前两天刚做了一个关于这个的~~~~~~

不过我做的是with respect to time dimention....感觉差不多，都应该是关于parameter stability的.....

我做的方法是，先对所有数据进行回归....call it regression equation 1.

然后再引入Indicator variable vector I_1(n)=1 if n belongs to larger company, 0 else..Another indicator variable vector I_2(n)=1 if n belongs to samll company, 0 elsewhere.

Now build new variables I_1*x and I_2*x, run regression y = \beta_1'I_1*x + \beta_2'I_2*x + \epsilon. Thus this regression is the resctricted model compared with equation 1.

More explicitly, we have a test: H_0: \beta_1=\beta_2, H_1: \beta_1 \neq \beta_2;

An LR test could be used here... this test statistic followed a Chi^2 distribution asymptotically...thus u could use the table to check for the critical value.....(What I did is with the unknown changing points...but here in ur case, I think u know the changing point..so it's simpler)

What I did in the assignment is then use the bootstrap to obtain an empirical(bootstrap) distribution of this LR statistic and derive the p-value with this empirical LR distribution....According to Diebold and Chen, this bootstrapped distribution performed better than assymptotical chi^2.....

Since u did not make it explicitly what u wanna do with bootstrap....to get the distribution for the statistic(what I described above)....or the statistic itself(in that case, the most simple case would be wild bootstrap)......thus cannot help u further~~~

good luck!

不好意思....这些东西我用英文写比用中文顺手....

虽然用英文写的也不清楚准确........

希望lz能看明白吧......

我是tex user...所以中间的equation用的是tex notation~~~

看平狄克的《计量经济模型与经济预测》，上面有介绍Chow test，就是用来检验两个回归模型的回归系数是否相等的。

第四版中译本，第82页。