Financial Econometrics
MSc Finance
Academic Year 2020/2021
1st Semester
Exercises
Multiple linear regression model
29/10/2020
Suppose a student is asked to study the determinants of the salaries of chief executive officers
(CEO) for U.S. corporations. Using information about 209 firms the student estimated by OLS
the following regressions:
ln salaryi = α0 + α1 ln salesi + α2 roei + α3 rosnegi + ui (1)
ln salaryi = β0 + β1 ln salesi + β2 roei + β3 f inancei + vi (2)
ln salaryi = θ0 + θ1 ln salesi + θ2 roei + θ3 rosnegi + θ4 ln salesi rosnegi
+θ5 roei rosnegi + wi (3)
where ln is the natural logarithm, salaryi = annual compensation in thousands of dollars,
salesi = annual sales in thousand dollars, roei = return on equity (in percentage), and rosnegi
is a binary variable that equals 1 if the return on firm’s stock (in percentage) is negative. The
variable f inancei is a dummy variable that equals 1 if the firm belongs to the financial sector.
The outputs obtained from the software gretl are the following:
Model 1: OLS, using observations 1–209
Dependent variable: l salary
Coefficient Std. Error t-ratio p-value
const 4.29760 0.293253 14.65 0.0000
l sales 0.288387 0.0336172 8.579 0.0000
roe 0.0166571 0.00396809 4.198 0.0000
rosneg −0.225675 0.109338 −2.064 0.0403
Mean dependent var 6.950386 S.D. dependent var 0.566374
Sum squared resid 46.93196 S.E. of regression 0.478473
R2 0.296606 Adjusted R2 0.286313
F (3, 205) 28.81470 P-value(F ) 1.37e–15
Model 2: OLS, using observations 1–209
Dependent variable: l salary
Coefficient Std. Error t-ratio p-value
const 4.30578 0.292083 14.74 0.0000
l sales 0.273784 0.0329345 8.313 0.0000
roe 0.0194431 0.00397804 4.888 0.0000
finance 0.182657 0.0810419 2.254 0.0253
Mean dependent var 6.950386 S.D. dependent var 0.566374
Sum squared resid 46.74884 S.E. of regression 0.477539
2 2
R 0.299351 Adjusted R 0.289097
F (3, 205) 29.19524 P-value(F ) 9.25e–16
1
Model 3: OLS, using observations 1–209
Dependent variable: l salary
Coefficient Std. Error t-ratio p-value
const 4.07448 0.306678 13.29 0.0000
l sales 0.314236 0.0351504 8.940 0.0000
roe 0.0172612 0.00404558 4.267 0.0000
rosneg 2.09463 1.00944 2.075 0.0392
lnsalesrosneg −0.257574 0.111999 −2.300 0.0225
roerosneg −0.00342803 0.0178213 −0.1924 0.8477
Mean dependent var 6.950386 S.D. dependent var 0.566374
Sum squared resid 45.71688 S.E. of regression 0.474559
2 2
R 0.314817 Adjusted R 0.297941
F (5, 203) 18.65426 P-value(F ) 3.08e–15
Coefficient covariance matrix
const l sales roe rosneg lnsalesrosneg
0.094051 −0.010414 −0.00041021 −0.094051 0.010414 const
0.0012356 1.4683e–05 0.010414 −0.0012356 l sales
1.6367e–05 0.00041021 −1.4683e–05 roe
1.0190 −0.10925 rosneg
0.012544 lnsalesrosneg
roerosneg
0.00041021 const
−1.4683e–05 l sales
−1.6367e–05 roe
−0.0031934 rosneg
−0.00011568 lnsalesrosneg
0.00031760 roerosneg
(a) Interpret the estimates of the regression coefficients of model (1).
(b) Is the elasticity of salaries with respect to the firm sales statistically significant? Use model
(1) to motivate your answer.
(c) Is there evidence that CEO’s of firms from the financial sector have higher compensations
than that of firms belonging to other sectors, all other factors being equal?
(d) Would you say that factors related to the firm performance determine the CEO salaries?
Answer to the question without using model (3).
(e) For firms with a negative ros, does an increase on the annual sales conduce to a drop on the
CEO compensation, holding other factors fixed? Motivate your answer.
(f) Describe how you can test for heteroskedasticity in model (3) using the White’s test. State
the test equation and assume that the R2 for this model is 0.0135. Can you conclude that the
error term of model (3) is heteroskedastic? What are the implications in terms of statistical
inference based on the results of that model?
