Using gret l for Principles of Econometrics, 4th Edition

Log-Log Model

Finally, a log-log model is used. This functional form is often used to estimate demand equations as it implies a constant price elasticity for the commodity in question. This example uses the newbroiler. gdt which is adapted from Epple and McCallum (2006). The variable Q is per capita consumption of chicken, in pounds and P is the real price in dollars. The sample is from 1950-2001. The estimated log-log model is

l_q = 3.71694 - 1.12136 l_p

(0.022359) (0.048756)

T = 52 R2 = 0.9119 F(1, 50) = 528.96 <7 = 0.11799
(standard errors in parentheses)

The coefficient on logarithm of P is 1.121 which means that a 1% increase in the real price of chicken will decrease quantity demanded by 1.121%.

Once again, the predictor of quantity needs to be corrected since the model is estimated in logarithms. Qc = exp (b + b2 ln(x) + 72/2) = eln(Q)e<J /2. The R2 statistic can be computed as the squared correlation between Q and Q. The script for this exercise is:

1 open "@gretldirdatapoenewbroiler. gdt"

2 logs q p

3 ols l_q const l_p

4 series yht=$yhat

5 series pred = exp(yht)

6 series corrected_pred=pred*exp($sigma"2/2)

7 scalar r2= corr(corrected_pred, q)"2

8 gnuplot corrected_pred q p

The results are

? scalar r2= corr(corrected_pred, q)"2 Generated scalar r2 = 0.881776

and the corresponding graph is found in Figure 4.20.

4.5

Script

set echo off

# estimate model by LS and predict food_exp open "@gretldirdatapoefood. gdt"

ols food_exp const income

scalar yhat0 = $coeff(const) + $coeff(income)*20

# prediction interval ols food_exp const income

scalar yhat0 = $coeff(const) + $coeff(income)*20 scalar f=8013.2941+(8013.2941/40)+4.3818*(20-19.6047)rt2 scalar ub=yhat0+2.0244*sqrt(f) scalar lb=yhat0-2.0244*sqrt(f)

# prediction interval using accessors ols food_exp const income

scalar yhat0=$coeff(const)+20*$coeff(income) scalar sig2 = $ess/$df

scalar f = sig2 + sig2/$nobs + ((20-mean(income))rt2)*($stderr(income)rt2) scalar lb = yhat0-critical(t,$df,0.025)*sqrt(f) scalar ub = yhat0+critical(t,$df,0.025)*sqrt(f)

# correlations

ols food_exp const income --anova c1 = corr(food_exp,$yhat)

# linear-log model

series l_income = ln(income) ols food_exp const l_income series yhat2 = $yhat gnuplot yhat2 food_exp income

# simple data plot

open "@gretldirdatapoech4sim1.gdt" gnuplot e x

# residual plot

open "@gretldirdatapoech4sim2.gdt" ols y const x series ehat = $uhat gnuplot ehat x

# normality tests

open "@gretldirdatapoefood. gdt"

ols food_exp const income

series uhat1 = $uhat

summary uhat1

normtest uhat1 —jbera

normtest uhat1 —all

modtest —normality

# polynomial

open "@gretldirdatapoewa-wheat. gdt" ols greenough const time gnuplot greenough time

setinfo greenough - d "Wheat yield in tonnes" - n "Yield in tonnes" gnuplot greenough —with-lines —time-series

series t3=timert3/1000000 ols greenough const t3

gnuplot greenough —with-lines —time-series

open "@gretldirdatapoewa-wheat. gdt" series lyield = log(greenough) ols lyield const time

# log-linear model

open "@gretldirdatapoecps4_small. gdt" logs wage

ols l_wage const educ

scalar lb = $coeff(educ) - 1.96 * $stderr(educ) scalar ub = $coeff(educ) + 1.96 * $stderr(educ) print lb ub

open "@gretldirdatapoecps4_small. gdt" logs wage

ols l_wage const educ series y = exp($yhat) scalar corrl = corr(y, wage) scalar Rsquare = corr1rt2 print corr1 Rsquare

# simple prediction in log-linear model open "@gretldirdatapoecps4_small. gdt" logs wage

ols l_wage const educ

scalar l_wage_12 = $coeff(const)+$coeff(educ)*12 scalar nat_pred = exp(l_wage_12) scalar corrected_pred = nat_pred*exp($sigmart2/2) print l_wage_12 nat_pred corrected_pred

smpl educ=12 --restrict summary wage smpl full

# prediction intervals using a loop open "@gretldirdatapoecps4_small. gdt" logs wage

ols l_wage const educ scalar sig2 = $ess/$df matrix sem = zeros(21,5) loop for i = 1..21 --quiet

scalar yh = ($coeff(const) + $coeff(educ)*i)

scalar f = sig2 + sig2/$nobs + ((i-mean(educ))rt2)*($stderr(educ)rt2)

sem[i,1]=i

sem[i,2] = yh

sem[i,3]=sqrt(f)

sem[i,4]=exp(yh-critical(t,$df,0.025)*sqrt(f)) sem[i,5]=exp(yh+critical(t,$df,.025)*sqrt(f)) endloop print sem

nulldata 21 —preserve series ed=sem[,1] series wage=exp(sem[,2]) series lb=sem[,4] series ub=sem[,5]

# corrected predictions in log-linear model open "@gretldirdatapoenewbroiler. gdt" logs q p

ols l_q const l_p series yht=$yhat series pred = exp(yht)

series corrected_pred=pred*exp($sigmart2/2) scalar r2= corr(corrected_pred, q)rt2 gnuplot corrected_pred q p