Module I: Issues with linear regression , overview on ACF, PACF, MA

processes
Assumptions of Linear regression analysis. Violations of the
assumptions of Classical Linear Regression
Analysis. Time series analysis - Preparing data for analysis, Univariate
time series analysis, Autocorrelation
function (ACF), Partial autocorrelation function (PACF), Moving Average
processes (MA), Auto Regressive
processes (AR), ARMA process,
Module II: Univariate time series modelling and forecasting
Box Jenkins approach – Building ARMA, ARIMA models in EVIEWS,
GRETL, R. Forecasting ARMA models using
EVIEWS, Exponential smoothing models, ARIMA models, applications in
financial decision making.
Module III: Multivariate models
Multiequation modelling- Simultaneous equation modelling. Vector Auto
Regression (VAR), VAR with
exogenous variables, VAR estimation in E Views, GRETL, R. Impulse
Response and variance decomposition
Module IV: Modelling long-run relationships
Stationarity and unit root testing, cointegration, equilibrium or error
correction models, testing for cointegration
using Johansen technique.
Module V: Modelling volatility in time series
Modelling time series volatility: Volatility - Historical volatility, Implied
volatility models, ARCH processes,
GARCH Processes, Estimation of ARCH, GARCH models in EVIEWS,
Extensions of GARCH models, Multivariate
GARCH models
MODULE 1

Issues with Linear Regression

Linear regression assumes a linear relationship between independent (XX) and
dependent (YY) variables. However, in real-world scenarios, this assumption may
not always hold. Below are some key issues:
1. Multicollinearity
Multicollinearity occurs when independent variables (X1,X2,X3,...X_1, X_2,
X_3, ...) are highly correlated. This makes it difficult to determine the individual
effect of each predictor.
 Detection: Variance Inflation Factor (VIF) is used to detect
multicollinearity.

Where R_j^2 is the R^2 value obtained by regressing X_j on the other
independent variables.
o If VIF > 10, multicollinearity is a problem.

o If VIF ≈ 1, no multicollinearity exists.

 Solution:
o Remove one of the correlated variables.

o Use Principal Component Analysis (PCA) to reduce dimensionality.

2. Heteroscedasticity
Heteroscedasticity means that the variance of residuals (ϵt\epsilon_t) is not
constant across all values of XX.
 Mathematical Representation:

This violates the assumption of constant variance.

 Detection:
o Breusch-Pagan Test: Checks if the variance of residuals is related
to independent variables.
o White’s Test: A more general test for heteroscedasticity.

 Solution:
o Use Weighted Least Squares (WLS) instead of OLS.

o Transform the dependent variable (e.g., using log

transformation).

3. Autocorrelation
Autocorrelation occurs when residuals (ϵt\epsilon_t) are correlated over time,
violating the assumption of independence.
 Detection:
o Durbin-Watson (DW) Test:

 Solution:
o Use ARMA (AutoRegressive Moving Average) models instead
of OLS.

4. Endogeneity
Endogeneity occurs when an independent variable (X) is correlated with the error
term (ϵ\epsilon), leading to biased and inconsistent estimates.
 Solution:
o Use Instrumental Variable (IV) Regression, where another
variable (instrument) is used to replace the endogenous X.

5. Non-Linearity
If the relationship between X and Y is non-linear, a simple linear regression
model is inappropriate.
 Solution:
o Use Polynomial Regression (e.g., add X^2, X^3 terms).

o Use Logarithmic Transformations.

Autocorrelation Function (ACF) and Partial Autocorrelation Function

(PACF)
Autocorrelation Function (ACF)
ACF measures how current values of a time series (Y_t) are correlated with past
values (Y_{t-k}) at different lags k.
  Formula:

o If ACF gradually declines, it suggests an AR (Auto-Regressive)

process.
o If ACF cuts off suddenly, it suggests an MA (Moving Average)
process.

Partial Autocorrelation Function (PACF)

PACF measures the correlation between Y_t and Y_{t-k}, removing the effect
of intermediate lags.
 Use Case: Helps determine the order of AR(p) processes in an ARIMA
model.
 Interpretation:
o If PACF cuts off after lag pp, the series follows an AR(p) process.

Moving Average (MA) Process

A Moving Average (MA) model represents a time series as a function of past
white noise (random errors).
MA(q) Process Equation

 ACF cuts off after lag qq, while PACF decays gradually.

Auto-Regressive (AR) Process

An Auto-Regressive (AR) model represents a time series as a function of its
past values.
AR(p) Process Equation

 PACF cuts off after lag pp, while ACF decays gradually.
ARMA Process (Auto-Regressive Moving Average)
A combination of AR(p) and MA(q) models.

Used when data exhibits both autoregressive (AR) and moving average
(MA) properties.

Assumptions of Linear Regression

1. Linearity:
o The relationship between independent and dependent variables is
linear.
2. Homoscedasticity:
o Constant variance of residuals (σ2).

3. Independence of Errors:
o No autocorrelation in residuals.

4. Normality of Errors:
o Residuals follow a normal distribution.

5. No Multicollinearity:
o Independent variables should not be highly correlated.

Violations of Classical Linear Regression Assumptions

Assumption Violation Solution

Linearity Non-linearity Polynomial Regression

Homoscedasticit Weighted Least Squares

Heteroscedasticity
y (WLS)

Independence Autocorrelation ARIMA, ARMA models

Normality Non-normal residuals Log transformations

No High correlation among independent Remove one variable or

Multicollinearity variables use PCA
Time Series Analysis - Preparing Data
1. Stationarity Check
 Use Augmented Dickey-Fuller (ADF) Test to check for stationarity.
 If p-value > 0.05, the series is non-stationary.
2. Differencing
 If non-stationary, apply first-order differencing:

3. Identify Model Order

 Use ACF for MA order selection.
 Use PACF for AR order selection.
Four Key Steps in Time Series Analysis
Time series analysis involves Identification, Estimation, Diagnostic
Checking, and Forecasting to develop reliable predictive models. Let's go
through each step in detail.

1. Identification
The first step is to determine the appropriate model for the time series data. This
involves analyzing the stationarity, trend, seasonality, and autocorrelation
structure.
Key Activities in Identification:
1. Stationarity Check
o A stationary time series has a constant mean, variance, and
autocovariance over time.
o Dickey-Fuller Test (ADF Test) is used to check stationarity.

o If the series is non-stationary, differencing is applied to make it

stationary.
ADF Test Example in Python:
from statsmodels.tsa.stattools import adfuller
result = adfuller(data['Price'])
print('ADF Statistic:', result[0])
print('p-value:', result[1])
2. Autocorrelation Analysis
o Autocorrelation Function (ACF): Identifies Moving Average
(MA) order (q).
 o Partial Autocorrelation Function (PACF): Identifies
AutoRegressive (AR) order (p).
Plotting ACF and PACF:
from statsmodels.graphics.tsaplots import plot_acf, plot_pacf
plot_acf(data['Price'])
plot_pacf(data['Price'])
plt.show()
3. Choosing the Right Model
o If the series is stationary → ARMA Model.

o If the series is non-stationary → ARIMA Model.

o If the series has seasonality → SARIMA Model.

2. Estimation
Once the model is selected, the next step is estimating the parameters of the
model.
Mathematical Representation of ARIMA Model:

where:
 p = Order of AutoRegressive (AR) component.
 q = Order of Moving Average (MA) component.
 d = Differencing order to make series stationary.
 ϵt = White noise (error term).
Estimating ARIMA Parameters in Python:
from statsmodels.tsa.arima.model import ARIMA
model = ARIMA(data['Price'], order=(p, d, q))
model_fit = model.fit()
print(model_fit.summary())
 The Maximum Likelihood Estimation (MLE) method is used for
parameter estimation.
  The model outputs coefficients, AIC (Akaike Information Criterion), and BIC
(Bayesian Information Criterion) to determine model fit.

3. Diagnostic Checking
After estimating the model, we need to check its validity by analyzing residuals.
Key Diagnostic Tests:
1. Residual Analysis
o The residuals should be normally distributed with zero mean
and no autocorrelation.
o A Q-Q Plot can check normality:

2. import scipy.stats as stats

3. import matplotlib.pyplot as plt
4. stats.probplot(model_fit.resid, dist="norm", plot=plt)
5. plt.show()
6. Ljung-Box Test (Checking for Serial Correlation)
o Tests whether residuals are uncorrelated.

7. from statsmodels.stats.diagnostic import acorr_ljungbox

8. print(acorr_ljungbox(model_fit.resid, lags=[10], return_df=True))
o If p-value > 0.05, residuals are uncorrelated → Model is valid.

o If p-value < 0.05, residuals have autocorrelation → Model needs

improvement.
9. Heteroskedasticity Test (ARCH Test for GARCH models)
o Checks whether residual variance is constant over time.

10.from statsmodels.stats.diagnostic import het_arch

11.print(het_arch(model_fit.resid))

4. Forecasting
After model validation, it is used for future predictions.
Forecasting with ARIMA:
forecast = model_fit.forecast(steps=10)
print(forecast)
Evaluating Forecast Accuracy:
1. Mean Absolute Error (MAE)
 2. Root Mean Squared Error (RMSE)

Python Implementation:
from sklearn.metrics import mean_squared_error
import numpy as np
rmse = np.sqrt(mean_squared_error(actual_values, forecasted_values))
print('RMSE:', rmse)

Conclusion
The Four Steps in Time Series Analysis:
1. Identification: Choose AR, MA, or ARIMA based on stationarity and
ACF/PACF plots.
2. Estimation: Estimate model parameters using Maximum Likelihood
Estimation (MLE).
3. Diagnostic Checking: Verify residuals are normal and uncorrelated.
4. Forecasting: Use the model to predict future values and evaluate
accuracy.

MODULE 2
Here’s a detailed breakdown of Module 2: Univariate Time Series
Modelling and Forecasting, including equations, model-building steps,
and forecasting techniques.

Univariate Time Series Modelling and Forecasting

A univariate time series consists of observations of a single variable
over time. The goal of univariate time series modeling is to understand
the data structure and make forecasts based on past observations.
1. Components of a Time Series
A time series consists of four components:
1. Trend (T_t) – The long-term movement in the series.
2. Seasonality (S_t) – Periodic fluctuations at regular intervals.
3. Cyclic (C_t) – Long-term wave-like movements.
4. Irregular (I_t) – Unpredictable noise or randomness.
A time series model can be additive or multiplicative:

2. Box-Jenkins Approach (ARMA, ARIMA Models)

The Box-Jenkins methodology is a systematic approach to building
ARMA and ARIMA models for time series forecasting.
Steps in the Box-Jenkins Approach
1. Identify
o Check stationarity using Augmented Dickey-Fuller (ADF)
Test.
o Use ACF & PACF plots to determine AR and MA orders.

2. Estimate
o Fit an ARMA(p, q) or ARIMA(p, d, q) model using maximum
likelihood estimation (MLE).
3. Diagnose
o Check residuals for autocorrelation using the Ljung-Box
test.
o Ensure residuals are normally distributed.

4. Forecast
o Use the estimated model to predict future values.

3. ARMA and ARIMA Models

Auto-Regressive Moving Average (ARMA) Model
ARMA combines Auto-Regressive (AR) and Moving Average (MA)
components.
  AR(p) Component: Uses past values of Y_t.
 MA(q) Component: Uses past forecast errors.

Auto-Regressive Integrated Moving Average (ARIMA) Model

ARIMA is an extension of ARMA that includes differencing to make the
series stationary.

Where:
 p = order of Auto-Regression (AR)
 d = degree of Differencing
 q = order of Moving Average (MA)
 L = Lag Operator
If d=1d = 1, the model is ARIMA(1,1,1), meaning one differencing is
applied.

4. Building ARMA and ARIMA Models in EVIEWS, GRETL, R

EVIEWS (Econometric Views)
1. Import the time series data.
2. Check for stationarity using ADF Test.
3. Use ACF and PACF plots to determine p, q.
4. Estimate ARMA or ARIMA using the "Estimate Equation" option.
5. Evaluate residual diagnostics.
 6. Forecast future values.

GRETL (Gnu Regression, Econometrics, and Time-Series Library)

1. Load time series data into GRETL.
2. Use ADF test to check stationarity.
3. Select Time Series → ARIMA and input p,d,qp, d, q values.
4. Analyze results and check residual diagnostics.
5. Generate forecasts.

R (Statistical Computing)
Load libraries:
1. library(forecast)
2. library(tseries)
Import data and check stationarity:
3. adf.test(time_series_data)
Fit ARIMA model:
4. model <- auto.arima(time_series_data)
Forecast future values:
5. forecast(model, h=10)
6. plot(forecast(model, h=10))

5. Forecasting ARMA Models using EVIEWS

Steps in Forecasting
1. Open estimated ARMA model in EVIEWS.
2. Go to Forecast → Specify future periods.
3. Choose Dynamic Forecasting to predict beyond sample.
4. Analyze forecast errors using Mean Absolute Error (MAE) and Root
Mean Squared Error (RMSE).

6. Exponential Smoothing Models

Exponential smoothing models are used for short-term forecasting,
especially when there is no strong seasonality or trend.
Simple Exponential Smoothing (SES)
Holt’s Linear Trend Model

 Handles both level (ltl_t) and trend (btb_t).

Holt-Winters Method (Seasonal Data)

 Multiplicative Seasonality for periodic trends.

7. Applications in Financial Decision Making

Time series forecasting is crucial in finance and business:
1. Stock Price Prediction
o ARIMA models forecast stock prices based on past
movements.
2. Interest Rate Forecasting
o ARMA models predict interest rate changes, helping banks
and policymakers.
3. Risk Management
o Volatility forecasting (e.g., using GARCH models) helps
manage financial risk.
4. Macroeconomic Forecasting
o ARIMA models predict GDP, inflation, and unemployment
rates.
Conclusion
 Box-Jenkins Approach is essential for time series modeling.
 ARMA & ARIMA models help forecast financial trends.
 EVIEWS, GRETL, R are commonly used for modeling.
 Exponential Smoothing is useful for short-term predictions.

MODULE 3
Module 3: Multivariate Time Series Models
In contrast to univariate models, where only one dependent variable is
analyzed, multivariate time series models examine multiple
interdependent variables simultaneously. These models help in
understanding causality, interrelationships, and forecasting when
multiple time-dependent variables interact over time.

1. Multivariate Time Series Models

Why Use Multivariate Models?
 Captures the interdependencies among multiple variables.
 Helps in policy analysis and economic forecasting.
 More robust than single-equation models as they control for
simultaneous effects.
Types of Multivariate Models:
1. Simultaneous Equation Models (SEM)
2. Vector Auto Regression (VAR)
3. VAR with Exogenous Variables (VARX)
4. Impulse Response & Variance Decomposition

2. Multiequation Modelling – Simultaneous Equation Models (SEM)

In a simultaneous equation model (SEM), multiple equations describe
the relationship between variables, where at least one dependent
variable is also an explanatory variable in another equation.
Example of a Two-Equation SEM
Where:
 Y_1 and Y_2 are endogenous variables (determined within the
system).
 X_1 and X_2 are exogenous variables (determined outside the
system).
 u_1, u_2 are error terms.
Identification Problem in SEM
 Underidentified: Insufficient data to estimate parameters.
 Exactly identified: The number of equations matches the number
of unknowns.
 Overidentified: More equations than unknowns, requiring
instrumental variables (IV) estimation.
Estimation Methods in SEM
 Two-Stage Least Squares (2SLS)
 Three-Stage Least Squares (3SLS)

3. Vector Auto Regression (VAR) Model

Definition of VAR
VAR models generalize univariate AR models to multiple time series
variables, where each variable is modeled as a function of past values
of itself and other variables.

Where:
 Y_t is a vector of endogenous variables.
 A_i are coefficient matrices.
 ϵt is a white noise error term.
Example of a VAR(1) Model (Two Variables)

 Y_1 and Y_2 influence each other based on past values.

 No strict distinction between dependent and independent
variables.
Steps for Estimating a VAR Model
1. Check Stationarity – Use Augmented Dickey-Fuller (ADF) test.
2. Select Lag Length – Use Akaike Information Criterion (AIC) or
Bayesian Information Criterion (BIC).
3. Estimate VAR Model – Fit the model using Ordinary Least Squares
(OLS) for each equation.
4. Model Diagnostics – Check for autocorrelation (Ljung-Box Test).
5. Impulse Response Analysis – Measure impact of shocks.
6. Forecasting – Use the fitted model for multi-step predictions.

4. VAR with Exogenous Variables (VARX Model)

VAR models can be extended by adding exogenous variables that
influence the system but are not influenced by it.

Where:
 X_t represents exogenous variables (e.g., policy shocks,
macroeconomic indicators).
 B is the coefficient matrix capturing their impact.

5. VAR Estimation in EVIEWS, GRETL, R

VAR Estimation in EVIEWS
1. Import Data → Open VAR Estimation from the menu.
2. Check Stationarity → Perform ADF test for each variable.
3. Select Lag Order → Use Lag Length Criteria (AIC, BIC, HQC).
4. Estimate VAR Model → Run regression and analyze coefficients.
5. Impulse Response & Variance Decomposition → Generate results.

VAR Estimation in GRETL

1. Load Data → Open time-series dataset.
2. Check Stationarity → Run unit root test (ADF or KPSS).
3. Specify VAR Model → Set variables and lag order.
4. Estimate Model → Obtain coefficient estimates.
 5. Perform Diagnostics → Check residuals and impulse response.

VAR Estimation in R
library(vars)

# Load dataset
data <- read.csv("timeseries.csv")

# Convert to time series object

ts_data <- ts(data, start=c(2000,1), frequency=12)

# Estimate VAR model with 2 lags

var_model <- VAR(ts_data, p=2, type="const")

# Summary of the model

summary(var_model)

# Forecasting
forecast <- predict(var_model, n.ahead=10)
plot(forecast)

6. Impulse Response and Variance Decomposition

Impulse Response Function (IRF)
Impulse response analysis examines how a shock to one variable affects
the others in the VAR system.
 Mathematical Formulation:

where C_h represents the response of Y at horizon h after a shock ϵt

 Graphical Representation:
o Plots show the response over time.

o Helps policymakers understand the impact of monetary or

fiscal shocks.
Variance Decomposition (VD)
Variance decomposition shows how much of the forecast error variance
of a variable is explained by shocks to itself and other variables.
If Var(Y_t) is decomposed into contributions from different variables, we
get:

 Higher percentage → Greater influence of that variable on the

system.
 Used in macroeconomic forecasting (e.g., GDP, inflation, interest
rates).

7. Applications in Financial Decision Making

VAR Models are Used in:
1. Macroeconomic Policy Analysis
o Examining the effect of monetary policy on GDP and
inflation.
2. Stock Market Forecasting
o Analyzing relationships between stock prices and interest
rates.
3. Risk Management
o Measuring the impact of oil price shocks on financial
markets.
4. Foreign Exchange Rate Forecasting
o Understanding exchange rate movements based on past
data.

Conclusion
 Multivariate models help analyze the interactions between
multiple time series.
 VAR models are powerful tools for economic forecasting and
policy analysis.
 Impulse Response & Variance Decomposition provide valuable
insights into shock transmission.
  EVIEWS, GRETL, and R are widely used for estimating and
forecasting VAR models.

MODULE 4
Module 4: Modelling Long-Run Relationships in Time Series
Long-run relationships in time series data are crucial for
macroeconomic forecasting, financial decision-making, and policy
formulation. Traditional time series models like ARMA and VAR are
suitable for stationary data, but real-world financial and economic data
often exhibit non-stationary behavior. This module covers:
 Stationarity and Unit Root Testing
 Cointegration and Long-Run Equilibrium
 Error Correction Models (ECM)
 Testing for Cointegration Using Johansen’s Technique

1. Stationarity and Unit Root Testing

What is Stationarity?
A time series is stationary if its mean, variance, and autocovariance
remain constant over time.
Mathematically, a time series Y_t is stationary if:

Example: White Noise Process

This process is strictly stationary as its mean and variance are

constant.

Unit Root and Non-Stationarity

A non-stationary series exhibits a unit root, implying that shocks have a
permanent impact on the series.
A unit root process follows:
Example: Random Walk Process

Here, Var(Yt)=tσ2, meaning variance increases over time → non-

stationary.

Unit Root Tests (Checking for Stationarity)

1. Augmented Dickey-Fuller (ADF) Test
The ADF test checks whether the series has a unit root by estimating:

 Null Hypothesis (H_0): γ=0\gamma = 0 (Unit root exists, non-

stationary).
 Alternative (H_1): γ<0\gamma < 0 (Stationary).
 If p-value < critical value, reject H_0, meaning the series is
stationary.

2. Phillips-Perron (PP) Test

 Similar to ADF but adjusts for heteroskedasticity and serial
correlation.
 Uses non-parametric correction rather than adding lags.

3. KPSS Test (Kwiatkowski-Phillips-Schmidt-Shin)

 Null Hypothesis (H_0): Series is stationary.
 Alternative (H_1): Series is non-stationary.
 If p-value is small, reject H_0, meaning series is non-stationary.
2. Cointegration and Long-Run Equilibrium
What is Cointegration?
Even if two variables are individually non-stationary, they can form a
stationary linear combination. This indicates a long-run equilibrium
relationship.
Mathematically, if:

then X_t and Y_t are cointegrated.

Example: Cointegrated Variables

Consider inflation rate (Y_t) and money supply (X_t).
Even if both are non-stationary, their difference may be stationary,
suggesting a long-term relationship.

3. Error Correction Models (ECM)

Why Use ECM?
 Cointegration implies a long-run equilibrium, but short-run
deviations may occur.
 ECM adjusts for these short-run dynamics while maintaining the
long-run equilibrium.
ECM Representation

Where:
 γ\gamma is the error correction term (ECT), capturing deviations
from equilibrium.
 If γ\gamma is negative and significant, the system returns to
equilibrium.

Example: ECM for Interest Rates and Inflation

  If inflation rises too much, the negative ECT pulls it back to
equilibrium.

4. Testing for Cointegration Using Johansen’s Technique

Why Johansen’s Test?
 The Engle-Granger test is limited to two variables.
 Johansen’s test allows for multiple cointegrating relationships.
Johansen’s Cointegration Test Procedure
It uses a Vector Error Correction Model (VECM):

 The matrix Π\Pi contains cointegrating relationships.

Steps in Johansen’s Test

1. Determine Optimal Lag Length
o Use AIC or BIC.

2. Estimate VAR Model

o Check if the system is stable.

3. Perform Johansen’s Test

o Two test statistics:

 Trace Test: Tests the number of cointegrating vectors.

 Max-Eigenvalue Test: Tests the rank of Π\Pi.
o If p-value < 0.05, cointegration exists.

Implementation in R
library(urca)
# Load data
data <- read.csv("financial_data.csv")
Y <- ts(data[,2:3]) # Select variables
# Johansen test
johansen_test <- ca.jo(Y, type="trace", ecdet="const", K=2)
summary(johansen_test)

5. Applications in Financial Decision-Making

1. Stock Market Analysis
 Identifying cointegration among NIFTY and SENSEX for pairs
trading.
2. Exchange Rate Forecasting
 USD/INR and interest rate relationships help in forex trading.
3. Interest Rate and Inflation Modelling
 ECM helps in monetary policy formulation.

Conclusion
 Unit root tests ensure data stationarity.
 Cointegration tests check for long-run relationships.
 Error Correction Models adjust short-run deviations while
maintaining equilibrium.
 Johansen’s technique is a powerful tool for multivariate
cointegration testing.

MODULE 5
Module 5: Modelling Volatility in Time Series
Volatility modelling is crucial in financial markets, risk management,
and derivative pricing. Unlike standard time series models, which
assume constant variance, financial time series often exhibit time-
varying volatility, commonly seen in stock returns, exchange rates, and
interest rates. This module covers:
 Historical Volatility and Implied Volatility
 Autoregressive Conditional Heteroskedasticity (ARCH) Models
 Generalized ARCH (GARCH) Models
 Extensions of GARCH Models
 Multivariate GARCH Models
 Estimation of ARCH/GARCH Models in EVIEWS

1. Volatility in Time Series

What is Volatility?
Volatility measures the dispersion of returns over a given period. It is
essential for:
 Risk management (VaR calculations).
 Derivatives pricing (Black-Scholes model).
 Portfolio allocation (minimizing risk).
Mathematically, volatility (σt\sigma_t) is defined as:

where R_t represents returns.

Types of Volatility
1. Historical Volatility (HV)
o Measures past fluctuations in asset prices.

o Estimated using the standard deviation of past returns:

o Example: NIFTY 50 return volatility over the past year.

2. Implied Volatility (IV)

o Derived from option prices using models like Black-Scholes.

o Forward-looking: Reflects market expectations of future

volatility.
Example: VIX Index (India VIX)
IV=f(option price, strike price, time to maturity)

2. ARCH (Autoregressive Conditional Heteroskedasticity) Models

Why Use ARCH Models?
Financial time series often show:
 Volatility clustering: Periods of high volatility followed by more
high volatility.
  Leptokurtosis: Returns have fatter tails than the normal
distribution.
Basic ARCH(q) Model
The ARCH model captures time-varying volatility by modeling the
conditional variance as a function of past squared errors.
Model Specification:

where:
 σt2\sigma_t^2 is the conditional variance.
 ϵt2\epsilon_t^2 represents past shocks.
Estimation in EVIEWS:
arch_model <- garchFit(~garch(1,0), data=returns, trace=FALSE)
summary(arch_model)
Limitation: ARCH(q) requires a large number of lags (q) to capture
volatility, making it inefficient.

3. GARCH (Generalized ARCH) Model

Why GARCH?
 ARCH models require many lags to model volatility.
 GARCH introduces lagged conditional variance terms, reducing
the number of parameters.
GARCH(p, q) Model

where:
 βj\beta_j represents past volatility’s impact on current variance.
 p and q are selected using AIC/BIC.
Example: GARCH(1,1) Model

 If α1+β1≈1, volatility is persistent, seen in financial crises.

Estimation in EVIEWS
1. Load time series data.
2. Run GARCH(1,1) in EVIEWS:
ARCH estimation -> Choose GARCH(1,1)

4. Extensions of GARCH Models

EGARCH (Exponential GARCH) Model
 Captures asymmetric effects (bad news increases volatility more
than good news).
 Logarithmic variance model:

 Used for stock market crashes and financial crises.

TGARCH (Threshold GARCH)
 Models leverage effect (negative returns impact volatility more
than positive returns).

 Used in options pricing.

GARCH-M (GARCH-in-Mean)

 Used in asset pricing models.

5. Multivariate GARCH Models

Why Multivariate GARCH?
 Models volatility spillovers between multiple assets.
  Example: How volatility in NIFTY 50 affects Bank NIFTY.
BEKK-GARCH Model

where H_t is the conditional variance matrix.

DCC-GARCH (Dynamic Conditional Correlation)
 Captures time-varying correlations between assets.

6. Estimation of ARCH/GARCH Models in EVIEWS

Steps for GARCH(1,1) in EVIEWS
1. Load Data
o Import return series.

2. Set Up GARCH Model

o In EVIEWS:

Quick -> Estimate Equation -> ARCH

o Choose GARCH(1,1).

3. Interpret Results
o Alpha (α1\alpha_1): Impact of past shocks.

o Beta (β1\beta_1): Volatility persistence.

o If α1+β1≈1→ Highly persistent volatility.

7. Applications in Financial Decision-Making

1. Risk Management
 VaR (Value at Risk) calculation using GARCH models.
2. Option Pricing
 Implied volatility estimation for option strategies.
3. Portfolio Optimization
 Multivariate GARCH for hedging strategies.

Conclusion
 ARCH models capture conditional variance.
 GARCH models improve efficiency by adding lagged variance.
 EGARCH, TGARCH, and GARCH-M enhance modeling capabilities.
 Multivariate GARCH is essential for portfolio risk management.
 EVIEWS, R, and Python help estimate these models efficiently.