Empirical Example: Returns to Scale in Electricity
Supply
Walter Sosa-Escudero
Econ 507. Econometric Analysis. Spring 2009
February 10, 2009
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Classic paper: Nerlove (1963), analyzed in detail in Hayashi
(2000).
Problem: returns to scale in a regulated industry (electric
power supply). See Hayashi for a discussion of the underlying
institutional framework.
Data: 145 in 44 states.
Variables: total costs, factor prices (wage rate, price of fuel,
and the rental price of capital), and output.
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Economic problem: cannot just look at costs vs. output.
Factor prices interact with output.
Cobb-Douglas technology:
Qi = A1 x1i1 x2i2 x3i4
Qi : output
xj : factors
Ai : firm heterogeneity: unobservable differences in
production efficiency.
r 1 + 2 + 3 measures degree of returns to scale.
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Cost function:
1/r /r /r /r
T Ci = r (11 22 33 )1/r Qi p1i1 p2i2 p3i3
Taking logs:
ln T C1 = i +
1
1
2
3
ln Qi +
ln p1i +
ln p2i +
ln p3i ,
r
r
r
r
i ln [r (11 22 33 )]
The linear econometric model is:
ln T C1 = 1 + 2 ln Qi + 3 ln p1i + 4 ln p2i + 5 ln p3i + i
with
2 = 1/r, 3 = 1 /r, 4 = 2 /r, 5 = 4 /r.
E(i ), i i , so E(i ) = 0.
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Role of classical assumptions
Linearity: is a consequence of the Cobb-Douglas tecnology (in
logs).
Strict exogeneity: factor prices are given to the firm
independently of efficiency. Ouput: if prices are set
independently of efficiency, Ok. If prices are set to cover
costs: unobserved efficiency is related to ouput.
No multicollinearity: Ok
Firm spillovers might harm the no-serial correlation
assumption.
Heteroskedasticity very likely to be present. Careful.
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
The Cobb-Douglas technology imposes the following restriction:
3 + 4 + 5 = 1,
a consequence of r 1 + 3 + 3 . This is a testable linear
restriction.
If we set 5 = 1 3 4 , replacing in the original model, the
restricted model is
p1i
p2i
T C1
= 1 + 2 ln Qi + 3 ln
+ 4 ln
+ i
ln
pi3
pi3
pi3
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Unrestricted model:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -3.56651
1.77938 -2.004
0.047 *
lq
0.72091
0.01743 41.352 < 2e-16 ***
lpl
0.45596
0.29980
1.521
0.131
lpf
0.42581
0.10032
4.244 3.97e-05 ***
lpk
-0.21515
0.33983 -0.633
0.528
--Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1
1
Residual standard error: 0.3923 on 140 degrees of freedom
Multiple R-squared: 0.926,
Adjusted R-squared: 0.9239
F-statistic: 437.9 on 4 and 140 DF, p-value: < 2.2e-16
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Restricted model:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -4.63607
0.89499 -5.180 7.52e-07
lq
0.72134
0.01739 41.477 < 2e-16
lpln
0.60647
0.20724
2.926
0.004
lpfn
0.41437
0.09878
4.195 4.81e-05
--Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1
***
***
**
***
1
Residual standard error: 0.3915 on 141 degrees of freedom
Multiple R-squared: 0.9272,
Adjusted R-squared: 0.9256
F-statistic: 598.3 on 3 and 141 DF, p-value: < 2.2e-16
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Returns to scale: 1.4 = 1/0.72091
How would you test H0 : 3 + 4 + 5 = 1? (do it).
R2 = 0.926 high or low?
Multicolinearity
cor(cbind(lq,lpl,lpf,lpk))
lq
lpl
lpf
lpk
lq
1.00000000 0.02350778 -0.1689466 -0.09877366
lpl 0.02350778 1.00000000 0.3337830 -0.19015184
lpf -0.16894664 0.33378301 1.0000000 0.13090722
lpk -0.09877366 -0.19015184 0.1309072 1.00000000
What do you think?
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
A FWLT approach to estimating the effect of output
Coefficients:
Estimate Std. Error t value Pr(>|t|)
lqs 0.72091
0.01719
41.94
<2e-16 ***
--Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1
Residual standard error: 0.3868 on 144 degrees of freedom
Multiple R-squared: 0.9243,
Adjusted R-squared: 0.9238
F-statistic: 1759 on 1 and 144 DF, p-value: < 2.2e-16
How are the dependent and independent variables defined?
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
log total cost vs. log output
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
The FWLT plot
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
FWLT Fit
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Residuals
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Test for heteroskedasticity: White test.
(Intercept)
lq
lpln
lpfn
I(lq^2)
I(lpln^2)
I(lpfn^2)
I(lq * lpln)
I(lq * lpfn)
I(lpln * lpfn)
Estimate Std. Error t value Pr(>|t|)
2.141043 14.231399
0.150 0.880638
-1.020659
0.374418 -2.726 0.007262 **
-0.986754
6.708450 -0.147 0.883279
0.438692
2.830667
0.155 0.877070
0.031975
0.004993
6.404 2.32e-09 ***
-0.187534
0.813741 -0.230 0.818084
-0.221966
0.176251 -1.259 0.210068
-0.060283
0.087892 -0.686 0.493970
-0.153918
0.044745 -3.440 0.000775 ***
0.061019
0.645043
0.095 0.924776
Residual standard error: 0.2805 on 135 degrees of freedom
Multiple R-squared: 0.493,
Adjusted R-squared: 0.4592
F-statistic: 14.58 on 9 and 135 DF, p-value: 2.653e-16
> 145*summary(auregw)$[Link]
[1] 71.48219
> qchisq(0.95,9)
[1] 16.91898
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply
Test for heteroskedasticity: Koenker test.
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.77486
0.09999
7.749 1.56e-12 ***
lq
-0.09544
0.01464 -6.517 1.14e-09 ***
Residual standard error: 0.3361 on 143 degrees of freedom
Multiple R-squared: 0.229,
Adjusted R-squared: 0.2236
F-statistic: 42.48 on 1 and 143 DF, p-value: 1.142e-09
> 145*summary(auregk)$[Link]
[1] 33.20744
> qchisq(0.95,1)
[1] 3.841459
Walter Sosa-Escudero
Empirical Example: Returns to Scale in Electricity Supply