Unit 4a: Introduction and Unit Root Testing

Financial Analytics

Ioannis Vrontos
Athens University of Economics and Business

MSc in Statistics, Department of Statistics

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Unit 4a: Introduction and Unit Root Testing

Outline

This chapter provides theory and practice of time series analysis

▶ Presents deterministic time series models
▶ Presents the basic theory of stationary/non-stationary
processes - unit root testing
▶ Describes and presents analytically AR(I)MA models and the
Box-Jenkins methodology
▶ Introduces the class of conditional heteroscedastic models
(ARCH/GARCH)
▶ Presents time series forecasting techniques
▶ Illustrative examples applying time series models/techniques
to actual economic and financial data

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Unit 4a: Introduction and Unit Root Testing

▶ Introduction: modeling approaches

▶ Basic concepts: Autocorrelation and stationarity
▶ Properties of stationary and non-stationary processes
▶ Unit root testing: Augmented Dickey-Fuller test
▶ Illustration of unit root testing using Eviews-R to economic
and financial data sets
▶ Example 1: unit root testing to financial time series, e.g.
stocks and indices (application and useful conclusions)
▶ Example 2: Unit root testing to exchange rate series
(application and useful conclusions)

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Introduction: Data

▶ Types of data
▶ Time series data, yt , t = 1, . . . , T , is a sequence of random
variables taking values at specific time periods (daily, weekly,
monthly, etc.)
▶ Cross-sectional data, yi , i = 1, . . . , N refer one or more
characteristics (variables) collected at the same point in time
▶ Pooled data/panel data/longitudinal data, yit , i = 1, . . . , N
and t = 1, . . . , T refer measures of one or more characteristics
collected at specific time periods (weekly, monthly, yearly, etc.)

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Introduction: Aim of the analysis

▶ Study the characteristics of the data

▶ in order to find the relationship between the variables or the
values of the time series
▶ with the aim to construct appropriate models
▶ to examine the theory (economic, financial)
▶ and to create predictions/forecasts

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Introduction: Modeling Approaches

▶ Regression-type models: models that use explanatory

variables, based on the economic/financial theory, or the
problem at hand
▶ Time series models: models that use the behavior -
characteristics of the series under consideration at previous
time periods
▶ Regression models with time series components
▶ Univariate models
▶ Multivariate models
▶ In this course we will focus on constructing univariate models

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Introduction: Regression Models

Use explanatory variables, based on the economic - financial
theory, or the problem at hand
Asset Pricing: built models with the aim to identify important
explanatory variables (risk factors) that explain financial series

yt = α + β1 x1,t + β2 x2,t + . . . + βk xk,t + εt

Return Predictability: built models with the aim to identify

important predictive variables that have the ability to forecast
financial returns

yt = α + β1 x1,t−1 + β2 x2,t−1 + . . . + βk xk,t−1 + εt

Assume (a) uncorrelated errors, (b) constant variance -

homoscedastic errors, (c)normal errors
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Introduction: Regression Models

When the assumptions of the errors are violated:

▶ Estimation of model parameters is valid [e.g. least squares]
▶ Statistical inference, which is theoretically based on the above
assumptions is not valid [e.g. hypothesis testing]
Consequences
▶ we can not identify accurately which risk factors are really
important to explain financial returns and to predict future
returns [model selection problem]
▶ we can not accurately estimate the constant, that is the alpha
in the regression model, which is a measure of the
performance evaluation or a measure of the skill of a manager

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Introduction: Time Series models

Use lagged data points or/and lagged errors

Autoregressive models [AR(p)]

yt = δ + ϕ1 yt−1 + ϕ2 yt−2 + . . . + ϕp yt−p + εt

Moving Average models [MA(q)]

yt = µ + θ1 εt−1 + θ2 εt−2 + . . . + θq εt−q + εt

Autoregressive Moving Average models [ARMA(p,q)]

yt = δ + ϕ1 yt−1 + . . . + ϕp yt−p + θ1 εt−1 + . . . + θq εt−q + εt

Assume (a)uncorrelated errors, (b)constant variance -

homoscedastic errors, (c)normal errors

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Introduction: Regression - time series models

Models that use explanatory variables and time series components

[due to autocorrelated regression errors]

yt = α + β1 x1,t + β2 x2,t + . . . + βk xk,t + ut

ut = δ + ϕ1 ut−1 + θ1 εt−1 + εt

εt ∼ N(0, σ 2 )

Account for autocorrelated errors, assume homoscedastic and

normal errors

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Introduction: Regression - time series - volatility models

Models that use explanatory variables, time series components [due
to autocorrelated regression errors] and volatility models [due to
heteroscedastic errors]
yt = α + β1 x1,t + β2 x2,t + . . . + βk xk,t + ut

ut = δ + ϕ1 ut−1 + θ1 εt−1 + εt

εt ∼ N(0, σt2 )

σt2 = α0 + α1 ε2t−1 + α2 σt−1

2

Account for (a)autocorrelated errors, (b)heteroscedasticity, i.e.

volatility clustering, fat tails, excess kurtosis
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Basic concepts: Stationarity

▶ Strictly Stationary process: the joint distribution of

(yi , yi+1 , . . . , yi+k ) and (yi+m , yi+m+1 , . . . , yi+m+k ) are the
same for all i, k, m
▶ Weakly Stationary process: the mean, the variance and the
autocovariance do not depend on time t
In other words, a process is said to be weakly stationary if:

E (yt ) = µ , for all t

V (yt ) = E (yt − µ)2 = σ 2 , for all t

γk = Cov (yt , yt−k ) = E [(yt − µ)(yt−k − µ)] , for all t and any k

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Basic concepts: Autocorrelation

Autocorrelation shows the interdependence - correlation between
different values of the series
Cov (yt ,yt−k ) γk
ρk = Corr (yt , yt−k ) = σyt σyt−k = γ0

E [(yt −µ)(yt−k −µ)]

ρk = √ √
E (yt −µ)2 E (yt−k −µ)2

Properties of autocorrelation:
ρk = ρ−k
−1 ≤ ρk ≤ 1

Sample estimate of autocorrelation:

PT −k
t=1 (yt −ȳ )(yt−k −ȳ )
ρˆk = PT 2
t=1 (yt −ȳ )

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Basic concepts: Test for autocorrelation I

Bartlett’s test (for a particular lag k):

H0 : ρk = 0
H1 : ρk ̸= 0
If the time series is random (white noise), then the sampling
distribution of ρˆk is approximately normal, i.e. ρˆk ∼ N(0, T1 )
ρˆk −0
test statistic: Z = √ ∼ N(0, 1)
1/T

Reject H0 , at level of significance α, if the observed value of the

test statistic Z < −Z1−α/2 or Z > Z1−α/2
100(1 − α)% pConfidence interval p
for ρk :
(ρˆk − Z1−α/2 1/T , ρˆk + Z1−α/2 1/T )

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Basic concepts: Test for all autocorrelations II

H0 : ρ1 = ρ2 = . . . = ρm = 0 , for a fixed m
H1 : ρi ̸= 0 , for at least one i ≤ m

Box-Pierce test statistic: Q = T m 2 2

P
k=1 ρˆk ∼ χm
Pm ρˆk 2
Ljung-Box test statistic: LB = T (T + 2) k=1 T −k ∼ χ2m

Ljung-Box test has better small sample properties

Reject H0 , at level of significance α, if the observed value of the

test statistic Q > χ2m,1−α (LB > χ2m,1−α )

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Understanding stationarity

Consider a time series yt , and assume an AR(1) model of the form:

yt = µ + ρyt−1 + ϵt , where ϵt are uncorrelated with mean zero and
variance σ 2 .
t = 1: y1 = µ + ρy0 + ϵ1
t = 2:
y2 = µ+ρy1 +ϵ2 = µ+ρ(µ+ρy0 +ϵ1 )+ϵ2 = µ+ρµ+ρ2 y0 +ρϵ1 +ϵ2
t = 3: y3 = µ + ρµ + ρ2 µ + ρ3 y0 + ρ2 ϵ1 + ρϵ2 + ϵ3
...
t = t:
yt = µ + ρµ + ρ2 µ + . . . + ρt−1 µ + ρt y0 + ρt−1 ϵ1 + ρt−2 ϵ2 + . . . + ϵt
yt = ρt y0 + µ t−1
P s
Pt t−s ϵ
s=0 ρ + s=1 ρ s

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Understanding stationarity
ϵt are the shocks at time t. Parameter ρ shows if the shocks are
permanent or temporary. Assume that at time t = 1 the shock is
ϵ1 . Which is the effect of ϵ1 on the series yt ?
∂yt
The effect can be given by computing: ∂ϵ1 = ρt−1
∂y1
t = 1: ∂ϵ1 = ρ1−1 = ρ0 = 1
∂y2
t = 2: ∂ϵ1 = ρ2−1 = ρ
∂y3
t = 3: ∂ϵ1 = ρ3−1 = ρ2 . . .

if |ρ| < 1, then ∂y

∂ϵ1 → 0, as t → ∞ : not permanent shocks, i.e.
t

the effect of ϵ1 vanish after some time period

if ρ = 1, then ∂yt
∂ϵ1 = 1 : permanent shocks, i.e. the random event
at time t = 1, ϵ1 , affect the series yt for a long time period
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Non-stationary process I: Random walk with drift

For ρ = 1 i.e. when the shocks are permanent, the model takes the
form: yt = µ + yt−1 + ϵt [Random walk with drift]
We will write the model in an equivalent form:
t = 1: y1 = µ + y0 + ϵ1
t = 2:
y2 = µ + y1 + ϵ2 = µ + (µ + y0 + ϵ1 ) + ϵ2 = µ + µ + y0 + ϵ1 + ϵ2
t = 3: y3 = µ + y2 + ϵ3 = µ + µ + µ + y0 + ϵ1 + ϵ2 + ϵ3
...
t = t: yt = tµ + y0 + ts=1 ϵs
P

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Non-stationary process I: Random walk with drift

Pt
Random walk with drift: yt = µ + yt−1 + ϵt = tµ + y0 + s=1 ϵs

We will prove that in this case, the series yt is a non-stationary

process. For simplicity assume that y0 = 0.

E (yt ) = E (tµ + y0 + ts=1 ϵs ) = E (tµ) + E (y0 ) + E ( ts=1 ϵs ) = tµ

P P

V (yt ) = V (tµ + y0 + ts=1 ϵs ) = V (tµ) + V (y0 ) + V ( ts=1 ϵs ) =

P P
tσ 2

γk = Cov (yt , yt−k ) = E [(yt − E (yt ))(yt−k − E (yt−k ))]

= E [(yt − tµ)(yt−k − (t − k)µ)] = E [(y0 + ts=1 ϵs )(y0 + t−k

P P
s=1 ϵs )]
Pt Pt−k
= E [( s=1 ϵs )( s=1 ϵs )] = (t − k)σ 2

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Non-stationary process I: Random walk with drift

Therefore, the Random walk with drift model: yt = µ + yt−1 + ϵt

▶ is a non-stationary process
▶ has permanent shocks
▶ its mean is not constant over time, E (Yt ) = tµ, i.e. has a
linear trend
▶ its variance is not constant over time, V (yt ) = tσ 2 , i.e.
increases over time
▶ its covariance i.e. the way the lagged values affect future
values change over time

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Non-stationary process II: Random walk without drift

Consider a time series yt and assume a model of the form:
yt = ρyt−1 + ϵt , where ϵt are uncorrelated with mean zero and
variance σ 2 .
For ρ = 1 i.e. when the shocks are permanent, the model takes the
form: yt = yt−1 + ϵt [Random walk without drift]
We will write the model in an equivalent form:
t = 1: y1 = y0 + ϵ1
t = 2: y2 = y1 + ϵ2 = (y0 + ϵ1 ) + ϵ2 = y0 + ϵ1 + ϵ2
t = 3: y3 = y2 + ϵ3 = y0 + ϵ1 + ϵ2 + ϵ3
...
t = t: yt = y0 + ts=1 ϵs
P

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Non-stationary process II: Random walk without drift

Pt
Random walk without drift: yt = yt−1 + ϵt = y0 + s=1 ϵs

We will prove that the series yt is a non-stationary process. For

simplicity assume that y0 = 0.

E (yt ) = E (y0 + ts=1 ϵs ) = E (y0 ) + E ( ts=1 ϵs ) = 0

P P

V (yt ) = V (y0 + ts=1 ϵs ) = V (y0 ) + V ( ts=1 ϵs ) = tσ 2

P P

γk = Cov (yt , yt−k ) = E [(yt − E (yt ))(yt−k − E (yt−k ))]

= E [yt yt−k ] = E [(y0 + ts=1 ϵs )(y0 + t−k

P P
s=1 ϵs )]

= E [( ts=1 ϵs )( t−k 2
P P
s=1 ϵs )] = (t − k)σ

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Non-stationary process II: Random walk without drift

Therefore, the Random walk without drift model: yt = yt−1 + ϵt

▶ is a non-stationary process
▶ has permanent shocks
▶ its mean is constant through time, E (Yt ) = 0, i.e. yt moves
around zero
▶ its variance is not constant over time, V (yt ) = tσ 2 , i.e.
increases over time
▶ its covariance i.e. the way the lagged values affect future
values change over time

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Stationarity through Differencing I

Consider a non-stationary process yt which follows a Random walk
model with drift, i.e. yt = µ + yt−1 + ϵt
By subtracting yt−1 we obtain:
yt = µ + yt−1 + ϵt ⇒ yt − yt−1 = µ + yt−1 + ϵt − yt−1 ⇒
Zt = ∆yt = µ + ϵt
E (Zt ) = E (∆yt ) = E (µ + ϵt ) = E (µ) + E (ϵt ) = µ
V (Zt ) = V (∆yt ) = V (µ + ϵt ) = V (µ) + V (ϵt ) = σ 2
γk = Cov (Zt , Zt−k ) = E [(Zt − E (Zt ))(Zt−k − E (Zt−k ))]
= E [(Zt − µ)(Zt−k − µ)] = E [ϵt ϵt−k ] = 0
That is Zt = ∆yt is a stationary process.
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Stationarity through Differencing II

Consider a non-stationary process yt which follows a Random walk

model without drift, i.e. yt = yt−1 + ϵt
By subtracting yt−1 we obtain:

yt = yt−1 + ϵt ⇒ yt − yt−1 = ϵt ⇒ Zt = ∆yt = ϵt

E (Zt ) = E (∆yt ) = E (ϵt ) = 0

V (Zt ) = V (∆yt ) = V (ϵt ) = σ 2

γk = Cov (Zt , Zt−k ) = E [(Zt − E (Zt ))(Zt−k − E (Zt−k ))]

= E [Zt Zt−k ] = E [ϵt ϵt−k ] = 0

That is Zt = ∆yt is a stationary process

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Stationarity through Differencing: Definitions

Consider a non-stationary process yt

If ∆yt = yt − yt−1 is a stationary process, then yt is called

Integrated of order one [I (1)]

Generally, if yt is non-stationary, and by taking iteratively d

differences, the yt becomes stationary, then yt is called I (d)

If yt is stationary, then it is I (0) process

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Stationary process: The AR(1) model

Consider a time series yt and assume an AR(1) model of the form:
yt = µ + ρyt−1 + ϵt , where ϵt are uncorrelated with mean zero and
variance σ 2 .
Recall that for |ρ| < 1, the shocks are not permanent and the the
effect of ϵ1 or generally of ϵt−i vanish after some time period.
Mean, variance and autocovariance at lag k of yt :
µ
E (yt ) = 1−ρ
σ2
V (yt ) = 1−ρ2
σ 2
γk = Cov (yt , yt−k ) = ρk γ0 = ρk 1−ρ 2

Observe that these characteristics are constant over time, i.e. the
process is stationary.
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Stationary process: The AR(1) model without constant

Consider a time series yt and assume an AR(1) model of the form:
yt = ρyt−1 + ϵt , where ϵt are uncorrelated with mean zero and
variance σ 2 .
Recall that for |ρ| < 1, the shocks are not permanent and the the
effect of ϵ1 or generally of ϵt−i vanish after some time period.
Mean, variance and autocovariance at lag k of yt :
E (yt ) = 0
σ2
V (yt ) = 1−ρ2
σ 2
γk = Cov (yt , yt−k ) = ρk γ0 = ρk 1−ρ 2

Observe that these characteristics are constant over time, i.e. the
process is stationary.
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Unit-Root test of Stationarity: Different tests

The stationary test of interest is:
H0 : ρ = 1
H1 : |ρ| < 1 usually H1 : ρ < 1
Under H0 , the process is non-stationary, the variance of the process
increases over time, the t-test is not valid.
Different testing approaches have been proposed in the literature:
▶ Dickey - Fuller test (Augmented Dickey-Fuller)
▶ Phillips - Perron test
▶ Kwiatkowski - Phillips - Schmidt - Shin test
▶ Ng - Perron test

The main problem of the stationarity tests is that the power of the
tests is not high.
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Unit-Root test of Stationarity: Different models

The stationary test of interest is:
H0 : ρ = 1
H1 : |ρ| < 1 usually H1 : ρ < 1
Different modeling approaches have been proposed in the
literature:
▶ AR(1) model with constant: yt = µ + ρyt−1 + ϵt
▶ AR(1) model without constant: yt = ρyt−1 + ϵt
▶ AR(1) model with constant and trend:
yt = µ + ρyt−1 + γt + ϵt
▶ AR(p) model with/without constant/trend
▶ Structural break-point AR(p) models, etc.

The idea is that in order to test if a process is stationary or not,

one needs to use a model that fits the data well.
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Dickey-Fuller test - Model with constant

Model under consideration: yt = µ + ρyt−1 + ϵt
H0 : ρ = 1 [Non-stationary process: Random walk with drift]
H1 : ρ < 1 [Stationary process: AR(1) with constant]
The model can be reparametrized as follows:

yt = µ + ρyt−1 + ϵt ⇒ yt − yt−1 = µ + ρyt−1 + ϵt − yt−1 ⇒

∆yt = µ + (ρ − 1)yt−1 + ϵt ⇒

∆yt = µ + βyt−1 + ϵt , where β = ρ − 1

H0 : β = 0 [Non-stationary process]
H1 : β < 0 [Stationary process]
The reparametrized model is used, but the test examines
stationarity of the yt process, not of the ∆yt process!!!
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Dickey-Fuller test - Model with constant

▶ Similar in spirit with an one-tailed regression-type test

▶ The test statistic is of the form: βb
s.e.(β)
b
▶ Due to non-stationarity under H0 , the distribution of the test
statistic is not Student-t
▶ Dickey - Fuller have provided ’corrected’ critical values
▶ Reject H0 if the test statistic is smaller than the critical value
in the left tail of the distribution
▶ Reject H0 if the significance level α is larger than the
corresponding p-value

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Dickey-Fuller test - Model without constant

Model under consideration: yt = ρyt−1 + ϵt
H0 : ρ = 1 [Non-stationary process: Random walk without drift]
H1 : ρ < 1 [Stationary process: AR(1) without constant]
The model can be reparametrized as follows:

yt = ρyt−1 + ϵt ⇒ yt − yt−1 = ρyt−1 + ϵt − yt−1 ⇒

∆yt = (ρ − 1)yt−1 + ϵt ⇒

∆yt = βyt−1 + ϵt , where β = ρ − 1

H0 : β = 0 [Non-stationary process]
H1 : β < 0 [Stationary process]
The reparametrized model is used, but the test examines
stationarity of the yt process, not of the ∆yt process!!!
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Dickey-Fuller test - Model without constant

▶ Similar in spirit with an one-tailed regression-type test

▶ The test statistic is of the form: βb
s.e.(β)
b
▶ Due to non-stationarity under H0 , the distribution of the test
statistic is not Student-t
▶ Dickey - Fuller have provided ’corrected’ critical values
▶ Reject H0 if the test statistic is smaller than the critical value
in the left tail of the distribution
▶ Reject H0 if the significance level α is larger than the
corresponding p-value

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Dickey-Fuller test - Model with constant and trend

Model under consideration: yt = µ + ρyt−1 + γt + ϵt
H0 : ρ = 1(γ = 0) [Non-stationary process: Stochastic trend]
H1 : ρ < 1(γ ̸= 0) [Stationary process: Deterministic trend]
The model can be reparametrized as follows:

yt = µ+ρyt−1 +γt +ϵt ⇒ yt −yt−1 = µ+ρyt−1 +γt +ϵt −yt−1 ⇒

∆yt = µ + (ρ − 1)yt−1 + γt + ϵt ⇒

∆yt = µ + βyt−1 + γt + ϵt , where β = ρ − 1

H0 : β = 0(γ = 0) [Non-stationary process]

H1 : β < 0(γ ̸= 0) [Stationary process]
The reparametrized model is used, but the test examines
stationarity of the yt process, not of the ∆yt process!!!
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Dickey-Fuller test - Model with constant and trend

Model under consideration: yt = µ + ρyt−1 + γt + ϵt
H0 : ρ = 1(γ = 0) [Non-stationary process: Stochastic trend]
H1 : ρ < 1(γ ̸= 0) [Stationary process: Deterministic trend]
The model can be reparametrized as follows:

yt = µ+ρyt−1 +γt +ϵt ⇒ yt −yt−1 = µ+ρyt−1 +γt +ϵt −yt−1 ⇒

∆yt = µ + (ρ − 1)yt−1 + γt + ϵt ⇒

∆yt = µ + βyt−1 + γt + ϵt , where β = ρ − 1

H0 : β = 0(γ = 0) [Non-stationary process]

H1 : β < 0(γ ̸= 0) [Stationary process]
The reparametrized model is used, but the test examines
stationarity of the yt process, not of the ∆yt process!!!
Ioannis Vrontos, Athens University of Economics and Business Financial Analytics
 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Dickey-Fuller test - Model without constant and trend

▶ Similar in spirit with an one-tailed regression-type test

▶ The test statistic is of the form: βb
s.e.(β)
b
▶ Due to non-stationarity under H0 , the distribution of the test
statistic is not Student-t
▶ Dickey - Fuller have provided ’corrected’ critical values
▶ Reject H0 if the test statistic is smaller than the critical value
in the left tail of the distribution
▶ Reject H0 if the significance level α is larger than the
corresponding p-value

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Augmented Dickey-Fuller test

H0 : β = 0 [Non-stationary process]
H1 : β < 0 [Stationary process]
If the errors ϵbt in the model under consideration are correlated, we
use the Augmented Dickey-Fuller test (ADF) to examine
stationarity. That is, the model takes the form:

∆yt = µ + βyt−1 + pj=1 λj ∆yt−j + ϵt

P

∆yt = βyt−1 + pj=1 λj ∆yt−j + ϵt

P

∆yt = µ + βyt−1 + γt + pj=1 λj ∆yt−j + ϵt

P

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

Application to financial and economic series

▶ Example 1: unit root testing to Johnson & Johnson quarterly

data
▶ Example 2: unit root testing to exchange rate series
▶ Detailed empirical analysis is presented in the application files
using R script routines

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics

 Introduction - Modeling Approaches
Basic concepts
Unit 4a: Introduction and Unit Root Testing
Characteristics of stationary/non-stationary processes
Unit Root Testing

References

Hamilton, James D. Time Series Analysis. Princeton, New Jersey: Princeton University Press, 1994.
Enders, Walter. Applied Econometric Time Series. New York: Wiley, 2010.
Cowpertwait, Paul S.P., and Metcalfe V. Andrew. Introductory Time Series with R. New York: Springer Texts in
Statistics, 2009.
Cryer, Jonathan D., and Chan Kung-Sik. Time Series Analysis with Applications in R. Springer Texts in Statistics,
2010.
Gujarati, Damodar N. Basic Econometrics. New York: McGraw-Hill, 2008.
Harvey, Andrew. Time Series Models. Cambridge: MIT Press, 1993.
Hendry, David F. Dynamic Econometrics. Oxford: Oxford University Press, 1995.
Shumway, Robert H. and David S. Stoffer. Time Series Analysis and Its Applications with R Examples. New York:
Springer Texts in Statistics, 2011.

Ioannis Vrontos, Athens University of Economics and Business Financial Analytics