A unit root test is a statistical test used in time series analysis to determine whether a time series

variable is non-stationary and possesses a unit root.

Key Concepts

 A stationary time series has properties (mean, variance, autocorrelation) that do not
change over time.
 A non-stationary time series has a time-dependent structure, often due to trends or
persistence.
 A unit root implies that shocks to the system have a permanent effect, and the series
does not return to a long-term mean.

Why It Matters

In econometrics and statistics, many models (like regression, ARIMA, etc.) assume stationarity.
Using non-stationary data can lead to spurious regression results, meaning misleading or
incorrect conclusions.

Common Unit Root Tests

1. Augmented Dickey-Fuller (ADF) Test

o Null Hypothesis (H₀): The time series has a unit root (non-stationary).
o Alternative Hypothesis (H₁): The time series is stationary.
2. Phillips-Perron (PP) Test
o Similar to ADF but more robust to serial correlation and heteroskedasticity.

Example

Suppose you have GDP over time:

 Run ADF test:

o If p-value > 0.05 → Fail to reject H₀ → GDP has a unit root (non-stationary).
o If p-value < 0.05 → Reject H₀ → GDP is stationary.

What to Do If Unit Root Is Present

 Difference the series: Convert it into the change from one time period to the next.
 Use models appropriate for non-stationary data (e.g., cointegration models if multiple
series have unit roots but are linked).

In time series analysis, integration refers to the number of times a non-stationary time series
must be differenced to become stationary. This concept is closely tied to the idea of a unit
root.
Definition of Integration

A time series YtY_t is said to be:

 Integrated of order 0, written as I(0), if it is stationary in its original form.

 Integrated of order 1, written as I(1), if it is non-stationary, but its first difference
ΔYt=Yt−Yt−1\Delta Y_t = Y_t - Y_{t-1} is stationary.
 Integrated of order 2, written as I(2), if you need to difference it twice to make it
stationary.

So, integration tells us how many differences are required to "remove" non-stationarity.

Example

Let's say you have a time series of stock prices:

 You run a unit root test (e.g., ADF test) on the level of the series → it’s non-stationary.
 You difference the series once and test again → it’s now stationary.

→ Then the original series is I(1).

Why Integration Matters

 If multiple series are I(1), we can test for cointegration, which implies a long-run
equilibrium relationship between them.

Choosing between the Johansen-Juselius cointegration test and the ARDL bounds testing
approach depends mainly on the stationarity properties (integration order) of your time series
variables.

� Key Differences: Johansen-Juselius vs. ARDL

ARDL (Autoregressive Distributed
Feature Johansen-Juselius (JJ)
Lag)

Data Variables can be a mix of I(0) and

All variables must be I(1)
requirement I(1)

Not allowed Cannot include I(0) or I(2) Cannot include I(2)

Tests for single cointegrating

Purpose Tests for multiple cointegration relationships
relationship
 ARDL (Autoregressive Distributed
Feature Johansen-Juselius (JJ)
Lag)

Works well even with small

Sample size Works better with large samples
samples

VAR-based system (Vector Error Correction

Model type Single equation model
Model - VECM)

Cointegration Only detects one long-run

Can detect more than one cointegrating vector
rank relationship

Ease of use More complex; requires system estimation Easier and more flexible

� When to Use Johansen-Juselius

Use Johansen-Juselius when:

 All your variables are non-stationary at level (I(1)).

 You suspect more than one cointegration relationship (multiple dependent variables).
 You are modeling a system of equations (e.g., multiple endogenous variables).
 Your sample size is reasonably large (50+ observations preferred).

Example:

You're studying the long-run relationship between GDP, investment, and exports. All are I(1),
and you're interested in their joint behavior.

� When to Use ARDL

Use ARDL when:

 Your variables are a mix of I(0) and I(1), but none are I(2).
 You have a small or medium sample size.
 You want to model the effect of one dependent variable and its relationship with one or
more independent variables.
 You need both short-run and long-run dynamics.

Example:

You're investigating the effect of inflation and interest rate on exchange rate. Inflation is I(0),
interest rate and exchange rate are I(1). Use ARDL.
� What You Must Avoid

 Don't use Johansen if variables are not all I(1).

 Don't use ARDL if any variable is I(2).

� Practical Tip: How to Choose

1. Run unit root tests (ADF, PP, or KPSS).

2. Check integration orders of all variables.
3. Based on the integration orders:
o All I(1)? → Use Johansen-Juselius.
o Mix of I(0) and I(1)? → Use ARDL.

� What is Cointegration?

Cointegration is a concept in time series analysis that refers to a long-run equilibrium

relationship between two or more non-stationary variables.

� When Does Cointegration Occur?

If you have two or more variables that are individually non-stationary (they wander over time),
but some linear combination of them is stationary, then the variables are said to be
cointegrated.

� In simple terms:
The variables may drift apart in the short run, but they move together over time in a stable
relationship.

� Example

Let’s say:

 XtX_t = income
 YtY_t = consumption

Both series are non-stationary, but you find:

Yt=α+βXt+utY_t = \alpha + \beta X_t + u_t

where utu_t is stationary → then Y and X are cointegrated.

This implies that consumption and income may fluctuate individually, but they maintain a stable
long-term proportion.
� Why It Matters

Cointegration implies:

1. A genuine long-term relationship between variables.

2. Avoids spurious regression (which happens when regressing two unrelated non-
stationary series).
3. If variables are cointegrated, their short-run deviations can be modeled using an Error
Correction Model (ECM).

� How to Test for Cointegration

1. Engle-Granger Test – For two variables.

2. Johansen Test – For more than two variables.
3. ARDL Bounds Test – For mixed I(0)/I(1) variables (single equation).

� Summary Table

Concept Description
Non-stationary Has trends or changing variance over time
Cointegration A linear combination of non-stationary variables is stationary
Implication Variables share a stable long-term relationship
Common Use Economics, finance, macro time series (GDP, interest rate, etc.)

� What is VECM (Vector Error Correction Model)?

VECM is a special form of the VAR model used when your variables are:

 Non-stationary at level (i.e., they are I(1))

 But cointegrated (they have a long-run relationship)

� In short:

VECM = VAR + Cointegration

� When to Use VECM

Use VECM when:

1. All your variables are I(1) (non-stationary in level, stationary in first difference).
2. They are cointegrated (use Johansen test to confirm).
3. You want to model both short-run and long-run relationships among variables.
� How is VECM Different from VAR?
Feature VAR VECM

Data requirement Stationary (I(0)) Non-stationary but cointegrated

Long-run dynamics Not captured Captured via error correction term

Used when No cointegration Cointegration exists

� Summary

Term Meaning

VAR Vector Autoregression (used when all variables are stationary)

VECM Vector Error Correction Model (used when variables are non-stationary but cointegrated)

"VER" Likely a typo or misunderstanding of VECM

In EViews, the cointegration test (specifically the Johansen test) is used to determine whether
two or more non-stationary time series variables have a long-run equilibrium relationship —
meaning they move together over time, even though they may drift individually in the short run.

ARDL stands for Autoregressive Distributed Lag model. It is a popular econometric model
used to analyze the relationship between a dependent variable and one or more independent
variables, particularly when the data series are a mix of I(0) (stationary) and I(1) (non-
stationary).

� Key Features of ARDL

1. Handles Mixed Integration Orders:

ARDL can be used when your variables are a mix of I(0) and I(1), but not I(2) or higher.
2. Short-Run and Long-Run Dynamics:
It allows for the simultaneous estimation of short-run dynamics and long-run
equilibrium relationships.
3. Error Correction Representation (ECM):
If the variables are cointegrated, ARDL can be reparameterized into an Error
Correction Model, which shows how quickly the system returns to equilibrium after a
shock.
� When to Use ARDL

 You suspect a long-run relationship (cointegration) between variables.

 The variables are not all stationary, but are I(0) or I(1).
 Your sample size is small or moderate (ARDL works well even with fewer observations
compared to other cointegration methods like Johansen).

� Steps in ARDL Modeling

1. Test for Unit Roots (ADF, PP, or KPSS) to confirm none of the variables are I(2).
2. Estimate ARDL Model with different lag orders.
3. Bound Test for Cointegration (Pesaran et al.):
o Null hypothesis: No long-run relationship.
o If F-stat > upper bound → cointegration exists.
4. Estimate Long-Run and Short-Run Coefficients.
5. Check Diagnostics (serial correlation, stability, etc.).

� Advantages

 Easy to use even with small samples.

 Applicable when variables have different integration orders (I(0) and I(1)).
 Provides rich dynamics for analysis.

� Caution

 Do not use ARDL if any variable is I(2) or higher.

 Results can be sensitive to lag length selection.

� What is Switching Regression?

Switching Regression (also known as Regime-Switching Regression or Threshold

Regression) is an econometric model where different equations (regimes) describe the
relationship between variables depending on the state or regime the system is in.

In a standard regression, we assume the same relationship holds at all times.

But in real life, economic behavior changes across different periods or conditions (e.g., during
a recession vs boom).

� Switching regression allows parameters (like slope or intercept) to switch depending on an

underlying regime.

� Basic Switching Regression Structure

Assume two regimes (1 and 2):

Regime 1: Yt=α1+β1Xt+ε1tif in Regime 1\text{Regime 1: } Y_t = \alpha_1 + \beta_1 X_t +

\varepsilon_{1t} \quad \text{if in Regime 1}
Regime 2: Yt=α2+β2Xt+ε2tif in Regime 2\text{Regime 2: } Y_t = \alpha_2 + \beta_2 X_t +
\varepsilon_{2t} \quad \text{if in Regime 2}

 Each regime has its own intercept, slope, and error.

 The model "switches" between them based on a rule or probability.

� Why Use It?

You use switching regression when you suspect:

 The relationship between variables changes over time.

 There are structural breaks or hidden regimes.
 The system behaves differently in low/high inflation, crisis periods, or policy shifts.

Volatility modelling refers to the statistical techniques used to model and forecast the
variability (or volatility) of a time series, usually financial data like stock returns, exchange
rates, or commodity prices.

� What Is Volatility?

 Volatility measures the degree of variation or risk in the returns of a financial asset.
 In simple terms, it’s how much prices "bounce around."
 High volatility = large, unpredictable price changes
 Low volatility = small, steady changes

� Common Volatility Models

1. ARCH (Autoregressive Conditional Heteroskedasticity)

 Proposed by Engle (1982).

 Models current volatility as a function of past squared returns:

σt2=α0+α1ϵt−12\sigma_t^2 = \alpha_0 + \alpha_1 \epsilon_{t-1}^2

2. GARCH (Generalized ARCH)

 Introduced by Bollerslev (1986).

 Adds past variances to ARCH:
 σt2=α0+α1ϵt−12+β1σt−12\sigma_t^2 = \alpha_0 + \alpha_1 \epsilon_{t-1}^2 + \beta_1
\sigma_{t-1}^2

 Very popular for modeling financial time series.

3. EGARCH, TGARCH, and GJR-GARCH

 These account for asymmetries and leverage effects (e.g., how negative news affects
volatility more than positive news).

� Interpretation

 The model doesn't predict prices—it predicts how variable the prices will be.
 You use it to understand risk dynamics, not direction.

These models are used to explain conditional heteroskedasticity — that is, situations where the
variance of a time series changes over time, often seen in financial data like stock returns.

� 1. ARCH: Autoregressive Conditional Heteroskedasticity

� Introduced by:

Engle (1982)

� Basic Idea:

 Volatility (variance) at time tt depends on the past squared errors (shocks).

 Clusters of volatility: periods of high variance tend to follow each other.

� Model:

yt=μ+εt,εt∼N(0,ht)y_t = \mu + \varepsilon_t,\quad \varepsilon_t \sim N(0, h_t) ht=α0+α1εt−12h_t =

\alpha_0 + \alpha_1 \varepsilon_{t-1}^2

 hth_t: conditional variance

 εt−12\varepsilon_{t-1}^2: past shock (squared residual)

� Key Feature: Lagged squared residuals determine today’s volatility.

� 2. GARCH: Generalized ARCH

� Introduced by:
Bollerslev (1986)

� Why GARCH?

ARCH requires many lags to capture volatility — GARCH simplifies this by adding a lagged
variance term.

� Model:

ht=α0+α1εt−12+β1ht−1h_t = \alpha_0 + \alpha_1 \varepsilon_{t-1}^2 + \beta_1 h_{t-1}

 Adds persistence in volatility (i.e., past variance matters).

 More realistic for financial time series.

� Key Feature: Combines shock effect and volatility persistence.

� 3. TGARCH: Threshold GARCH

� Introduced by:

Glosten, Jagannathan, and Runkle (1993)

� Motivation:

In reality, negative shocks increase volatility more than positive shocks of the same
magnitude (leverage effect).

� Model:

ht=α0+α1εt−12+γDt−1εt−12+β1ht−1h_t = \alpha_0 + \alpha_1 \varepsilon_{t-1}^2 + \gamma D_{t-1}

\varepsilon_{t-1}^2 + \beta_1 h_{t-1}

 Dt−1=1D_{t-1} = 1 if εt−1<0\varepsilon_{t-1} < 0, 0 otherwise.

 γ\gamma: captures the asymmetric effect of bad news.

� Key Feature: Volatility reacts differently to positive vs negative shocks.

� 4. EGARCH: Exponential GARCH

� Introduced by:

Nelson (1991)

� Motivation:
EGARCH allows:

 Asymmetry (like TGARCH)

 No need to restrict coefficients to be positive (log specification ensures ht>0h_t > 0)

� Model:

ln⁡ht=ω+βln⁡ht−1+α∣εt−1ht−1∣+γεt−1ht−1\ln h_t = \omega + \beta \ln h_{t-1} + \alpha \left|

\frac{\varepsilon_{t-1}}{\sqrt{h_{t-1}}} \right| + \gamma \frac{\varepsilon_{t-1}}{\sqrt{h_{t-1}}}

 γ\gamma: captures asymmetry

 Logarithmic variance means variance is always positive, no need for parameter
constraints

� Key Feature: Allows for asymmetric and exponential effects of shocks on volatility.

� Summary Table
Model Key Feature Volatility Depends On Asymmetry Allowed?

ARCH Basic time-varying volatility Past shocks (ε2\varepsilon^2) � No

GARCH Adds persistence Past shocks + past variance � No

TGARCH Captures leverage effect Past shocks + negative shock dummy � Yes

EGARCH Asymmetric + log variance Signed standardized shocks � Yes

� When to Use Which?

Scenario Use This Model

Volatility clusters but no asymmetry ARCH or GARCH

Negative shocks increase volatility more TGARCH or EGARCH

You want log variance + no non-negativity constraints EGARCH

The GARCH-M model includes the conditional variance (or standard deviation) as a regressor in the
mean equation, to account for the idea that volatility (risk) affects returns.
After Estimation

 The coefficient on the GARCH term in the output tells you how risk (volatility) affects
mean returns.
 If significant and positive → Higher risk leads to higher expected return.

� What Does the News Impact Curve (NIC) in EViews Tell Us?

The News Impact Curve in EViews shows how new information (shocks)—positive or
negative—affects the conditional variance (volatility) of a time series in nonlinear volatility
models like:

 GARCH
 EGARCH
 GJR-GARCH
 TARCH

� Intuition Behind the NIC

 The curve plots the impact of a standardized shock (residual) on volatility.

 It helps determine whether negative shocks ("bad news") have a different effect on
volatility compared to positive shocks ("good news") of the same size.

� What It Shows
Aspect Meaning

In standard GARCH, the curve is symmetric — good and bad news have the
Symmetry
same effect on volatility.

Asymmetry (Leverage In models like EGARCH or GJR-GARCH, bad news increases volatility more than
Effect) good news → curve is asymmetric.

Shows the sensitivity of volatility to shocks. A steeper curve = larger response to

Steepness
news.

� How to View NIC in EViews

1. Estimate a nonlinear GARCH-type model (e.g., EGARCH, GJR).

2. In the results window, go to:
o View → News Impact Curve.
3. A graph will appear showing:
 o X-axis: size of the shock (positive or negative residual).
o Y-axis: impact on conditional variance.

� Interpreting the NIC

Shape Interpretation

U-shape (symmetric) GARCH model: positive and negative shocks affect volatility equally.

Asymmetric (skewed EGARCH, GJR: Negative shocks increase volatility more than positive ones
right) (leverage effect).

Flat curve Volatility is not much affected by shocks (rare).

� Example Use

In financial data (e.g., stock returns), you may observe:

 Bad news (e.g., a market crash) increases volatility significantly.

 Good news increases it less — this is captured in the asymmetric shape of the NIC in
models like EGARCH.

� Summary
Term Meaning

News Impact Curve Graph of how shocks affect volatility

Symmetric NIC Shocks affect volatility equally (GARCH)

Asymmetric NIC Negative shocks affect volatility more (EGARCH, GJR-GARCH)

Use in EViews View → News Impact Curve after estimating a volatility model