect in different time and CPI series arehighly relevant to its previous value,the thesis presents a method of calculatingcombined weights, basing on the previous absolute error to calculate the combined weight,samely,the prediction effect is superior to single forecasting model.Using thetree established combination forecasting methods to forecast the CPI index fromJanuary to October of2011, predicted and actual values are close,among the the treecombination forecasting methods, the rciprocal of error sum squares combinationforecasting model is the best model. In the prediction, estimating parameters byperiod and recalculating the combined weights, ensure the timeliness of the model insome extent.Through the study of this thesis, the rciprocal of error sum squares combinationforecasting model have better accuracy comparing with other the individualforecasting models, and have some reference value in CPI forecast,the predicted valueof October is November4.5%.