R Square太小说明模型识别不好；

不同时期回归系数变化较大，说明回归系数可能存在是时变性，

这也正说明了R Square也有变化

拟和优度太小说明模型建立的不正确

或者数据不能直接建立线性回归分析

可尝试先检验自相关性或异方差

Please read Greene's "Econometric Analysis" first. There are totally 6 assumptions for the classical linear regression.

One of the important note: only caculate R-square when the model intercept is included, since the sum of error may not be equal to zeo.

Additionally, R-square is increasing if a regressor is added. So R-square is not a good thing to measure the regression, and R-square is not comparable between different moels.

R小就像医学中的发烧一样，单凭这个看不出什么毛病，建议你更丰富的描述你的问题！“不同时期的数据的回归系数变化较大”，可能是你的数据自由度太小，也可能是你的模型非线性，而你硬用线性回归！