Time Series Analysis (1)
Basic concepts and ARMA
(Chapter 5)
Main Points
(of univariate time series models)
Time series data
Stationary Process
Why important?
autocorrelation function (acf)
White noise and tests
ARMA Processes
Definitions
Behaviour of the acf and pacf
Building ARMA Models
Forecasting
SAS program
Univariate time series models
The nature of time series
A time series is an ordered set of observations of
random variables at different time frames (e.g.,
hour, daily, monthly, and annually, etc.)
A time series is also called a stochastic process
(“stochastic” is an synonym for random)
(used for modelling purpose)
Examples of financial time series
◦ Stock prices/returns
◦ Interest rates
◦ Exchange rates
An example of time series data
Univariate time series models
Univariate time series models are a class of
specifications
where one attempts to model and to predict
financial variables
using only information contained in their own
past values and possibly current and past
values of an error term
The family of Autoregressive Moving Average
(ARMA) models, or known as BoxJenkins (BJ)
methodology
Where we attempt to predict returns using only information
contained in their past values.
Some Notation and Concepts
A Strictly Stationary Process
A strictly stationary process is one where
i.e. the probability measure for the sequence {y
t
} is
the same as that for {y
t+m
} ¬ m.
Univariate Time Series Models
P y b y b P y b y b
t t n t m t m n
n n
{ ,..., } { ,..., }
1 1
1 1
s s = s s
+ +
Why do we need stationary process in financial analysis?
Strictly stationary:
Formula (5.1) of the textbook which means that
the probability distribution function, F, relative to
its last T periods history remain the same as
time progress (or over the whole sample time).
For example: Y
t
=F(Y
t1
, Y
t2
) (here T=2), which
means …… Y
5
, Y
4
, Y
3
, Y
2
, Y
1
, Y
0
,
Y
1
, Y
2
, Y
3
, Y
4
Y
3
, Y
2
=>Y
1
Y
2
, Y
1
=>Y
0
Y
1
, Y
0
=>Y
1
……
Stationary Process
Y
1
=F(Y
2
, Y
3
)
Y
0
=F(Y
1
, Y
2
)
Y
1
=F(Y
0
, Y
1
)
Y
2
=F(Y
1
, Y
0
)
The relation F remains the same
over three consecutive times.
A Weakly Stationary Process
If a series satisfies the next three equations, it is
said to be weakly or covariance stationary
1. E(y
t
) = µ , t = 1,2,...,·
2.
3. ¬ t
1
, t
2
Univariate Time Series Models
E y y
t t t t
( )( )
1 2 2 1
÷ ÷ =
÷
µ µ ¸
E y y
t t
( )( ) ÷ ÷ = < · µ µ o
2
A practical way of analysis
(Simplified and enough for practical applications)
Stationarity
A time series is said to be weakly stationary if its
mean and variance are constant over time and the
value of the covariance between the two time
periods depends on only the distance or gap or lag
between the two time periods and not the actual
time at which the covariance is computed
Why is stationarity so important?
Univariate time series models
Why is stationarity so important?
To model and to predict financial variables
Using information contained in their own past values
and possibly current and past values of an error term
The stationarity is a necessary
condition for the existence of such a
model
So if the process is covariance stationary, all the
variances are the same and all the covariances depend
on the difference between t
1
and t
2
. The moments
, s = 0,1,2, ...
are known as the covariance function.
These covariances, ¸
s
, are also known as
autocovariances (same variable, over a time period).
Univariate Time Series Models (cont’d)
E y E y y E y
t t t s t s s
( ( ))( ( )) ÷ ÷ =
+ +
¸
However, the value of the autocovariances depend on the units
of measurement of y
t
.
It is thus more convenient to use the autocorrelations which
are the autocovariances normalised by dividing by the
variance:
, s = 0,1,2, ...
If we plot t
s
against s=0,1,2,... then we obtain the
autocorrelation function or correlogram.
Univariate Time Series Models (cont’d)
t
¸
¸
s
s
=
0
A white noise process is one with (virtually) no
discernible structure. A definition of a white noise
process is
Thus the autocorrelation function will be zero apart
from a single peak of 1 at s = 0. t
s
~ approximately
N(0,1/T) where T = sample size
A White Noise Process
E y
Var y
if t r
otherwise
t
t
t r
( )
( )
=
=
=
=
¦
´
¹
÷
µ
o
¸
o
2
2
0
An example of stationary time series
A white noise process:
2
~ (0, )
t t t
Y IID = µ µ o
0 50 100 150 200 250 300 350 400 450 500
4
3
2
1
0
1
2
3
4
Time
u
(
t
)
white noise process
Why white noise process is important?
We are interested to find the relations (in models).
◦ One extreme is that we can identify a clear pattern of relation.
◦ The other extreme is that there is no relation at all (which is a
white noise process).
◦ It is based upon the condition of white noise process that we
can start to measure the relations.
A White Noise Process
Thus the autocorrelation function will be zero apart from a single peak
of 1 at s = 0. t
s
~ approximately N(0,1/T) where T = sample size
We can use this to do significance tests for the
autocorrelation coefficients by constructing a confidence
interval.
For example, a 95% confidence interval would be given by
[from N(0,1/T)]
If the sample autocorrelation coefficient, , falls outside
this region for any value of s, then we reject the null
hypothesis that the true value of the coefficient at lag s is
zero.
A White Noise Process
t
s
T
1
96 . 1 × ±
We can also test the joint hypothesis that all m of the t
k
correlation coefficients
are simultaneously equal to zero using the Qstatistic developed by Box and
Pierce:
where T = sample size, m = maximum lag length
The Qstatistic is asymptotically distributed as a .
However, the Box Pierce test has poor small sample properties, so a variant
has been developed, called the LjungBox statistic:
This statistic is very useful as a portmanteau (general) test of linear dependence
in time series.
Joint Hypothesis Tests (of white noise)
_
m
2
¿
=
=
m
k
k
T Q
1
2
t
( )
2
1
2
~ 2
m
m
k
k
k T
T T Q _
t
¿
=

÷
+ =
Question:
Suppose that a researcher had estimated the first 5 autocorrelation
coefficients using a series of length 100 observations, and found them to be
(from 1 to 5): 0.207, 0.013, 0.086, 0.005, 0.022.
Test each of the individual coefficient for significance, and use both the
BoxPierce and LjungBox tests to establish whether they are jointly
significant.
Solution:
A coefficient would be significant if it lies outside (19.6,+1.96) at the 5%
level, so only the first autocorrelation coefficient is significant.
Q=5.09 and Q*=5.26
Compared with a tabulated _
2
(5)=11.1 at the 5% level, so the 5 coefficients
are jointly insignificant.
An ACF Example
E views Example
Consider/run the following model for NZSE
,
= +
,−1
+
,
The e views output is as follows;
Selecting View/Residual Diagnostics/CorrelogramQ
statistics for the first 12 lags from this equation produces the
following view:
The third column gives us the Qstatistic and the 4. column
gives us the probability of the Q statistic.
Let u
t
(t=1,2,3,...) be a sequence of independently and identically
distributed (iid) random variables with E(u
t
)=0 and Var(u
t
)= , then
y
t
= µ + u
t
+ u
1
u
t1
+ u
2
u
t2
+ ... + u
q
u
tq
is a q
th
order moving average model MA(q).
Its properties are
E(y
t
)=µ; Var(y
t
) = ¸
0
= (1+ )o
2
Covariances
Moving Average Processes
o
c
2
u u u
1
2
2
2 2
+ + + ...
q
¦
¹
¦
´
¦
>
= + + + +
=
÷ + +
q s f or
q s f or
s q q s s s
s
0
,..., 2 , 1 ) ... (
2
2 2 1 1
o u u u u u u u
¸
1. Consider the following MA(2) process:
where c
t
is a zero mean white noise process with
variance .
(i) Calculate the mean and variance of X
t
(ii) Derive the autocorrelation function for this process (i.e.
express the autocorrelations, t
1
, t
2
, ... as functions of the
parameters u
1
and u
2
).
(iii) If u
1
= 0.5 and u
2
= 0.25, sketch the acf of X
t
.
Example of an MA Problem
2 2 1 1 ÷ ÷
+ + =
t t t t
u u u X u u
2
o
(i) If E(u
t
)=0, then E(u
ti
)=0 ¬ i.
So
E(X
t
) = E(u
t
+ u
1
u
t1
+ u
2
u
t2
)= E(u
t
)+ u
1
E(u
t1
)+ u
2
E(u
t2
)=0
Var(X
t
) = E[X
t
E(X
t
)][X
t
E(X
t
)]
but E(X
t
) = 0, so
Var(X
t
) = E[(X
t
)(X
t
)]
= E[(u
t
+ u
1
u
t1
+ u
2
u
t2
)(u
t
+ u
1
u
t1
+ u
2
u
t2
)]
= E[ +crossproducts]
But E[crossproducts]=0 since Cov(u
t
,u
ts
)=0 for s=0.
Solution
2
2
2
2
2
1
2
1
2
÷ ÷
+ +
t t t
u u u u u
Solution (cont’d)
So Var(X
t
) = ¸
0
= E [ ]
=
=
We have the autocovariances, now calculate the autocorrelations:
(iii) For u
1
= 0.5 and u
2
= 0.25, substituting these into the formulae
above gives t
1
= 0.476, t
2
= 0.190.
t
¸
¸
0
0
0
1 = =
t
¸
¸
3
3
0
0 = =
t
¸
¸
s
s
s = = ¬ >
0
0 2
) 1 (
) (
) 1 (
) (
2
2
2
1
2 1 1
2 2
2
2
1
2
2 1 1
0
1
1
u u
u u u
o u u
o u u u
¸
¸
t
+ +
+
=
+ +
+
= =
) 1 ( ) 1 (
) (
2
2
2
1
2
2 2
2
2
1
2
2
0
2
2
u u
u
o u u
o u
¸
¸
t
+ +
=
+ +
= =
2
2
2
2
2
1
2
1
2
÷ ÷
+ +
t t t
u u u u u
2 2
2
2 2
1
2
o u o u o + +
2 2
2
2
1
) 1 ( o u u + +
Thus the ACF plot will appear as follows:
ACF Plot
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5 6
s
a
c
f
An autoregressive model of order p, an AR(p) can be expressed as
Or using the lag operator notation:
Ly
t
= y
t1
L
i
y
t
= y
ti
or
or where
.
Autoregressive Processes
    ( ) ( ... ) L L L L
p
p
= ÷ + + 1
1 2
2
t p t p t t t
u y y y y + + + + + =
÷ ÷ ÷
   µ ...
2 2 1 1
¿
=
÷
+ + =
p
i
t i t i t
u y y
1
 µ
¿
=
+ + =
p
i
t t
i
i t
u y L y
1
 µ
t t
u y L + = µ  ) (
The condition for stationarity of a general AR(p) model is that the
roots of all lie outside the unit circle.
A stationary AR(p) model is required for it to have an MA(·)
representation.
Example 1: Is y
t
= y
t1
+ u
t
stationary?
The characteristic root is 1, so it is a unit root process (so non
stationary)
Example 2: Is y
t
= 3y
t1
 0.25y
t2
+ 0.75y
t3
+u
t
stationary?
The characteristic roots are 2, 1/3, and 2. Since only two of these lies
outside the unit circle, the process is nonstationary.
The Stationary Condition for an AR Model
1 0
1 2
2
÷ ÷ ÷ ÷ =    z z z
p
p
...
Measures the correlation between an observation k periods ago
and the current observation, after controlling for observations
at intermediate lags (i.e. all lags < k).
So t
kk
measures the correlation between y
t
and y
tk
after
removing the effects of y
tk+1
, y
tk+2
, …, y
t1
.
At lag 1, the acf = pacf always
At lag 2, t
22
= (t
2
t
1
2
) / (1t
1
2
)
For lags 3+, the formulae are more complex.
The Partial Autocorrelation Function (denoted t
kk
)
The pacf is useful for telling the difference between an AR process and an
ARMA process.
In the case of an AR(p), there are direct connections between y
t
and y
ts
only
for ss p.
So for an AR(p), the theoretical pacf will be zero after lag p.
In the case of an MA(q), this can be written as an AR(·), so there are
direct connections between y
t
and all its previous values.
For an MA(q), the theoretical pacf will be geometrically declining.
The Partial Autocorrelation Function (denoted t
kk
)
(cont’d)
By combining the AR(p) and MA(q) models, we can obtain an ARMA(p,q)
model:
where
and
or
with
ARMA Processes
    ( ) ... L L L L
p
p
= ÷ ÷ ÷ ÷ 1
1 2
2
q
q
L L L L u u u u + + + + = ... 1 ) (
2
2 1
t t
u L y L ) ( ) ( u µ  + =
t q t q t t p t p t t t
u u u u y y y y + + + + + + + + + =
÷ ÷ ÷ ÷ ÷ ÷
u u u    µ ... ...
2 2 1 1 2 2 1 1
s t u u E u E u E
s t t t
= = = = , 0 ) ( ; ) ( ; 0 ) (
2 2
o
AR(1): y
t
=a*y
t1
+μ
t
(1aL) y
t
= μ
t
and so (to a MV form)
y
t
= μ
t
/(1aL) = μ
t
+ (aL) μ
t
+(aL)
2
μ
t
……+(aL)
j
μ
t
+……
= μ
t
+ (a) μ
t1
+(a)
2
μ
t2
……+(a)
j
μ
t
j
+……
which is MA(∞)
MA(1): y
t
=μ
t
+b*μ
t1
y
t
= (1bL) μ
t
and similarly (to an AR form)
y
t
/(1bL) = μ
t
y
t
/(1bL) = y
t
+(bL) y
t1
+(bL) y
t1
……+(bL)
j1
μ
t
j
+……
So y
t
= μ
t
(b) y
t1
(b) y
t1
……(b)
j
μ
t
j
……
which is AR(∞)
ARMA Processes
t t
u L y L ) ( ) ( u µ  + =
Similar to the stationarity condition, we typically require the MA(q) part of
the model to have roots of u(z)=0 greater than one in absolute value.
The mean of an ARMA series is given by
The autocorrelation function for an ARMA process will display
combinations of behaviour derived from the AR and MA parts, but for lags
beyond q, the acf will simply be identical to the individual AR(p) model.
The Invertibility Condition
E y
t
p
( )
...
=
÷ ÷ ÷ ÷
µ
   1
1 2
An autoregressive process has
a geometrically decaying acf
number of spikes of pacf = AR order
A moving average process has
Number of spikes of acf = MA order
a geometrically decaying pacf
Summary of the Behaviour of the acf for
AR and MA Processes
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
The acf and pacf are not produced analytically from the relevant formulae for a model of that
type, but rather are estimated using 100,000 simulated observations with disturbances drawn
from a normal distribution.
ACF and PACF for an MA(1) Model: y
t
= – 0.5u
t1
+ u
t
Some sample acf and pacf plots
for standard processes
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0.05
1 2 3 4 5 6 7 8 9 10
Lag
a
c
f
a
n
d
p
a
c
f
acf
pacf
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
ACF and PACF for an MA(2) Model:
y
t
= 0.5u
t1

0.25u
t2
+ u
t
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
1 2 3 4 5 6 7 8 9 10
Lags
a
c
f
a
n
d
p
a
c
f
acf
pacf
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
Lags
a
c
f
a
n
d
p
a
c
f
acf
pacf
ACF and PACF for a slowly decaying AR(1) Model:
y
t
= 0.9y
t1
+ u
t
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
ACF and PACF for a more rapidly decaying AR(1) Model: y
t
= 0.5y
t1
+ u
t
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
1 2 3 4 5 6 7 8 9 10
Lags
a
c
f
a
n
d
p
a
c
f
acf
pacf
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
ACF and PACF for a more rapidly decaying AR(1) Model
with Negative Coefficient: y
t
= 0.5y
t1
+ u
t
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
1 2 3 4 5 6 7 8 9 10
Lags
a
c
f
a
n
d
p
a
c
f
acf
pacf
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
ACF and PACF for a Nonstationary Model
(i.e. a unit coefficient): y
t
= y
t1
+ u
t
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
Lags
a
c
f
a
n
d
p
a
c
f
acf
pacf
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
ACF and PACF for an ARMA(1,1):
y
t
= 0.5y
t1
+ 0.5u
t1
+ u
t
0.4
0.2
0
0.2
0.4
0.6
0.8
1 2 3 4 5 6 7 8 9 10
Lags
a
c
f
a
n
d
p
a
c
f
acf
pacf
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Box and Jenkins (1970) were the first to approach the task of estimating an
ARMA model in a systematic manner. There are 3 steps to their approach:
1. Identification
2. Estimation
3. Model diagnostic checking
Step 1:
 Involves determining the order of the model.
 Use of graphical procedures
 A better procedure is now available
Building ARMA Models
 The Box Jenkins Approach
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Step 2:
 Estimation of the parameters
 Can be done using least squares or maximum likelihood depending
on the
model.
Step 3:
 Model checking
Box and Jenkins suggest 2 methods:
 deliberate overfitting
 residual diagnostics
Building ARMA Models
 The Box Jenkins Approach (cont’d)
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Identification would typically not be done using acf’s.
We want to form a parsimonious model.
Reasons:
 variance of estimators is inversely proportional to the number of degrees of
freedom.
 models which are profligate might be inclined to fit to data specific features
This gives motivation for using information criteria, which embody 2 factors
 a term which is a function of the RSS
 some penalty for adding extra parameters
The object is to choose the number of parameters which minimises the
information criterion.
Some More Recent Developments in
ARMA Modelling
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
The information criteria vary according to how stiff the penalty term is.
The three most popular criteria are Akaike’s (1974) information criterion
(AIC), Schwarz’s (1978) Bayesian information criterion (SBIC), and the
HannanQuinn criterion (HQIC).
where k = p + q + 1, T = sample size. So we min. IC s.t.
SBIC embodies a stiffer penalty term than AIC.
Which IC should be preferred if they suggest different model orders?
◦ SBIC is strongly consistent but (inefficient).
◦ AIC is not consistent, and will typically pick
“bigger” models.
Information Criteria for Model Selection
AIC k T = + ln(
) / o
2
2
p p q q s s ,
T
T
k
SBIC ln ) ˆ ln(
2
+ = o
)) ln(ln(
2
) ˆ ln(
2
T
T
k
HQIC + = o
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
As distinct from ARMA models. The I stands for integrated.
An integrated autoregressive process is one with a characteristic
root on the unit circle.
Typically researchers difference the variable as necessary and then
build an ARMA model on those differenced variables.
An ARMA(p,q) model in the variable differenced d times is
equivalent to an ARIMA(p,d,q) model on the original data.
ARIMA Models
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Forecasting = prediction.
An important test of the adequacy of a model. e.g.
 Forecasting tomorrow’s return on a particular share
 Forecasting the price of a house given its characteristics
 Forecasting the riskiness of a portfolio over the next year
 Forecasting the volatility of bond returns
We can distinguish two approaches:
 Econometric (structural) forecasting
 Time series forecasting
The distinction between the two types is somewhat blurred (e.g, VARs).
Forecasting in Econometrics
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Expect the “forecast” of the model to be good insample.
Say we have some data  e.g. monthly FTSE returns for 120 months: 1990M1 –
1999M12. We could use all of it to build the model, or keep some observations
back:
A good test of the model since we have not used the information from
1999M1 onwards when we estimated the model parameters.
InSample Versus OutofSample
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Models for Forecasting
Structural models
e.g. y = X + u
To forecast y, we require the conditional expectation of its future
value:
=
But what are , etc.? We could use , so
=
t kt k t t
u x x y + + + + =   
2 2 1
( ) ( )
t kt k t t t
u x x E y E + + + + = O
÷
  
2 2 1 1
( ) ( )
kt k t
x E x E    + + +
2 2 1
( )
k k t
x x y E    + + + =
2 2 1
y
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Models for Forecasting (cont’d)
Time Series Models
The current value of a series, y
t
, is modelled as a function only of its previous
values and the current value of an error term (and possibly previous values of the
error term).
Models include:
• simple unweighted averages
• exponentially weighted averages
• ARIMA models
• Nonlinear models – e.g. threshold models, GARCH,
bilinear models, etc.
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
The forecasting model typically used is of the form:
where f
t,s
= y
t+s
, ss 0; u
t+s
= 0, s > 0
= u
t+s
, s s 0
Forecasting with ARMA Models
¿ ¿
=
÷ +
=
÷
+ + =
q
j
j s t j
p
i
i s t i s t
u f f
1 1
, ,
u  µ
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
An MA(q) only has memory of q.
e.g. say we have estimated an MA(3) model:
y
t
= µ + u
1
u
t1
+ u
2
u
t2
+ u
3
u
t3
+ u
t
y
t+1
= µ + u
1
u
t
+ u
2
u
t1
+ u
3
u
t2
+ u
t+1
y
t+2
= µ + u
1
u
t+1
+ u
2
u
t
+ u
3
u
t1
+ u
t+2
y
t+3
= µ + u
1
u
t+2
+ u
2
u
t+1
+ u
3
u
t
+ u
t+3
We are at time t and we want to forecast 1,2,..., s steps ahead.
We know y
t
, y
t1
, ..., and u
t
, u
t1
Forecasting with MA Models
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
f
t, 1
= E(y
t+1 , t
) = E(µ + u
1
u
t
+ u
2
u
t1
+ u
3
u
t2
+ u
t+1
)
= µ + u
1
u
t
+ u
2
u
t1
+ u
3
u
t2
f
t, 2
= E(y
t+2 , t
) = E(µ + u
1
u
t+1
+ u
2
u
t
+ u
3
u
t1
+ u
t+2
)
= µ + u
2
u
t
+ u
3
u
t1
f
t, 3
= E(y
t+3 , t
) = E(µ + u
1
u
t+2
+ u
2
u
t+1
+ u
3
u
t
+ u
t+3
)
= µ + u
3
u
t
f
t, 4
= E(y
t+4 , t
) = µ
f
t, s
= E(y
t+s , t
) = µ
¬ s > 4
Forecasting with MA Models (cont’d)
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Say we have estimated an AR(2)
y
t
= µ + 
1
y
t1
+ 
2
y
t2
+ u
t
y
t+1
= µ + 
1
y
t
+ 
2
y
t1
+ u
t+1
y
t+2
= µ + 
1
y
t+1
+ 
2
y
t
+ u
t+2
y
t+3
= µ + 
1
y
t+2
+ 
2
y
t+1
+ u
t+3
f
t, 1
= E(y
t+1 , t
) = E(µ + 
1
y
t
+ 
2
y
t1
+ u
t+1
)
= µ + 
1
E(y
t
) + 
2
E(y
t1
)
= µ + 
1
y
t
+ 
2
y
t1
f
t, 2
= E(y
t+2 , t
) = E(µ + 
1
y
t+1
+ 
2
y
t
+ u
t+2
)
= µ + 
1
E(y
t+1
) + 
2
E(y
t
)
= µ + 
1
f
t, 1
+ 
2
y
t
Forecasting with AR Models
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
f
t, 3
= E(y
t+3 , t
) = E(µ + 
1
y
t+2
+ 
2
y
t+1
+ u
t+3
)
= µ + 
1
E(y
t+2
) + 
2
E(y
t+1
)
= µ + 
1
f
t, 2
+ 
2
f
t, 1
We can see immediately that
f
t, 4
= µ + 
1
f
t, 3
+ 
2
f
t, 2
etc., so
f
t, s
= µ + 
1
f
t, s1
+ 
2
f
t, s2
Can easily generate ARMA(p,q) forecasts in the same way.
Forecasting with AR Models (cont’d)
‘Introductory
Econometrics for Finance’
© Chris Brooks 2008
Statistical Versus Economic or Financial loss functions
Statistical evaluation metrics may not be appropriate.
How well does the forecast perform in doing the job we wanted it for?
Limits of forecasting: What can and cannot be forecast?
All statistical forecasting models are essentially extrapolative
Forecasting models are prone to break down around turning points
Series subject to structural changes or regime shifts cannot be forecast
Predictive accuracy usually declines with forecasting horizon
Forecasting is not a substitute for judgement
SAS programs provided
(in the Stream web page)
The SAS program for the UKHP example is
available: UKHP.SAS.
(Data: UKHP1.xls)