Are firms’ forecast errors driven by rational or irrational behaviour?

cube-no-animation-2

Are firms’ forecast errors driven by rational or irrational behaviour?

Topic:

Firms

By Alexandros Botsis, Christoph Görtz and Plutarchos Sakellaris

Expectations of economic actors are the quintessence of modern economic theory. Consumers, firms and policy makers take expectations about future developments into account when making economic decisions. For example, firms’ expectations about future sales and profits are important components in a businesses’ decision making on hiring and investment policies. If businesses use their information rationally, then their forecast errors should display no systematic behaviour or bias – they should be completely random. How firms form expectations and whether these are formed rationally is an open question. Our new research provides important insights. You can access our full ESCoE Discussion Paper here.

Do firms make errors in forecasting future sales that are systematic and not random? If so, what does that reveal about firm behaviour and the way they form expectations? We address these questions using a novel panel dataset that combines survey data on Greek manufacturing firms’ own sales forecasts, as well as corresponding balance-sheet data on realised sales. The study documents that only when firms make major forecast errors on sales growth, these are not completely random and behave in a systematic manner. Hence, firms’ forecasts violate what economists call the ‘Full Information Rational Expectations’ (FIRE) hypothesis. In our work, major forecast errors are defined as those that imply firms’ expected sales growth to be 14.3% (8.6%) higher (lower) than the subsequent realisation. In contrast, minor forecast errors do not violate the FIRE hypothesis. These differences in firms’ forecasting behaviour have not been documented before. It is an important finding as it documents that minor forecast errors are made under rational expectations while some irrational behaviour drives major forecast errors.

Key to demonstrating this empirical result is a novel methodology we developed that can extract quantitative forecasts when survey-based data on expectations is only available in categorical form. In other words, this methodology derives continuous forecasts on sales growth based on qualitative survey data. This is an important contribution as most surveys are qualitative in the sense that respondents only provide categorical answers that indicate a variable is either higher, unchanged or lower. This quantification methodology builds on the work of Pesaran (1987) and Smith and McAleer (1995) and extends it to retain the panel nature of the dataset, that is, it provides firm-level quantitative forecasts. We use higher-frequency (monthly) qualitative survey data on expected sales growth together with lower-frequency (annual) quantitative data on realised sales growth to estimate quantified expected annual sales growth.

The methodology can be applied to a wide range of applications and datasets and is not limited to quantifying firm forecasts. The only requirement is that the researcher has access to high-frequency categorical survey data on expectations, together with lower-frequency quantitative realisations of the corresponding variables. The methodology allows obtaining firm-level quantitative (continuous) expectations based on widely available qualitative survey responses.

We provide evidence of external validity and accuracy for our methodology in four ways. First, we show that the quantified estimates on sales growth expectations are fully consistent in terms of sign with the corresponding qualitative survey-based expectations. In a horse race, our methodology also substantially outperforms alternative models for obtaining quantified predictions. Second, we construct a small dataset of UK manufacturing firms that contains monthly qualitative survey expectations from the Confederation of British Industries (CBI) and the corresponding annual realisations from balance sheets, which allow use of the methodology to derive estimates for annual forecast errors. Importantly, for each firm, the dataset also includes annual quantitative survey expectations from ONS’ Management Practices Survey, which is employed as a benchmark. Comparing the estimated annual forecast errors with the directly observable benchmark forecast errors confirms the accuracy of our quantification methodology. Such an exercise can only be conducted using a dataset that includes quantitative forecasts, made by the same forecaster at a high and a lower frequency. In practice, data on quantitative firm-level expectations are rarely available, highlighting the need for, and value of, our quantification methodology, which allows researchers to utilise the large number of qualitative surveys to quantify expectations. Third, we perform a Monte Carlo exercise that provides a benchmark based on simulated data. We find forecast errors based on our methodology are highly accurate when compared with forecast errors based on the underlying artificial ‘true’ data. In a fourth exercise, we show that the main empirical results on the systematic behaviour of only the major forecast errors are also a feature of the observed qualitative survey data. This suggests that the key findings are not driven by the quantification methodology.

To explain our empirical findings, we provide a model of rational inattention. When operating in market environments where information processing is more costly, firms optimally limit their degree of attention to new information. Limited attention results in major forecast errors that are predictable and autocorrelated.

Read the full ESCoE Discussion Paper here.

Notes

Pesaran, M. H. (1987). The limits to rational expectations. Blackwell Publishers.

Smith, J. and McAleer, M. (1995). Alternative procedures for converting qualitative response data to quantitative expectations: an application to Australian manufacturing. Journal of Applied Econometrics, 10(2):165185.

Alexandros Botsis is a Research Economist at the University of Birmingham.
Christoph Görtz is a Senior Lecturer in Macroeconomics at the University of Birmingham.
Plutarchos Sakellaris is a Professor at the Athens University of Economics and Business.

ESCoE blogs are published to further debate. Any views expressed are solely those of the author(s) and so cannot be taken to represent those of the ESCoE, its partner institutions or the Office for National Statistics.

About the authors

Alex Botsis

Christoph Görtz

Plutarchos Sakellaris