哈尔滨消费函数的构建-中国统计教育学会.DOC

基于时序模型的哈尔滨市居民消费相关性研究 东北林业大学 邹敏、宋丽影、王艳 摘 要 本文从对西方消费函数理论假说出发，分别构建了哈尔滨市居民人均消费支出、居民活期存款利率、居民家庭人均可支配收入、居民人均储蓄余额、居民消费价格指数、房屋销售价格指数的ARIMA模型。由于自变量之间具有高度的相关性，则首先构建以人均消费支出为因变量，其余五项指标分别为自变量的五个协整回归模型，然后建立五项指标与因变量的岭回归模型，进行综合因素的分析。 在模型构建成功的基础上，利用ARIMA模型分别预测2010年1季度——2012年4季度六项指标的预测值，Theil不相等系数都在0.05以下，预测协方差比例值都在0.9以上，模型预测结果非常理想。将ARIMA模型预测值与真实值组合成新序列，运用新序列进行BP神经网络训练预测，其训练实际精度高达0.001064。运用BP神经网络预测到2012年第四季度哈尔滨市人均消费支出将达到4596.1791元。而2011年正处于“十二五”规划的开局之年，经济等各方面才处于起步阶段，因此，BP神经网络对哈尔滨市居民人均消费支出的预测结果与ARIMA模型预测结果相比更具有现实意义。 关键词：ARIMA 协整回归模型 岭回归 BP神经网络 Based on Time Series Model of Capita Consumption-related Research in Harbin Abstract This article starts from the hypothesis of the Western theory of consumption function, and respectively built ARIMA model in Harbin city from six parts which are per capita consumption expenditure, capita demand deposit interest rates, per capita disposable income of households, per capita savings balances, consumer price index, home sales price index. Because with high degree of correlation between these six variables, first building cointegrated regression model, choose per capita consumption expenditure as the dependent variable, and the remaining five indexes respectively as independent variable, then establishing Ridge regression model for comprehensive analysis of the factors. On the basis of the successful model, using ARIMA model respectively predicting six indicators for the 8 quarters from 2011 to 2012, by the result, we can know, the value of Theil is below 0.05, and predicted covariance ratio value above 0.9, that showing us the model predictions is ideal. Put the predict values and actual values into a new sequence, using BP neural network training to predict new sequences, their training real precision is up to 0.001064, and per capita consumption expenditure on the forth quarter in 2012 which will amount to ￥ 4596.1791. Because 2011 is the first year in the twelfth five years plan ,economic and other sectors is in its infancy, so the predict