Economic forecasting models help central bank policy-makers to make decisions and measure whether their policies have had the effect they intended. Developing these models is a complex job, and their predictive value is challenged by the fact that economic activity changes constantly. Professor George Kapetanios has been working with analysts at the Bank of England and the European Central Bank (ECB) to produce new methods to deal with the problems of ‘time variation’ in economic data, improving the banks’ analytical capability and leading to more robust policy decisions.

As soon as an economic measurement is captured and fed into an economic model, it is already out of date. Over time, the individual industries, occupations or types of transaction being measured may become more or less important than they were when the model was designed. There is also a constant interaction between the responses of businesses, individuals and government. As a result, the measurements we use to help us to interpret data on particular type of economic activity, such as the typical variation we would expect to see within the dataset, change too.

At times of such rapid technological and political change as we have experienced recently, significant economic shifts can occur in a relatively short space of time.

To ensure that their forecasts keep pace, economists use rolling averages rather than the average over a whole period. The challenge with this approach is that sometimes change is unpredictable, and even random. Economists worry that if a model is too rigid, forecasts may be inaccurate because they become distorted by a single random change.

Professor George Kapetanios has been working with analysts at the Bank of England and the ECB to produce new methods of using rolling averages. The models developed by Professor Kapetanios and his collaborators enable policy makers to see how shocks from outside a particular country’s economy, or which are non-financial in nature, can drive change. They achieve this through the use of ‘non-parametric estimation techniques’ that do not assume that different elements of the economic picture move within set parameters. This enables the incorporation of realistic assumptions about the sources of change in the economy.

Brexit has thrown up important examples of the effectiveness of these new techniques. For example, the techniques helped the Bank of England to build its understanding of how the sudden exchange-rate movements prompted first by the outcome of the referendum in June 2016, and then by the political turmoil over the final Brexit deal in September 2019 would affect import prices and feed into overall inflation.

Professor Kapetanios’s work can also be applied in conjunction with the new data sources and techniques being used by central banks, such as very large data sets and machine learning models. This means that is likely to play a key part in the evolution of forecasting activities for some time to come.