Financial Econometrics Assignment 2

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Financial Econometrics Assignment 2

Functional Forms of Regression Models


Submitted by:
Name: Shruti Gupta
Class: BBA (FIA) 2B
Roll No: 18373

Functional Form of Regression Model


Functional form of a Regression Model is defined as the algebraic form of relationship
between the dependent variable and the independent (explanatory) variable(s). Choosing
the right functional form for the regression analysis is extremely crucial to maintain the
robustness the regression model.
As per the assumptions of the Classical Linear Regression Model, the regression model
should be linear in parameters i.e. the parameters (βs) should only be raised to the power 1.
But the model may or may not be linear in variables.
The simplest functional form is the linear functional form, where the relationship between
the dependent variable and an independent variable is graphically represented by a straight
line. But the linear functional form is not suitable for all regression models as there might be
a non – linearity in the dependent and independent variables.
The relationship between the variables is non – linear if the rate of increase or decrease
changes due to changes in one variable. A non – linear function such as a quadratic or cubic
function is better suitable to depict the curved pattern in the data.

Linear – in – Variables Model


This is the simplest functional form in which the dependent and independent variables are
linearly related. This model is of the form:
Yi = β0 + β1Xi + µi
In this, β represents a linear relationship i.e. for a unit change in Xi, Yi changes by β units,
keeping all other variables constant. This relationship between the variables is graphically
represented by a straight line.
The slope of the model (dy/dx) = β1 and the elasticity (δY/Y)/ (δX/X) = β1.(X/Y)
a) When β > 0
When β > 0, there is a positive
relationship between the
dependent and independent
variables and this is depicted by
an upward sloping straight line

b) When β < 0
When β < 0, there is a negative
relationship between the
dependent and independent
variables and this is depicted by a
downward sloping straight line

Analysis: To study the research problem of whether capital structure affects a firm’s
performance, we consider the following econometric model:

Firm Performance = β0 + β1*Capital Structure + μ


ROCE = β0 + β1*D/E Ratio + μ

This is an example of the Linear – in – Variables Model, as the dependent variable (ROCE)
and independent variable (Capital Structure) are linearly related.
Interpretation of slope coefficient: As per the OLS regression results, the value of β1 =
-12.98158 which indicates that due to a unit change in D/E, the absolute value of ROCE will fall
by 12.98158.

Log – Linear Model (Log – Log Model)


Log – Log model is obtained by taking a natural log for variables on both sides of the
econometric equation. This log transformation can be used to transform the non – linearity
in parameters to a linear relationship. Under this model, both the dependent and the
independent variable(s) are in the log form.
Consider the following model:
Yi = β0.Xi^β1.e^ui
Taking natural log on both sides
ln Yi = ln β0 + β1.(ln Xi) + µi
ln Yi = α + β1.(ln Xi) + µi
where α = ln β0

This is the structural form of a Log – Log model. By taking log on both sides, the non – linear
relationship between the variables is transformed into a linear one. After the log
transformation, the model is linear in the parameters and also linear in the log of variables
and thus is called a Double – log or log – linear model.

ln Yi = α + β1.(ln Xi) + µi
On differentiating both sides, we observe that
Rate of change in ln Y = β1.(Rate of change in ln X)
δY/Y = β1.(δX/X)
Elasticity = (δY/Y)/ (δX/X) = β1
This implies that the elasticity in a log transformation model is constant throughout the
model and its value is equal to β1. Thus, this model is also termed as a Constant Elasticity
Model.
It can be inferred that the coefficients in a log-log model represent the elasticity of Y
variable with respect to the X variable. In other words, the coefficient is the estimated
percent change in your dependent variable for a percent change in your independent
variable.

Slope = Change in Y / Change in X = β1.(Y/X)


This implies that the value of the slope is not constant throughout the model and is
dependent on the variables. The value of slope changes with change in the absolute values
of X and Y.
To check whether the log – log model is suitable for the data, we plot the values of Ln Y and
Ln X. If the data points fall on a straight line, it implies that the log – log model fits the data.

Interpretation of coefficient β1:


The value of the slope coefficient β1 in a Log – log model is taken to be the percentage
change in the dependent variable due to a percentage change in the independent variable.
For instance, if β = 1.5, then this implies that a 1% change in the independent variable leads
to a 1.5% change in the dependent variable.
If the value of β1 > 1, the elasticity between the dependent and independent variable is
greater that one. This implies that the percentage change in the dependent variable is
proportionately higher than the percentage change in independent variable. Thus, the
impact of the independent variable is positive and becomes larger as its value increases. We
obtain an upward sloping curving which increases at an increasing rate.
If the value of 0 < β1 < 1, the elasticity between the dependent and independent variable is
between 0 and 1. This implies that the percentage change in the dependent variable is
lower than the percentage change in independent variable but is positive. Thus, the impact
of the independent variable is negative and becomes smaller as its value increases. We
obtain an upward sloping curving which increases at a diminishing rate.
If the value of β1 < 0, the elasticity between the dependent and independent variable is less
than 0. This implies that a change in the independent variable leads to a change in the
dependent variable in the opposite direction. Thus, the relationship between the variables is
negative and we obtain a downward sloping curve.
Analysis: To study the log – log model, we study the impact of the independent variable (log
of Stock price of Maruti) on the dependent variable (log of Price of Nifty50 Index). The data
for nifty50 and Maruti has been taken for 1 year. The regression equation has been taken to
be
Ln (Nifty) = β0 + β1*(Ln Maruti) + µi
Log has been taken for both the dependent and independent variables in order to analyse
the relative change in the variables

The value of the coefficient β1 is 0.3249. This implies that a 1% change in the Ln Maruti
causes a 32.49% change in the Ln Nifty. Since β1 is positive, it indicates that there is a
positive relationship between Ln Maruti and Ln Nifty

Log – Lin Model


Log – Lin Model is obtained by taking log of the dependent variable while keeping the
independent variable in its absolute form. This model is suitable when the dependent
variable grows exponentially with respect to the independent variable. Thus, the log – lin
model can be used to depict the rate of growth in the dependent variable due to a change in
the independent variable.
For instance, the value of the population of a country exponentially grows every year. In
order to study the relationship between the population and the time (year), a log – lin
model will be used.
Consider the following model to study the relationship between the population and time
Y(t)=Yₒ∙(1+r) ͭ
where Yₒ= the initial value of Y, Y(t)=Y’s value at time t, r= the rate of growth of Y.
Taking natural log on both sides,
lnY(t)=lnYₒ+t∙ln(1+r)
lnY(t)= β0 +β1∙t + µ
where β0 = lnYₒ, β1 = ln(1+r)
This is the structural form of the Log – Lin model which is also known as the semi – log
model. In this model, the parameters are linear after the transformation.
On differentiating both sides, we observe that
Rate of change in ln Y = β1.(Rate of change in t)
δY/Y = β1.δX
Elasticity = (δY/Y)/ (δX/X) = β1.X
Slope = Change in Y / Change in X = β1.Y
It can be observed that the elasticity and the slope of the model are not constant and vary
throughout the model. The elasticity changes with the change in the value of X and the
slope changes with the change in the value of Y. This implies that elasticity and slope are not
independent of the variables. Regression coefficients do not represent the elasticity or slope
in the log – lin model.
After estimation of the log – lin model, the growth rate (r) can be estimated by taking
antilog of β1 = ln(1+r) which gives r = e^ β1 – 1

Interpretation of coefficient β1:

The coefficient β1 in a log-linear model represent the estimated percent change in the
dependent variable for a unit change in the independent variable.
If β1 > 0, it indicates that the dependent variable increases exponentially which a unit
increase in the independent variable. Thus, there is a positive relationship between the
dependent and the independent variable which is depicted by an upward sloping curve.
If β1 < 0, it indicates that the dependent variable decreases exponentially which a unit
increase in the independent variable. Thus, there is a negative relationship between the
dependent and the independent variable which is depicted by a downward sloping curve.
Analysis: To study the log – lin model, we study the impact of the independent variable (P/E
ratio, P/E ratio and Dividend Yield of Nifty) on the dependent variables (log of Nifty). The
data for all the variables has been taken for 5 years. The regression equation has been taken
to be
Ln (Nifty) = β0 + β1*PE + β2*PB + β3*Div_Yield + µi

The value of β1 is 0.0628 which implies that a unit increase in P/E ratio will cause a 6.28%
increase in the Ln nifty. The value of β2 is -0.0679 which implies that a unit increase in P/B
ratio will cause a 6.79% fall in Ln Nifty. The value of β3 is 0.0156 which implies that a unit
increase in Dividend Yield will cause a 1.56% increase in the Ln nifty.
Lin – Log Model
Lin – Log model is another form of semi – log models, which is obtained by taking the
dependent variable in its linear form while taking the log of the independent variable. The
lin – log model is suitable when the impact of your independent variable on your dependent
variable decreases as the value of your independent variable increases. It is also suitable in
the cases when the dependent variable is extremely sensitive to changes in the independent
variable.
A suitable example of the Lin – Log model is the impact of interest rates on the amount of
investment in the economy. Any change in the rates of interest will have a significant
impact on the amount of investment. As the interest rate increases, the cost of the
investments increases and thus the amount of investment falls.
This can be depicted by the following model
Yi = β0 + β1*(ln Xi) + µi
Where Yi = Amount of Investment, Xi = Rate of Interest
On differentiating both sides, we observe that
Rate of change in Y = β1.(Rate of change in ln Xi)
δY = β1.(δX/X)
Elasticity = (δY/Y)/ (δX/X) = β1.(1/X)
Slope = Change in Y / Change in X = β1.(1/Y)
It can be observed that the elasticity and the slope of the model are not constant and vary
throughout the model. The elasticity is inversely proportional to the independent variable
whereas the slope is inversely proportional to the dependent variable. This implies that
elasticity and slope are not independent of the variables. Regression coefficients do not
represent the elasticity or slope in the lin – log model.

Interpretation of coefficient β1:


The coefficient β1 in a linear-log model represent the estimated unit change in the
dependent variable for a percentage change in the independent variable.
If β1 > 0, it indicates that there is an increase in the absolute value of the dependent
variable due to a percentage increase in the independent variable. Thus, there is a positive
relationship between the dependent and the independent variable which is depicted by an
upward sloping curve.
If β1 < 0, it indicates that there is a decrease in the absolute value of the dependent variable
due to a percentage increase in the independent variable. Thus, there is a negative
relationship between the dependent and the independent variable which is depicted by a
downward sloping curve.
It can be observed that there is an asymptotic relationship between the dependent and
independent variables.

Reciprocal Model / Inverse Model

This is a mathematical form in which the inverse of the independent variable is taken and its
impact is estimated on the dependent variable. The reciprocal model is suitable in case the
dependent variable approaches the intercept term when the independent variable increases
indefinitely. As the independent variable approaches to infinity, its inverse i.e. 1/Xi
approaches zero. And thus, the dependent variable approaches the limiting value of β0.

This is depicted by the following model


Yi = β0 + β1*(1/Xi) + µi

As Xi approaches ∞, 1/Xi approaches 0. Yi approaches to β0

As Xi approaches 0, 1/Xi approaches ∞


• Yi tends to ∞ if β1 > 0
• Yi tends to - ∞ if β1 < 0

When β1 > 0, the value of Y falls and approaches β0, whereas when β1 < 0, the value of Y
increases and approaches β0. Because the slope is a function of 1/X, the slope gets flatter as
X increases.
The reciprocal function is sometimes used to estimate demand curves. The reciprocal model
is also suitable in case of the Engel expenditure curve

Polynomial Model
Polynomial Regression is a form of linear regression in which the relationship between the
independent variable x and dependent variable y is modelled as an nth degree polynomial.
The independent variables which are raised to a power such as square (2) or cube (3) are
termed as polynomial variables.
Under the polynomial model, the dependent variable is regressed on a number of
independent variables which are higher powers of the same variable. It is of the form
Yi = β0 + β1xi + β2x2i + β3x3i + µi
This is the structural form of the third order polynomial model or the cubic model
The polynomial model is suitable in cases where the relationship between dependent and
explanatory variables is curvilinear i.e. it is not uniform.

An example of the quadratic model is the short run marginal cost curve. As we know,
initially the marginal cost falls with increase in the quantity produced as the marginal
product increases (due to increasing returns). Then as marginal product falls due to the law
of diminishing returns, marginal cost increases with increase in quantity produced, which
leads to a U - shaped Marginal cost curve.
The equation of the marginal cost curve can be formulated with a polynomial model of
second order:
MC = β0 + β1.Q + β2.Q2 + µi

Linear Quadratic Cubic

Interpretation of slope coefficient: It is not possible to interpret the values of β1, β2 as


there is high degree of multicollinearity because of the inclusion of the polynomial variables.
β1 is the impact X1 has on Y holding all other factors constant. If X1 is related to X1 then β1
will also capture the impact of changes in X2. In other words, interpretation of the
parameters becomes difficult.

Choosing the correct functional form


Choosing the correct functional form is extremely important since it has a significant
implication on the interpretation of the parameters. The following points can be taken into
consideration while choosing the correct functional form for the regression analysis:
1. Plotting the observations:
One important way of understanding the correct functional form for the model is to
plot the observations in the data and observe the trend followed by the data points.
For example, if the data points form a parabolic curve then a second order
polynomial model (quadratic model) is suitable, if the data points follow an
exponential curve, then a log – lin model may be suitable. If the values seem to fall
on a straight line, then a linear model may be suitable

2. Underlying Theory:
The functional form of the regression model can also be chosen on the basis of the
underlying theory. For example, in the study of price elasticity of demand, elasticity
measures the percentage change in quantity demanded due to percentage change in
price. This type of relationship can be best suited for a Log – Log model. Similarly, the
production function first increases at an increasing rate and then decrease at a
diminishing rate and this relationship can be suited with a third - degree polynomial
form (cubic model).
3. Slope and Elasticity:
The rate of change in the dependent variable due to the independent variable
(slope) and the elasticity between the two variables can also be used to choose the
right functional form for the model. For instance, of the slope of the regression
model is constant, then the linear functional form is suitable whereas if the value of
elasticity is constant throughout the model, then the Log – Log model is suitable for
the regression analysis

4. Adjusted R2:
Different functional forms can be fitted to the data to select the functional form
which best suites the data. The best functional form can be selected on the basis of
the value of adjusted R2. The functional form which gives a higher value of adjusted
R2 is better suited for the model.
But it should be taken into consideration that while comparing the values of
adjusted R2 of the different functional forms, the dependent variable of the
functional forms should be the same. This implies that the log – log model and the lin
– log model cannot be compared on the basis of value of adjusted R2 because the
dependent variables in the two functional forms differ.

You might also like