ARCH (Autoregressive Conditional Heteroskedasticity) stationary time series is one whose statistical

model to capture changing variance over time, a properties, such as mean, variance, and
phenomenon known as conditional heteroskedasticity. autocorrelation, remain constant over time.
This model is particularly useful when working with
financial and economic data where volatility, or the
variability of returns, is not constant. AutoRegressive Model (AR Model):
Conditional Variance: The core idea of an ARCH model • describes a time series as a linear combination of
is to model the conditional variance of a time series. It its own past values. In an AR model, the future
recognizes that the variance of a series is not constant values of the time series are modeled as a
but varies with past observations. weighted sum of its previous observations.
Autoregressive Structure: The conditional variance is • Order of the AR Model: denoted as "p," specifies
modeled as a function of past squared observations. In how many previous observations are included in
essence, the model captures how the past shocks or the model. For example, an AR(1) model includes
volatilities affect the current volatility. The model only the immediately preceding observation,
typically involves a lag structure, where the past while an AR(2) model includes the two most
squared observations, or residuals, are used to predict recent observations.
the current conditional variance. • Stationarity: For an AR model to be applicable,
Parameters: most critical parameter is the order of the the time series should be stationary, meaning
model, often denoted as p. This parameter determines that its statistical properties do not change over
the number of lagged squared terms to include in the time. If the time series is not stationary,
model. For example, in an ARCH(1) model, only the differencing may be applied to achieve
squared term at the previous time step is used to stationarity.
predict the current variance. • Parameter Estimation: using statistical methods
Residuals: The model assumes that the series is such as maximum likelihood estimation.
stationary, and it models the squared residuals as a • INCREASE in the lagged order => decreases the df
function of past squared residuals. This accounts for the . ex – AR(3) => n-3 =df
clustering of large and small residuals over time. • PACF => order of AP
Conditional Heteroskedasticity: The term ACF => MA order
"heteroskedasticity" refers to the changing variance
over time, and "conditional" signifies that this variance Box and Jenkins, First model for modeling and
depends on past observations. forecasting univariate time series data. 3 main stages:
Estimation: Estimating the parameters of an ARCH • Model Identification: In this initial stage, the goal
model is typically done using maximum likelihood is to identify the appropriate model for the time
estimation (MLE). This involves finding the parameter series data. This involves assessing the data for its
values that maximize the likelihood of observing the stationarity (or transforming it into a stationary
data given the model. series), identifying the order of differencing
Use Cases: ARCH models are commonly used in finance required, and selecting autoregressive (AR) and
to model the volatility of financial returns, such as stock moving average (MA) orders using
prices. They are particularly helpful in understanding autocorrelation and partial autocorrelation
and predicting periods of high and low volatility in plots. The chosen model is often referred to as an
financial markets. ARIMA (Autoregressive Integrated Moving
Extensions- GARCH incorporate not only past squared Average) model.
residuals but also past conditional variances in the • Model Estimation: After identifying the ARIMA
model. model, the next step is to estimate the model's
parameters, which involves finding the
coefficients for the autoregressive and moving
ACF measures the correlation between a time series average terms. Estimation can be done using
and its own past values at different time lags. identify maximum likelihood estimation or other
the presence of autocorrelation, techniques. The estimation process also provides
ACF plot displays the correlation coefficients at various information about the model's goodness of fit.
lags. A sharp drop in autocorrelation after a certain lag • Model Diagnostic Checking: Once the model is
suggests a seasonal pattern, while a slow decay suggests estimated, it's crucial to assess its validity.
a trend or a more complex dependence structure. Diagnostic checks are performed to ensure that
the model residuals are white noise, meaning
PACF measures the correlation between a time series they are independent and identically
and its own past values at different time lags while distributed. Various diagnostic tests, including
controlling for the intermediate lags. residual autocorrelation plots and the Ljung-Box
helps identify the direct relationship between a time test, are used to evaluate the model's adequacy.
series and its past values, excluding the indirect
influence of intermediate lags. useful in determining the
order of autoregressive (AR) processes . PACF plot
displays the partial correlation coefficients at various
lags. Significant values at specific lags indicate the order
of the AR process. you can build an appropriate
autoregressive model.

Chow test: detecting shifts or structural changes in the

relationship between variables at a specific point in
time. STEPS- Data Preparation: The first step is to
organize your data into two subsets: one before the
suspected structural break point and one after it. You
also need to specify the point in time at which you
suspect a structural break occurred. Regression
Modeling: Next, you fit separate linear regression
models to the two subsets of data Calculation: Calculate
the sum of squared residuals for both models. F-
Statistic: Compute: H0: there is no structural break F-
Statistic Table: Compare the calculated F-statistic to a
critical value from an F-distribution table. The critical
value depends on your desired level of significance (e.g.,
0.05). If calculated F-statistic > critical value => there is
a significant structural break in the data. coefficients are
not the same
 4. Stationarity Assumption: ARCH models
assume that the time series is stationary. If the data is
not stationary, it may lead to spurious results.
5. Misspecification: ARCH models may not
capture more complex patterns in volatility, such as
long memory or structural breaks, which are common
in financial time series data.
6. GARCH as an Extension: To overcome some
of these limitations, the Generalized Autoregressive
Conditional Heteroskedasticity (GARCH) model was
introduced. GARCH models are more flexible and
widely used in practice. They allow for more
parameters while still maintaining a parsimonious
form, and they can capture time-varying volatility more
effectively.
7. Model Complexity: ARCH models primarily
focus on conditional heteroskedasticity in the data,
which might be an oversimplification in some cases.
More advanced models, like stochastic volatility
models or Bayesian methods, can capture additional
complexities in volatility dynamics.

(ARCH) introduced by Robert Engle in 1982, modeling

conditional volatility in time series data. Limitations-
1. Choice of Lag Order (q): can be challenging.
no one-size-fits-all answer, requires trial and error.
Models with too few lags might fail to capture the
complexity of volatility dynamics, while models with
too many lags may become overly complex.
2. Large Model Sizes: If the optimal model
requires a high number of lags (large q), it can result in
a large conditional variance model with many
parameters. lack of parsimony can make interpretation
and estimation of the model parameters more difficult.
3. Non-negativity constraints might be
violated.: bWhen you have many parameters, as can
happen with high q values, there's a higher likelihood
of some parameter estimates being negative, which
may not make sense in the context of volatility
modeling.