Firms’ subjective uncertainty and forecast errors: survey evidence from Japan

ABSTRACT

Using original survey data linked with the financial statements of Japanese listed firms, this study presents an ex post evaluation of firms’ ex ante subjective uncertainty. Ex ante forecast uncertainty in terms of sales and employment growth is derived from firms’ subjective confidence interval around their point forecasts. Ex post forecast error is calculated as the deviation of the realized figures from the point forecasts. The results indicate that ex ante subjective uncertainty has a significant positive association with realized absolute forecast error. The subjective confidence interval for a firm’s own business forecast, in comparison with that for macroeconomic variables, is reliable as a measure of uncertainty. These findings indicate that the subjective probability distribution of business outlook captured by firm surveys contains valuable information for measuring economic uncertainty at the micro-level.

KEYWORDS:

• Uncertainty
• confidence interval
• probability distribution
• forecast error

JEL CLASSIFICATION:

• D84
• E22

Acknowledgments

I would like to thank two anonymous reviewers, seminar participants at RIETI for their helpful comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ This study is also related to the behavioural explanations of forecast errors, such as optimistic bias (e.g. Ito 1990; Bachmann and Elstner 2015; Massenot and Pettinicchi 2018), rational inattention (e.g. Coibion, Gorodnichenko, and Kumar 2018; Massenot and Pettinicchi 2018; Morikawa 2019; Tanaka et al. 2020), and herding (e.g. Pierdzioch and Stadtmann 2011).

² Real GDP growth rate was 1.4% in 2012 and 1.2% (on an annual basis) for the subsequent three years, meaning that the survey was conducted at normal economic condition.

³ In calculating 90% confidence intervals, 10% is assigned to the ±5% or over category.

⁴ Since the calculation of σ is a simple linear transformation, assuming other distributions does not alter the results of this study. 40% is assigned to the ±20% or over category (GDP and CPI).

⁵ Even if the absolute forecast error is replaced by the squared forecast error, we obtain essentially the same results.

Additional information

Funding

This research is supported by the JSPS Grants-in-Aid for Scientific Research [26285063, 16H06322, 18H00858, 20H00071].