Professional Documents
Culture Documents
L Dhliwayo
Department of Statistics
University of Zimbabwe
Contents
1 The Model Building Strategy 1
1.1 What this Unit is all About . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The Model Building Strategy . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Time Series and Stochastic Processes . . . . . . . . . . . . . . . . . . . . 2
i
Unit 1
The Model Building Strategy
Unit Objectives
Finding an appropiate model is not an easy task. We will develop a model building
strategy which was developed by Box and Jenkins in 1976.
The Box-Jenkins methodology consist of a four step iterative procedure.
Step 2. Estimation: historical data is used to estimate the parameters of the tentatively
identified model.
Step 3. Diagnostic Checking: various diagnostics are used to check the adequacy of
the tentatively identified model, and ifneed be, to suggest an improved model,
which is then regarded as a new tentatively identified model.
Step 4. Forecasting: once a final model is obtained, it is used to forecast future time
series values.
The following diagram depicts the steps in the Box-Jenkins iterative approach to models
building.
*** diagram
In the process of model selection, we shall try to adhere to the principle of parsimony:
Definition 1.1 (The Principle of Parsimony) The model used should require the
smallest possible number of parameters that will adequately represent the data.
1
2 The Model Building Strategy
Definition 1.2 (Mean Function) Let {Zt } be a stochatic process, then the mean
function is defined by
Note:
2. |ρk | ≤ 1
The acf is used to determine the moving average part for an ARMA model [to be
discussed latter].
The PACF is used to determine the autoregressive part for an ARMA model [to be
discussed latter].
Stationarity
Whenever we make statistical inference, it is important that the laws governing the
process do not change, that is, the process is in statistical equilibrium. This brings us
to the definition of stationarity.
γt,s = Cov(Zt , Zs )
= Cov(Zt−s , Z0 ) = Cov(Z0 , Zs−t )
(1.6)
= Cov(Z0 , Z|s−t| )
= γ0,|s−t|
Thus, the covariance depends on time only through the time difference k = |s − t| and
not on the actual times t and s. This means for a stationary process we can rewrite
above as
γk = Cov(Zt , Zt−k )
(1.7)
ρk = Corr(Zt , Zt−k )
4 The Model Building Strategy
Properties of γ
1. γ0 = V ar(Zt ), ρ0 = 1
2. γk = γ−k , ρk = ρ−k
3. |γk | ≤ γ0 , |ρk | ≤ 1
This means that if strictly stationary process has a finite variance, then the covariance
function must depend only on the time lag.
Definition 1.7 (Second Order Stationary) A time series {Zt } having finite sec-
ond moment is said to be second order stationary or weakly stationary if
1. E(Zt ) = µ
Note: From the above definition, we see that both the mean and the autocovariance
function do not depend on time.
Definition 1.8 (Ergodic) A time series {Zt } is ergodic in some parameter θ if an ap-
propriate parameter t̂heta constructed from a single realisation converges almost surely
to theta, that is to say, ensample averaging is equivalent to sample averaging in the
above sense.
Note: In this course the term stationary (without word strictly) means weakly station-
ary.