话说时间序列的预测一般是一步预测（就是预测后一期）这个感觉是滞后一期。不知你预测的步骤如何。通常想eviews，是先要扩充样本（增加新的自变量值）然后再做一步预测的……

我用的是matlab的forecast函数。不知道是动态预测还是静态预测

时间序列预测，也就是移动加权平均的一种而已，只不过显得有“统计学基础”，实际与如下重多移动平均线没啥子实质区别，预测均是向后瞧得，只不过你的图，只说明，预测仅仅是X(t)=X(t-1)。昨天的就是今天的，常人常这样想的。 好点该是：X(t)=a*X(t-1)+(a-1)*X(t-2)， a在[0,1]。再好点  向后瞧更远点。。
SMA calculates the arithmetic mean of the series over the past n observations.
EMA calculates an exponentially-weighted mean, giving more weight to recent observations. See
Warning section below.
WMA is similar to an EMA, but with linear weighting if the length of wts is equal to n. If the length
of wts is equal to the length of x, the WMA will use the values of wts as weights.
DEMA is calculated as: DEMA = (1 + v) * EMA(x,n) -EMA(EMA(x,n),n) * v (with the corresponding
wilder and ratio arguments).
EVWMA uses volume to define the period of the MA.
ZLEMA is similar to an EMA, as it gives more weight to recent observations, but attempts to remove
lag by subtracting data prior to (n-1)/2 periods (default) to minimize the cumulative effect.
VWMA and VWAP calculate the volume-weighted moving average price.
VMA calculate a variable-length moving average based on the absolute value of w. Higher (lower)
values of w will cause VMA to react faster (slower).
HMA a WMA of the difference of two other WMAs, making it very reponsive.
ALMA inspired by Gaussian filters. Tends to put less weight on most recent observations, reducing
tendency to overshoot

我也遇到相同问题，楼主解决了吗？想请教一下

很大可能模型采用单步预测，而每次输出只是记住了上一步真实值的结果，所以该预测方式是失效的