SPECIFICATION: CHOOSING A FUNCTIONAL

FORM

β 7-1
Alternative Functional Forms
• Distinction between an equation linear in the coefficients
and one that is linear in the variables.
• An equation is linear in the variables if plotting the
function in terms of of Y and X generates a straight line.
• Equation 7.1 is linear in variables:
𝑖 = + + 𝜀𝑖 (7.1)

• Equation 7.2 is not linear in variables:

𝑖 = + + 𝜀𝑖 (7.2)

β 7-2
Alternative Functional Forms (continued)
• An equation is linear in the coefficients only if the
coefficients (the βs):
1. Appear in their simplest form (not raised to a power)
2. Are not multiplied or divided by other coefficients
3. Do not themselves include some sort of function
• Equation 7.1 is linear in the coefficients:
𝑖 = + + 𝜀𝑖 (7.1)
• Equation 7.3 is not linear in the coefficients:
= + 𝛽1 + 𝜀𝑖 (7.1)
𝑖

β 7-3
Alternative Functional Forms (continued)
• Of all possible equations for a single explanatory
variable, only functions of the general form in Equation
7.4 are linear in the coefficients β0 and β1.

𝑓 = + 𝑓 (7.4)

• OLS requires the equation to be linear in the

coefficients.
• There is a wide variety of functional forms that are linear
in the coefficients but not linear in the variables.

β 7-4
Linear Form
• The linear regression model is based on the assumption
that slope of the relationship between the independent
variable and dependent variable is constant:
∆
= 𝑘 = , ,..,𝐾
∆ 𝑘

• If the slope is constant, then the elasticity of Y respect to

X is:
∆ ∆ 𝑘 𝑘
𝑎 𝑖𝑐𝑖 𝑦 , 𝑘 = = ∙ = 𝑘
∆ 𝑘 𝑘 ∆ 𝑘

• The linear model is sometimes referred to as the default

β functional form.
7-5
Linear Form

β 7-6
Linear Form
• Example:

𝑃 𝑡 = + 𝑃 𝐼𝑡 + 𝑡 (a)

Where:
PCE = Personal cons expenditure (US$ billion)
PDI = Personal disposible income (US$ billion).

β 7-7
Linear Form

. reg pce pdi

Source SS df MS Number of obs = 88

F( 1, 86) =14369.09
Model 18548225.7 1 18548225.7 Prob > F = 0.0000
Residual 111012.434 86 1290.84225 R-squared = 0.9941
Adj R-squared = 0.9940
Total 18659238.1 87 214474.002 Root MSE = 35.928

pce Coef. Std. Err. t P>|t| [95% Conf. Interval]

pdi .9672499 .0080691 119.87 0.000 .9512091 .9832907

_cons -171.4412 22.91726 -7.48 0.000 -216.9992 -125.8832

β 7-8
Linear Form
𝑃 𝑡 =− . + . 𝑃 𝐼𝑡
se (22.9172) (0.0080))
t (-7.48) (119.87)

𝑅 = . n = 87

Interpretation:
An increase in PDI of US$ 1 billion, on average,
leads to increase in PCE by US$0.9672 billion.

β 7-9
Double-Log Form
• In double-log functional form the natural log of Y is the
dependent variable and the natural log of X is
independent variable:
ln = + ln + ln + 𝜀 (7.5)

• An individual coefficient can be interpreted as elasticity:

∆ × ∆ ln
𝑎 𝑖𝑐𝑖 𝑦 , 𝑘 = = = 𝑘 (7.6)
∆ 𝑘 𝑘 × ∆ ln 𝑘

• The shape of double-log functional form curve depends

on the sign and magnitude of β1 (see Figure 7.2).
β 7-10
Double-Log Form (continued)

β 7-11
Double-Log Form
• Before we can regress model Eq(a), we need to create new
variables in log form for variable PCE and PDI.

. gen lnpce = log(pce)

. gen lnpdi = log(pdi)

β 7-12
Double-Log Form

. reg lnpce lnpdi

Source SS df MS Number of obs = 88

F( 1, 86) =15482.66
Model 2.9216373 1 2.9216373 Prob > F = 0.0000
Residual .016228525 86 .000188704 R-squared = 0.9945
Adj R-squared = 0.9944
Total 2.93786582 87 .033768573 Root MSE = .01374

lnpce Coef. Std. Err. t P>|t| [95% Conf. Interval]

lnpdi 1.060262 .008521 124.43 0.000 1.043322 1.077201

_cons -.5781124 .0675262 -8.56 0.000 -.7123501 -.4438748

β 7-13
Double-Log Form

ln 𝑃 𝑡 =− . + . ln 𝑃 𝐼𝑡
se (0.0675) (0.0052)
t (-8.56) (124.43)

𝑅 = . n = 87

Interpretation:
An increase in PDI of 1 percent, on average, leads to
increase in PCE by 1.06 percent.

β 7-14
Semilog Form
• Semi-log functional form is a variant of the double-log
equation in which some but not all of the variables are
expressed in natural log.
• Logarithm used on an independent variable (Lin-
Log):
𝑖 = + ln 𝑖 + ln + 𝜀 (7.7)

∆ ∆
𝑎 𝑖𝑐𝑖 𝑦 , 𝑘 = = = 𝑘
∆ 𝑘 𝑘 × ∆ ln 𝑘

β 7-15
Semilog Form
• Give absolute change in Y for percentage change in X.
• If ∆ changes by 0.01 unit (or 1 percent), the absolute change in Y is
(0.01 x 𝑘 )

Example:
𝑃 𝑡 = + ln 𝑃 𝐼𝑡 + 𝑡

PCE = Personal cons expenditure (US$ billion)

lnPDI = log of disposible income (US$ billion).

β 7-16
Semilog Form

. reg pce lnpdi

Source SS df MS Number of obs = 88

F( 1, 86) = 4814.93
Model 18331811.4 1 18331811.4 Prob > F = 0.0000
Residual 327426.786 86 3807.28821 R-squared = 0.9825
Adj R-squared = 0.9822
Total 18659238.1 87 214474.002 Root MSE = 61.703

pce Coef. Std. Err. t P>|t| [95% Conf. Interval]

lnpdi 2655.844 38.27434 69.39 0.000 2579.757 2731.931

_cons -18504.75 303.3123 -61.01 0.000 -19107.71 -17901.78

β 7-17
Semilog Form
𝑃 𝑡 =− . + . ln 𝑃 𝐼𝑡
se (303.3123) (38.2743)
t (-61.01) (69.39)

𝑅 = . n = 87

Interpretation:
An increase in PDI of 1 percent, on average, leads to
increase in PCE about 2655.844 x 0.01 = US$26.58 billion.

β 7-18
Semilog Form

• Logarithm used on the dependent variable (Log-Lin):

ln 𝑖 = + 𝑖 + 𝑖 +𝜀 (7.9)

∆ × ∆ ln
𝑎 𝑖𝑐𝑖 𝑦 , 𝑘 = = = 𝑘
∆ 𝑘 𝑘 ∆ 𝑘

β 7-19
Semilog Form
• If we multiply the relative change in Y by 100, will give the
percentage change, or growth rate , in Y for an obsolute
change in X.

ln 𝑃 𝑡 = + 𝑃 𝐼𝑡 + 𝑡

lnPCE = log of personal cons expenditure (US$ billion)

PDI = Disposible income (US$ billion).

β 7-20
Semilog Form

. reg lnpce pdi

Source SS df MS Number of obs = 88

F( 1, 86) =13524.60
Model 2.9193026 1 2.9193026 Prob > F = 0.0000
Residual .018563219 86 .000215851 R-squared = 0.9937
Adj R-squared = 0.9936
Total 2.93786582 87 .033768573 Root MSE = .01469

lnpce Coef. Std. Err. t P>|t| [95% Conf. Interval]

pdi .0003837 3.30e-06 116.30 0.000 .0003772 .0003903

_cons 6.747635 .0093714 720.03 0.000 6.729005 6.766265

β 7-21
Semilog Form
ln 𝑃 𝑡 = . + . 𝑃 𝐼𝑡
se (0.0093) (3.30x −6 )
t (720.03) (116.3)

𝑅 = . n = 87

Interpretation:
An increase in PDI of US$1 billion, on average, leads
to increase in PCE by 0.00038 x 100 = 0.038 percent.

β 7-22
Semilog Form
• If we regress lnPCE by the time, we can get the growth rate
of PCE.

ln 𝑃 𝑡 = + + 𝑡

lnPCE = log of personal cons expenditure (US$ billion)

t = time (year)

β 7-23
Semilog Form

. reg lnpce t

Source SS df MS Number of obs = 88

F( 1, 86) = 6679.80
Model 2.90052264 1 2.90052264 Prob > F = 0.0000
Residual .037343179 86 .000434223 R-squared = 0.9873
Adj R-squared = 0.9871
Total 2.93786582 87 .033768573 Root MSE = .02084

lnpce Coef. Std. Err. t P>|t| [95% Conf. Interval]

t .0071471 .0000874 81.73 0.000 .0069733 .007321

_cons 7.504107 .0044808 1674.72 0.000 7.495199 7.513015

β 7-24
Semilog Form
ln 𝑃 𝑡 = . + .
se (0.0044) (0.000087)
t (1674.72) (81.73)

𝑅 = . n = 87

Interpretation:
One quarterly year increase, on average, leads to
0.00714 x 100 = 0.714 percent increase in PCE.
The growth rate of PCE is 0.714 percent quarterly.

β 7-25
Semilog Form (continued)

β 7-26
 Polynomial Form
• Polynomial functional forms express Y as a function
of independent variables, some of which are raised to
powers other than one.
• A second-degree polynomial (also called a quadratic):
𝑖 = + 𝑖 + 𝑖+ 𝑖 + 𝜀𝑖 (7.10)

• The slope of Y with respect to X1 in Equation 7.10:

∆
= + (7.11)
∆ 𝑘

• Figure 7.4 depicts the U or inverted U shape of

β quadratic functional forms. 7-27
Polynomial Form (continued)

β 7-28
Polynomial Form (continued)
Example: Model of annual employee earnings as a
function of age of employee and a other measures
• What is expected impact of age on earnings?

• Logical relationship between earnings and age: earnings

would rise, level off, and then fall as age increased.
• This relationship can be modeled by a quadratic
equation:
2
Earningsi = b 0 + b 1 Agei + b 2 Agei +... + e i (7.12)

β 7-29
Choosing a Functional Form
• Best way to choose a functional form is to select
specification that best matches underlying theory.

β 7-30