Advanced Search

Nobel Prize Winners

2003: Robert Engle & Clive Granger

This is an occasional series introducing the latest in thinking in economics by looking at the work of Nobel Prize winners in Economics. This starts with the 2003 winners, Robert F. Engle and Clive W.J. Granger who were joint winners of the Nobel Prize in economics for their work in developing the understanding of time series analysis in economic modelling.

Robert Engle is currently Professor of Finance, Michael Armellino Professorship in the Management of Financial Services at Stern School of Business, New York University. He is a citizen of the United States and has research interests in Financial Econometrics, Time Series Analysis, Volatility and Risk Management and Empirical Market Microstructure.

Image Source: Reproduced with kind permission of Robert F. Engle (http://weber.ucsd.edu/~mbacci/engle)

Clive Granger was born in Wales and educated at Nottingham University where he gained his PhD in 1959. He currently works at the Department of Economics, University of California, San Diego. His research interests include statistics and econometrics, especially time-series analysis, forecasting, finance, demographics and methodology.

Image Source: Copyright The Nobel Foundation (http://www.nobel.se). Reproduced with kind permission.

Area of Research: Autoregressive Conditional Heteroskedasticity (ARCH) and Cointegration

UK government investment decisions in a wide range of activities, for example the NHS, are based on medium/long term economic growth rates. Forecasts of these growth rates are vitally important in planning the likely revenue the government might receive from taxes and thus allows the government to plan how much, if any, they will have to borrow to finance those spending plans.

A key part of any economic research therefore is dealing with data. Much of this data is what is called 'Time-Series' data; this simply means information or variables that change over a period of time. Such data might include GDP, retail prices, interest rates, share prices and so on. When looking at such data, economists might be interested in developing hypotheses about the behaviour of the variables being analysed. Such hypotheses can then be used to assess relationships between variables and contribute to policy and decision-making.

Looking at any time-series data, it is clear that there is often significant volatility in the relationship between the variables. You can get further information about time-series analysis in our data section. Economists have traditionally looked at such data using what are called 'stochastic' methods. What this means is that over time, some random process has generated the data being investigated. If researchers can identify the nature of that randomness then it can be used to infer something about what might happen in the future.

In addition, the volatility present in much time-series data presents more problems. Some years, the variables concerned may be subject to rapid and violent changes whereas in other years such volatility can be much less violent. Is there some way in which we could predict periods of volatility in future time series data, understanding or identifying underlying trends thus allowing us to be able to make decisions and assess risk in the future?

Much of the data handled by economists consisted of what is called non-stationary series. This means that the variables are generated randomly and are not subject to returning to some fixed value. We might expect GDP, for example, to fluctuate between -2% and +5% over a period of 50 years but not that it would, of necessity, eventually return to 2%. The main problem therefore is how to apply statistical techniques to analyse such data and get results that are accurate and reliable such that we can base decision-making or predictions on it.

Engle developed the concept of 'autoregressive conditional heteroskedasticity' (ARCH) which enabled economists to understand the nature of volatility in time series data. The assumption had been that volatility was 'constant' over a period of time; in other words it might be possible to identify patterns of volatility over time.

Engle showed that this could not be assumed and his work allowed researchers to better understand the nature of volatility so that their results would be more accurate and reliable.

Granger identified a phenomenon he termed 'cointegration'. He observed that much data analysis in economics was based on the assumption that the variables were non-stationary. (We could usefully think of this as not returning to any equilibrium). The normal method of analysing such data was to use linear regression. This technique involves averaging out variations in the relationship between variables to give a straight line. For example, it is unlikely that if we collected data showing the relationship between interest rates and exchange rates that we would get a nice neat correlation between the two - a straight line! Linear regression averages out the variations to give that straight line showing the relationship between the two - a line of best fit.

Granger suggested that the underlying assumption of non-stationary variables might lead to results that were misleading because some of the variables might exhibit stationary characteristics. In other words, it might be possible with some data that is randomly generated to exhibit a tendency to return to an equilibrium position or to fluctuate around a given value. Economic growth for example might vary significantly over time showing periods of high positive growth and periods of negative growth but GDP might eventually return (although not naturally) to a mean rate consistent with a country's potential growth rate - say 2%.

Granger therefore developed statistical systems that could be used to take account of these possibilities in time series data and help to generate results that were, again, more reliable and accurate. Prior to his work, economists may have generated data that suggested some significant relationship between two variables when in fact that relationship may not have been that strong at all!

If we refer back to our starting paragraph relating to government investment and planning, if the projections are wrong (because the statistical method was wrong) then the Government are making very poor/inefficient decisions that have an impact on resource allocation both today and in the future. Therefore, this work is making a contribution as it helps us better understand the how the world works, and helps us tackle the economic problem.

Biz/ed wishes to extend our thanks to Professor Robert Engle for his comments on the above article.