You are on page 1of 14

Time Series Models for Business

and Economic Forecasting


Hung-Pin Lai
March 18, 2007

The six key features of economic


time series
Many time series data have at least one of the first
five features:
Trends
Seasonality
Somehow influential data points (aberrant
observations)
A variance that changes because of past
observations (conditional heteroskedasticity)
non-linearity
Common features

1. Trends

Modeling trend
By a Linear trend

Use growth rate

Changing trends
1946-1960
increasing trend
1960-1973
decreasing trend
1974 -upward again

2. Seasonality
Observations in certain seasons display
strikingly different features to those in other
Seasons.

Modeling seasonality

(See figure 2.3 for the original data)

R2=0.296

R2=0.254

R2=0.965

R2=0.971

(increasing variance; changing seasonality)

(mean shift; regime shift; structural break)

television

radio

3. Aberrant observation
Definition: When a single observation have a major
impact on time series modeling and forecasting, it is called
aberrant observation.
yt =log(pricet)

Features of Figure 2.11:


1. The differenced yt is not a good approximation to
the inflation when the quarterly inflation rate is high.
2. The data in 1989 (extreme value) seem to be
quite different from those observations the years
before. Consider the model:

10

4. Conditional heteroskedasticity

11

Aberrant observations tend to emerge in


clusters
Clusters of observations with large variances
(volatility clustering or conditional
heteroskedasticity)
Modeling conditional heteroskedasticity:

5. Non-linearity
Everything other than nonlinearity
Regime switching or state dependency

12

Modeling regime switching:


Model

6. Common features
Common trend (cointegration); common seasonality,
Multivariate models

13

14

You might also like