Chapter 10 模型选择的标准及检验
Chapter 10 模型选择的标准及检验
判断模型好坏的标准( A C. Harvey 1.简约性( Parsimony) 2.可识别性( Identifiability)参数的估计唯一; 3. Goodness of fit越高越好 4.理论一致性( Theoretical consistency)与理论 或常识要一致。如在消费函数中,可支配收 入的系数一般为正; 5. Predictive power
判断模型好坏的标准(A. C. Harvey): 1. 简约性(Parsimony); 2. 可识别性(Identifiability)参数的估计唯一; 3. Goodness of fit 越高越好; 4. 理论一致性(Theoretical consistency) 与理论 或常识要一致。如在消费函数中,可支配收 入的系数一般为正; 5. Predictive power 2 R
Model specification errors(模型设定失误) 1. Missing key independent variables true model: y=B2x2+B3x3+8 specified model y=B,x,+E (here suppose x,=x=y=0 Then ECB*)=B,+B In general, the estimator is biased Example( USing Eviews):Y支出,X—可支 配收入,Z时间趋势变量
Model specification errors(模型设定失误): 1. Missing key independent variables true model: specified model: (here suppose ) Then In general, the estimator is biased. Example (Using Eviews):Y—支出,X—可支 配收入,Z—时间趋势变量。 2 2 3 3 y x x = + + * * 2 2 y x = + 2 3 x x y = = = 0 * 2 3 2 2 3 2 2 ˆ ( ) x x E x = +
2. Including irrelevant variables true model y=B,x,+8 specified model: y=B2x2+B33+e It is can be proved that var(>var(B) 3.不正确的函数形式
2. Including irrelevant variables true model: specified model: It is can be proved that 3. 不正确的函数形式 2 2 y x = + * * * 2 2 3 3 y x x = + + * 2 2 var( ) var( ) 垐
设定误差的检验 1.诊断非相关变量的存在 Use t-test or f-test 2.遗漏变量和不正确的函数形式的检验 1)R2 2) t-statistic 3)与预期比较,估计系数的符号 4) DW-statistic 5)预测误差
设定误差的检验 1. 诊断非相关变量的存在 Use t-test or F-test 2. 遗漏变量和不正确的函数形式的检验 1) ; 2)t-statistic 3) 与预期比较,估计系数的符号 4)DW-statistic 5) 预测误差 2 R
残差检验法与DW统计量检验法 Example(Using eviews to show) There are other tests for model specification such as Ramsey resettest: likelihood ratio test Wald test. hausman test and so on 建模既是一门科学也是一门艺术!! —C.W.J. Granger
残差检验法与DW统计量检验法 Example (Using Eviews to show) There are other tests for model specification such as Ramsey RESET test; likelihood ratio test Wald test; Hausman test and so on. 建模既是一门科学也是一门艺术!!! ——C. W. J. Granger