Hello,

I have panel data for a set of companies for the years 2001-2019. I am running fixed effect regression to check the impact of Geographical Diversification (GS) on company performance (Return on Assets or ROA.) I have divided the data between crisis years (pre 2007) and post crisis years (post 2007.) I noticed that the quadratic lagged GS term has a positive relationship with ROA in the pre-crisis years (regular U) but a negative relationship post crisis (inverted U.) I am trying to understand why crisis has this kind of moderation effect (i.e. it converts a U shaped relationship into an inverted U relationship.) I am pasting below my data for your reference (- with the two coefficients in RED font. Apologies that I could not format the pasted portion of the XTREG output better.) Any insight would be very welcome. Thanks.

Fixed-effects (within) regression Number of obs = 1,632
Group variable: ID Number of groups = 176

R-sq: Obs per group:
within = 0.1463 min = 1
between = 0.1141 avg = 9.3
overall = 0.1042 max = 15

F(8,1448) = 31.01
corr(u_i, Xb) = -0.5743 Prob > F = 0.0000

------------------------------------------------------------------------------------
ROA_win05 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------------+----------------------------------------------------------------
LnRev | 1.519683 .1349646 11.26 0.000 1.254936 1.78443
TDTE | -.000868 .0005299 -1.64 0.102 -.0019074 .0001714
Co_Age | -.2007525 .0397295 -5.05 0.000 -.278686 -.1228189
|
GS |
L1. | -10.90764 2.109404 -5.17 0.000 -15.04546 -6.769828
|
cL.GS#cL.GS | 8.629529 2.179336 3.96 0.000 4.354535 12.90452
|
1.Crisis | -1.665125 .5070016 -3.28 0.001 -2.659661 -.6705883
|
Crisis#cL.GS |
1 | 10.60625 2.250596 4.71 0.000 6.191469 15.02102
|
Crisis#cL.GS#cL.GS |
1 | -9.339311 2.26808 -4.12 0.000 -13.78839 -4.890236
|
_cons | 2.545409 .8957072 2.84 0.005 .7883866 4.302432
-------------------+----------------------------------------------------------------
sigma_u | 4.7423593
sigma_e | 3.8260412
rho | .60573235 (fraction of variance due to u_i)
------------------------------------------------------------------------------------
F test that all u_i=0: F(175, 1448) = 5.82 Prob > F = 0.0000