Summary of Week 1Lecture

Time series is a stochastic process

indexed by time
Time interval much be constant
Monthly, quarterly and annually
Annual % change from quarterly change
400*(log y log x)
Level, 1st lag, 2nd lag and so on
Forecasting, estimating causal effect
Components of time series

Chapter 2 Week
2
1 Issues facing time series analysis
2 Stationarity -Autoregressive processes
3 Non-stationarity
4 Unit roots and integrated order of series

Technical issues
encountered in Time series
data

Time lags - past information may be useful for present forecasting.

Can we treat time lag variable as independent variables?
Correlation over time (serial correlation, a.k.a. autocorrelation)
Forecasting models built on regression methods:
autoregressive (AR) models
autoregressive distributed lag (ADL) models
need not (typically do not) have a causal interpretation
Conditions under which dynamic effects can be estimated, and how
to estimate them
Calculation of standard errors when the errors are serially
correlated

Autocorrelation
The correlation of a series with its own lagged values is called
autocorrelation or serial correlation.
The first autocorrelation of Yt is corr(Yt,Yt1)
The first autocovariance of Yt is cov(Yt,Yt1)
Thus, corr(Yt,Yt1) =
corr(Yt,Yt1) =

cov(Yt , Yt 1 )
var(Yt ) var(Yt 1 )=1

These are population correlations they describe the

population joint distribution of (Yt,Yt1)

Example: Autocorrelations of:

(1) the quarterly rate of U.S. inflation
(2) the quarter-to-quarter change in the quarterly rate of
inflation

Sample autocorrelations
Sample autocorrelations
The j th sample autocorrelation is an estimate of the j th
population
autocorrelation

Y ,Y )
cov(
t
t j
=
Y)
var(
t

wh
ere
Y ,Y )
cov(
t
t j
=

where

Y j 1,T

1 T
(Yt Y j 1,T )(Yt j Y1,T j )

T t j 1
is the sample average of Yt computed over
observations t = j+1,,T. NOTE:

o the summation is over t=j+1 to T (why?)

o The divisor is T, not T j (this is the conventional definition used for
time series data)

Two main uses for Time

Series
Forecasting and estimation of causal
effects are quite different objectives.
For forecasting,
R2

matters (a lot!)

Omitted variable bias isnt a problem!

We will not worry about interpreting coefficients in
forecasting models
External validity is paramount: the model estimated
using historical data must hold into the (near) future

Main Criterion for

Forecasting
Historical Data
Using historical data for forecasting the future
Variation of the data must be maintained the
same for the past, present and future.
The probability distribution of past data,
present data and future data must be the
same - STATIONARY.
Stationary behaviour of the data set simply
means that its probability distribution does
not change with time

Stationarity

Autoregressive Process
AR(p)
A natural starting point for a forecasting model is to use past
values of Y (that is, Yt1, Yt2,) to forecast Yt.
An autoregression is a regression model in which Yt is
regressed against its own lagged values.
The number of lags used as regressors is called the order of
the autoregression.
AR(1) model -In a first order autoregression, Yt is
regressed against Yt1
AR(p) model - In a pth order autoregression, Yt is
regressed against Yt1,Yt2,,Ytp.

AR(1) model
The First Order Autoregressive (AR(1)) Model
The population AR(1) model is
Yt = 0 + 1Yt1 + ut
0 and 1 do not have causal interpretations
if 1 = 0, Yt1 is not useful for forecasting Yt
The AR(1) model can be estimated by OLS regression of Yt against
Yt1
Testing 1 = 0 v. 1 0 provides a test of the hypothesis that Yt1 is
not useful for forecasting Yt

Estimated using data from 1962:I

2004:IV:
Inf = 0.017

t
R 2 = 0.05
0.238Inft1
(0.126)

(0.096)

Is the lagged change in inflation a useful predictor of the

current change in inflation?
t = .238/.096 = 2.47 > 1.96 (in absolute value)
Reject H0: 1 = 0 at the 5% significance level
Yes, the lagged change in inflation is a useful predictor of current change in
inflation
2
R
is pretty low!
but the

However, adjusted R square is very low!

Forecasts: terminology and notation

15

Forecast errors

16

Example: forecasting inflation using an

AR(1)

17

The AR(p) model: using multiple lags for

forecasting

18

Example: AR(4) model of inflation

19

Digression: we used Inf, not Inf, in the

ARs. Why?

20

So why use Inft, not Inft?

21

Nonstationarity I: Trends
(SW Section 14.6)

22

Outline of discussion of trends in time

series data:

23

1. What is a trend?

24

25

26

What is a trend (contd)

27

Deterministic and stochastic trends

28

Deterministic and stochastic trends,

ctd.

29

Deterministic and stochastic trends,

ctd.

30

Deterministic and stochastic trends,

ctd.

31

Stochastic trends and unit

autoregressive roots

32

Unit roots in an AR(2)

33