CHAPTER 5

DYNAMIC ECONOMETRIC MODELS

T. Sunde

1
Chapter Contents
Dynamic econometric models
Reasons for lags
Types of dynamic models
Calculation of impact multipliers
ARDL without cointegration
ARDL with cointegrated variables
ARDL with no cointegration and Granger causality
ARDL with cointegration and Granger causality

2
Representation

Autoregressive model

In economics very often Y responds to X with a lapse of time. Such a

lapse of time is called a lag.

Example:
3
 DYNAMIC ECONOMETRIC MODELS
Distributed lag models

If central bank is worried about the economy slowing down, it might want to
cut interest rates. The impact of such a policy would no doubt take more than a
year to feed through the economy and affect other important variables such as
output and unemployment.
Fiscal and monetary policies used by government have impacts that are felt
only in some future period.
For e.g., the decision by a firm to carry out a new investment (e.g., the purchase
of new computers will not immediately affect production. It takes time to
purchase the computers, install them, and train workers to use them. Thus, such
an investment will only influence production in the future.
The value of the dependent variable at a given point in time should depend
values of the explanatory variables in the past 4
 DYNAMIC ECONOMETRIC MODELS…
The simplest model to incorporate such dynamic effects is known as the distributed lag
model. It is a regression model of the form:

Where , are the parameters and p is the lag order or the lag length.

 measures the effect of the explanatory variables two periods ago on the dependent variable
ceteris paribus.

Example. The effects of safety training on industrial accidents. Losses due to industrial
accidents can be quite substantial in large companies.

Accordingly, many companies provide safety training to their workers in an effort to try and
5
reduce industrial accidents.
DYNAMIC ECONOMETRIC MODELS…
Data could be could collected monthly for some years on the
following variables:

losses due to accidents (in $s per month)

 hours of safety training provided to each worker per month.

What can the company conclude about the effectiveness of its safety
training programmes
6
 Reasons for lags

Distributed lag models

short run or impact multiplier

Long run or total distributed lag multiplier is as follows:

this is the proportion of the long run felt by a certain period.

7
 Estimation of the distributed lag models
Consider the model

Since the variables , …. Are non stochastic, the model can be estimated by
using Ordinary Least Squares (OLS).

First regress on , then on , , then on , , and so on until coefficients start to

become statistically insignificant.

This how you determine the number of lags of X to include in the distributed
lag model. 8
 Regression with time series variables
(ARDL)
• Effect of an explanatory variable or variables on a dependent variable.

• May be complicated by the fact that sometimes the explanatory variable may
influence the dependent variable with a time lag. This often necessitates the
inclusion of time lags of the explanatory variable in the regression.

• Sometimes the dependent variable may be correlated with lags of itself

suggesting that lags of the dependent variable should also be included in the
regression equation.

• These considerations motivate the commonly used Autoregressive Distributed

Lag (ARDL) model 9
 TYPES OF DYNAMIC MODELS
Autoregressive Model: When only lagged dependent variables are used as explanatory variables.

Distributed Lag Model: If explanatory variables are purely of the exogenous type.

Autoregressive Distributed Lag (ARDL) Model: When both lagged dependent variables and
exogenous variables are included.

10
Interpretation of Dynamic Model: Short Run and Long Run Impact Multipliers [I(0) Variables]

Consider the Dynamic Model:

ARDL(1):
The parameters, , , and are short run impact multipliers. The coefficient measures the
instantaneous impact, while indicates the impact one period later.
To get the Long Run Impact, all the time subscripts can be removed. The relationship in the
steady state is:

[( ]
The long run multiplier of
The long run multiplier of (
11
 EXAMPLE

• Consider the Dynamic Model:

• ARDL(1):
• The instantaneous impact multiplier of X is 0.2
• The instantaneous impact multiplier of Z is 0.3
• The cumulative short run multiplier of X after 1 period is 0.3
• The cumulative short run multiplier of Z after 1 period is 0.4

• For long run impact, all the time subscripts can be removed. The relationship in the
steady state is:

Collecting like terms to the LHS, and then solving the equation gives:

• The long run multiplier 12

ARDL (p, q) Multiplier Calculation for I(1) Variables
•

• We can interpret the multiplier of the above model.

• It can be shown that the LR multiplier for ARDL(p, q) model is: .

• In other words the coefficients of and matter in the long run behaviour.

• If the multiplier is 1.042, then the long run multiplier effect of computer
purchases on sales is 1.042%. This means that if X permanently increases by 1%
the equilibrium value of Y will increase by 1.042%
13
 Estimation of impact multipliers

• The estimations in slides 11 and 12 are possible if the variables are integrated
of order zero (stationary in levels).

• Its possible to have integrated of order zero variables if they are transformed
to growth rates.

• If using variables that are converted to natural logarithms the multipliers

become elasticities.

14
 Stationary ARDL(p, q) Model
•

• Y is the dependent variable

• Y depends on p lags of itself, the current value of the explanatory variable (X), as
well as q lags of X.
• The model also allows for a deterministic trend (T)
• Since the model contains p lags of Y and q lags of X we denote it by .
• In this chapter we focus on the case where there is one explanatory variable, X. It
is possible to use more than one explanatory variable in the model.
• Estimation and interpretation of the model depends on whether the series X and Y
are stationary or not.
• NB we assume that X and Y have the same stationarity properties, that is, they
either must both be stationary or both have a unit root 15
 Time series regression when X and Y have unit roots: Spurious
Regression
• Assume we have

• If X and Y have unit roots then all OLS regression results might be
misleading and incorrect. This is called SPURIOUS regression.

• R2 will be very high

• t-values will be very high signifying significance
• F-test will be very high
• DW will be small

16
• If R > DW, the regression model is Spurious [Rule of Thumb]
2
 Time series regression when X and Y have unit
roots but are not cointegrated
• Using first differenced data

17
The Engle-Granger Two Step Procedure: Time series regression
when X and Y have unit roots and are cointegrated
• STEP1: Run a regression of Y on X and save the residuals
• STEP 2: Run a regression of ∆Y, ∆X and STEP 1 residuals lagged once.
• Before carrying out the two-step procedure for ECM, you must verify that Y and X have unit
roots and are cointegrated (by testing for the stationarity of the residuals).
• In practice just as the ARDL(p, q) model has lags of the dependent and explanatory variables
the ECM may also have lags. It may also have a deterministic trend. Incorporating these
features into the ECM yields.
•

18
 The Engle-Granger Two Step Procedure: Time series regression
when X and Y have unit roots and are cointegrated..

• The above model can still be estimated using the two-step procedure.

• The adjustment towards equilibrium intuition still holds for this model.

• The precise value of p and q can be determined using the t and p-values in the
same manner as the ARDL

• The ECM is closely related to the ARDL model in that it is the restricted
version of the ARDL.

19
Granger Causality in a Simple ARDL Model

• The coefficient measures the influence of on .

• If then past values of X have no effect on Y and there is no way X could Granger Cause
Y. In other words, if then X does not Granger cause Y.

HYPOTHESIS
X does not Granger cause Y (
X Granger causes Y

DECISION RULE
20
Granger Causality in an ARDL(p, q) When Variables are not Cointegrated
• ARDL (p, q)

HYPOTHESIS
(X does not Granger cause Y)
(X Granger causes Y)

DECISION RULE
Reject the null hypothesis ) if the p-values are less than 5% (X Granger causes Y).

21
Joint Granger Causality with Cointegrated Variables

HYPOTHESIS
(X does not Granger cause Y)
(X Granger causes Y)

DECISION RULE
Reject the null hypothesis ) if the p-values are less than 5% (X Granger causes
Y)

22
The End

23