lec%2010-2003

# lec%2010-2003 - ACTL2003 Week 10 Time Series - An...

This preview shows pages 1–7. Sign up to view the full content.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ACTL2003 Week 10 Time Series - An Introduction I Residual analysis I Non-stationarity I ARIMA I SARIMA I Unit root test I Conintegrated processes I Markov property I Simulation Readings: Chan, Chapters 4 and 8 Reference: ACTED Chapter 12 and 13 CT6 1/40 Model building For a time series, model building can be classified into three stages: 1. Model selection (choosing ARIMA) 2. Model identification (estimation of parameters) 3. Model checking (diagnostic) (This Week) 2/40 Model checking-Residual Analysis I Principle : The residuals Z t = X t- X t will form a good approximation to a white noise process, if the proposed model X t is a good approximation to the underlying time series process. Here, X t are the fitted values, which are computed using the estimated parameters. I Frequently used checks to see whether a set of data is a likely realisation of a white noise process. I Plot the residuals Z t against t : any evident pattern, whether in the average level of the residuals (trend) or in the magnitude of the fluctuations about 0, means that the model is inadequate. 3/40 Frequently used checks to see whether a set of data is a likely realisation of a white noise process (continued): I Plot the sample ACF (SACF) of Z t I Rationale: Under the null hypothesis that Z t WN (0 , 1), it is can be shown that the ACF of { Z t } Z ( h ) N (0 , 1 n ) . So the confidence intervals at significance 95% of the sample autocorrelations are 0 1 . 96 n . (Recall that for white noise process, the ACF is 1 at lag 0, and 0 at other lags.) I As a rule of thumb, one can use 2 n to determine if the SACF is significantly different from zero . (Note: only short lags of SACF are meaningful.) These bounds are usually provided automatically by the estimation software (such as Minitab, R, S-Plus). I : If too many values of SACF are outside the range (- 2 n , + 2 n ), we can conclude that the fitted model does not fit well (more parameters will be needed). 4/40 Frequently used checks to see whether a set of data is a likely realisation of a white noise process (continued): Portmanteau Test : I Portmanteau statistic: Q = n ( n + 2) h X j =1 2 Z ( j ) n- j . I Theorem: Under the null hypothesis that Z t WN (0 , 1), the Portmanteau statistic Q = n ( n +2) h X j =1 2 Z ( j ) n- j 2 ( n- p- q ) approximately for large n , Note: In practice, h is chosen between 15 and 30, and n should be large, say n 100. I Reject the null if Q &gt; 2 1- ( n- p- q ). 5/40 Non-stationarity In practice, different kinds of non-stationarity are often encountered. Roughly speaking, a non-stationary time series may exhibit non-stationarity in the level of mean, variance, or both....
View Full Document

## This note was uploaded on 06/12/2011 for the course ASB 2003 taught by Professor Kim during the Three '11 term at University of New South Wales.

### Page1 / 33

lec%2010-2003 - ACTL2003 Week 10 Time Series - An...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online