Cointegration Copenhagen

university of copenhagen department of economics
Econometrics II
Cointegration and Spurious Regression
Morten Nyboe Tabor
Econometrics II
Course Outline: Cointegration and Spurious Regression

1 Regression with Non-Stationarity
Example of Spurious Regression
Simulation Example
2 Definitions and Concepts
Cointegration Defined
Cointegration and Economic Equilibrium
Cointegration and Error Correction
3 Engle-Granger Two-Step Cointegration Analysis
Static Regression with Cointegrated Series
Super-Consistency
Residual-Based Tests for No-Cointegration
Engle-Granger Two-Step Analysis
4 Empirical Example Based on the Engle-Granger Two-Step Approach
5 Cointegration Analysis Based on Dynamic Models
Estimation in the Unrestricted ADL/ECM Model
PcGive Test for No-Cointegration
6 Empirical Examples Based on the ADL One-Step Approach
7 Comparison of the Engle-Granger and ADL Approaches
Econometrics II — Cointegration and Spurious Regression — Slide 2/51

Learning Outcomes
After completing this topic, you should be able to:

1 Explain the concept of spurious regression.
2 Give a precise definition and interpretation of the concepts cointegration
and error correction.
3 Give an account of statistical models based on cointegration and error
correction.
4 Construct statistical tests for cointegration and error correction in
economic time series.

1. Regression with Non-Stationarity
Motivation: Regression with Non-Stationarity
• What happens to the properties of OLS if variables are non-stationary?

• Consider two presumably unrelated variables:
CONS Danish private consumption in 1995 prices.

BIRD Number of breeding cormorants (skarv) in Denmark.
And consider a static regression model
log(CONSt ) = µ + β2 · log(BIRDt ) + ut .
We would expect (or hope) to get βb2 ≈ 0 and R 2 ≈ 0.

• Applying OLS to yearly data 1982 − 2001 gives the result:
log(CONSt ) = 12.145 + 0.095 · log(BIRDt ) + ut ,

(80.90) (6.30)
with R 2 = 0.688.
• It looks like a reasonable model.
But it is complete nonsense: spurious regression.

• The variables are non-stationary. The residual, ut , is non-stationary and

standard
• The variablesresults for OLS do not hold.
are non-stationary.
•The general, regression
In residual,  , is non-stationary andnon-stationary
models for standard results for OLSgive
variables do not hold.
spurious
• In results.
general, regression models for non-stationary variables give spurious results.
Only exception is if the model eliminates the stochastic trends to produce
Only exception is if the model eliminates the stochastic trends to produce stationary
stationary residuals: Cointegration.
residuals: Cointegration.
• For non-stationary variables we should always think in terms of
• Forcointegration.
non-stationaryOnly
variables
look we
at should always
regression thinkifinthe
output terms of cointegration.
variables are
Only look at regression
stationary output if the variables are stationary or cointegrate.
or cointegrate.
Consumption and breeding birds, logs Regression residuals (cons on birds)

Consumption
13.2 Number of breeding birds 1
13.1
0
13.0
-1
12.9
1985 1990 1995 2000 1985 1990 1995 2000
3 of 31

Simulation: Stationary Case

Simulation: Stationary Case
• Consider first two independent IID variables:
• Consider first
yt =two independent IID
yt
variables:
yt

0 1 0

where µ ¶ ∼ N µµ ¶ µ , ¶¶ ,
xt = =xt  xt
 00 1 00 1
where ∼  
 =   0 0 1
for T = 50, 100, 500.
for  = 50 100 500.
• Here, we get standard results for the regression model
• Here, we get standard results for the regression model
yt = µ + β2 xt + ut .
 =  +  2 + 
IID Case, estimates IID Case, t-ratios

50 0.4 N(0,1)
100 Note the 50 Looks like
7.5 500 convergence 100 a N(0,1)
to  2=0 500 for all T.
as T diverges. Standard testing.
5.0
0.2
2.5
-0.50 -0.25 0.00 0.25 0.50 -4 -2 0 2 4
4 of 31

Simulation: I(1) Spurious Regression

• Now consider two independent random walks
Simulation: I(1) Spurious
Regression

yt = yt−1 + yt yt 0 1 0
• Now consider where ∼N , ,
xt = xtwo independent
t−1 + xt
random walks
xt 0 0 1
µ ¶ µµ ¶ µ ¶¶
 = −1 +   0 1 0
for T = 50, 100, 500. where ∼  
• Again,we  = −1 +   0 0 1
estimate the regression model
for  = 50 100 500.
yt = µ + β2 xt + ut .
• Under the null hypothesis,  2 = 0, the residual is I(1). The condition for consistency
• Under the null hypothesis, β2 = 0, the residual is I(1). The condition for
is not fulfilled. is not fulfilled.
consistency
I(1) case, estimates I(1) case, t-ratios
0.75 50 50
100 Looks unbiased, 100 The distribution
500 but NO 0.075 500
gets increasingly
convergence. dispersed.
0.50
0.050
0.25 0.025 Note the scale

as compared
to a N(0,1)
-3 -2 -1 0 1 2 3 -75 -50 -25 0 25 50 75
Econometrics II — Cointegration and Spurious Regression — 5Slide

of 31
8/51
2. Definitions and Concepts
Cointegration Defined
• Let xt = ( x1t x2t )0 be two I(1) variables:
x1t = τ1t + initial value + stationary process

x2t = τ2t + initial value + stationary process.
where τ1t and τ2t are stochastic trends.

• In general, a linear combination of x1t and x2t will also have a random
walk.
• Define β = ( 1 −β2 )0 and consider the linear combination:

x1t
β 0 xt =

zt = 1 −β2 = x1t − β2 x2t
x2t
= τ1t − β2 τ2t + initial value + stationary process.
Definition of Cointegration
Let xt = ( x1t x2t )0 be two I(1) variables. If there exists a vector β such
that zt = β 0 xt is stationary (i.e., zt ∼ I(0)), then x1t and x2t are said to
co-integrate with cointegration vector β.

Remarks
1 Cointegration occurs if the stochastic trends in x1t and x2t are the same,
so that they cancel out: τ1t − β2 τ2t = 0.
This is called a common stochastic trend.
2 You can think of an equation eliminating the random walks in x1t and x2t :
x1t = µ + β2 x2t + ut . (♦)
If ut is I(0) (mean zero) then β = ( 1 −β2 )0 is a cointegrating vector.

Remarks (Continued)
3 The cointegrating vector is only unique up to a constant factor:

If β 0 xt ∼I(0), then so is βe0 xt = bβ 0 xt for b 6= 0.
We can choose a normalization

1 −βe1
β= or βe = .
−β2 1
This corresponds to different variables on the left hand side of (♦)
4 Cointegration is easily extended to more variables.

0
The variables in xt = x1t x2t · · · xpt cointegrate if
zt = β 0 xt = x1t − β2 · x2t − ... − βp · xpt ∼ I(0).

Cointegration and Economic Equilibrium
• Consider a regression model for two I(1) variables, x1t and x2t , given by
x1t = µ + β2 x2t + ut . (∗)
The term, ut , is interpretable as the deviation from the relation in (∗).
• If x1t and x2t cointegrate, then the deviation
ut = x1t − µ − β2 x2t
is a stationary process with mean zero.
• Shocks to x1t and x2t have permanent effects. x1t and x2t co-vary and
ut ∼I(0). We can think of (∗) as defining an equilibrium between x1t and
x2t .
• If x1t and x2t do not cointegrate, then the deviation ut is I(1).

There is no natural interpretation of (∗) as an equilibrium relation.

Empirical Example: Consumption and Income

Empirical Example:
• Time series for Consumption
log consumption, c , and log income, and
t y , are Income
likely to be t
integrated of order 1, I(1).
• Time series for log consumption, , and log income, , are likely to be I(1).
• Define a vector xt = ( ct 0 yt )0 . Consumption and income are
Define a vector  = (   ) .
cointegrated with cointegration vector β = ( 1 −1 )0 if the (log-)
  
consumption-income
• Consumption and income ratio,
are cointegrated with cointegration vector  = ( 1 −1 )0
if the (log-) consumption-income ratio,

0 ct
zt = β xt = ( 1 −1 )µ ¶ = ct − yt ,
yt
 =  0 = ( 1 −1 ) =  − 

is a stationary process. The consumption-income ratio is an equilibrium
is a relation.
stationary process. The consumption-income ratio is an equilibrium relation.
Consumption and income, logs Income ratio, l o g (C t )lo g (Y t )

Income 0.00
Consumption
6.25 -0.05
-0.10
6.00
-0.15
1970 1980 1990 2000 1970 1980 1990 2000
10 of 31
How is the Equilibrium Sustained?
• There must be forces pulling x1t or x2t towards the equilibrium.
Engle and Granger’s theorem (1987)

x1t and x2t cointegrate if and only if there exists an error correction model for
either x1t , x2t , or both.

More on Error-Correction
• Cointegration is a system property. Both variables could error correct, e.g.:
∆x1t = δ1 + Γ11 ∆x1t−1 + Γ12 ∆x2t−1 + α1 (x1t−1 − β2 x2t−1 ) + 1t

∆x2t = δ2 + Γ21 ∆x1t−1 + Γ22 ∆x2t−1 + α2 (x1t−1 − β2 x2t−1 ) + 2t ,
• We may write the model as the so-called vector error correction model,

∆x1t δ1 Γ11 Γ12 ∆x1t−1 α1 1t
= + + (x1t−1 − β2 x2t−1 ) + ,
∆x2t δ2 Γ21 Γ22 ∆x2t−1 α2 2t
or simply
∆xt = δ + Γ∆xt−1 + αβ 0 xt−1 + t .
• Note that β 0 xt−1 = x1t−1 − β2 x2t−1 appears in both equations.
• For x1t to error correct we need α1 < 0.

For x2t to error correct we need α2 > 0.

Considera asimple
Consider simple model
model for for
twotwo cointegrated
cointegrated variables:
variables:
µ ¶ µ ¶
µ ¶
∆x∆
1t 1 −0.2
−02 11t
== (
(x1−1
1t−1 −
− x2t−1
2−1 ) +
) +  .
∆x∆
2t 2 01
0.1 22t
(A) Two cointegrated variables (B) Deviation from equilibrium

x1t
0 x2t  'x t =x 1t x 2t
2.5
-5
0.0
-10
-2.5
0 20 40 60 80 100 0 20 40 60 80 100
(C) Speed of adjustment (D) Cross-plot

12.5
x1t  x2t
10.0 0
7.5
5.0 -5
 'x t
2.5
0.0 -10 x
100
-2.5 13 of 31
0 20 40 60 80 100 -12.5 -10.0 -7.5 -5.0 -2.5 0.0

3. Engle-Granger Two-Step Cointegration
Analysis
Static Regression with Cointegrated Series
• In the cointegration case there exists a β2 so that the error term, ut , in
x1t = µ + β2 x2t + ut . (∗∗)
is stationary.
• OLS applied to (∗∗) gives consistent results, so that βb2 → β2 for T → ∞.
• Note that consistency is obtained even if potential dynamic terms are

neglected. This is because the stochastic trends in x1t and x2t dominate.
We can even get consistent estimates in the reverse regression
x2t = δ + γ1 x1t + vt .
(See Section 2.3 in the lecture note.)
• Unfortunately, it turns out that βb2 is not asymptotically normal in general.

The normal inferential procedures do not apply to βb2 !
We can use (∗∗) for estimation but not for testing.

Super-Consistency Super-Consistency
• For stationary series, the variance of βb2 declines at a rate of T −1 .
• For stationary series, the b2 declines bat a rate of  −1.
of 
• For cointegrated I(1)variance
series, the variance of β2 declines at a faster rate of
T −2 . b
• For cointegrated I(1) series, the variance of  declines at a faster rate of  −2.
• Intuition: If βb2 = β2 then ut is stationary.2 If βb2 6= β2 then the error is I(1)
• Intuition: b2have
and Ifwill
 =  2a then
large variance. b
The ‘information’ on the parameter grows
 is stationary. If  2 6=  2 then the error is I(1) and will
very fast.
have a large variance. The ’information’ on the parameter grows very fast.
True β2 = 1.
Stationary, T=50 Stationary, T=100 Stationary, T=500

30 30 30
20 20 20
10 10 10
0 0 0
0.5 1.0 1.5 0.5 1.0 1.5 0.5 1.0 1.5
Non-Stationary, T=50 Non-Stationary, T=100 Non-Stationary, T=500
30 30 30
20 20 20
10 10 10
0 0 0
0.5 1.0 1.5 0.5 1.0 1.5 0.5 1.0 1.5

Test for No-Cointegration, Known β2
• Suppose that x1t and x2t are I(1).

Also assume that β = ( 1 −β2 )0 is known.
• The series cointegrate if

zt = x1t − β2 x2t
is stationary.
• This can be tested using an ADF unit root test, e.g. the test for
H0 : π = 0 in
k
X
∆zt = δ + ci ∆zt−i + πzt−1 + ηt .
i=1
The usual DF critical values apply to tπ=0 .
• Note, that the null, H0 : π = 0, is a unit root, i.e. no cointegration.

Test for No-Cointegration, Estimated β2

• Engle-Granger (1987) two-step procedure.
• If β = ( 1 −β2 )0 is unknown, it can be (super-consistently) estimated
in
x1t = µ + β2 x2t + ut . (∗ ∗ ∗)
ut = x1t − µ
βb is a cointegration vector if b b − βb2 x2t is stationary.
• This can be tested using a DF unit root test, e.g. the test for H0 : π = 0 in
k
X
∆b
ut = ci ∆b
ut−i + πb
ut −1 + ηt .
i=1
Remarks:
1 The residual b
ut has mean zero. No deterministic terms in DF regression.
2 The critical value for tπ=0 still depends on the deterministic regressors in
(∗ ∗ ∗).
3 The fact that βb2 is estimated also changes the critical values.
OLS minimizes the variance of b ut . Look ‘as stationary as possible’.
Critical value depends on the number of regressors.

Dickey-Fuller distribution for test for no-cointegration

0.6
Estimated parameters in cointegrating
regression (with constant in the regression (***))
76 54
3 2 1
0.5
DF(constant)
0.4
N(0,1)
0.3
0.2
0.1
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5
18 23/51
Econometrics II — Cointegration and Spurious Regression — Slide of 31
Critical values for the Dickey-Fuller test for no-cointegration

Table 1 A constant term in (∗ ∗ ∗)
Number of estimated Significance level

parameters 1% 5% 10%
0 −3.43 −2.86 −2.57
1 −3.90 −3.34 −3.04
2 −4.29 −3.74 −3.45
3 −4.64 −4.10 −3.81
4 −4.96 −4.42 −4.13
Table 2 A constant and a trend in (∗ ∗ ∗)
Number of estimated Significance level

parameters 1% 5% 10%
0 −3.96 −3.41 −3.13
1 −4.32 −3.78 −3.50
2 −4.66 −4.12 −3.84
3 −4.97 −4.43 −4.15
4 −5.25 −4.72 −4.43

Models for the Dynamic Adjustment
• Given the estimated cointegrating vector we can define the error

correction term
ecmt = but = x1t − µ
b − βb2 x2t ,
which is, per definition, a stationary stochastic variable.
• Since βb2 converges to β2 very fast we can treat it as a fixed regressor and
formulate an error correction model conditional on ecmt−1 , i.e.
∆x1t = δ + λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + α · ecmt−1 + t ,
where α < 0 is consistent with error-correction.
• Given cointegration, all terms are stationary, and normal inference applies
to δ, λ1 , κ0 , κ1 , and α.
• We could estimate ECM models for all variables.

Outline of an Engle-Granger Analysis
1 Test individual variables, e.g. x1t and x2t , for unit roots.
2 Run the static cointegrating regression
x1t = µ + β2 x2t + ut .
Note that the t−ratios cannot be used for inference.
3 Test for no-cointegration by testing for a unit root in the residuals, b

ut .
4 If cointegration is not rejected, estimate a dynamic (ECM) model like
∆x1t = δ + λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + αb

ut−1 + t .
All terms are stationary. Remaining inference is standard.

4. Empirical Example Based on the
Engle-Granger Two-Step Approach
Empirical Example: Danish Interest Rates

Bond Yield and money market interest rate Interest rate spread
0.2 B ond yield 0.05
M oney ma rket inte rest rate
0.00
0.1
-0.05
1970 1980 1990 2000 1970 1980 1990 2000
R esiduals from rt =   b t   t Impulse responses from b t to r t

1.50
1.25 1.19928
0.05
1.00
0.75 0.866611
0.00
0.50
0.25
-0.05
0.00
1980 1990 2000 0 5 10
Econometrics II — Cointegration and Spurious Regression — Slide

23 28/51
of 31

• Consider two Danish interest rates:
rt : Money market interest rate (IMM)
bt : Bond Yield (IBZ)
for the period t = 1972 : 1 − 2003 : 2.

• Test for unit roots in rt and bt (5% critical value is −2.86):
cr t = 0.00638118 − 0.126209 · ∆rt−1 − 0.234330 · ∆rt−4 − 0.0826987 · rt−1

∆
(1.35) (−2.39) (−2.70) (−1.80)
c t = 0.00116558 + 0.395115 · ∆bt−1 − 0.0128941 · bt−1

∆b
(0.658) (4.67) (−0.909)
• We cannot reject unit roots. Test if st = rt − bt is I(1)

(5% critical value is −2.86):
cs t = −0.00848594 + 0.207606 · ∆st−3 − 0.379449 · st−1 .

∆
(−3.71) (2.56) (−5.35)
It is easily rejected that bt and rt are not cointegrating.

• Instead of assuming β2 = 1 we could estimate the coefficient
Modelling IMM by OLS

The estimation sample is: 1974 (3) to 2003 (2)
Coefficient Std.Error t-value t-prob Part.Rˆ2

Constant -0.00468506 0.005545 -0.845 0.400 0.0062
IBZ 0.845524 0.04495 18.8 0.000 0.7563
sigma 0.0224339 RSS 0.0573738644

Rˆ2 0.756314 F(1,114) = 353.8 [0.000]**
log-likelihood 276.885 DW 0.82
no. of observations 116 no. of parameters 2
mean(IMM) 0.0919727 var(IMM) 0.00202967

• We could test for a unit root in the residuals (5% critical value is −3.34):
t = 0.230210 · ∆b
∆b t−3 − 0.499443 · b
t−1 .
(2.95) (−6.77)
Again we reject no-cointegration.

• Finally we could estimate the error correction models based on the spread:
cr t = −0.00774026 + 1.17725 · ∆bt − 0.406456 · (rt−1 − bt−1 )

∆
(−3.23) (4.55) (5.22)
c t = −0.00181602 + 0.438970 · ∆bt−1 − 0.0673997 · ∆rt

∆b
(−2.11) (4.16) (−2.01)
− 0.0638286 · (rt−1 − bt−1 )

(2.22)
Note that the short-rate, rt , error corrects, while the bond-yield, bt , does
not.
Note that we could have used ˆt−1 instead of rt−1 − bt−1 in the ECM.

5. Cointegration Analysis Based on Dynamic
Models
Estimation of β in the ADL/ECM Model
• The estimator of β2 from a static regression is super-consistent.

...but it is often biased (due to ignored dynamics) and hypotheses cannot
be tested.
• An alternative estimator is based on an unrestricted ADL model, e.g.
x1t = δ + θ1 x1t−1 + θ2 x1t−2 + φ0 x2t + φ1 x2t−1 + φ2 x2t−2 + t ,
where t is IID. This is equivalent to a linear regression error correction

model:
∆x1t = δ + λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + γ1 x1t−1 + γ2 x2t−1 + t ,
or the non-linear regression model
∆x1t = λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + α (x1t−1 − µ − β2 x2t−1 ) + t .

Estimation of β in the ADL/ECM Model (Continued)
• Setting x1t = x1t−1 = x1t−2 and x2t = x2t−1 = x2t−2 in the ADL and ECM
models yields the long-run solution:
x1t = µ + β2 x2t + t ,
with
−γ2 φ0 + φ1 + φ2 δ δ
β2 = = , and µ= = .
γ1 1 − θ1 − θ2 −γ1 1 − θ1 − θ2
• Hence, an estimate of β2 can be found from
−bγ2 φ
b0 + φb1 + φb2 δ
b δ
b
βb2 = = , and µ
b= = .
bγ1 1 − θb1 − θb2 −b
γ1 1 − θ1 − θb2
b
• The main advantage is that the analysis is undertaken in a well-specified

(dynamic) model. Inference on βb2 is possible. The approach is optimal if
only x1t error corrects.

Simulation Example: Static Regression vs. ADL

Consider M = 10, 000 simulated replications from the data-generating process,
yt = bxt + b1 xt−1 + yt yt ∼ N(0, 1)

xt = xt−1 + xt xt ∼ N(0, 1),
for t = 1, 2, ..., T and with b = 0.2 and b1 = 0.8. The variables are
cointegrated with cointegration vector β = (1, −1).
For each replication, we estimate:
1 The static regression,
yt = µ + β2 xt + ut ,
which is misspecied due the omitted dynamic term, b1 xt−1 .
2 The ADL(1,1) model,
yt = µ + θyt−1 + φ0 xt + φ1 xt−1 + t ,
which nests the DGP.

We plot the distribution of βb2 from the static regression and βb2 = φb0 +φb1 from
1−θ b
the ADL model.

Testing for No-Cointegration
• Due to Engle and Granger’s theorem, the null hypothesis of

no-cointegration corresponds to the null of no-error-correction. Several
tests have been designed in this spirit.
• The most convenient is the so-called PcGive test for no-cointegration.
• Consider the unrestricted ADL or ECM:
∆x1t = δ + λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + γ1 x1t−1 + γ2 x2t−1 + t . (#)
Test the hypothesis

H 0 : γ1 = 0
against the cointegration alternative, γ1 < 0.
• This is basically a unit root test (not a N(0, 1)). The distribution of the
t−ratio,
γ1
tγ1 =0 = ,
b
se(b
γ1 )
depends on the deterministic terms and the number of regressors in (#).

• Remark: The test for no-cointegration relies on the assumption that x2t
does not error-correct.
The error-correction model might be formulated based on economic
theory.

• Critical values for the PcGive test for no-cointegration are given by:
Table 3 A constant term in (#)
Number of variables Significance level

in xt 1% 5% 10%
2 −3.79 −3.21 −2.91
3 −4.09 −3.51 −3.19
4 −4.36 −3.76 −3.44
5 −4.59 −3.99 −3.66
Table 4 A constant and a trend in (#)
Number of variables Significance level

in xt 1% 5% 10%
2 −4.25 −3.69 −3.39
3 −4.50 −3.93 −3.62
4 −4.72 −4.14 −3.83
5 −4.93 −4.34 −4.03

Outline of a (One-Step) Cointegration Analysis
1 Test individual variables, e.g. x1t and x2t , for unit roots.
2 Estimate an ADL or ECM model
x1t = δ + θ1 x1t−1 + θ2 x1t−2 + φ0 x2t + φ1 x2t−1 + φ2 x2t−2 + t

∆x1t = δ + λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + γ1 x1t−1 + γ2 x2t−1 + t .
The lag-length is determined based on the general-to-specific approach.
3 Test for no-cointegration with tγ1 =0 .

If cointegration is found, the cointegrating relation is the long-run solution.
4 Derive the long-run solution
x1t = µ
b + βb2 x2t .
Inference on β2 is standard (under some conditions).
5 Rest of the analysis is standard: All N(0, 1). Look at dynamics.

How to determine the lag-length in ADL/ECM models?
• From Box 2 (p.15): An estimated parameter in an ADL/ECM model

follows a normal distribution (asymptotically) if the parameter can be
written as the coefficient to a mean zero stationary variable - possibly
after a linear transformation of the model.
• Consider the ADL and ECM models,
x1t = δ + θ1 x1t−1 + θ2 x1t−2 + φ0 x2t + φ1 x2t−1 + φ2 x2t−2 + εt

∆x1t = δ + λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1
+γ1 x1t−1 + γ2 x2t−1 + εt .
Here x1t and x2t are I(1) with no drift.

Can we do standard inference on the parameters φ2 and κ0
in the sense that the t-ratios, tφ2 =0 and tκ0 =0 , are asymptotically N(0, 1)
under the null hypothesis?

How to test a hypothesis about β2 ?
• The argument using linear transformations does not apply, as β2 is a

non-linear function of the model parameters.
• From Box 2 (p.15): If x1t is the only variable that error-corrects, then all
information about β2 is contained in
x1t = δ + θ1 x1t−1 + θ2 x1t−2 + φ0 x2t + φ1 x2t−1 + φ2 x2t−2 + εt .
Here
φ0 + φ1 + φ2
β2 = .
1 − θ1 − θ2
a
• In that case, tβ2 =b ∼ N(0, 1).

6. Empirical Examples Based on the ADL
One-Step Approach
Empirical Example 1: Consumption, Income, and Wealth

• Estimation based on a ADL model with two lags.
EQ( 1) Modelling c by OLS
The dataset is: ConsumptionData.in7
The estimation sample is: 1973(3) - 2003(2)

c_1 0.553265 0.09655 5.73 0.0000 0.2283
c_2 0.197843 0.09534 2.08 0.0403 0.0373
Constant -0.0570216 0.09606 -0.594 0.5540 0.0032
y 0.240160 0.06411 3.75 0.0003 0.1122
y_1 -0.149070 0.07359 -2.03 0.0452 0.0356
y_2 0.0273620 0.06974 0.392 0.6956 0.0014
w 0.427883 0.1286 3.33 0.0012 0.0907
w_1 -0.265189 0.1831 -1.45 0.1503 0.0185
w_2 -0.0603298 0.1328 -0.454 0.6506 0.0019
sigma 0.0154533 RSS 0.0265072643

Rˆ2 0.983383 F(8,111) = 821.1 [0.000]**
Adj.Rˆ2 0.982186 log-likelihood 334.797
mean(c) 6.13834 se(c) 0.115781
AR 1-5 test: F(5,106) = 0.91098 [0.4769]

ARCH 1-4 test: F(4,112) = 1.3514 [0.2554]
Normality test: Chiˆ2(2) = 13.943 [0.0009]**
Hetero test: F(16,103) = 1.9555 [0.0232]*

• The PcGive test for no-cointegration is given by (5% critical value -3.51):
PcGive Unit-root t-test: -3.5295
• The long-run solution is given by:

Solved static long-run equation for c
Coefficient Std.Error t-value t-prob
Constant -0.229103 0.3718 -0.616 0.5390
y 0.475918 0.1534 3.10 0.0024
w 0.411280 0.1373 3.00 0.0033
Long-run sigma = 0.0620885
ECM = c + 0.229103 - 0.475918*y - 0.41128*w;

WALD test: Chiˆ2(2) = 382.791 [0.0000] **

• Plot of the estimated cointegration relation:
ECM
0.05
0.00
-0.05
1970 1975 1980 1985 1990 1995 2000 2005
ECM = c + 0.229103 - 0.475918*y - 0.41128*w;


• Plot of the dynamic multipliers, ∂c∂yt+k for k ≥ 0, and the accumulated
t
dynamic multipliers. Note, PcGive scales the multipliers by the long-run
multipliers given by the cointegration coefficients.
1.0
y y(cum)
0.4
0.8
0.2
0.6
0.0
0 10 20 30 40 0 10 20 30 40
1.0 w w(cum)
1.025
0.5
1.000
0.0 0.975
0 10 20 30 40 0 10 20 30 40


• Estimation based on a linear ECM model.
EQ( 2) Modelling dc by OLS
The dataset is: ConsumptionData.in7
The estimation sample is: 1973(3) - 2003(2)

dc_1 -0.197843 0.09534 -2.08 0.0403 0.0373
Constant -0.0570216 0.09606 -0.594 0.5540 0.0032
dy 0.240160 0.06411 3.75 0.0003 0.1122
dy_1 -0.0273620 0.06974 -0.392 0.6956 0.0014
dw 0.427883 0.1286 3.33 0.0012 0.0907
dw_1 0.0603298 0.1328 0.454 0.6506 0.0019
c_1 -0.248891 0.07052 -3.53 0.0006 0.1009
y_1 0.118452 0.04795 2.47 0.0150 0.0521
w_1 0.102364 0.04774 2.14 0.0342 0.0398
sigma 0.0154533 RSS 0.0265072643

Rˆ2 0.367368 F(8,111) = 8.057 [0.000]**
Adj.Rˆ2 0.321773 log-likelihood 334.797
mean(dc) 0.00310834 se(dc) 0.0187643
AR 1-5 test: F(5,106) = 0.91098 [0.4769]

ARCH 1-4 test: F(4,112) = 1.3514 [0.2554]
Normality test: Chiˆ2(2) = 13.943 [0.0009]**
Hetero test: F(16,103) = 2.5544 [0.0023]**

Empirical Example 2: Interest Rates Revisited
• Estimation based on a ADL model. The significant terms are:
Modelling r by OLS (using DanishBondRate2003.in7)

The estimation sample is: 1973 (4) to 2003 (2)

r_1 0.615152 0.07909 7.78 0.000 0.3447
Constant -0.00250456 0.004573 -0.548 0.585 0.0026
b 1.19928 0.2347 5.11 0.000 0.1851
b_1 -0.865763 0.2648 -3.27 0.001 0.0851
sigma 0.0182398 RSS 0.0382594939

Rˆ2 0.841437 F(3,115) = 203.4 [0.000]**
log-likelihood 309.674 DW 2.16
mean(r) 0.092754 var(r) 0.00202764

Empirical Example 2: Interest Rates Revisited
• The test for no-cointegration is given by (5% critical value −3.21):

PcGive Unit-root t-test: -4.8661
(Obtained via: Test→Dynamic Analysis→Lag Structure.)
• The long-run solution is given in PcGive as
Solved static long-run equation for r
Coefficient Std.Error t-value t-prob
Constant -0.00650791 0.01184 -0.550 0.584
b 0.866611 0.09491 9.13 0.000
Long-run sigma = 0.0473948
(Obtained via: Test→Dynamic Analysis→Static long-run solution.
Note: PcGive assumes that the estimated model is and ADL.)
• Here the t−values can be used for testing! β2 is not significantly different
from unity.
Moreover, note that the estimate of β2 differs from the estimate obtained
from estimating the static regression model.
• The dynamic multipliers, ∂x1t /∂x2t , ∂x1t /∂x2t−1 , . . . and the cumulated
P
∂x1t /∂x2t−i can be graphed.

7. Comparison of the Engle-Granger and
ADL Approaches
Pros and Cons of Engle-Granger vs. ADL/ECM Approach
Pros Cons
Engle-Granger Error-correction Static regression in general

of several variables. biased due to misspecification.
Multiple cointegration Inference on β not possible.

relations possible.
ADL/ECM Estimation in dynamically Only optimal (and efficient)

well-specified model. if only one variable is
Unbiased estimates of β. error-correcting.
Inference on β. Only one cointegration relation

can be identified and estimated.

Cointegration Copenhagen

Uploaded by

Copyright:

Available Formats

Cointegration Copenhagen

Uploaded by

Document Information

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

Cointegration Copenhagen

Uploaded by

Copyright:

Available Formats

university of copenhagen department of economics

Course Outline: Cointegration and Spurious Regression

Econometrics II — Cointegration and Spurious Regression — Slide 2/51

After completing this topic, you should be able to:

Econometrics II — Cointegration and Spurious Regression — Slide 3/51

Motivation: Regression with Non-Stationarity

• What happens to the properties of OLS if variables are non-stationary?

CONS Danish private consumption in 1995 prices.

And consider a static regression model

We would expect (or hope) to get βb2 ≈ 0 and R 2 ≈ 0.

log(CONSt ) = 12.145 + 0.095 · log(BIRDt ) + ut ,

Econometrics II — Cointegration and Spurious Regression — Slide 5/51

• The variables are non-stationary. The residual, ut , is non-stationary and

Consumption and breeding birds, logs Regression residuals (cons on birds)

Econometrics II — Cointegration and Spurious Regression — Slide 6/51

Simulation: Stationary Case

IID Case, estimates IID Case, t-ratios

-0.50 -0.25 0.00 0.25 0.50 -4 -2 0 2 4

Econometrics II — Cointegration and Spurious Regression — Slide 7/51

Simulation: I(1) Spurious Regression

0.25 0.025 Note the scale

-3 -2 -1 0 1 2 3 -75 -50 -25 0 25 50 75

Econometrics II — Cointegration and Spurious Regression — 5Slide

x1t = τ1t + initial value + stationary process

where τ1t and τ2t are stochastic trends.

Econometrics II — Cointegration and Spurious Regression — Slide 10/51

x1t = µ + β2 x2t + ut . (♦)

If ut is I(0) (mean zero) then β = ( 1 −β2 )0 is a cointegrating vector.

Econometrics II — Cointegration and Spurious Regression — Slide 11/51

3 The cointegrating vector is only unique up to a constant factor:

This corresponds to different variables on the left hand side of (♦)

4 Cointegration is easily extended to more variables.

zt = β 0 xt = x1t − β2 · x2t − ... − βp · xpt ∼ I(0).

Econometrics II — Cointegration and Spurious Regression — Slide 12/51

Cointegration and Economic Equilibrium

x1t = µ + β2 x2t + ut . (∗)

The term, ut , is interpretable as the deviation from the relation in (∗).

• If x1t and x2t cointegrate, then the deviation

is a stationary process with mean zero.

• If x1t and x2t do not cointegrate, then the deviation ut is I(1).

Econometrics II — Cointegration and Spurious Regression — Slide 13/51

Empirical Example: Consumption and Income

Consumption and income, logs Income ratio, l o g (C t )lo g (Y t )

1970 1980 1990 2000 1970 1980 1990 2000

How is the Equilibrium Sustained?

• There must be forces pulling x1t or x2t towards the equilibrium.

Engle and Granger’s theorem (1987)

Econometrics II — Cointegration and Spurious Regression — Slide 15/51

• Cointegration is a system property. Both variables could error correct, e.g.:

∆x1t = δ1 + Γ11 ∆x1t−1 + Γ12 ∆x2t−1 + α1 (x1t−1 − β2 x2t−1 ) + 1t

• Note that β 0 xt−1 = x1t−1 − β2 x2t−1 appears in both equations.

• For x1t to error correct we need α1 < 0.

Econometrics II — Cointegration and Spurious Regression — Slide 16/51

(A) Two cointegrated variables (B) Deviation from equilibrium

(C) Speed of adjustment (D) Cross-plot

Econometrics II — Cointegration and Spurious Regression — Slide 17/51

Static Regression with Cointegrated Series

• In the cointegration case there exists a β2 so that the error term, ut , in

x1t = µ + β2 x2t + ut . (∗∗)

• OLS applied to (∗∗) gives consistent results, so that βb2 → β2 for T → ∞.

∆x1t = δ1 + Γ11 ∆x1t−1 + Γ12 ∆x2t−1 + α1 (x1t−1 − β2 x2t−1 ) + 1t

∆x1t = δ + λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + α · ecmt−1 + t ,

x1t = δ + θ1 x1t−1 + θ2 x1t−2 + φ0 x2t + φ1 x2t−1 + φ2 x2t−2 + t ,

where t is IID. This is equivalent to a linear regression error correction

∆x1t = δ + λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + γ1 x1t−1 + γ2 x2t−1 + t ,

∆x1t = λ1 ∆x1t−1 + κ0 ∆x2t + κ1 ∆x2t−1 + α (x1t−1 − µ − β2 x2t−1 ) + t .