‘Modelling non-stationary economic time series’

1
Book review Simon P. Burke & John Hunter, Modelling non- stationary economic time series: A multivariate approach, Palgrave texts in Econometrics, Mac- millan, 2005, ISBN: 1-4039-0202-X (cloth) or 1- 4039-0203-8 (pbk), 253 pp., £19.99 Along with many others, I have always found it difficult to construct multivariate models for time series data. The seminal paper by Box and Newbold (1971) showed that even eminent statisticians like Sir Maurice Kendall could be fooled by spuriously large cross- correlations contaminated by autocorrelations within the individual series. I am also aware that statisticians and economists have often worked with little knowl- edge of what each other is doing (with occasional exceptions such as Clive Granger). This is a book written by economists for economists and so it is not obvious that a statistician like me should be reviewing it. Yet, it is only by cross-fertilization that barriers can be broken down. As I rather expected, this book refers to much econometric research that is unfamiliar to me. I also notice the absence of references to statistical work that I would regard as fundamental (e.g. Reinsel’s (1997) book and the work of George Tiao and Ruey Tsay, the latter represented by just one joint paper). The book concentrates on the search for co-integration and for exogeneity, often guided by economic theory. VAR and VARMA models are employed but there is no mention of statistical tools such as the cross-correlation function and the cross-spectrum. Readers of this journal will want to know that there is rather little emphasis explicitly on forecasting, apart from Section 6.3 on forecasting cointegrated systems. They should also be aware that there is no reference whatsoever to research that has appeared in the Journal of Forecasting or the International Journal of Forecasting. There are no examples that take the reader right through from the initial examination of real multivariate data through the modelling process and onto forecasting, but rather the examples use data published elsewhere to illustrate particular tests or procedures. One bivariate simulated sample in Chapter 3 is used to illustrate cointegration. The text has not changed my view that multivariate modelling is hard! I was also somewhat disturbed by apparent flaws in the text that hinder understanding. For example (concentrating on Chapter 2, Univariate Time Series), I spotted non-trivial typos in equations (2.9b), (2.9c), (2.11a), (2.24), the sentence after (2.10b), the expan- sion of Dd on page 31, and in the legends for Fig. 2.1 (surely T =124) and 2.4. More generally, the graphics software used in the book generally labels scales in rather peculiar ways, and the drawing in Fig. 2.7 appears to have been erroneously copied from Fig. 2.6, albeit with a different legend. As regards the text, I found it difficult to follow in places and I was particularly concerned to read on page 17 that, for an MA(2) process, dthe autocovariances depend not only on the time gap, but on the time itself. The process is therefore stationary... .T!! No doubt these are all slips of the pen that can be corrected in any later printings, but they make it difficult to recommend the book in its present form. References Box, G.E.P., & Newbold, P. (1971). Some comments on a paper of Coen, Gomme and Kendall. Journal of the Royal Statistical Society, Series A, 134, 229 – 240. Reinsel, G. C. (1997). Elements of multivariate time series analysis (2nd edn.). New York7 Springer-Verlag. Chris Chatfield Department of Mathematical Sciences, University of Bath, Bath, BA2 7AY, United Kingdom E-mail address: [email protected]. doi:10.1016/j.ijforecast.2005.12.003 International Journal of Forecasting 22 (2006) 819 www.elsevier.com/locate/ijforecast

Transcript of ‘Modelling non-stationary economic time series’

www.elsevier.com/locate/ijforecast

International Journal of For

Book review

Simon P. Burke & John Hunter, Modelling non-

stationary economic time series: A multivariate

approach, Palgrave texts in Econometrics, Mac-

millan, 2005, ISBN: 1-4039-0202-X (cloth) or 1-

4039-0203-8 (pbk), 253 pp., £19.99

Along with many others, I have always found it

difficult to construct multivariate models for time series

data. The seminal paper by Box and Newbold (1971)

showed that even eminent statisticians like Sir Maurice

Kendall could be fooled by spuriously large cross-

correlations contaminated by autocorrelations within

the individual series. I am also aware that statisticians

and economists have often worked with little knowl-

edge of what each other is doing (with occasional

exceptions such as Clive Granger). This is a book

written by economists for economists and so it is not

obvious that a statistician like me should be reviewing

it. Yet, it is only by cross-fertilization that barriers can

be broken down. As I rather expected, this book refers

to much econometric research that is unfamiliar to me. I

also notice the absence of references to statistical work

that I would regard as fundamental (e.g. Reinsel’s

(1997) book and the work of George Tiao and Ruey

Tsay, the latter represented by just one joint paper). The

book concentrates on the search for co-integration and

for exogeneity, often guided by economic theory. VAR

and VARMA models are employed but there is no

mention of statistical tools such as the cross-correlation

function and the cross-spectrum. Readers of this

journal will want to know that there is rather little

emphasis explicitly on forecasting, apart from Section

6.3 on forecasting cointegrated systems. They should

also be aware that there is no reference whatsoever to

research that has appeared in the Journal of Forecasting

or the International Journal of Forecasting. There are no

examples that take the reader right through from the

initial examination of real multivariate data through the

doi:10.1016/j.ijforecast.2005.12.003

modelling process and onto forecasting, but rather the

examples use data published elsewhere to illustrate

particular tests or procedures. One bivariate simulated

sample in Chapter 3 is used to illustrate cointegration.

The text has not changed my view that multivariate

modelling is hard!

I was also somewhat disturbed by apparent flaws in

the text that hinder understanding. For example

(concentrating on Chapter 2, Univariate Time Series),

I spotted non-trivial typos in equations (2.9b), (2.9c),

(2.11a), (2.24), the sentence after (2.10b), the expan-

sion of Dd on page 31, and in the legends for Fig. 2.1

(surely T=124) and 2.4. More generally, the graphics

software used in the book generally labels scales in

rather peculiar ways, and the drawing in Fig. 2.7

appears to have been erroneously copied from Fig.

2.6, albeit with a different legend. As regards the text,

I found it difficult to follow in places and I was

particularly concerned to read on page 17 that, for an

MA(2) process, dthe autocovariances depend not only

on the time gap, but on the time itself. The process is

therefore stationary. . ..T!! No doubt these are all slips

of the pen that can be corrected in any later printings,

but they make it difficult to recommend the book in its

present form.

References

Box, G.E.P., & Newbold, P. (1971). Some comments on a paper of

Coen, Gomme and Kendall. Journal of the Royal Statistical

Society, Series A, 134, 229–240.

Reinsel, G. C. (1997). Elements of multivariate time series analysis

(2nd edn.). New York7 Springer-Verlag.

Chris Chatfield

Department of Mathematical Sciences,

University of Bath, Bath, BA2 7AY, United Kingdom

E-mai l address: c [email protected] h.ac.uk.

ecasting 22 (2006) 819