CH 09

Chapter 9
Regression with Time Series Data:

Stationary Variables
Walter R. Paczkowski
Rutgers University
Chapter 9: Regression with Time Series Data:
Principles of Econometrics, 4th Edition Page 1
Chapter Contents
 9.1 Introduction
 9.2 Finite Distributed Lags
 9.3 Serial Correlation
 9.4 Other Tests for Serially Correlated Errors
 9.5 Estimation with Serially Correlated Errors
 9.6 Autoregressive Distributed Lag Models
 9.7 Forecasting
 9.8 Multiplier Analysis

9.1
Introduction

9.1
Introduction
When modeling relationships between variables,

the nature of the data that have been collected has
an important bearing on the appropriate choice of
an econometric model
– Two features of time-series data to consider:
1. Time-series observations on a given
economic unit, observed over a number of
time periods, are likely to be correlated
2. Time-series data have a natural ordering
according to time

9.1
Introduction
There is also the possible existence of dynamic

relationships between variables
– A dynamic relationship is one in which the
change in a variable now has an impact on that
same variable, or other variables, in one or
more future time periods
– These effects do not occur instantaneously but
are spread, or distributed, over future time
periods

9.1
Introduction FIGURE 9.1 The distributed lag effect

9.1
Introduction
9.1.1
Dynamic Nature of
Relationships
Ways to model the dynamic relationship:
1. Specify that a dependent variable y is a
function of current and past values of an
explanatory variable x
Eq. 9.1 yt  f ( xt , xt 1 , xt 2 ,...)
• Because of the existence of these lagged

effects, Eq. 9.1 is called a distributed lag
model

9.1
Introduction
9.1.1
Dynamic Nature of
Relationships
Ways to model the dynamic relationship (Continued):
2. Capturing the dynamic characteristics of time-series
by specifying a model with a lagged dependent
variable as one of the explanatory variables
Eq. 9.2 yt  f ( yt 1 , xt )
• Or have:
Eq. 9.3 yt  f ( yt 1 , xt , xt 1 , xt  2 )
– Such models are called autoregressive
distributed lag (ARDL) models, with
‘‘autoregressive’’ meaning a regression of yt on
its own lag or lags
9.1
Introduction
9.1.1
Dynamic Nature of
Relationships
Ways to model the dynamic relationship (Continued):
3. Model the continuing impact of change over
several periods via the error term
Eq. 9.4 yt  f ( xt )  et et  f (et 1 )
• In this case et is correlated with et - 1

• We say the errors are serially correlated or
autocorrelated

9.1
Introduction
9.1.2
Least Squares
Assumptions
The primary assumption is Assumption MR4:
cov  yi , y j   cov  ei , e j   0 for i  j

• For time series, this is written as:
cov  yt , ys   cov  et , es   0 for t  s
– The dynamic models in Eqs. 9.2, 9.3 and 9.4

imply correlation between yt and yt - 1 or et and et
- 1 or both, so they clearly violate assumption
MR4
9.1
Introduction
9.1.2a
Stationarity
A stationary variable is one that is not explosive,

nor trending, and nor wandering aimlessly without
returning to its mean

9.1
Introduction FIGURE 9.2 (a) Time series of a stationary variable
9.1.2a
Stationarity

9.1 FIGURE 9.2 (b) time series of a nonstationary variable that is ‘‘slow-turning’’
Introduction
or ‘‘wandering’’
9.1.2a
Stationarity

9.1
Introduction FIGURE 9.2 (c) time series of a nonstationary variable that ‘‘trends”
9.1.2a
Stationarity

9.1 FIGURE 9.3 (a) Alternative paths through the chapter starting with finite
Introduction
distributed lags
9.1.3
Alternative Paths
Through the
Chapter

9.1 FIGURE 9.3 (b) Alternative paths through the chapter starting with
Introduction
serial correlation
9.1.3
Alternative Paths
Through the
Chapter

9.2
Finite Distributed Lags

9.2
Finite Distributed
Lags
Consider a linear model in which, after q time

periods, changes in x no longer have an impact on
y
Eq. 9.5 yt    0 xt  1 xt 1   2 xt  2     q xt  q  et
– Note the notation change: βs is used to denote

the coefficient of xt-s and α is introduced to
denote the intercept

9.2
Finite Distributed
Lags
Model 9.5 has two uses:

– Forecasting
Eq. 9.6 yT 1    0 xT 1  1 xT   2 xT 1     q xT  q 1  eT 1
– Policy analysis
• What is the effect of a change in x on y?
E ( yt ) E ( yt  s )
Eq. 9.7   s
xt  s xt

9.2
Finite Distributed
Lags
Assume xt is increased by one unit and then

maintained at its new level in subsequent periods
– The immediate impact will be β0
– the total effect in period t + 1 will be β0 + β1, in
period t + 2 it will be β0 + β1 + β2, and so on
• These quantities are called interim
multipliers
– The total multiplier is the final effect on y of
the sustained increase
q
after q or more periods
have elapsed  β s
s 0

9.2
Finite Distributed
Lags
The effect of a one-unit change in xt is distributed

over the current and next q periods, from which we
get the term ‘‘distributed lag model’’
– It is called a finite distributed lag model of
order q
• It is assumed that after a finite number of
periods q, changes in x no longer have an
impact on y
– The coefficient βs is called a distributed-lag
weight or an s-period delay multiplier
– The coefficient β0 (s = 0) is called the impact
multiplier
9.2
Finite Distributed
Lags
ASSUMPTIONS OF THE DISTRIBUTED LAG MODEL
9.2.1
Assumptions
TSMR1. yt    β 0 xt  β1 xt 1  β 2 xt  2    β q xt  q  et , t  q  1, , T
TSMR2. y and x are stationary random variables, and et is independent of
current, past and future values of x.
TSMR3. E(et) = 0
TSMR4. var(et) = σ2
TSMR5. cov(et, es) = 0 t ≠ s
TSMR6. et ~ N(0, σ2)

9.2
Finite Distributed
Lags
9.2.2
An Example:
Okun’s Law
Consider Okun’s Law
– In this model the change in the unemployment rate
from one period to the next depends on the rate of
growth of output in the economy:
Eq. 9.8 U t  U t 1    Gt  GN 
– We can rewrite this as:

Eq. 9.9 DU t    β 0Gt  et
where DU = ΔU = Ut - Ut-1, β0 = -γ, and α=

γGN
9.2
Finite Distributed
Lags
9.2.2
An Example:
Okun’s Law
We can expand this to include lags:
Eq. 9.10 DU t    β 0Gt  β1Gt 1  β 2Gt 2    β q Gt q  et
We can calculate the growth in output, G, as:
GDPt  GDPt 1
Eq. 9.11 Gt  100
GDPt 1

9.2
Finite Distributed FIGURE 9.4 (a) Time series for the change in the U.S. unemployment rate:
Lags 1985Q3 to 2009Q3
9.2.2
An Example:
Okun’s Law

9.2
Finite Distributed
Lags
FIGURE 9.4 (b) Time series for U.S. GDP growth: 1985Q2 to 2009Q3
9.2.2
An Example:
Okun’s Law

9.2
Finite Distributed
Lags
Table 9.1 Spreadsheet of Observations for Distributed Lag Model
9.2.2
An Example:
Okun’s Law

9.2
Finite Distributed
Lags
Table 9.2 Estimates for Okun’s Law Finite Distributed Lag Model
9.2.2
An Example:
Okun’s Law

9.3
Serial Correlation

9.3
Serial Correlation
When is assumption TSMR5, cov(et, es) = 0 for

t ≠ s likely to be violated, and how do we assess
its validity?
– When a variable exhibits correlation over time,
we say it is autocorrelated or serially
correlated
• These terms are used interchangeably

9.3
Serial Correlation FIGURE 9.5 Scatter diagram for Gt and Gt-1
9.3.1
Serial Correlation
in Output Growth

9.3
Serial Correlation
9.3.1a
Computing
Autocorrelation
Recall that the population correlation between two

variables x and y is given by:
cov  x, y 
ρ xy 
var  x  var  y 

9.3
Serial Correlation
9.3.1a
Computing For the Okun’s Law problem, we have:
Autocorrelation
cov  Gt , Gt 1  cov  Gt , Gt 1 
ρ1  
var  Gt 
Eq. 9.12
var  Gt  var  Gt 1 
The notation ρ1 is used to denote the population

correlation between observations that are one period apart
in time
– This is known also as the population autocorrelation
of order one.
– The second equality in Eq. 9.12 holds because
var(Gt) = var(Gt-1) , a property of time series that are
stationary
9.3
Serial Correlation
9.3.1a
Computing
Autocorrelation
The first-order sample autocorrelation for G is

obtained from Eq. 9.12 using the estimates:
1 T
 t t 1 
 G ,G
cov  
T  1 t 2
 Gt  G   Gt 1  G 
T
1
var  Gt     Gt  G 
 2
T  1 t 1

9.3
Serial Correlation
9.3.1a
Computing
Autocorrelation
Making the substitutions, we get:
 G  G  Gt t 1  G
Eq. 9.13 r1  t 2
T
 G  G
2
t
t 1

9.3
Serial Correlation
9.3.1a
Computing
Autocorrelation
More generally, the k-th order sample

autocorrelation for a series y that gives the
correlation between observations that are k periods
apart is: T
Eq. 9.14
  yt  y   yt k  y 
rk  t  k 1 T
  yt  y 
2
t 1

9.3
Serial Correlation
9.3.1a
Computing
Autocorrelation
Because (T - k) observations are used to compute

the numerator and T observations are used to
compute the denominator, an alternative that leads
to larger estimates in finite samples is:
T
1

T  k t  k 1
 yt  y   yt k  y 
Eq. 9.15 rk 
1 T
  yt  y 
2
T t 1

9.3
Serial Correlation
9.3.1a
Computing
Autocorrelation
Applying this to our problem, we get for the first

four autocorrelations:
Eq. 9.16 r1  0.494 r2  0.411 r3  0.154 r4  0.200

9.3
Serial Correlation
9.3.1a
Computing
Autocorrelation
How do we test whether an autocorrelation is

significantly different from zero?
– The null hypothesis is H0: ρk = 0
– A suitable test statistic is:
rk  0
Eq. 9.17 Z  T rk  N  0,1
1T

9.3
Serial Correlation
9.3.1a
Computing
Autocorrelation
For our problem, we have:
Z1  98  0.494  4.89, Z 2  98  0.414  4.10

Z 3  98  0.154  1.52, Z 4  98  0.200  1.98
– We reject the hypotheses H0: ρ1 = 0 and H0: ρ 2
=0
– We have insufficient evidence to reject H0: ρ 3
=0
– ρ4 is on the borderline of being significant.
– We conclude that G, the quarterly growth rate in
U.S. GDP, exhibits significant serial correlation at
lags one and two
9.3
Serial Correlation
9.3.1b
The Correlagram
The correlogram, also called the sample

autocorrelation function, is the sequence of
autocorrelations r1, r2, r3, …
– It shows the correlation between observations
that are one period apart, two periods apart,
three periods apart, and so on

9.3
Serial Correlation FIGURE 9.6 Correlogram for G
9.3.1b
The Correlagram

9.3
Serial Correlation
9.3.2
Serially Correlated
Errors
The correlogram can also be used to check

whether the multiple regression assumption
cov(et, es) = 0 for t ≠ s is violated

9.3
Serial Correlation
9.3.2a
A Phillips Curve
Consider a model for a Phillips Curve:
Eq. 9.18 INFt  INFt E  γ  U t  U t 1 
– If we initially assume that inflationary

expectations are constant over time (β1 = INFEt)
set β2= -γ, and add an error term:
Eq. 9.19 INFt  β1  β 2 DU t  et

9.3
Serial Correlation FIGURE 9.7 (a) Time series for Australian price inflation
9.3.2a
A Phillips Curve

9.3 FIGURE 9.7 (b) Time series for the quarterly change in the Australian
Serial Correlation
unemployment rate
9.3.2a
A Phillips Curve

9.3
Serial Correlation
9.3.2a
A Phillips Curve
To determine if the errors are serially correlated,

we compute the least squares residuals:
Eq. 9.20 eˆt  INFt  b1  b2 DU t 

9.3 FIGURE 9.8 Correlogram for residuals from least-squares estimated
Serial Correlation
Phillips curve
9.3.2a
A Phillips Curve

9.3
Serial Correlation
9.3.2a
A Phillips Curve The k-th order autocorrelation for the residuals can
be written as: T
 eˆ eˆ t t k
Eq. 9.21 rk  t  k 1
T
t
ˆ
e 2
t 1
– The least squares equation is:

INF  0.7776  0.5279 DU
Eq. 9.22
 se   0.0658  0.2294 

9.3
Serial Correlation
9.3.2a
A Phillips Curve
The values at the first five lags are:
r1  0.549 r2  0.456 r3  0.433 r4  0.420 r5  0.339

9.4
Other Tests for Serially Correlated
Errors

9.4
Other Tests for
Serially Correlated
Errors
9.4.1
A Lagrange
Multiplier Test
An advantage of this test is that it readily

generalizes to a joint test of correlations at more
than one lag

9.4
Other Tests for
Serially Correlated
Errors
9.4.1
A Lagrange
Multiplier Test
If et and et-1 are correlated, then one way to model

the relationship between them is to write:
Eq. 9.23 et  ρet 1  vt
– We can substitute this into a simple regression
equation:
Eq. 9.24 yt  β1  β 2 xt  ρet 1  vt

9.4
Other Tests for
Serially Correlated
Errors
9.4.1
A Lagrange
Multiplier Test
We have one complication: ê0 is unknown

– Two ways to handle this are:
1. Delete the first observation and use a total
of T observations
2. Set eˆ0  0 and use all T observations

9.4
Other Tests for
Serially Correlated
Errors
9.4.1
A Lagrange
Multiplier Test
For the Phillips Curve:
 i t  6.219 F  38.67 p-value  0.000

 ii  t  6.202 F  38.47 p-value  0.000
– The results are almost identical
– The null hypothesis H0: ρ = 0 is rejected at all
conventional significance levels
– We conclude that the errors are serially
correlated

9.4
Other Tests for
Serially Correlated
Errors
9.4.1
A Lagrange
Multiplier Test
To derive the relevant auxiliary regression for the
autocorrelation LM test, we write the test equation
as:
Eq. 9.25 yt  β1  β 2 xt  ρeˆt 1  vt
– But since we know that yt  b1  b2 xt  eˆt , we

get:
b1  b2 xt  eˆt  β1  β 2 xt  ρeˆt 1  vt

9.4
Other Tests for
Serially Correlated
Errors
9.4.1
A Lagrange
Multiplier Test
Rearranging, we get:
eˆt   β1  b1    β 2  b2  xt  ρeˆt 1  vt
Eq. 9.26
 γ1  γ 2 xt  ρeˆt 1  v
– If H0: ρ = 0 is true, then LM = T x R2 has an
approximate χ2(1) distribution
• T and R2 are the sample size and goodness-
of-fit statistic, respectively, from least
squares estimation of Eq. 9.26

9.4
Other Tests for
Serially Correlated
Errors
9.4.1
A Lagrange
Multiplier Test
Considering the two alternative ways to handle ê0 :
 
iii LM   T  1  R 2
 89  0.3102  27.61
 iv  LM  T  R 2  90  0.3066  27.59
– These values are much larger than 3.84, which
is the 5% critical value from a χ2(1)-distribution
• We reject the null hypothesis of no
autocorrelation
– Alternatively, we can reject H0 by examining
the p-value for LM = 27.61, which is 0.000

9.4
Other Tests for
Serially Correlated
Errors
9.4.1a
Testing Correlation
at Longer Lags
For a four-period lag, we obtain:
 iii  LM   T  4   R 2  86  0.3882  33.4

 iv  LM  T  R 2  90  0.4075  36.7
– Because the 5% critical value from a χ2(4)-
distribution is 9.49, these LM values lead us to
conclude that the errors are serially correlated

9.4
Other Tests for
Serially Correlated
Errors
9.4.2
The Durbin-
Watson Test
This is used less frequently today because its

critical values are not available in all software
packages, and one has to examine upper and lower
critical bounds instead
– Also, unlike the LM and correlogram tests, its
distribution no longer holds when the equation
contains a lagged dependent variable

9.5
Estimation with Serially Correlated
Errors

9.5
Estimation with
Serially Correlated
Errors
Three estimation procedures are considered:

1. Least squares estimation
2. An estimation procedure that is relevant when
the errors are assumed to follow what is
known as a first-order autoregressive model
et  ρet 1  vt
3. A general estimation strategy for estimating
models with serially correlated errors

9.5
Estimation with
Serially Correlated
Errors
We will encounter models with a lagged

dependent variable, such as:
yt  δ  θ1 yt 1  δ0 xt  δ1 xt 1  vt

9.5
Estimation with
Serially Correlated ASSUMPTION FOR MODELS WITH A LAGGED DEPENDENT VARIABLE
Errors
TSMR2A In the multiple regression model yt  β1  β 2 xt 2    β K xK  vt

Where some of the xtk may be lagged values of y, vt is uncorrelated with all
xtk and their past values.

9.5
Estimation with
Serially Correlated
Errors
9.5.1
Least Squares
Estimation
Suppose we proceed with least squares estimation
without recognizing the existence of serially
correlated errors. What are the consequences?
1. The least squares estimator is still a linear
unbiased estimator, but it is no longer best
2. The formulas for the standard errors usually
computed for the least squares estimator are
no longer correct
• Confidence intervals and hypothesis tests
that use these standard errors may be
misleading
9.5
Estimation with
Serially Correlated
Errors
9.5.1
Least Squares
Estimation
It is possible to compute correct standard errors

for the least squares estimator:
– HAC (heteroskedasticity and autocorrelation
consistent) standard errors, or Newey-West
standard errors
• These are analogous to the heteroskedasticity
consistent standard errors

9.5
Estimation with
Serially Correlated
Errors
9.5.1
Least Squares
Estimation
Consider the model yt = β1 + β2xt + et
– The variance of b2 is:
var  b2    wt2 var  et     wt ws cov  et , es 

t ts
Eq. 9.27    wt ws cov  et , es  

  wt2 var  et  1  ts 
t

  t
wt var  et 
2 

where
wt   xt  x    x  x
2
t t

9.5
Estimation with
Serially Correlated
Errors
9.5.1
Least Squares
Estimation
When the errors are not correlated, cov(et, es) = 0,
and the term in square brackets is equal to one.
– The resulting expression
var  b2    t wt2 var  et 
is the one used to find heteroskedasticity-

consistent (HC) standard errors
– When the errors are correlated, the term in
square brackets is estimated to obtain HAC
standard errors
9.5
Estimation with
Serially Correlated
Errors
9.5.1
Least Squares
Estimation
If we call the quantity in square brackets g and its

estimate ĝ , then the relationship between the two
estimated variances is:
Eq. 9.28 
varHAC  b2   
varHC  b2   g
ˆ

9.5
Estimation with
Serially Correlated
Errors
9.5.1
Least Squares
Estimation
Let’s reconsider the Phillips Curve model:

INF  0.7776  0.5279 DU
Eq. 9.29  0.0658   0.2294   incorrect se 
 0.1030   0.3127   HAC se 

9.5
Estimation with
Serially Correlated
Errors
9.5.1
Least Squares
Estimation
The t and p-values for testing H0: β2 = 0 are:

t  0.5279 0.2294  2.301 p  0.0238  from LS standard errors 
t  0.5279 0.3127  1.688 p  0.0950  from HAC standard errors 

9.5
Estimation with
Serially Correlated
Errors
9.5.2
Estimating an
AR(1) Error Model
Return to the Lagrange multiplier test for serially

correlated errors where we used the equation:
Eq. 9.30 et  ρet 1  vt
– Assume the vt are uncorrelated random errors

with zero mean and constant variances:
Eq. 9.31 E  vt   0 var  vt    v2 cov  vt , vs   0 for t  s

9.5
Estimation with
Serially Correlated
Errors
9.5.2
Estimating an
AR(1) Error Model
Eq. 9.30 describes a first-order autoregressive
model or a first-order autoregressive process for
et
– The term AR(1) model is used as an
abbreviation for first-order autoregressive
model
– It is called an autoregressive model because it
can be viewed as a regression model
– It is called first-order because the right-hand-
side variable is et lagged one period

9.5
Estimation with
Serially Correlated
Errors
9.5.2a
Properties of an
AR(1) Error
We assume that:
Eq. 9.32 1  ρ  1
The mean and variance of et are:
 2
Eq. 9.33 E  et   0 var  et     2
e
v
1  ρ2
The covariance term is:
ρ k 2
Eq. 9.34 cov  et , et  k   , k 0 v
1 ρ 2

9.5
Estimation with
Serially Correlated
Errors
9.5.2a
Properties of an The correlation implied by the covariance is:
ρ k  corr  et , et  k 
AR(1) Error
cov  et , et  k 

var  et  var  et  k 
cov  et , et  k 
Eq. 9.35 
var  et 

ρ k v2 1  ρ 2 
 v2  1  ρ 2 
 ρk
9.5
Estimation with
Serially Correlated
Errors
9.5.2a
Properties of an Setting k = 1:
AR(1) Error
Eq. 9.36 ρ1  corr  et , et 1   ρ
– ρ represents the correlation between two errors that are

one period apart
• It is the first-order autocorrelation for e,
sometimes simply called the autocorrelation
coefficient
• It is the population autocorrelation at lag one for a
time series that can be described by an AR(1) model
• r1 is an estimate for ρ when we assume a series is
AR(1)
9.5
Estimation with
Serially Correlated
Errors
9.5.2a
Properties of an
AR(1) Error
Each et depends on all past values of the errors vt:
Eq. 9.37 et  vt  ρvt 1  ρ vt  2  ρ vt 3 

2 3
– For the Phillips Curve, we find for the first five

lags:
r1  0.549 r2  0.456 r3  0.433 r4  0.420 r5  0.339
– For an AR(1) model, we have:

ρˆ 1  ρˆ  r1  0.549

9.5
Estimation with
Serially Correlated
Errors
9.5.2a
Properties of an
AR(1) Error
For longer lags, we have:
ρˆ 2  ρˆ   0.549   0.301
2 2
ρˆ 3  ρˆ 3   0.549   0.165
3
ρˆ 4  ρˆ   0.549   0.091
4 4
ρˆ 5  ρˆ 5   0.549   0.050
5

9.5
Estimation with
Serially Correlated
Errors
9.5.2b
Nonlinear Least
Squares Estimation
Our model with an AR(1) error is:
Eq. 9.38 yt  β1  β 2 xt  et with et  ρet 1  vt
with -1 < ρ < 1

– For the vt, we have:
Eq. 9.39 E  vt   0 var  vt    v2 cov  vt , vt 1   0 for t  s

9.5
Estimation with
Serially Correlated
Errors
9.5.2b
Nonlinear Least
Squares Estimation
With the appropriate substitutions, we get:
Eq. 9.40 yt  β1  β 2 xt  ρet 1  vt
– For the previous period, the error is:
Eq. 9.41 et 1  yt 1  β1  β 2 xt 1
– Multiplying by ρ:
Eq. 9.42 ρet 1  et yt 1  ρβ1  ρβ 2 xt 1

9.5
Estimation with
Serially Correlated
Errors
9.5.2b
Nonlinear Least
Squares Estimation
Substituting, we get:
Eq. 9.43 yt  β1  1  ρ   β 2 xt  ρyt 1  ρβ 2 xt 1  vt

9.5
Estimation with
Serially Correlated
Errors
9.5.2b
Nonlinear Least
Squares Estimation
The coefficient of xt-1 equals -ρβ2

– Although Eq. 9.43 is a linear function of the
variables xt , yt-1 and xt-1, it is not a linear
function of the parameters (β1, β2, ρ)
– The usual linear least squares formulas cannot
be obtained by using calculus to find the values
of (β1, β2, ρ) that minimize Sv
• These are nonlinear least squares estimates

9.5
Estimation with
Serially Correlated
Errors
9.5.2b
Nonlinear Least
Squares Estimation
Our Phillips Curve model assuming AR(1) errors
is:
Eq. 9.44 INFt  β1  1  ρ   β 2 DU t  ρINFt 1  ρβ 2 DU t 1  vt
– Applying nonlinear least squares and presenting

the estimates in terms of the original
untransformed model, we have:

INF  0.7609  0.6944 DU et  0.557et 1  vt
Eq. 9.45
 se   0.1245  0.2479   0.090 
9.5
Estimation with
Serially Correlated
Errors
9.5.2c
Generalized Least
Squares Estimation
Nonlinear least squares estimation of Eq. 9.43 is

equivalent to using an iterative generalized least
squares estimator called the Cochrane-Orcutt
procedure

9.5
Estimation with
Serially Correlated
Errors
9.5.3
Estimating a More
General Model
We have the model:
Eq. 9.46 yt  β1  1  ρ   β 2 xt  ρyt 1  ρβ 2 xt 1  vt
– Suppose now that we consider the model:
Eq. 9.47 yt  δ  θ1 yt 1  δ0 xt  δ1 xt 1  vt
• This new notation will be convenient when

we discuss a general class of autoregressive
distributed lag (ARDL) models
– Eq. 9.47 is a member of this class
9.5
Estimation with
Serially Correlated
Errors
9.5.3
Estimating a More
General Model
Note that Eq. 9.47 is the same as Eq. 9.47 since:
Eq. 9.48 δ  β1  1  ρ  δ0  β 2 δ1  ρβ 2 θ1  ρ
– Eq. 9.46 is a restricted version of Eq. 9.47 with

the restriction δ1 = -θ1δ0 imposed

9.5
Estimation with
Serially Correlated
Errors
9.5.3
Estimating a More
General Model
Applying the least squares estimator to Eq. 9.47

using the data for the Phillips curve example
yields:
 t  0.3336  0.5593INF  0.6882DU  0.3200DU
INF t 1 t t 1
Eq. 9.49
 se   0.0899  0.0908  0.2575  0.2499

9.5
Estimation with
Serially Correlated
Errors
9.5.3
Estimating a More
General Model
The equivalent AR(1) estimates are:

δˆ  βˆ 1  1  ρˆ   0.7609   1  0.5574   0.3368
θˆ1  ρˆ  0.5574
δˆ  βˆ  0.6944
0 2
ˆ ˆ 2  0.5574   0.6944   0.3871

δˆ1  ρβ
– These are similar to our other estimates

9.5
Estimation with
Serially Correlated
Errors
9.5.3
Estimating a More
General Model
The original economic model for the Phillips

Curve was:
Eq. 9.50 INFt  INFt E  γ  Ut  Ut 1 
– Re-estimation of the model after omitting DUt-1

yields:
 t  0.3548  0.5282INF  0.4909DU
INF t 1 t
Eq. 9.51
 se  0.0876  0.0851  0.1921
9.5
Estimation with
Serially Correlated
Errors
9.5.3
Estimating a More
General Model
In this model inflationary expectations are given

by:
INFt E  0.3548  0.5282INFt 1
– A 1% rise in the unemployment rate leads to an

approximate 0.5% fall in the inflation rate

9.5
Estimation with
Serially Correlated
Errors
9.5.4
Summary of
Section 9.5 and
We have described three ways of overcoming the
Looking Ahead
effect of serially correlated errors:
1. Estimate the model using least squares with
HAC standard errors
2. Use nonlinear least squares to estimate the
model with a lagged x, a lagged y, and the
restriction implied by an AR(1) error
specification
3. Use least squares to estimate the model with a
lagged x and a lagged y, but without the
restriction implied by an AR(1) error
specification
9.6
Autoregressive Distributed Lag
Models

9.6
Autoregressive
Distributed Lag
Models
An autoregressive distributed lag (ARDL) model

is one that contains both lagged xt’s and lagged yt’s
Eq. 9.52 yt    0 xt  1 xt 1     q xt q  1 yt 1     p yt  p  vt
– Two examples:
ADRL  1,1 : INF


t  0.3336  0.5593 INFt 1  0.6882 DU t  0.3200 DU t 1
ADRL  1, 0  : INF

t  0.3548  0.5282 INFt 1  0.4909 DU t

9.6
Autoregressive
Distributed Lag
Models
An ARDL model can be transformed into one with

only lagged x’s which go back into the infinite
past:
yt    0 xt  β1 xt 1  β 2 xt  2  β 3 xt 3    et
Eq. 9.53 
    β s xt  s  et
s 0
– This model is called an infinite distributed

lag model

9.6
Autoregressive
Distributed Lag
Models
Four possible criteria for choosing p and q:

1. Has serial correlation in the errors been
eliminated?
2. Are the signs and magnitudes of the estimates
consistent with our expectations from
economic theory?
3. Are the estimates significantly different from
zero, particularly those at the longest lags?
4. What values for p and q minimize information
criteria such as the AIC and SC?

9.6
Autoregressive
Distributed Lag
Models
The Akaike information criterion (AIC) is:

 SSE  2 K
Eq. 9.54 AIC  ln  
 T  T
where K = p + q + 2
The Schwarz criterion (SC), also known as the
Bayes information criterion (BIC), is:
 SSE  K ln  T 
Eq. 9.55 SC  ln  
 T  T
– Because Kln(T)/T > 2K/T for T ≥ 8, the SC

penalizes additional lags more heavily than
does the AIC
9.6
Autoregressive
Distributed Lag
Models
9.6.1
The Phillips Curve
Consider the previously estimated ARDL(1,0)

model:
 t  0.3548  0.5282INF  0.4909DU , obs  90

INF
Eq. 9.56 t 1 t
 se  0.0876  0.0851  0.1921

9.6
Autoregressive
Distributed Lag FIGURE 9.9 Correlogram for residuals from Phillips curve ARDL(1,0) model
Models
9.6.1
The Phillips Curve

9.6
Autoregressive
Distributed Lag Table 9.3 p-values for LM Test for Autocorrelation
Models
9.6.1
The Phillips Curve

9.6
Autoregressive
Distributed Lag
Models
9.6.1
The Phillips Curve
For an ARDL(4,0) version of the model:
 t  0.1001  0.2354INF  0.1213INF  0.1677INF

INF t 1 t 2 t 3
Eq. 9.57
 se  0.0983  0.1016  0.1038  0.1050
 0.2819INFt -4  0.7902DUt
 0.1014  0.1885 obs  87

9.6
Autoregressive
Distributed Lag
Models
9.6.1
The Phillips Curve
Inflationary expectations are given by:
INFt E  0.1001  0.2354INFt 1  0.1213INFt 2  0.1677INFt 3  0.2819INFt -4

9.6
Autoregressive
Distributed Lag Table 9.4 AIC and SC Values for Phillips Curve ARDL Models
Models
9.6.1
The Phillips Curve

9.6
Autoregressive
Distributed Lag
Models
9.6.2
Okun’s Law
Recall the model for Okun’s Law:
 t  0.5836  0.2020G  0.1653G  0.0700G , obs  96

DU t t 1 t 2
Eq. 9.58
 se  0.0472  0.0324  0.0335  0.0331

9.6
Autoregressive
Distributed Lag FIGURE 9.10 Correlogram for residuals from Okun’s law ARDL(0,2) model
Models
9.6.2
Okun’s Law

9.6
Autoregressive
Distributed Lag Table 9.5 AIC and SC Values for Okun’s Law ARDL Models
Models
9.6.2
Okun’s Law

9.6
Autoregressive
Distributed Lag
Models
9.6.2
Okun’s Law
Now consider this version:
 t  0.3780  0.3501DU  0.1841G  0.0992G , obs  96

DU t 1 t t 1
Eq. 9.59
 se  0.0578  0.0846  0.0307  0.0368

9.6
Autoregressive
Distributed Lag
Models
9.6.3
Autoregressive
Models
An autoregressive model of order p, denoted

AR(p), is given by:
Eq. 9.60 yt  δ  θ1 yt 1  θ2 yt 2    θ p yt  p  vt

9.6
Autoregressive
Distributed Lag
Models
9.6.3
Autoregressive
Models
Consider a model for growth in real GDP:
 t  0.4657  0.3770G  0.2462G

G t 1 t 2
Eq. 9.61
 se  0.1433  0.1000  0.1029 obs = 96

9.6
Autoregressive
Distributed Lag FIGURE 9.11 Correlogram for residuals from AR(2) model for GDP
Models
growth
9.6.3
Autoregressive
Models

9.6
Autoregressive
Distributed Lag Table 9.6 AIC and SC Values for AR Model of Growth in U.S. GDP
Models
9.6.3
Autoregressive
Models

9.7
Forecasting

9.7
Forecasting
We consider forecasting using three different

models:
1. AR model
2. ARDL model
3. Exponential smoothing model

9.7
Forecasting
9.7.1
Forecasting with
an AR Model
Consider an AR(2) model for real GDP growth:
Eq. 9.62 Gt  δ  θ1Gt 1  θ2Gt 2  vt
The model to forecast GT+1 is:
GT 1  δ  θ1GT  θ2GT 1  vT 1

9.7
Forecasting
9.7.1
Forecasting with
an AR Model
The growth values for the two most recent

quarters are:
GT = G2009Q3 = 0.8
GT-1 = G2009Q2 = -0.2
The forecast for G2009Q4 is:
GˆT 1  δˆ  θˆ1GT  θˆ 2GT 1

Eq. 9.63  0.46573  0.37700  0.8  0.24624   0.2
 0.7181

9.7
Forecasting
9.7.1
Forecasting with
an AR Model
For two quarters ahead, the forecast for G2010Q1 is:
GˆT 2  δˆ  θˆ1GT 1  θˆ 2GT

Eq. 9.64  0.46573  0.37700  0.71808  0.24624  0.8
 0.9334
For three periods out, it is:
GˆT 3  δˆ  θˆ1GT 2  θˆ 2GT 1

Eq. 9.65  0.46573  0.37700  0.93343  0.24624  0.71808
 0.9945

9.7
Forecasting
9.7.1
Forecasting with
an AR Model
Summarizing our forecasts:

– Real GDP growth rates for 2009Q4, 2010Q1,
and 2010Q2 are approximately 0.72%, 0.93%,
and 0.99%, respectively

9.7
Forecasting
9.7.1
Forecasting with
an AR Model
A 95% interval forecast for j periods into the

future is given by:
GˆT  j  t 0.975,df  σ̂ j
where σ̂ j is the standard error of the forecast
error and df is the number of degrees of freedom
in the estimation of the AR model

9.7
Forecasting
9.7.1
Forecasting with
an AR Model
The first forecast error, occurring at time T+1, is:
    
u1  GT 1  GˆT 1  δ  δˆ  θ1  θˆ1 GT  θ2  θˆ 2 GT 1  vT 1 
Ignoring the error from estimating the coefficients,
we get:
Eq. 9.66 u1  vT 1

9.7
Forecasting
9.7.1
Forecasting with
an AR Model
The forecast error for two periods ahead is:
Eq. 9.67  
u2  θ1 GT 1  GˆT 1  vT  2  θ1u1  vT  2  θ1vT 1  vT  2
The forecast error for three periods ahead is:
Eq. 9.68  
u3  θ1u2  θ 2u1  vT 3  θ12  θ 2 vT 1  θ1vT  2  vT 3

9.7
Forecasting
9.7.1
Forecasting with
an AR Model
Because the vt’s are uncorrelated with constant

 2
variance v , we can show that:
σ12  var  u1   σ v2
σ 22  var  u2   σ v2 1  θ12  
σ  var  u3   σ
2
3
2
v  θ  θ2
2
1 
2

 θ12  1

9.7
Forecasting Table 9.7 Forecasts and Forecast Intervals for GDP Growth
9.7.1
Forecasting with
an AR Model

9.7
Forecasting
9.7.2
Forecasting with
an ARDL Model
Consider forecasting future unemployment using

the Okun’s Law ARDL(1,1):
Eq. 9.69 DU t  δ  θ1DU t 1  δ0Gt  δ1Gt 1  vt

The value of DU in the first post-sample quarter
is:
Eq. 9.70 DU T 1  δ  θ1 DU T  δ0GT 1  δ1GT  vT 1
– But we need a value for GT+1

9.7
Forecasting
9.7.2
Forecasting with
an ARDL Model
Now consider the change in unemployment

– Rewrite Eq. 9.70 as:
U T 1  UT  δ  θ1  UT  UT 1   δ0GT 1  δ1GT  vT 1
– Rearranging:
U T 1  δ   θ1  1 U T  θ1U T 1  δ0GT 1  δ1GT  vT 1

Eq. 9.71
 δ  θ1*U T  θ*2U T 1  δ 0GT 1  δ1GT  vT 1

9.7
Forecasting
9.7.2
Forecasting with
an ARDL Model
For the purpose of computing point and interval

forecasts, the ARDL(1,1) model for a change in
unemployment can be written as an ARDL(2,1)
model for the level of unemployment
– This result holds not only for ARDL models
where a dependent variable is measured in
terms of a change or difference, but also for
pure AR models involving such variables

9.7
Forecasting
9.7.3
Exponential
Smoothing
Another popular model used for predicting the

future value of a variable on the basis of its history
is the exponential smoothing method
– Like forecasting with an AR model, forecasting
using exponential smoothing does not utilize
information from any other variable

9.7
Forecasting
9.7.3
Exponential
Smoothing
One possible forecasting method is to use the

average of past information, such as:
yT  yT 1  yT  2
yˆT 1 
3
– This forecasting rule is an example of a simple
(equally-weighted) moving average model with
k=3

9.7
Forecasting
9.7.3
Exponential
Smoothing
Now consider a form in which the weights decline

exponentially as the observations get older:
yˆT 1  αyT  α  1  α  yT 1  α  1  α  yT 2 
1 2
Eq. 9.72
– We assume that 0 < α < 1

– Also, it can be shown that:
s
 α  1 α  1

s 0

9.7
Forecasting
9.7.3
Exponential
Smoothing
For forecasting, recognize that:
 1 α yˆT  α 1 α yT 1  α 1 α yT 2  α  1 α yT 3 
2 3
Eq. 9.73
– We can simplify to:
Eq. 9.74 yˆT 1  αyT   1 α yˆT

9.7
Forecasting
9.7.3
Exponential
Smoothing
The value of α can reflect one’s judgment about
the relative weight of current information
– It can be estimated from historical information
by obtaining within-sample forecasts:
Eq. 9.75 yˆt  αyt 1   1 α yˆt 1 t  2,3,,T
• Choosing α that minimizes the sum of

squares of the one-step forecast errors:
Eq. 9.76 vt  yt  yˆt  yt   αyt 1 + 1 α yˆt 1 

9.7 FIGURE 9.12 (a) Exponentially smoothed forecasts for GDP growth with α
Forecasting
= 0.38
9.7.3
Exponential
Smoothing

9.7 FIGURE 9.12 (b) Exponentially smoothed forecasts for GDP growth with α
Forecasting
= 0.8
9.7.3
Exponential
Smoothing

9.7
Forecasting
9.7.3
Exponential
Smoothing
The forecasts for 2009Q4 from each value of α

are:
α  0.38: GˆT 1  αGT   1 α GˆT  0.380.8   1 0.38   0.403921
= 0.0536
α  0.8: GˆT 1  αGT   1 α GˆT  0.8 0.8   1 0.8   0.393578
= 0.5613

9.8
Multiplier Analysis

9.8
Multiplier Analysis
Multiplier analysis refers to the effect, and the

timing of the effect, of a change in one variable on
the outcome of another variable

9.8
Multiplier Analysis
Let’s find multipliers for an ARDL model of the

form:
Eq. 9.77 yt    1 yt 1     p yt  p  0 xt  1 xt 1     q xt  q  vt
– We can transform this into an infinite

distributed lag model:
Eq. 9.78 yt  α  β 0 x t + β1 xt 1  β 2 xt  2  β 3 xt 3    et

9.8
Multiplier Analysis
The multipliers are defined as:

yt
βs   s period delay multiplier
xt  s
β
j 0
j  s period interim multiplier
β
j 0
j  total multiplier

9.8
Multiplier Analysis
The lag operator is defined as:

Lyt  yt 1
– Lagging twice, we have:
L  Lyt   Lyt 1  yt  2
– Or:
L2 yt  yt  2
– More generally, we have:

Ls yt  yt  s

9.8
Multiplier Analysis
Now rewrite our model as:
yt    1 Lyt  2 L2 yt     p Lp yt  0 xt  1Lxt   2 L2 xt
Eq. 9.79
   q Lq xt  vt

9.8
Multiplier Analysis
Rearranging terms:
Eq. 9.80  1 2
1   L   L2
    p Lp
 t
y     0 1 2
   L   L2
    q Lq
 xt  vt

9.8
Multiplier Analysis
Let’s apply this to our Okun’s Law model

– The model:
Eq. 9.81 DU t  δ  θ1DU t 1  δ0Gt  δ1Gt 1  vt
can be rewritten as:
Eq. 9.82  1  θ1L  DU t  δ   δ0  δ1L  Gt  vt

9.8
Multiplier Analysis
Define the inverse of (1 – θ1L) as (1 – θ1L)-1 such

that:
 1  θ1L   1  θ1L   1
1

9.8
Multiplier Analysis
Multiply both sides of Eq. 9.82 by (1 – θ1L)-1:
DUt   1  θ1L  δ   1  θ1L   δ0  δ1L  Gt   1  θ1L 

1 1 1
Eq. 9.83 vt
– Equating this with the infinite distributed lag

representation:
DU t  α  β0Gt  β1Gt 1  β 2Gt 2  β3Gt 3    et

 
Eq. 9.84
 α  β0  β1L  β 2 L2  β3 L3   Gt  et

9.8
Multiplier Analysis
For Eqs. 9.83 and 9.84 to be identical, it must be

true that:
α=  1  θ1L  δ
1
Eq. 9.85
β0  β1L  β2 L  β3 L     1  θ1L   δ0  δ1L 

2 3 1
Eq. 9.86
et   1  θ1L  vt
1
Eq. 9.87

9.8
Multiplier Analysis
Multiply both sides of Eq. 9.85 by (1 – θ1L) to

obtain (1 – θ1L)α = δ.
– Note that the lag of a constant that does not
change so Lα = α
– Now we have:
δ
 1  θ1  α  δ and α 
1  θ1

9.8
Multiplier Analysis
Multiply both sides of Eq. 9.86 by (1 – θ1L):

δ0  δ1L   1 θ1L β0  β1L  β2 L2  β3L3  
 β0  β1L  β2 L2  β3L3 
Eq. 9.88
 β0θ1L  β1θ1L2  β2θ1L3 
 β0   β1  β0θ1  L   β2  β1θ1  L2   β3  β2θ1  L3  

9.8
Multiplier Analysis
Rewrite Eq. 9.86 as:
Eq. 9.89 δ0  δ1L  0L2  0L3  β0   β1  β0θ1  L   β2  β1θ1  L2   β3  β2θ1  L3 
– Equating coefficients of like powers in L yields:
δ0 = β0
δ1  β1  β0θ1
0  β2  β1θ1
0  β3  β2θ1
and so on
9.8
Multiplier Analysis
We can now find the β’s using the recursive

equations:
β0 = δ0
Eq. 9.90 β1  δ1  β0θ1
β j  β j 1θ1 for j  2

9.8
Multiplier Analysis
You can start from the equivalent of Eq. 9.88

which, in its general form, is:

δ0  δ1L  δ2 L2  δq Lq  1 θ1L  θ2 L2  θ p Lp 
Eq. 9.91

 β0  β1L  β2 L2  β3L3  
– Given the values p and q for your ARDL
model, you need to multiply out the above
expression, and then equate coefficients of like
powers in the lag operator

9.8
Multiplier Analysis
For the Okun’s Law model:

 t  0.3780  0.3501DU  0.1841G  0.0992G
DU t 1 t t 1
– The impact and delay multipliers for the first

four quarters are:
βˆ 0 = δˆ 0  0.1841
βˆ  δˆ  βˆ θˆ  0.099155  0.184084  0.350116  0.1636
1 1 0 1
βˆ 2  βˆ1θˆ1  0.163606  0.350166  0.0573

βˆ  βˆ θˆ  0.057281 0.350166  0.0201
3 2 1
βˆ 4  βˆ 3θˆ1  0.020055 0.350166  0.0070

9.8
Multiplier Analysis FIGURE 9.13 Delay multipliers from Okun’s law ARDL(1,1) model

9.8
Multiplier Analysis
We can estimate the total multiplier given by:


β
j 0
j
and the normal growth rate that is needed to

maintain a constant rate of unemployment:

GN  α β
j 0
j

9.8
Multiplier Analysis
We can show that:

 ˆ  δˆ θˆ
δ 0.163606

j 0
ˆ
β j  δ0  1
ˆ
1 θ
0 1
 0.184084 
1 0.350116
 0.4358
1
– An estimate for α is given by:
δ̂ 0.37801
α̂    0.5817
1 θˆ1 0.649884
– Therefore, normal growth rate is:
0.5817
ĜN   1.3% per quarter
0.4358

Key Words

Keywords
AIC criterion dynamic models LM test

AR(1) error exponential multiplier analysis
AR(p) model smoothing nonlinear least squares
ARDL(p,q) model finite distributed lag out-of-sample
forecast error forecasts
autocorrelation
forecast intervals sample
Autoregressive
forecasting autocorrelations
distributed lags
HAC standard errors serial correlation
autoregressive error standard error of
autoregressive impact multiplier
forecast error
model infinite distributed lag
SC criterion
BIC criterion interim multiplier
total multiplier
correlogram lag length
T x R2 form of LM
delay multiplier lag operator test
lagged dependent within-sample
distributed lag
variable forecasts
weight
Appendices

9A
The Durbin-
Watson Test
For the Durbin-Watson test, the hypotheses are:

H0 :   0 H1 :   0
The test statistic is:
T
  eˆt  eˆt 1 
2
Eq. 9A.1 d t 2
T
t
ˆ
e 2
t 1

9A
The Durbin-
Watson Test
We can expand the test statistic as:

T T T
 t  t 1  2 eˆt eˆt 1
ˆ
e 2
 ˆ
e 2
d t 2 t 2
T
t 2
t
ˆ
e 2
t 1
T T T
Eq. 9A.2
 eˆ 2
t  eˆ 2
t 1  eˆt eˆt 1
 t 2
T
 t 2
T
2 t 2
T
t
ˆ
e 2
t
ˆ
e 2
t
ˆ
e 2
t 1 t 1 t 1
 1  1  2r1
9A
The Durbin-
Watson Test
We can now write:
Eq. 9A.3 d  2  1  r1 
– If the estimated value of ρ is r1 = 0, then the
Durbin-Watson statistic d ≈ 2
• This is taken as an indication that the model
errors are not autocorrelated
– If the estimate of ρ happened to be r1 = 1 then d ≈
0
• A low value for the Durbin-Watson statistic
implies that the model errors are correlated, and
ρ>0
9A
The Durbin-
Watson Test FIGURE 9A.1 Testing for positive autocorrelation

9A
The Durbin- FIGURE 9A.2 Upper and lower critical value bounds for the Durbin-
Watson Test
Watson test
9A.1
The Durbin-
Watson Bounds
Test

9A
The Durbin-
Watson Test
9A.1
The Durbin-
Watson Bounds
Test
Decision rules, known collectively as the Durbin-

Watson bounds test:
– If d < dLc: reject H0: ρ = 0 and accept H1:
ρ>0
– If d > dUc do not reject H0: ρ = 0
– If dLc < d < dUc, the test is inconclusive

9B
Properties of the
AR(1) Error
Note that:
et  ρet 1  vt
 ρ  ρet  2  vt 1   vt
Eq. 9B.1
 ρ 2 et  2  ρvt 1  vt
Further substitution shows that:
et  ρ  ρet 3  vt  2   ρvt 1  vt
2
Eq. 9B.2
 ρ3et 3  ρ 2 vt  2  ρvt 1  vt

9B
Properties of the
AR(1) Error
Repeating the substitution k times and

rearranging: k 1
Eq. 9B.3 et  ρ et  k  vt  ρvt 1  ρ vt  2    ρ vt  k 1
k 2
If we let k → ∞, then we have:

Eq. 9B.4 et  vt  ρvt 1  ρ 2 vt  2  ρ3vt 3  

9B
Properties of the
AR(1) Error
We can now find the properties of et:

E  et   E  vt   ρE  vt 1   ρ 2 E  vt  2   ρ3 E  vt 3   
 0  ρ  0  ρ 2  0  ρ3  0  
0
var  et   var  vt   ρ2var  vt 1   ρ4 var  vt 2   ρ6 var  vt 3  
v2  ρ2v2  ρ4v2  ρ6v2 
  1 ρ  ρ  ρ 
2
v
2 4 6
v2

1 ρ2
9B
Properties of the
AR(1) Error
The covariance for one period apart is:

cov  et , et 1   E  et et t 
 E  vt  ρvt 1  ρ vt 2  ρ vt 3  
 2 3
 v  ρv  ρ v  ρ v   
t 1 t 2
2
t 3
3
t 4
 ρE  v   ρ E  v   ρ E  v   
2
t 1
3 2
t 2
5 2
t 3
 ρ  1  ρ  ρ  
2
v
2 4
ρv2

1  ρ2
9B
Properties of the
AR(1) Error
Similarly, the covariance for k periods apart is:
ρk v2
cov  et , et k   k 0
1  ρ2

9C
Generalized Least
Squares Estimation
We are considering the simple regression model

with AR(1) errors:
yt  1  2 xt  et et  et 1  vt
To specify the transformed model we begin with:
Eq. 9C.1 yt  1  2 xt  yt 1  1   2 xt 1  vt

– Rearranging terms:
Eq. 9C.2 yt  yt 1  1  1     2  xt  xt 1   vt

9C
Generalized Least
Squares Estimation
Defining the following transformed variables:
yt  yt  yt 1 xt2  xt  xt 1 xt1  1  
Substituting the transformed variables, we get:
Eq. 9C.3 yt  xt11  xt22  vt

9C
Generalized Least
Squares Estimation
There are two problems:

1. Because lagged values of yt and xt had to be
formed, only (T - 1) new observations were
created by the transformation
2. The value of the autoregressive parameter ρ is
not known

9C
Generalized Least
Squares Estimation
For the second problem, we can write Eq. 9C.1 as:
Eq. 9C.4 yt  1  2 xt  ( yt 1  1  2 xt 1 )  vt
For the first problem, note that:

y1  1  x12  e1
and that
1  2 y1  1  2 1  1   2 x1 2  1   2 e1

9C
Generalized Least
Squares Estimation
Or:
Eq. 9C.5 y1  x11 1  x12 2  e1
where
y1  1  2 y1 x11  1  2
Eq. 9C.6
x12  1  2 x1 e1  1  2 e1

9C
Generalized Least
Squares Estimation
To confirm that the variance of e*1 is the same as

that of the errors (v2, v3,…, vT), note that:
 2
var(e1 )  (1  2 ) var(e1 )  (1  2 ) v 2  v2
1 


CH 09

Uploaded by

Copyright:

Available Formats

CH 09

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CH 09

Uploaded by

Copyright:

Available Formats

Chapter 9

Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

When modeling relationships between variables,

Chapter 9: Regression with Time Series Data:

There is also the possible existence of dynamic

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

• Because of the existence of these lagged

Chapter 9: Regression with Time Series Data:

Eq. 9.4 yt  f ( xt )  et et  f (et 1 )

• In this case et is correlated with et - 1

Chapter 9: Regression with Time Series Data:

cov  yi , y j   cov  ei , e j   0 for i  j

cov  yt , ys   cov  et , es   0 for t  s

– The dynamic models in Eqs. 9.2, 9.3 and 9.4

A stationary variable is one that is not explosive,

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Consider a linear model in which, after q time

– Note the notation change: βs is used to denote

Chapter 9: Regression with Time Series Data:

Model 9.5 has two uses:

Chapter 9: Regression with Time Series Data:

Assume xt is increased by one unit and then

Chapter 9: Regression with Time Series Data:

The effect of a one-unit change in xt is distributed

Chapter 9: Regression with Time Series Data:

– We can rewrite this as:

where DU = ΔU = Ut - Ut-1, β0 = -γ, and α=

We can expand this to include lags:

Eq. 9.10 DU t    β 0Gt  β1Gt 1  β 2Gt 2    β q Gt q  et

We can calculate the growth in output, G, as:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

When is assumption TSMR5, cov(et, es) = 0 for

Chapter 9: Regression with Time Series Data:

Chapter 9: Regression with Time Series Data:

Recall that the population correlation between two

Chapter 9: Regression with Time Series Data:

The notation ρ1 is used to denote the population

The first-order sample autocorrelation for G is

Chapter 9: Regression with Time Series Data:

Making the substitutions, we get:

Chapter 9: Regression with Time Series Data:

More generally, the k-th order sample

Chapter 9: Regression with Time Series Data:

Because (T - k) observations are used to compute