Scribd
Upload
36 views

Forecasting Models

This document discusses two methods for forecasting: ARIMA (autoregressive integrated moving average) models and VAR (vector autoregression) models. It focuses on explaining the ARIMA methodology, which involves analyzing time series data to identify patterns and differences for modeling. The ARIMA process includes identifying an appropriate ARIMA(p,d,q) model, estimating its parameters, checking the model fit, and using the model for forecasting within and outside the sample data. Static forecasts use actual past values while dynamic forecasts use previously forecasted values, which can compound errors.

Uploaded by

Varun Vaishnav
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
36 views

Forecasting Models

This document discusses two methods for forecasting: ARIMA (autoregressive integrated moving average) models and VAR (vector autoregression) models. It focuses on explaining the ARIMA methodology, which involves analyzing time series data to identify patterns and differences for modeling. The ARIMA process includes identifying an appropriate ARIMA(p,d,q) model, estimating its parameters, checking the model fit, and using the model for forecasting within and outside the sample data. Static forecasts use actual past values while dynamic forecasts use previously forecasted values, which can compound errors.

Uploaded by

Varun Vaishnav
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 13

Forecasting Models

Introduction
• Based on past and current information, the objective of forecasting is to
provide quantitative estimate(s) of the likelihood of the future course of the
object of interest (e.g. personal consumption expenditure).
• There are several methods of forecasting, we will consider two prominent
methods of forecasting in this chapter:
• (1) the autoregressive integrated moving average (ARIMA) models, popularized
by statisticians Box and Jenkins and known as the Box–Jenkins (BJ)
methodology, and
• (2) the vector autoregression (VAR) models
The Box–Jenkins methodology: ARIMA
modeling
• The basic idea underlying the BJ methodology to forecasting is to analyze the
probabilistic, or stochastic, properties of economic time series on their own
under the philosophy “let the data speak for themselves”.
• Unlike traditional regression models, in which the dependent variable Yt is
explained by k explanatory variables X1, X2, X3, ..., Xk, the BJ time series
models allow Yt to be explained by the past, or lagged, values of Yt itself and
the current and lagged values of ut, which is an uncorrelated random error term
with zero mean and constant variance – that is, a white noise error term.
• The BJ methodology is based on the assumption that the time series under
study is stationary.
• The autoregressive (AR) model
• Consider the following model:
• ---------- (16.6)
• where ut is a white noise error term.
• Model (16.6) is called an autoregressive model of order p, AR (p), for it involves
regressing Y at time t on its values lagged p periods into the past, the value of p
being determined empirically using some criterion, such as the Akaike
information criterion.
• The moving average (MA) model
• We can also model Yt as follows:
• ---------------- (16.7)
• That is, we express Yt as a weighted, or moving, average of the current and past
white noise error terms. Model (16.7) is known as the MA(q) model, the value
of q being determined empirically.
• The autoregressive moving average (ARMA) model We can combine the AR and
MA models and form what is called the ARMA (p,q) model, with p
autoregressive terms and q moving average terms. Again, the values of p and q
are determined empirically.
• The autoregressive integrated moving average (ARIMA) model
• The BJ methodology is based on the assumption that the underlying time series
is stationary or can be made stationary by differencing it one or more times.
• This is known as the ARIMA (p,d,q) model, where d denotes the number of
times a time series has to be differenced to make it stationary.
• In most applications d = 1 – that is, we take only the first differences of the time
series.
• Of course, if a time series is already stationary, then an ARIMA (p,d,q) becomes
an ARMA (p,q) model.
• The practical question is to determine the appropriate model in a given
situation. To answer this question, the BJ methodology follows a four-step
procedure:
• Step 1: Identification: Determine the appropriate values of p, d and q. The main
tools in this search are the correlogram and partial correlogram.
• Step 2: Estimation: Once we identify the model, the next step is to estimate the
parameters of the chosen model.
• In some cases we can use the method ordinary least-squares (OLS), but in some
cases we have to resort to nonlinear (in parameter) estimation methods.
• Since several statistical packages have built-in routines, we do not have to
worry about the actual mathematics of estimation.
• Step 3: Diagnostic checking: One simple test to check that the chosen ARIMA
model is correct one or not , we need to see if the residuals from the fitted
model are white noise; if they are, we can accept the chosen model, but if they
are not, we will have to start afresh.
• That is why the BJ methodology is an iterative process.
• Step 4: Forecasting: The ultimate test of a successful ARIMA model lies in its
forecasting performance, within the sample period as well as outside the
sample period.
Forecasting with ARIMA
• Once a particular ARMA model is fitted, we can use it for forecasting, for this is
the primary objective of such models.
• There are two types of forecast: static and dynamic.
• In static forecasts, we use the actual current and lagged values of the forecast
variable,
• whereas in dynamic forecasts, after the first period forecast, we use the
previously forecast values of the forecast variable.
• This figure gives the actual and forecast values of logs of closing IBM prices, as
well as the confidence interval of forecast.
• The accompanying table gives the same measures of the quality of the forecast
that we saw before, namely, the root mean square, mean absolute error, mean
absolute percentage error and the Theil Inequality Coefficient.
• For our example, this coefficient is practically zero, suggesting that the fitted
model is quite good.
• This can also be seen from Figure 16.5, which shows how closely the actual and
forecast values track each other.
• The picture of the dynamic forecast is given in Figure 16.6.
• On the basis of the Theil coefficient, the dynamic forecast does not do as well as
the static forecast.
• Also the 95% confidence band increases rapidly as we travel along the time axis.
• The reason for this is that we use the previous forecast values in computing
subsequent forecasts and if there is an error in the previously forecast value(s),
that error will be carried forward

You might also like