Functional forms of regression models

We will discuss the following regression

models:
Lecture 7
1. The log – log model
FUNCTIONAL FORMS OF 2. Semilog models: lin-log and log-lin models

REGRESSION MODELS 3. Reciprocal models

 The special features of each model , when

they are appropriate, and how they are
Gujarati D., Basic Econometrics, (2004) estimated
Chapter 6, p. 175  201

The LOG – LOG model The LOG – LOG model

Consider the following model: If the assumptions of the classical linear
regression model are fulfilled, the parameters of
Yi  1 X  2 e ui (6.5.1) (6.5.3) can be estimated by the OLS method by
letting
which may be expressed alternatively as
Yi *     2 X i*  ui , where
ln Yi  ln 1   2 ln X i  ui (6.5.2)
Yi *  ln Yi and X i*  ln X i
where ln=natural log (i.e. Log to the base e, and
where e = 2.718)
To obtain model which is linear in the parameters, The OLS estimators ˆ and ˆ2 obtained will
denote  = ln1 be best linear unbiased estimators of  and 2
respectively.
ln Yi     2 ln X i  ui (6.5.3)

Econometrics 1
 The LOG – LOG model The LOG – LOG model

^ ^ If Y represents the quantity of commodity demanded and

ln Y i  ˆ  ˆ2 ln X i , uˆi  ln Yi  ln Y i X its unit price, 2 measures the price elasticity of
demand, a parameter of considerable economic interest.
One attractive feature of the log-log model, which has
made it popular in applied work, is that the slope
coefficient 2 measures elasticity of Y with respect
to X:

2 shows the percentage change in Y for a given

(small) percentage change in X.

Example 2160
LOG -LIN model
Purchase price  Production 
of wheat  of wheat  2140
(lats per t) (thsd t)  PRF in case of log-lin model:

Yi  β1eβ 2 X i e ui
1995 70.34 2065.34 2120
1996 94.72 2090.72
1997 78.57 2075.57 2100
1998 63.84 2061.84 where β1, β2  population regression coefficients, ui 
1999 60.3 2059.3 2080 residual term.
For sample data: Yˆ  ˆ e  2 X i e uˆi
2000 61.19 2061.19 ˆ
2001
2002
59.14
59.58
2060.14
2061.58
2060
y = 1693.7x0.0476
R² = 0.9652

i 1
2003 63.1 2066.1
2004 67.91 2071.91
2040
where ˆ1 and ˆ2  OLS estimates for unknown
40 60 80 100 120 140 160
2005 61.56 2066.56 population parameters, û i  estimate for residual term
2006 77.83 2083.83
Increase in purchase price of wheat by 1%  Take log from both sides:
2007
2008
132.11
108.76
2139.11
2116.76 leads to about 0.0479% change in production  
ln Yi  ln 1e  2 X i eui  ln 1   2 X i  ui
2009 79.75 2088.75 of whaet.  2 shows: if value of X increases by 1 unit,
2010 110.21 2120.21 Data source: www.csb.gov.lv Y mean increases by 2 *100%
2011 138.82 2149.82

Econometrics 2
 LOG-LIN model How to measure the Growth rate
How to measure the Growth rate: application of
The LOG-LIN model  r is the compound (i.e., over time) rate of growth of Y:
Economists, businesspeople, and governments Yt = Y0 (1+r)t,
are often interested in finding out the rate of
growth of certain economic variables, such as whereY0 is the initial value of Y.
population, GDP, money supply, employment,  Taking log from both sides:
productivity, and trade deficit.  
ln Yt  ln Y0 (1  r) t  ln Y0  t  ln(1  r)
Denote: ln Y0 = ln β1 un β2 = ln(1+r), so we obtain log-lin
Let have values Y0; Y1; ....,Yt, where t - time model
Suppose that we want to estimate compound  2 shows: β2 = ln (1+r) = d(ln Y)/dX= (1/Y) dY/Y/dt =
(i.e., over time) rate of growth of Y (dY/Y)/dt
or β2 ≈( Y/Y )/ t
 Y/Y ≈ β2 t and 100 β2 shows mean change (percentage )
over the given time period
 Or 100 times β2 gives the growth rate in Y.

Example Example: interpretation

 The coefficient of the trend variable in the growth model 2
gives the instantaneous (at a point of time) rate of growth and
not the compound (over a period of time) rate of growth.

 But the latter can be easily found from model by taking

antilog of the estimated 2 and subtracting 1 from it and
multiplying the difference by 100.
 Thus, for our illustrative example:
ˆ2  0,0621
therefore
exp(0,0621)  1  0,0641
Thus, in our example the compound rate of growth on GDP
per capita in Latvia was about 6.41 % per year.

Econometrics 3
 LIN-LOG model
Example
PRF is following:
Yi = β1 + β2 ln Xi + ui
Sample regression function is:

Yˆi  ˆ1  ˆ2 ln X i , uˆi  Yi  Yˆi

Slope coefficient:
β2 =dY/d(ln X)=dY/(dX/X) or β2 ≈ Y/(X/X)

Thus: Y ≈ β2 (X/X) and 0,01 β2 shows change in Regression slope coefficient is ˆ2  10359
Y value, if X changes by 1% Interpretation: Increase in M2 by 1 percent, on
average, leads to about 103.57 milj. of euro increase
in GDP.

Reciprocal model Reciprocal model

Models of following type are known as
reciprocal models:
Yi = β1 + β2 (1/ Xi) + ui
Features:
o As X increases indefinitely, the term β2 (1/ Xi)
approaches zero and Y approaches the limiting or
asymptotic value β1
o Therefore: reciprocal models have built in them
an asymptote or limit value that the dependent
variable will take when the value of X increases
indefinitely.

Econometrics 4
 Reciprocal model
Example: Philips curve for Latvia
The Most popular application of reciprocal 40
WG=-17.97+298.9*1/(UNEMPL)
model is The Philips curve.
30

20

WG
10

-10

-20
4 6 8 10 12 14 16 18 20 22

UNEMPL

Choice of Functional Form

Choice of Functional Form The choice of a particular functional form may be comparatively easy in the
two-variables case, because we can plot the variables and get rough idea
about appropriate model.
Some guidelines:
1. The underlying theory.
2. Change in units vs. %
3. The coefficients of the model chosen should satisfy certain a
priori expectations.
4. Sometimes more than one model may fit a given set of data
reasonably well. Then we can use R2 value. But make sure
that in comparing two R2 values the dependent variable, or
regresand, of the two models is the same; the regressor(s)
can take any form.
5. What is of greater importance is the theoretical underpinning
of the chosen model, the signs of the estimated coefficients
and their statistical significance. If a model is good on these
criteria, a model with a lower R2 may be quite acceptable.

Econometrics 5
 Comparing two R2
Summary R2 values for two models are comparables only if
Model Interpretation of ̂ 2 dependent variable is the same for both models.
LOG- … model:
  ln Y  ln Y 
^
When X1 unit, Y mean by ̂ 2 2
Yˆi  ˆ1  ˆ2 X i ESS
units R2  
 ln Y  ln Y 
2
TSS
When X1 %, Y mean  by
Yˆi  ˆ1  ˆ2 ln X i 0.01* ̂ 2 units LIN- … model:
1 Has no direct interpretation R 
2 ESS  Yˆ  Y


2

Yˆi  ˆ1  ˆ2 TSS  Y  Y 2
Xi
 When X1 unit, Y mean by
ln Y i  ˆ1  ˆ2 X i 100* ̂2 % These values are not directly comparable. If we
 want to compare, we have to make additional
ln Y i  ˆ1  ˆ2 ln X i When X1 %, Y mean by ̂ 2 %
calculations.

Comparing two R2

Comparable R value we can obtain = CORREL( Y; antilog )^2

Econometrics 6