Futureskills Scotland Expert Briefing: Uses and Abuses of Forecasting

The purpose of the Expert Briefings Series is to make available to our colleagues in Scotland the knowledge and experience of people who are expert in their fields. The series covers issues concerning the labour market, education and training and their links to the economy.

Each briefing involves an invited expert providing a personal briefing to an invited audience under Chatham House Rules and the publication of the briefing paper. In providing this service, Futureskills Scotland is pursuing two of its aims:

• to improve the availability, quality and consistency of labour market information; and
• to analyse the Scottish labour market to inform policy making.

The views expressed in the briefing papers are those of the authors and not necessarily those of Futureskills Scotland.

This report examines the different methods by which forecasts can be prepared, the risks to forecasting and how to interpret the results. It also examines the potential for preparing forecasts for employment in Scotland. Forecasts are not just about how we expect the world to be but also inform how we might change it. It is important to understand the difference and the  ability to form any views about what will or might or could happen.

Forecasts need not require statistical analysis of past data. The use of expert judgement is common in forecasting, both where no data exists and where it is judged that data may not tell the whole story. Both expert and group judgement can be shown to produce accurate and useful results in some circumstances. However, most economic forecasts are based on statistical analysis of some form. First, an assessment of the accuracy of any data is needed, and particularly how consistent it is over time. Economic data is quite imperfect and measurement error may indeed be increasing as more variability in kinds of output or types of employment occur.

Second, not all data series can be forecast. If there is a lot of random noise in the series, it may be difficult to extract the movement which has some pattern. This applies whether or not it is clear what ought to drive movement in the series. Understanding the drivers of a system may not help much in forecasting it. Many complex and non-linear systems can be quite simple to describe and give straightforward behaviour on average but cannot be predicted from day to day. Alternatively, short term predictability may collapse as time passes. Finally, even if the system has a straightforward relationship between its drivers and the variable of interest - for example output drives employment - this pushes the problem back to forecasting output and in turn the drivers of output and so on and so on.

Interpreting the results of any forecasting exercise means being aware of these limitations and how important they are. Judging the performance of a forecast can be tricky if it is used as a guide to policy making. The purpose of the policy will be to adjust the existing parameters of the system, by providing more training for example, which will change the outcome in comparison with what was previously expected.

In an uncertain system, looking at ranges of potential outcomes can be helpful. The Bank of England does this in preparing its fan charts. This approach stresses that several possibilities are inherent in the existing data. This can be distinguished from scenarios in which a more detailed story of how policy makers react to particular shocks ought to be included.

Employment in Scotland is not measured with certainty - different measurement systems have produced different results. A  merged series can be created, and does show an upward trend. However an examination of the rate of change shows great variability. This suggests that levels are fairly stable but change can go in either direction in any year. It will require judgement to address the extent to which stable trends could be affected by new forms of structural change which are not already embedded in the system. Year to year changes will much more difficult to model at any level.

	Download
Full Report (PDF 305Kb) The reports are presented as Adobe Acrobat PDF files. You may acquire Acrobat Reader to view these files. This software is free.