4、ARCH模型是对的吗?
如果ARCH模型选取正确,即回归残差的条件方差是按规律变化的,那么标准化残差就会服从标准正态分布,即不会有ARCH效应了。
对q为3的ARCH模型做LM test,发现没有了ARCH效应。
36
注意,虽然是同一个检验名称,但是ARCH过程后是对标准化残差进行检验。注意观察被解释变量或者依赖变量是什么?
ARCH Test: F-statistic 0.238360 Probability 0.992099 Obs*R-squared 2.470480 Probability 0.991299
Test Equation:
Dependent Variable: STD_RESID^2
37
Method: Least Squares
Date: 10/21/04 Time: 21:56 Sample(adjusted): 2010 2254
Included observations: 245 after adjusting endpoints Variable Coefficient Std. Error t-Statistic Prob. C
1.102371 0.264990 4.160043 0.0000 STD_RESID^2(-1) -0.038545 0.065360 -0.589741 0.5559 STD_RESID^2(-2) -0.003804 0.065308 -0.058252 0.9536 STD_RESID^2(-3) -0.057313 0.065303 -0.877649 0.3810 STD_RESID^2(-4) -0.010325 0.065277 -0.158169 0.8745 STD_RESID^2(-5) 0.003537 0.065280 0.054185 0.9568 STD_RESID^2(-6) -0.007420 0.065274 -0.113670 0.9096 STD_RESID^2(-7) 0.063317 0.065264 0.970165 0.3330 STD_RESID^2(-8)
-0.012167
0.065293
-0.186340
0.8523
38
STD_RESID^2(-9) -0.010653 0.065278 -0.163194 0.8705 STD_RESID^2(-10) -0.020211 0.065228 -0.309845 0.7570 R-squared 0.010084 Mean dependent var 1.007544 Adjusted R-squared -0.032221 S.D. dependent var 2.112747 S.E. of regression 2.146514 Akaike info criterion 4.409426 Sum squared resid 1078.160 Schwarz criterion 4.566625 Log likelihood -529.1546 F-statistic
0.238360 Durbin-Watson stat 2.000071 Prob(F-statistic) 0.992099 方程整体是不显著的。
还可以观察标准化残差
39
ARCH建模以后,procs/make residual series/可以产生残差?t和标准化残差?t/?t,以下分别是残差和标准化残差。可以看出没有了集群现象。
40