Speaking Volumes: Time Series Analysis: Forecasting and Control

Denise Osborn on George E. P. Box and Gwilym Jenkins's Time Series Analysis: Forecasting and Control .

I first heard of Box and Jenkins in about 1971 as a graduate student in economic statistics at the University of Sydney. Their book had only recently been published but it was already clear it would mark a watershed by demystifying the statistical analysis of time series data and making the methods more applicable in practice. But I failed to understand the delightful simplicity of their approach.

In October 1972 I arrived at the London School of Economics as a research student. One of the first things I was set to read was Box and Jenkins. Suddenly, what had seemed abstract and complicated was revealed as practical and straightforward. My doctoral thesis examined one aspect of how their methodology might be applied in econometrics, the idea being that economics would provide the static relationships between variables while statistical time series would provide the dynamics. Two decades later, I believe the marriage of economics and time series analysis needs to take place at a much more fundamental level than the bolting together we explored. Nevertheless, Box and Jenkins played a vital role in shaping my agenda and in promoting the study of dynamics in economics.

By the early 1970s econometric models had been constructed for a number of countries using time series data. They were, however, largely static, with response to a change assumed to take place within one period, irrespective of whether it was a year or a quarter. The time series models of Box and Jenkins stood in stark contrast: they took a single variable in isolation and dynamics played the central role. A few studies compared the two approaches and virtually all concluded that univariate time series models provided the more accurate short-term forecasts. This was a turning point because it implied that dynamics were more important for understanding short-run movements than economic relationships as then understood. The emphasis in time series econometrics therefore shifted from modelling large simultaneous systems to taking account of dynamic interactions.

Box and Jenkins were influential not so much because what they said was new, but because they said it well. Their main contribution was to show how the dynamic properties of observed series could be matched to those of theoretical models. Their book advocates parsimony to produce simple empirical approximations that capture the dynamic relationships revealed by observed data. I still find this approach persuasive.

Denise Osborn is professor of econometrics and director of research, school of economic studies, University of Manchester.