Ever since Keynes’ famous quote about animal spirits (Keynes 1936), there has been interest in the importance of firm expectations for business actions and outcomes. More recent work has established a key role for beliefs in determining resource allocations (e.g. David et. al. 2016) and in driving business cycle dynamics (e.g. Beaudry and Portier 2004, Schmitt-Grohe and Uribe 2012, Ilut and Schneider 2014). However, empirical evidence on the link between beliefs and economic performance has been limited due to a lack of firm-level panel data containing both expectations and outcomes. Key outstanding questions include: how much do beliefs matter for economic outcomes, under what circumstances, and to what extent does forecast accuracy vary across firms?
In a recent paper, we explore these questions by building a dataset that combines a survey of Japanese firms’ quantitative forecasts of future GDP growth with detailed accounting data for over 1,000 large Japanese firms over 25 years (Tanaka et. al. 2018). We find four main results:
- First, optimistic forecasts of GDP growth are positively associated with employment, investment and revenue growth.
- Second, larger forecast errors (of either sign) lower profitability.
- Third, positive forecast errors reduce measured firm-level productivity, while negative forecast errors have insignificant effects. We provide a simple model of firm input choice under uncertainty and costly adjustment that rationalizes these patterns.
- Finally, larger and more cyclically sensitive firms make more accurate forecasts, as do more productive, older and bank-owned firms, suggesting that forecasting ability is linked to the return to forecast accuracy, as well as management ability, experience and governance.
Data linking beliefs and outcomes
Our data on firm expectations come from the “Annual Survey of Corporate Behavior” conducted by the Economic and Social Research Institute in the Cabinet Office of Japan. The data cover the period 1989-2015. The survey questionnaire was sent to all firms listed on the Tokyo and Nagoya Stock Exchanges and garnered a response rate averaging about 40%. The items we use are forecasts of the real GDP growth rate in Japan over one-year horizon and the firm’s employment and investment growth rates over three-year horizon. We match the survey data to two other firm-level datasets. We use the Development Bank of Japan’s “Financial Data of Listed Firms” to obtain detailed accounting data for these firms and the “Nikkei Needs Financial Quest” data for information on stock market prices, firm age and ownership structure. Our final sample contains forecasts, outcomes and characteristics for an average of over 1,000 firms per year spanning the period 1989-2015.1
Expectations and firm performance
Our first main finding is that firm optimism/pessimism strongly affects input choices and growth rates. We summarise this result in Figure 1. The figure displays binned scatterplots of firm-level forecasts of one-year-ahead GDP growth versus employment, investment and sales growth in the subsequent year. Each observation reflects a ‘bin’ of firms sorted by their level of optimism (or pessimism). Due to the longitudinal nature of our data, we are able to control for persistent firm characteristics and the realization of macroeconomic shocks (as well as any other time-varying factors that are common across firms, such as fluctuations in aggregate uncertainty) by extracting firm and time fixed-effects. The plots reveal a strong association between input and sales growth and firm expectations. The effects are statistically and economically significant – a 1% higher forecast of GDP growth is associated with a 0.2 percentage point increase in employment growth, a 2.8 percentage point increase in investment growth, and a 0.4 percentage point increase in sales growth.
Figure 1
Next, we explore the relationship between forecast accuracy and firm performance, i.e. measures of profitability and productivity. Figure 2 summarises our findings, again using binned scatterplots after controlling for firm and time fixed-effects. The figure shows that larger forecast errors (of either sign) reduce firm profitability. The effects are substantial – a 1% increase in the forecast error reduces realized profits by about 8%. Grouping positive and negative forecast errors gives a similar pattern for firm-level productivity – a 1% increase in the forecast error reduces measured productivity by 0.6%. However, splitting positive and negative forecast errors reveals an asymmetry in the relationship – positive forecast errors (over-optimism) are associated with significantly lower productivity, but negative forecast errors (over-pessimism) do not have a significant effect on productivity. This asymmetry is not present when examining profitability.
Figure 2
We present a simple model of firm input choice under uncertainty and costly adjustment that rationalises these patterns. Consider a set of monopolistically competitive firms. At the beginning of each period, firms form expectations over future economic conditions and frictionlessly choose an initial level of inputs. They are able to further adjust these inputs upon the realisation of shocks, but this latter adjustment is subject to a disruption cost that reduces output (e.g. by requiring a pause in the production line, etc.). The model predicts that firm input choices and sales are increasing in their beliefs regarding economic conditions. Larger forecast errors reduce firm profitability. Finally, measured productivity is decreasing in the size of the forecast error for optimistic firms. This is true for two reasons: first, because the firm over-accumulates inputs, it produces at a higher level, which reduces prices and revenue; second, larger forecast errors entail larger disruption costs, which also reduce revenue for a given level of inputs. Both of these forces lead to lower revenue productivity. However, the effects of a larger (absolute) forecast error for pessimistic firms is ambiguous – the pricing effect would tend to increase revenue productivity (the firm under-accumulates inputs and produces at a low level, raising prices and revenue) but the disruption cost reduces revenue productivity (the cost must be paid whether the firm is scaling production up or down). Thus, the model can reproduce the main patterns from the data, including the asymmetric effect on measured productivity.2
Forecast quality and firm characteristics
In our last set of results, we explore the relationship between forecast quality and a set of observable firm characteristics. We use two measures of forecast quality: difference from realised GDP growth and difference from forecasts by professional forecasters. Our findings reveal substantial variation in forecast accuracy across different types of firms. First, we find that more cyclically sensitive firms (firms whose stock prices are sensitive to announcements of realized GDP growth rates) make more accurate forecasts, presumably reflecting a higher return to accuracy for these firms. Additionally, we find that larger, more productive, older and bank-owned firms tend to have the most accurate forecasts. This latter set of results suggests that experience, management ability and governance may also play a role in determining firm-level forecast accuracy.
Conclusions
Relying on a novel dataset with information on firm expectations and outcomes, our work reveals a strong connection between beliefs regarding macroeconomic conditions and subsequent firm performance. Our results point to an important role for expectations in determining economic performance, both at the firm and aggregate level. Further work quantifying these relationships and uncovering the sources of forecasting ability would be fruitful in understanding the determinants of the information environment in which firms operate and may provide guidance as to whether there are actions that could be taken to better forecast accuracy. Our findings suggest that improving information quality could yield substantial economic benefits.
References
Beaudry, P and F Portier (2004) “An Exploration into Pigou’s Theory of Cycles,” Journal of Monetary Economics, 51(6): 1183-1216.
David, J M, H A Hopenhayn and V Venkateswaran (2016) “Information, Misallocation and Aggregate Productivity,” The Quarterly Journal of Economics, 132(2): 943-1005.
Ilut, C L and M Schneider (2014) “Ambiguous Business Cycles,” The American Economic Review, 104(8): 2368-2399.
Keynes, J M (1936) The General Theory of Employment, Interest, and Money, Harcourt, Brace.
Schmitt-Grohe, S and M Uribe (2012): “What’s News in Business Cycles,” Econometrica, 80(6): 2733-2764.
Tanaka, M, N Bloom, J M David and M Koga (2018) “Firm Performance and Macro Forecast Accuracy,” NBER Working Paper 24776.
Endnotes
[1] We perform a number of exercises to verify the quality of the forecast data, i.e. that these forecasts accurately capture firms’ expectations of future growth rates. We also address issues of sample selection.
[2] These predictions hinge on two key features of the environment: first, that the econometrician measures revenue productivity, which includes the effect of prices, and not physical productivity. Second, that the disruption costs reduce output and are not pure financial adjustment costs.
