You are on page 1of 53

Chapter 12

Regression with Time-Series


Data:
Nonstationary Variables

Walter R. Paczkowski
Rutgers University
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 1

12.3
Unit Root Tests for Stationarity

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 2

12.3
Unit Root Tests for
Stationarity

There are many tests for determining whether a


series is stationary or nonstationary
The most popular is the DickeyFuller test

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 3

12.3
Unit Root Tests for
Stationarity

12.3.1
Dickey-Fuller Test 1
(No constant and No
Trend)

The AR(1) process yt = yt-1 + vt is stationary when


|| < 1
But, when = 1, it becomes the nonstationary
random walk process
We want to test whether is equal to one or
significantly less than one
Tests for this purpose are known as unit root
tests for stationarity

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 4

12.3
Unit Root Tests for
Stationarity

12.3.1
Dickey-Fuller Test 1
(No constant and No
Trend)

Consider again the AR(1) model:


yt yt 1 vt

Eq. 12.4

We can test for nonstationarity by testing the


null hypothesis that = 1 against the alternative
that || < 1
Or simply < 1

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 5

12.3
Unit Root Tests for
Stationarity

12.3.1
Dickey-Fuller Test 1
(No constant and No
Trend)

A more convenient form is:


yt yt 1 yt 1 yt 1 vt
yt 1 yt 1 vt

Eq. 12.5a

yt 1 vt

The hypotheses are:

Principles of Econometrics, 4th


Edition

H0 : 1

H0 : 0

H1 : 1

H1 : 0

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 6

12.3
Unit Root Tests for
Stationarity

12.3.2
Dickey-Fuller Test 2
(With Constant but
No Trend)

The second DickeyFuller test includes a constant


term in the test equation:
yt yt 1 vt

Eq. 12.5b

The null and alternative hypotheses are the


same as before

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 7

12.3
Unit Root Tests for
Stationarity

12.3.3
Dickey-Fuller Test 3
(With Constant and
With Trend)

The third DickeyFuller test includes a constant


and a trend in the test equation:
yt yt 1 t vt

Eq. 12.5c

The null and alternative hypotheses are


H0: = 0 and H1: < 0

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 8

12.3
Unit Root Tests for
Stationarity

12.3.4
The Dickey-Fuller
Critical Values

To test the hypothesis in all three cases, we simply


estimate the test equation by least squares and
examine the t-statistic for the hypothesis that
=0
Unfortunately this t-statistic no longer has the
t-distribution
Instead, we use the statistic often called a
(tau) statistic

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 9

12.3
Unit Root Tests for
Stationarity

Table 12.2 Critical Values for the DickeyFuller Test

12.3.4
The Dickey-Fuller
Critical Values

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 10

12.3
Unit Root Tests for
Stationarity

12.3.4
The Dickey-Fuller
Critical Values

An important extension of the DickeyFuller test


allows for the possibility that the error term is
autocorrelated
Consider the model:
m

Eq. 12.6

where

yt yt 1 as yt s vt
s 1

yt 1 yt 1 yt 2 , yt 2 yt 2 yt 3 , K

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 11

12.3
Unit Root Tests for
Stationarity

12.3.4
The Dickey-Fuller
Critical Values

The unit root tests based on Eq. 12.6 and its


variants (intercept excluded or trend included) are
referred to as augmented DickeyFuller tests
When = 0, in addition to saying that the series
is nonstationary, we also say the series has a
unit root
In practice, we always use the augmented
DickeyFuller test

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 12

12.3
Unit Root Tests for
Stationarity

Table 12.3 AR processes and the Dickey-Fuller Tests

12.3.5
The Dickey-Fuller
Testing Procedures

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 13

12.3
Unit Root Tests for
Stationarity

12.3.5
The Dickey-Fuller
Testing Procedures

The Dickey-Fuller testing procedure:


First plot the time series of the variable and select a
suitable Dickey-Fuller test based on a visual
inspection of the plot
If the series appears to be wandering or fluctuating
around a sample average of zero, use test equation
(12.5a)
If the series appears to be wandering or fluctuating
around a sample average which is nonzero, use test
equation (12.5b)
If the series appears to be wandering or fluctuating
around a linear trend, use test equation (12.5c)

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 14

12.3
Unit Root Tests for
Stationarity

12.3.5
The Dickey-Fuller
Testing Procedures

The Dickey-Fuller testing procedure (Continued):


Second, proceed with one of the unit root tests
described in Sections 12.3.1 to 12.3.3

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 15

12.1
Stationary and
Nonstationary
Variables

Principles of Econometrics, 4th


Edition

FIGURE 12.1 U.S. economic time series

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 16

12.3
Unit Root Tests for
Stationarity

12.3.6
The Dickey-Fuller
Tests: An Example

As an example, consider the two interest rate


series:
The federal funds rate (Ft)
The three-year bond rate (Bt)
Following procedures described in Sections 9.3
and 9.4, we find that the inclusion of one lagged
difference term is sufficient to eliminate
autocorrelation in the residuals in both cases

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 17

12.3
Unit Root Tests for
Stationarity

12.3.6
The Dickey-Fuller
Tests: An Example

The results from estimating the resulting equations


are:
F 0.173 0.045 F 0.561F

t
t 1
t 1
(tau )

( 2.505)

B 0.237 0.056 B 0.237B

t
t 1
t 1
(tau )

( 2.703)

The 5% critical value for tau (c) is -2.86


Since -2.505 > -2.86, we do not reject the null
hypothesis
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 18

12.3
Unit Root Tests for
Stationarity

12.3.7
Order of Integration

Recall that if yt follows a random walk, then = 0


and the first difference of yt becomes:
yt yt yt 1 vt

Series like yt, which can be made stationary by


taking the first difference, are said to be
integrated of order one, and denoted as I(1)
Stationary series are said to be integrated of
order zero, I(0)
In general, the order of integration of a series is
the minimum number of times it must be
differenced to make it stationary
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 19

12.3
Unit Root Tests for
Stationarity

12.3.7
Order of Integration

The results of the DickeyFuller test for a random


walk applied to the first differences are:
F 0.447 F

t
t 1
(tau )

( 5.487)

B 0.701 B

t
t 1
(tau )
Principles of Econometrics, 4th
Edition

( 7.662)

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 20

12.3
Unit Root Tests for
Stationarity

12.3.7
Order of Integration

Based on the large negative value of the tau


statistic (-5.487 < -1.94), we reject the null
hypothesis that Ft is nonstationary and accept the
alternative that it is stationary
We similarly conclude that Bt is stationary
(-7:662 < -1:94)

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 21

12.3
Unit Root Tests for
Stationarity

12.3.4
The Dickey-Fuller
Critical Values

To carry out a one-tail test of significance, if c is


the critical value obtained from Table 12.2, we
reject the null hypothesis of nonstationarity if
c
If > c then we do not reject the null
hypothesis that the series is nonstationary
Ex Nelson Plosser Data

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 22

12.2
Spurious Regressions

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 23

12.2
Spurious
Regressions

The main reason why it is important to know


whether a time series is stationary or nonstationary
before one embarks on a regression analysis is that
there is a danger of obtaining apparently
significant regression results from unrelated data
when nonstationary series are used in regression
analysis
Such regressions are said to be spurious

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 24

12.2
Spurious
Regressions

Consider two independent random walks:


rw1 : yt yt 1 v1t
rw2 : xt xt 1 v2t

These series were generated independently and,


in truth, have no relation to one another
Yet when plotted we see a positive relationship
between them

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 25

12.2
Spurious
Regressions

FIGURE 12.3 Time series and scatter plot of two random walk variables

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 26

12.2
Spurious
Regressions

FIGURE 12.3 (Continued) Time series and scatter plot of two random
walk variables

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 27

12.2
Spurious
Regressions

A simple regression of series one (rw1) on series


two (rw2) yields:
17.818 0.842 rw ,
rw
1t
2t
(t )

R 2 0.70

(40.837)

These results are completely meaningless, or


spurious
The apparent significance of the relationship
is false
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 28

12.2
Spurious
Regressions

When nonstationary time series are used in a


regression model, the results may spuriously
indicate a significant relationship when there is
none
In these cases the least squares estimator and
least squares predictor do not have their usual
properties, and t-statistics are not reliable
Since many macroeconomic time series are
nonstationary, it is particularly important to
take care when estimating regressions with
macroeconomic variables
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 29

12.4
Cointegration

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 30

12.4
Cointegration

As a general rule, nonstationary time-series


variables should not be used in regression models
to avoid the problem of spurious regression
There is an exception to this rule

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 31

12.4
Cointegration

There is an important case when


et = yt - 1 - 2xt is a stationary I(0) process
In this case yt and xt are said to be cointegrated
Cointegration implies that yt and xt share
similar stochastic trends, and, since the
difference et is stationary, they never diverge
too far from each other

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 32

12.4
Cointegration

Eq. 12.7

The test for stationarity of the residuals is based on


the test equation:
et et 1 vt
The null hypothesis is that the gamma is zero i.e.
the residual are non stationary i.e. no cointegration. The alternative is cointegration.
The regression has no constant term because the
mean of the regression residuals is zero.
We are basing this test upon estimated values of
the residuals

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 33

12.4
Cointegration

Principles of Econometrics, 4th


Edition

Table 12.4 Critical Values for the Cointegration Test

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 34

12.4
Cointegration

There are three sets of critical values


Which set we use depends on whether the
residuals are derived from:
Eq. 12.8a

Equation 1: et yt bxt

Eq. 12.8b

Equation 2 : et yt b2 xt b1

Eq. 12.8c

Equation 3 : et yt b2 xt b1 t

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 35

Engle-Granger Test of Co-integration


Three Steps Procedure:
(1) Ensure that the two (or more) series to be
tested for co-integration are all I(1)
(2) Estimate the long run (co-integration
regression) in levels, store residuals
(3) Apply ADF test with one of the three sets of
EG critical values determined according to
whether the long run regression in (2) has no
intercept, intercept only or intercept and trend

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 36

12.4
Cointegration

12.4.1
An Example of a
Cointegration Test

Consider the estimated model:


Bt 1.140 0.914 Ft , R 2 0.881

Eq. 12.9

(t ) (6.548) (29.421)

The unit root test for stationarity in the


estimated residuals is:
et 0.225et 1 0.254et 1

(tau ) ( 4.196)

As -4.195 <-3.37 (note that the long run relation


includes an intercept) we reject the null
hypothesis of no co-integration. So the short
rate and bond rate are co-integrated.
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 37

12.4
Cointegration

12.4.1
An Example of a
Cointegration Test

The null and alternative hypotheses in the test for


cointegration are:
H 0 : the series are not cointegrated residuals are nonstationary
H1 : the series are cointegrated residuals are stationary

Similar to the one-tail unit root tests, we reject


the null hypothesis of no cointegration if c,
and we do not reject the null hypothesis that the
series are not cointegrated if > c
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 38

Cointegration and Long-run Equilibrium


Relationship
Although cointegration is a purely statistical
concept, it has important economic
implications.
It represents the existence of a long-run or a
steady state equilibrium relationship between
variables which is a result of economic
forces and behavior rules.
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

39

Page 39

Cointegration Relationships in Economic and Finance:

(1) Consumption-Income Relationship


Although both consumption and income exhibit a unit root,
over the long run consumption tends to be a roughly
constant proportion of income so that the difference
between log of consumption and log of income appear to
be a stationary process.
(2) Government Debt and GNP
Kremers (1989) suggested that governments are forced
politically to maintain their debt at a roughly constant
multiple of GNP, so that log(debt)-log(GNP) is stationary
even though each component individually is not stationary.
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

40

Page 40

Purchasing Power Parity

(3)
Apart from transaction cost, goods should sell for the same
effective price in two countries.
Let Pt : Price level index in the US (dollar per good)
Pt* : Index of price level in UK (pound per good)
St=Exchange rate between currencies (dollar per pund)
Then PPP states that
Pt=StPt* or taking log
pt=st+pt* (where small letter represent variables in log)
In practice error in measurement, transaction cost, and quality
differences prevent PPP from holding exactly. A weakrer
version of the hypothesis states that the variables defined by
yt pt-st-pt* is stationary even though individual elements
are all I(1).
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

41

Page 41

(4) A variable and its rational future forecast


Campbell and Shiller (1988) argue that if if y 1t is I(1) and y2t is a rational
forecast of future value of y1t , the y1t and y2t will be cointegrated. A
particular example is that stock prices is the present value of discounted
future stream of dividend so stock price and dividend should be
cointegrated.
Also if stock price is a true signal of economic activity, stock market
index should be cointegrated with macro variables e.g output. (direction
of causality from stock price to macro variables).
Spot and Forward market prices should be cointegrated since forward
rate is predictor of future spot rate. Short and Long maturity bonds
should be cointegrated.
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

42

Page 42

(5) Real Wages of Two Competing Industries


Intuition suggests that if a random shock disturbs the wage differential
between these industries, economic forces, such as labor migration
should cause real wages in these industries to return to their optimal
relationship over time. In a time series framework, wage differentials
should be a stationary time series process.

(6) Asset prices in two efficient markets cannot be cointegrated


E.g. two spot exchange rates cannot be cointegrated if they are
determined in an efficient market. If a pair of spot rates are
cointegrated, then by the Granger Representation Theorem there exists
an error correction representation, implying that at least one exchange
rate can be used to forecast (i.e. Granger-causes) the other. The ability
of one spot rate to forecast another is inconsistent with a weak form
efficiency. If forex markets are efficient spot rates should embody all
the relevant information so that it is not possible to forecast one as a
function of another.
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

43

Page 43

Market forces arising from no arbitrage conditions should ensure an


equilibrium relationship. Variables in a cointegration relationship can
each wander around over time such that they are non-stationary. As
Banerjee et al. (1993) state the divergence from a stable equilibrium
state must be stochastically bounded and, at some point diminishing
over time
But there is some economic/financial force that stops them wandering
too far from each other at any point in time.
In finance, this force holding variables is usually an arbitrage
relationship.
When deviations from the arbitrage relationship exist but are not
profitable to exploit the variables may act approximately
independently
But once they hit the point when arbitrage is profitable, they are forced
to behave in certain ways to remove the profit opportunity.

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

44

Page 44

Example: PPP Test between US$ and British


Aggregated over all goods, the law of one price
yields the purchasing power parity: Pt = St.Pt*
It is known that transportation costs, tariffs and
non-tariff barriers will entail market segmentation
and create a wedge among price across countries.
However, if these factors remain constant over
time, PPP can be restated, using a positive
constant A:
Pt = AStPt* which after rearranging in log testable
form
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 45

Unit Root Testing of st and (p-p*)t


Step 1: Unit Root Testing
N=603 obs

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 46

Unit Root in Residuals rejected (with EG critical values)


at 10%. Some evidence in favour of PPP

Step 2: Run the long run Regression in Levels


Save residuals ut

Step 3: Run ADF Test on Residuals


with EG critical values.
The test is significant at 10% level
h
Principles of Econometrics,
Implying 4tthat
Edition

Chapter
12: Regression
there
is a weak
evidencewith Time-Series Data:
Nonstationary Variables

Page 47

12.4
Cointegration

12.4.2
The Error Correction
Model

What is an ECM? Its benefit (Granger


Representation Theorem)
Consider a general model that contains lags of y
and x
yt 1 yt 1 0 xt 1 xt 1 vt
Namely, the autoregressive distributed lag
(ARDL) model, except the variables are
nonstationary:

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 48

12.4
Cointegration

12.4.2
The Error Correction
Model

If y and x are cointegrated, it means that there is a


long-run relationship between them
To derive this exact relationship, we set
yt = yt-1 = y, xt = xt-1 = x and vt = 0
Imposing this concept in the ARDL, we obtain:
y 1

This can be rewritten in the form:


y 1 2 x

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 49

12.4
Cointegration

12.4.2
The Error Correction
Model

Add the term -yt-1 to both sides of the equation:


yt yt 1 1 1 yt 1 0 xt 1 xt 1 vt

Add the term 0xt-1+ 0xt-1:


yt 1 1 yt 1 0 xt xt 1 0 1 xt 1 vt

Manipulating this we get:

0 1

yt 1 1
yt 1
xt 1 0 xt vt
1

1
1

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 50

12.4
Cointegration

12.4.2
The Error Correction
Model

Or:
yt yt 1 1 2 xt 1 0 xt vt

Eq. 12.10

This is called an error correction equation


This is a very popular model because:
It allows for an underlying or fundamental link
between variables (the long-run relationship)
It allows for short-run adjustments (i.e.
changes) between variables, including
adjustments to achieve the cointegrating
relationship
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 51

12.4
Cointegration

12.4.2
The Error Correction
Model

For the bond and federal funds rates example, we have:

Bt 0.142 Bt 1 1.429 0.777 Ft 1 0.842Ft 0.327Ft 1

2.857

9.387 3.855

The estimated residuals are

et 1 Bt 1 1.429 0.777 Ft 1

Interpretation of the error correction term: 14.2% of the short run deviation
from the long run equilibrium is corrected in one time period (here one
quarter). The t-statistic of the associated coefficient is significantly different
from zero. So the convergence towards equilibrium is statistically significant
providing further evidence in support of co-integration. Data File usa.wf
(See Iqbal and Najma Uddin (2013) for a forecast comparison of different
error correction models).
Principles of Econometrics, 4th
Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 52

12.5
Regression When
There is No
Cointegration

FIGURE 12.4 Regression with time-series data: nonstationary variables

12.5.3
Summary

Principles of Econometrics, 4th


Edition

Chapter 12: Regression with Time-Series Data:


Nonstationary Variables

Page 53

You might also like