Pooled OLS regression with low R-squared

[Yuqi WANG]

Hi,
I'm trying to run a regression model using panel data (daily data from 2007 to 2010 over 97 companies). My dependant variable is the difference in the price of one derivative.
I've tried to use xtreg, fe and re and conduct both Hausman test and Breusch-Pagan Lagrange Multiplier test (results are attached). It seems I need to use pooled OLS regression.
So I simply type command reg y x1 x2 x3... in stata which returns me a regression with only 6% adjusted R-squared. But according to the literature, the variables that I use should have accounted for over 20% variations.

Could you help me on this issue? Did I do sth wrong? Many thanks!!




. hausman fixed random

---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fixed random Difference S.E.
-------------+----------------------------------------------------------------
L.dbid_ask | .1817405 .1818394 -.000099 .0002398
L.dequity | -.4908646 -.4915901 .0007254 .0007976
L.div | .2564792 .2564289 .0000503 .0001617
L.dleve | 121.7294 121.6831 .0462955 .1208107
L.dspx | -.0238714 -.0238692 -2.21e-06 .0000502
L.dvix | -.1120313 -.1120491 .0000178 .000341
L.drf | -3.895379 -3.896595 .0012162 .0069157
clearing | -.6685403 -.6600459 -.0084944 .0054932
------------------------------------------------------------------------------
b = consistent under Ho and Ha; obtained from xtreg
B = inconsistent under Ha, efficient under Ho; obtained from xtreg

Test: Ho: difference in coefficients not systematic

chi2(8) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 3.48
Prob>chi2 = 0.9011



. xttest0

Breusch and Pagan Lagrangian multiplier test for random effects

dcds[company,t] = Xb + u[company] + e[company,t]

Estimated results:
| Var sd = sqrt(Var)
---------+-----------------------------
dcds | 54.84847 7.405975
e | 51.61045 7.184041
u | 0 0

Test: Var(u) = 0
chibar2(01) = 0.00
Prob > chibar2 = 1.0000

[Joshua D Merfeld]

If you think your specification is correct, you shouldn't worry about the R2. Also, there is no a priori reason to expect your R2 to be the same as previous studies unless the data are identical. That is, if the observations are different or if the variables are different in any way from previous studies, you will likely find differences in your results.

I don't run LM tests for RE vs OLS very often, so I won't make any comments on what the problem actually is, but the fact that the test statistic on the LM test is zero suggests a problem with the command. Could you please post ALL of the commands you ran, including the xtreg and the saved estimates?

[Yuqi WANG]

If you think your specification is correct, you shouldn't worry about the R2. Also, there is no a priori reason to expect your R2 to be the same as previous studies unless the data are identical. That is, if the observations are different or if the variables are different in any way from previous studies, you will likely find differences in your results.

I don't run LM tests for RE vs OLS very often, so I won't make any comments on what the problem actually is, but the fact that the test statistic on the LM test is zero suggests a problem with the command. Could you please post ALL of the commands you ran, including the xtreg and the saved estimates?

Thank you for your reply. I've attached my commands.

xtreg dcds L.dbid_ask L.dequity L.div L.dleve L.dspx L.dvix L.drf clearing, fe
estimates store fixed
xtreg dcds L.dbid_ask L.dequity L.div L.dleve L.dspx L.dvix L.drf clearing, re
estimates store random
hausman fixed random

xtreg dcds L.dbid_ask L.dequity L.div L.dleve L.dspx L.dvix L.drf clearing, re
xttest0

[Yuqi WANG]

And my regression suffers heteroskedasticity and violates normality assumption. How can I deal with this issue or improve it a little bit in stata?

[Yuqi WANG]

Could anyone help me?? Many thanks!!!!!

[Carlo Lazzaro]

Yuqi:
- if the distribution of residuals in your regression suffers from heteroskedsticity, use robust standard errors (see -vce- options under -xtreg- entries);
- normality is required for residual distribution only and its violation bites hardly when sample size is small;
- as Joshua underlines, please post what you typed and what Stata gave you back via code delimiters (see the FAQ on both these topics).