Type author name/s here

Dougherty

Introduction to Econometrics,
5th edition
Chapter heading
Chapter 4: Nonlinear Models and
Transformations of Variables

© Christopher Dougherty, 2016. All rights reserved.

 LINEARITY AND NONLINEARITY

Linear in variables and parameters:

Y  1   2 X 2   3 X 3   4 X 4  u

This sequence introduces the topic of fitting nonlinear regression models. First we need a
definition of linearity.

1
 LINEARITY AND NONLINEARITY

Linear in variables and parameters:

Y  1   2 X 2   3 X 3   4 X 4  u

The model shown above is linear in two senses. The right side is linear in variables
because the variables are included exactly as defined, rather than as functions.

2
 LINEARITY AND NONLINEARITY

Linear in variables and parameters:

Y  1   2 X 2   3 X 3   4 X 4  u

It is also linear in parameters since a different parameter appears as a multiplicative factor

in each term.

3
 LINEARITY AND NONLINEARITY

Linear in variables and parameters:

Y  1   2 X 2   3 X 3   4 X 4  u

Linear in parameters, nonlinear in variables:

Y   1   2 X 22   3 X 3   4 log X 4  u

The second model above is linear in parameters, but nonlinear in variables.

4
 LINEARITY AND NONLINEARITY

Linear in variables and parameters:

Y  1   2 X 2   3 X 3   4 X 4  u

Linear in parameters, nonlinear in variables:

Y   1   2 X 22   3 X 3   4 log X 4  u
Z 2  X 22 , Z3  X3 , Z 4  log X 4

Such models present no problem at all. Define new variables as shown.

5
 LINEARITY AND NONLINEARITY

Linear in variables and parameters:

Y  1   2 X 2   3 X 3   4 X 4  u

Linear in parameters, nonlinear in variables:

Y   1   2 X 22   3 X 3   4 log X 4  u
Z 2  X 22 , Z3  X3 , Z 4  log X 4
Y  1   2 Z 2   3 Z 3   4 Z4  u

With these cosmetic transformations, we have made the model linear in both variables and
parameters.

6
 LINEARITY AND NONLINEARITY

Linear in variables and parameters:

Y  1   2 X 2   3 X 3   4 X 4  u

Linear in parameters, nonlinear in variables:

Y   1   2 X 22   3 X 3   4 log X 4  u
Z 2  X 22 , Z3  X3 , Z 4  log X 4
Y  1   2 Z 2   3 Z 3   4 Z4  u

Nonlinear in parameters:

Y  1   2 X 2   3 X 3   2  3 X 4  u

This model is nonlinear in parameters since the coefficient of X4 is the product of the
coefficients of X2 and X3. As we will see, some models which are nonlinear in parameters
can be linearized by appropriate transformations, but this is not one of those.
7
 LINEARITY AND NONLINEARITY

Average annual percentage growth rates

Employment GDP Employment GDP

Australia 2.57 3.52 Korea 1.11 4.48

Austria 1.64 2.66 Luxembourg 1.34 4.55
Belgium 1.06 2.27 Mexico 1.88 3.36
Canada 1.90 2.57 Netherlands 0.51 2.37
Czech Republic 0.79 5.62 New Zealand 2.67 3.41
Denmark 0.58 2.02 Norway 1.36 2.49
Estonia 2.28 8.10 Poland 2.05 5.16
Finland 0.98 3.75 Portugal 0.13 1.04
France 0.69 2.00 Slovak Republic 2.08 7.04
Germany 0.84 1.67 Slovenia 1.60 4.82
Greece 1.55 4.32 Sweden 0.83 3.47
Hungary 0.28 3.31 Switzerland 0.90 2.54
Iceland 2.49 5.62 Turkey 1.30 6.90
Israel 3.29 4.79 United Kingdom 0.92 3.31
Italy 0.89 1.29 United States 1.36 2.88
Japan 0.31 1.85

We will begin with an example of a simple model that can be linearized by a cosmetic
transformation. The table reproduces the data in Exercise 1.5 on average annual rates of
growth of employment and GDP for 31 OECD countries.
8
 LINEARITY AND NONLINEARITY

2
e  1  u
g
3
Employment growth rate

0
0 1 2 3 4 5 6 7 8 9
GDP growth rate

A plot of the data reveals that the relationship is clearly nonlinear. We will consider various
nonlinear specifications for the relationship in the course of this chapter, starting with the
hyperbolic model shown.
9
 LINEARITY AND NONLINEARITY

2 1
e  1  u z e  1   2 z  u
g g
3
Employment growth rate

0
0 1 2 3 4 5 6 7 8 9
GDP growth rate

This is nonlinear in g, but if we define z = 1/g, we can rewrite the model so that it is linear in
variables as well as parameters.

10
 LINEARITY AND NONLINEARITY

Average annual percentage growth rates

e g z e g z

Australia 2.57 3.52 0.2841 Korea 1.11 4.48

0.2235
Austria 1.64 2.66 0.3757 Luxembourg 1.34 4.55
0.2199
Belgium 1.06 2.27 0.4401 Mexico 1.88 3.36
0.2976
Canada 1.90 2.57 0.3891 Netherlands 0.51 2.37
0.4221
Czech Republic 0.79 5.62 0.1781 New Zealand 2.67 3.41
0.2929
Denmark 0.58 2.02 0.4961 Norway 1.36 2.49
0.4013
Estonia 2.28 8.10 0.1234 Poland 2.05 5.16
0.1938
Finland 0.98 3.75 0.2664 Portugal 0.13 1.04
0.9603
France 0.69 2.00 0.5004 Slovak Republic 2.08 7.04
0.1420
Here is the data table 0.84
Germany a second time, showing
1.67 0.5980 theSlovenia
values of z computed
1.60from4.82
those of g.
There is no need in practice to perform the calculations oneself. Regression applications
0.2075
always have a facility1.55
Greece for generating
4.32 new variables
0.2315 as functions of existing
Sweden 0.83 ones.
3.47
11
0.2885
 LINEARITY AND NONLINEARITY

. gen z = 1/g
. reg e z
----------------------------------------------------------------------------
Source | SS df MS Number of obs = 31
-----------+------------------------------ F( 1, 29) = 13.68
Model | 5.80515811 1 5.80515811 Prob > F = 0.0009
Residual | 12.3041069 29 .424279548 R-squared = 0.3206
-----------+------------------------------ Adj R-squared = 0.2971
Total | 18.109265 30 .603642167 Root MSE = .65137
----------------------------------------------------------------------------
e | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-----------+----------------------------------------------------------------
z | -2.356137 .6369707 -3.70 0.001 -3.658888 -1.053385
_cons | 2.17537 .249479 8.72 0.000 1.665128 2.685612
----------------------------------------------------------------------------

Here is the output for a regression of e on z.

12
 LINEARITY AND NONLINEARITY

eˆ  2.18  2.36 z
3 ------------------------
e | Coef.
-----------+------------
z | -2.356137
Employment growth rate

2
_cons | 2.17537
------------------------

0
0.0 0.2 0.4 0.6 0.8 1.0

z=1/g

-1

The figure shows the transformed data and the regression line for the regression of e on z.

13
 LINEARITY AND NONLINEARITY

2.36
eˆ  2.18  2.36 z  2.18 
g
3
Employment growth rate

0
0 1 2 3 4 5 6 7 8 9

-1

-2

GDP growth rate

Substituting 1/g for z, we obtain the nonlinear relationship between e and g. The figure
shows this relationship plotted in the original diagram. The linear regression of e on g
reported in Exercise 1.5 is also shown, for comparison.
14
 LINEARITY AND NONLINEARITY

2.36
eˆ  2.18  2.36 z  2.18 
g
3
Employment growth rate

0
0 1 2 3 4 5 6 7 8 9

-1

-2

GDP growth rate

In this case, it was easy to see that the relationship between e and g was nonlinear. In the
case of multiple regression analysis, nonlinearity might be detected using the graphical
technique described in a previous slideshow.
15
Copyright Christopher Dougherty 2016.

These slideshows may be downloaded by anyone, anywhere for personal use.

Subject to respect for copyright and, where appropriate, attribution, they may be
used as a resource for teaching an econometrics course. There is no need to
refer to the author.

The content of this slideshow comes from Section 4.1 of C. Dougherty,

Introduction to Econometrics, fifth edition 2016, Oxford University Press.
Additional (free) resources for both students and instructors may be
downloaded from the OUP Online Resource Centre
www.oxfordtextbooks.co.uk/orc/dougherty5e/.

Individuals studying econometrics on their own who feel that they might benefit
from participation in a formal course should consider the London School of
Economics summer school course
EC212 Introduction to Econometrics
http://www2.lse.ac.uk/study/summerSchools/summerSchool/Home.aspx
or the University of London International Programmes distance learning course
EC2020 Elements of Econometrics
www.londoninternational.ac.uk/lse.

2016.05.01