Lec 7

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 39

CHAPTER 7

Functional
Form and
Logarithm
Menu for Functional Form
Today

Logarithms in Econ
Review
• Regression with Constant Term

• So far we have only considered linear regression


model with a constant term in it.

• Usually, the intercept has no economic meaning and thus it


should not be used for inference.
• Even though the constant term is of no economic use, it
should not be eliminated from our regression.
Different Functional Forms (refer to page 15

1. Linear
2. Log-Log
3. Semilog
• Linear-Log or Log-Linear
4. Reciprocal (or inverse)
1. Linear Regression Model
• The relationship between dependent variable and
independent variables ’s in the linear form can be expressed as

• It is linear in coefficients as well as variables. The (partial) slope


of a linear model is a constant. constant
j = 1, 2, …, k.
• The elasticity of Y with respect X jis not a constant.
to


The elasticity of Y with respect to X j is the percentage
change in caused by a 1% change in X j .

This elasticity is actually a point elasticity, which has
different value for different values of ji and Yi.
X

Linear model may be estimated by OLS. Linear model
should be used unless strong evidence that it is inappropriate
is found.
2. Log –Log Regression Model
• The log-log form is also called the double-log form
ln    1 ln X  ...  ln X i
Yi 0 1i k ki
•It is nonlinear in variables but linear in coefficients.
•The elasticities of the model are constant.

•Thus, log-log model is also known as constant elasticity model


•In a log-log  j can be interpreted as elasticity.
model,
• The slope of
j-Ygraph are not constant.
X

• Log-log model is transformed from the exponential regression


model 
Y   
i 1i 1...  2
k
e i
e X
0

X 
2i ki

where e = 2.718 is the base of the natural logarithm.


• The exponential regression model is nonlinear in coefficients and
variables. By taking natural logarithm to both sides, we have log-
log model
ln Yi   0  1 ln X 1i  ...   k
ln X ki  i
• The resulting model is linear in coefficients and thus may be
estimated by OLS.
A constant elasticity model.
Functional Forms of Regression models
Y
lnY

Y   1X  2
lnYln1 2 lnX
Quantity

Take
log

X lnX
Y price
lnY
Y   1X 2
Quantity

Take X
log

lnYln1 2 lnX
lnX
price
3. Semilog: Log-Lin Model
Log-lin model has log form in the dependent variable. Log-
lin model may be estimated by the OLS as it is linear in
coefficients.

lnYi  0   ... k X  i
1X1i  ki

Slope of log-lin model


X -j Ygraph is not constant.
in

Elasticity of log-lin model is not constant.


When to used Log-lin model…
• should be used when the impact
of
Xj Y is hypothesized to be
on
• increasing at an increasing rate
• decreasing at an increasing rate
• Log-lin model is applied in situation whereby Y adjusts in
percentage form to a unit change in X j.
• Example: Salary of a worker rises ( Y ) as a percentage of his/her
previous salary every year ( ).
j
X
• Also been used in the growth model. See example 6.4 (page
163, Gujarati and Porter 2009)
Semilog: Lin-Log Model
• Lin-log model has log form in the independent
variables. Slope of lin-log model is not constant.

Yi  0  1 ln  ...  ln X  i
X1i k ki

• Elasticity of log-lin model is not constant.


When to used lin-log model…
• should be used when the impact X j on Y is
of hypothesized to be
• increasing at an decreasing rate
• decreasing at an decreasing rate
• Example: consumption increase with the decreasing
rate with respect to the income or we could apply the
total utility theory.
• See example 6.5 (page 165: Gujarati and Porter, 2009)
4. Reciprocal Model
• In the reciprocal model, Y is expressed as a function of
the reciprocal (or inverse) of one or more of the X’s
variables.

Yi   1 
 0  1    2 X  i
 X1 2i
i 
• The slope of the X1–Y graph is not constant.

• Elasticity of the reciprocal model is not constant.


Example 6.6: (page 167: Gujarati and Porter, 2009)

The Phillips curve for the United States, 1958-1969;


(a) Reciprocal model; (b) linear model.
Summary of Functional Forms
Comparison between the functional form models

Linear Model

GDP  1.628M1
92.242
Double-log model
LOGGDP  1.318M1
1.667
Log-lin model
LOGGDP  4.798 0.02 M1

Lin-log model

GDP   727.056LOGM1
3592.667
Where M= 1/M1

Reciprocal model

GDP  1426.620  207060.3M


The Choice of Functional Form
• In our regression analysis, if wrong functional form is estimated,
• We commit specification error – A1&A9- RESET
• Estimated results are invalid for policy conclusion
• Estimated results are invalid for forecasting
• The choice of functional form should be based on the underlying
economic theory. A form that is linear in the variables should be
used until a specific hypothesis suggests otherwise.
• The rule of thumb are based on
1. information criteria: AIC, SC , FPE
2. forecast accuracy criteria –RMSE – Root mean square error
3. Residuals diagnostic. Correct functional form
should passed a battery of diagnostic tests –
JB, RESET
4. Ramsey’s RESET test of general specification
error (supplement of CHAP6).
5. MWD test
6. coefficient of variation(cv)
Alternative Criteria for comparing two different functional models:
The coefficient of variation:
^

It measures the average error of the sample regression


function (SE of regression) relative to the mean of Y (mean
dependent variable).

Linear, log-linear, and log-log equations can


be meaningfully compared.
The smaller C.V. of the model,
the more preferred equation (functional model).
Remember the 5 alternative

Slides 20 – 24, using CV, wh


How can we use a linear
estimator to fit a non-linear
relationship?
Logarithms in Econometrics
We can do something
“nonlinear” BEFORE plugging
into OLS.
OLS doesn’t know (or care)
where the Y ’s and X ’s
came from.
Instead of plugging X in
directly, we can do stuff to
it first.
• We can do “stuff” to X and Y before we plug
them into our OLS formula.
• If we do something nonlinear to Y and/or X, then our
linear regression of new-nonlinear-Y on new-
nonlinear-X will plot a nonlinear relationship
between Y and X.
• What can we do to X and Y to fit a curve?
• LOGARITHMS!!!!!!!!!!!!
• Regress

log(Y )   0  đlog(X
1
)
Our fitted value for b1 no longer tells us
the effect of a 1-unit change in X on Y. It
tells us the effect of a 1-unit change in
log(X) on log(Y).
Unit changes in log-X translate into
PERCENTAGE changes in X.

The percentage change in Y from a


1-percent change in X has a special name
in economics—elasticity!

Taking logs of both Y and X lets


us estimate elasticities!
Example: The Phillips Curve (before log)

• The US data from 1958–1969 suggest a


trade-off between inflation and unemployment.

Unemploymentt  0.06 - 0.55đInflationt


ˆ
 0  0.06
ˆ
  0.55
1
• How do we interpret these numbers?
• If Inflation were 0, our best guess of
Unemployment would be
0.06
percentage points.
• A one percentage point increase of Inflation
decreases our predicted Unemployment level by
0.55 percentage points.
Results (before log)
• If we log both Unemployment and Inflation,
then we can predict the percentage change in
unemployment resulting from a one percent
change in inflation.

• Percentage changes are nonlinear.

• Changing inflation from 0.01 to 0.02 is a 100%


increase. Changing inflation from 0.02 to 0.03
is only a 50% increase.
A Logarithmic Phillips Curve
log(Unemployment) =
-4.20 – 0.30·log(Inflation)
• A 1-percent increase in Inflation leads to a
-0.30% change in Unemployment.
• Caution: do not confuse percentage points
changes and percentage changes.
• If inflation is 15%, a 10 percentage
point change moves inflation to
25%
• If inflation is 15%, a 10 percent change
moves inflation to 16.5%

You might also like