Adequate Number of Data in a Window in Rolling Regression

[bok1234]

Dear Mr.Doan,

I am trying to test the model as below through rolling regression.
Total number of sample is 44 semi-annual data (22 years).

Yt = b*Xt + a

I want to know the adequate or minimum number of data in a window in this rolling regression.
Even in this case, should I input at least 30 data in unit window?
I guess that adequate number of data in a window in rolling regression should depend on the number of parameters just like general or usual regression test.

[TomDoan]

What are you "testing"?

Obviously the window size can't be too small relative to the number of regressors. Regardless of the size of the overall sample, the wider is the window, the less the estimates can change from window to window.

[bok1234]

I am testing the model as below through rolling regression.
Total number of sample is 44 semi-annual data (22 years)

Yt = b*Xt + a

Y : inflation expectations
X : actual inflation

What about 10 data (5 years) in each window?

[TomDoan]

Sorry. How are you "testing" anything? 10 observation estimates won't tell you much of anything.

[bok1234]

I know that 10 data-regression is stupid. But I have only 44 data and if I should select 30 in each window, then I will get only 14 windows as a rolling regression result. Is this good?

[TomDoan]

Since I have no idea what you think the rolling regressions are going to show, I really can't answer that. You have a regression with noisy proxies on both the left and the right, and probably highly serially correlated residuals. If you do rolling estimates, you'll probably discover that the sampling error is considerable, but you can probably tell that without running rolling estimates. You might want to read the excerpt on rolling causality tests.