Do US Macroeconomic Forecasters Exaggerate their Differences?

• Author: Michael P. Clements
• DOI: http://doi.org/10.1002/for.2358
• Published date: 01 December 2015
• Date: 01 December 2015

Journal of Forecasting,J. Forecast. 34, 649–660 (2015)

Published online 18 August 2015 in Wiley Online Library (wileyonlinelibrary.com)DOI: 10.1002/for.2358

Do US Macroeconomic Forecasters Exaggerate

their Differences?

MICHAEL P. CLEMENTS

ICMA Centre, Henley Business School, University of Reading, UK

ABSTRACT

Application of the Bernhardt et al. (Journal of Financial Economics 2006; 80(3): 657–675) test of herding to the

calendar-year annual output growth and inflation forecasts suggests forecasters tend to exaggerate their differences,

except at the shortest horizon, when they tend to herd. We consider whether these types of behaviour can help to

explain the puzzle that professional forecasters sometimes make point predictions and histogram forecasts which are

mutually inconsistent. Copyright © 2015 John Wiley & Sons, Ltd.

KEY WORDS macro-forecasting; imitative behaviour; histogram forecasts; point predictions

INTRODUCTION

It is generally understood that economic forecasters may have incentives to act strategically in the sense of seeking

to enhance their reputations (see, among others, Lamont, 2002; Ehrbeck and Waldmann, 1996; Laster et al., 1999;

Ottaviani and Sorensen, 2006). For example, Lamont (2002, p. 265) notes that besides minimizing squared forecast

errors agents may set their forecasts to ‘optimize profits or wages, credibility, shock value, marketability, political

power...’.Muchoftheempirical literature on testing the influence of factors other than accuracy on the determi-

nation of agents’ forecasts rests on the notion of herding—whether forecasters take into account the views of others

when they produce their forecasts. This may be manifest in forecasters skewing their optimal forecast towards the

consensus view, or artificially exaggerating the difference between their forecast and the consensus, where optimal

is to be understood in the narrow sense of maximizing the expected accuracy of the forecast (e.g. minimizing the

expected squared forecast error). Indeed, Lamont (2002) supposes that forecasters’ actual loss functions may contain

terms in the difference between the forecast and the consensus, as well as conventional accuracy measures such as

the absolute forecast error.

The focus of this paper is whether the respondents to the US Survey of Professional Forecasters (US-SPF) take into

account the consensus view when they issue their forecasts of US output growth and inflation. Compared to studies

such as Lamont (2002), where the individual forecasters are identified and their track performance is public knowl-

edge, the SPF respondents are anonymous (although each respondent has a unique identifier, so that individuals can

be tracked over time). One might suspect this would weaken the extent to which the forecasters behave strategically,

but alternatively, if the respondents report the same forecasts to the SPF as they make public through other spheres,

these issues remain pertinent. We regard it as a matter that can only be determined by an empirical study. We wish to

discover whether herding behaviour depends on the forecast horizon. Forecasters may behave differently when pro-

viding their expectations of relatively distant events compared to short-horizon forecasts. In our study the forecast

horizons span one-quarter to 1-year-ahead forecasts. As well as exploring forecast behaviour across different hori-

zons, we also explore behaviour by type of forecaster and assess whether forecasters working in firms characterized

as financial service providers systematically differ from those in non-financial service firms.

The behaviour of the forecasters that take part in the US-SPF is of interest in itself as part of the endeavour to

better understand the actual expectations formation process of economic agents, given the pre-eminence of the US-

SPF.1But, in addition, of particular interest is whether strategicbehaviour is responsible in part for the inconsistencies

between the respondents’ reported probability distributions and point predictions, as documented by Engelberg et al.

(2009) and Clements (2009, 2010).

Testing for herding is not straightforward, as individuals’ forecasts will tend to cluster together to the extent that

they share the same information, irrespective of whether they consider the consensus view when they form their

forecasts. There have been a number of influential papers on herding. Batchelor and Dua (1992) find that forecasters

Correspondence to: Michael P. Clements, ICMA Centre, Henley Business School, University of Reading, Reading RG6 6BA, UK. E-mail:

m.p.clements@reading.ac.uk

1The US SPF has been an important source of data for research on expectations formation. As of 21 April 2014, the Academic Bibliography main-

tained by the Philadelphia Fed listed 79 research papers based on the SPF forecast data: see http://www.philadelphiafed.org/research-and-data/

real-time-center/survey-of- professional-forecasters/academic-bibliography.cfm.

Copyright © 2015 John Wiley & Sons, Ltd