我想做样本外预测,但是发现根据回归出来的方程得到的,估计的误差很大。但是如下表所示,eviews输出的样本内预测值偏差不大,我想知道eviews样本内预测值是怎么算出来的,因为和用回归出来的方程算的不一样,回归方程就是表格最后一行的方程,各参数均显著。请高手指点,谢谢!
| | 实际值 | eviews样本内预测值 | 相对偏差=(预测值-实际值)/预测值 | 相对偏差的绝对值 |
| | 3203.106 | 3245.586443 | 0.013262 | 0.013262 |
| | 3256.754 | 3201.636385 | -0.01692 | 0.016924 |
| | 3276.891 | 3292.758319 | 0.004842 | 0.004842 |
| | 3177.368 | 3278.899963 | 0.031955 | 0.031955 |
| | 3114.071 | 3216.231088 | 0.032806 | 0.032806 |
| | 3219.059 | 3113.108531 | -0.03291 | 0.032913 |
| | 3308.393 | 3250.545464 | -0.01749 | 0.017485 |
| | 3319.475 | 3311.848358 | -0.0023 | 0.002298 |
| | 3317.141 | 3354.965047 | 0.011403 | 0.011403 |
| | 3389.395 | 3322.125071 | -0.01985 | 0.019847 |
| | | | | |
| | | | | |
| 样本外测试值 | 3511.05 | | | |
| 3512.07 | | | |
| 3535.68 | | | |
| 3451.11 | | | |
| 3442.66 | | | |
| | | | |
| | y=2004.515+1.030023y(t-2)+0.973686e(t-1) |
下面是回归方程时输出的数据
Dependent Variable: AINDEX
Method: Least Squares
Date: 07/01/15 Time: 10:39
Sample (adjusted): 3 245
Included observations: 243 after adjustments
Convergence achieved after 12 iterations
MA Backcast: 2
Variable Coefficient Std. Error t-Statistic Prob.
C 2004.515 177.1359 11.31625 0.0000
AR(2) 1.030023 0.011746 87.69497 0.0000
MA(1) 0.973686 0.015638 62.26484 0.0000
R-squared 0.991239 Mean dependent var 2344.657
Adjusted R-squared 0.991166 S.D. dependent var 303.8797
S.E. of regression 28.56209 Akaike info criterion 9.554307
Sum squared resid 195790.3 Schwarz criterion 9.597431
Log likelihood -1157.848 Hannan-Quinn criter. 9.571677
F-statistic 13576.47 Durbin-Watson stat 1.994518
Prob(F-statistic) 0.000000
Inverted AR Roots 1.01 -1.01
Estimated AR process is nonstationary
Inverted MA Roots -.97