Problem with standardized coefficients in rollling regressions

[Lisa Wilson]

Hi

I want to compare two coefficients in a regression to group observations based on the magnitudes of the coefficients. The regression is estimated for each firm on market variables using a firm time series in a rolling fashion.

I present two approaches where the market variables are standardized first before running the regressions or when the betas are standardized.

* Approach A:

bys firmID: asreg firm_ret market1 market2 , wind(qdate 30) minimum (20) 

egen std_b_market1=std(_b_market1)
egen std_b_market2=std(_b_market2)

sum std_b_market1 std_b_market2

* Approach B:

egen std_market1=std(market1)
egen std_market2=std(market2)

bys firmID: asreg firm_ret std_market1 std_market2 , wind(qdate 30) minimum (20) 

sum _b_std_market1 _b_std_market2

I think the norm is to standardize the variables before running the regression?! However, in this case, doing so and then using rolling regressions does not bring the betas to means of zero and standard deviations of 1 (as the market variables were initially standardized for the full sample). However, approach B will lead to betas with means 0 and standard deviations 1.


My questions

1- Does it make sense to use approach B so that I can subsequently compare between the standardized betas?

2- If approach B is incorrect, how to fix the issue in approach A?

Thank you

[Lisa Wilson]

Put it simply, how to get standardized coefficients in a rolling regression?

[Rich Goldstein]

I am not familiar with -asreg- and my response ignores that part; you can use the official -rolling- command and use standard regression options to get what you want; see

h rolling
h regress