Firms’ economic growth expectations and reactions to micro and macro news in Japan

ABSTRACT

This study investigates whether firms’ expectations of aggregate output growth overreact or underreact to micro and macro news by examining firm surveys conducted by a Japanese newspaper company, Nikkei. We capture micro news measured by forecast revisions and macro news measured by a new variable, the Japanese government’s growth forecast and that of professional forecasters’ consensus. We found that firms’ economic growth expectations overreact only to macro news. Our results indicate the heterogeneity of firm expectations formation across target variables and news types. Our findings suggest that firms excessively rely on government forecasts in turbulent times. The overreaction of firms’ output growth expectations to the government’s forecast may be associated with their lower performance, and that the government’s macroeconomic forecasts play a role in macroeconomic performance through firms’ expectation channels.

KEYWORDS:

• Firm expectations
• overreaction
• information rigidity
• government forecast
• private forecast

JEL CLASSIFICATION:

• D84
• E17
• E32

1. Introduction

Economic agents’ expectations play a pivotal role in shaping their economic decisions, and researchers are thus interested in what shapes their expectations. The full-information rational expectations (FIRE) model, where agents adjust their expectations promptly and accurately by incorporating new information, serves as the benchmark model. An influential work by Coibion and Gorodnichenko (2015) found evidence that professional forecasters’ consensus forecasts underreact to news. There is also evidence that contradicts the FIRE model, suggesting that the expectations of individual forecasters are likely to overreact to news. Thus, there is increasing attention on providing models that accommodate both underreactions and overreactions (Bordalo et al. 2020; Broer and Kohlhas 2022).

Extant studies have shed light on the type of news that plays a role in firms’ expectations formation and how firms react to various types of news. Dovern et al. (2023) showed that firms’ aggregate output expectations overreact to firm-specific local information. By differentiating news into firm-specific micro news and economy-wide macro news, Born, Enders, Menkhoff, et al. (2022, Born, Enders, Müller, et al. 2022). found that firms’ expectations of their production and business circumstances, overreact to micro news and underreact to macro news.

Therefore, this study examines whether firms’ expectations about real GDP growth in Japan overreact or underreact to both micro and macro news, as these expectations influence management and investment planning in firms, consequently influencing firm performance (Kumar, Gorodnichenko, and Coibion 2023; Tanaka et al. 2020). We introduce a new variable for macro news; that is, the Japanese government’s and professional forecasts, motivated by the finding that overly optimistic growth expectations of authoritative institutions induce economic contractions (Beaudry and Willems 2022). Micro news is defined as forecast revisions in line with the previous studies. Our findings show that firms’ expectations of economic growth overreact to both micro and macro news, and suggest that firms demonstrate an excessive reliance on the government forecasts. Additionally, they suggest that overreaction to the government forecast may be associated with lower performance in firm activity and that the government’s macroeconomic forecasts play a role in macroeconomic performance through firms’ expectation channels.

The remainder of this paper is organized as follows. The data used in the analysis are presented in the following section. Section III introduces statistical approaches. Section IV presents the results and discusses their implications. The final section concludes the paper.

2. Data

Our study is based on a dataset of 20 leading Japanese firms’ real GDP growth forecasts for the following fiscal year from New Year’s surveys that have been conducted by Nikkei newspaper since December 1962. The Nikkei editors select the leading firms for these surveys in December, and the survey responses are returned by the end of December. To measure forecast revisions, we used surveys by Toyo Keizai Inc. Approximately 100 firms are surveyed about real GDP growth forecasts in the middle of November each year, and the forecasts are reported in Toyo Keizai monthly statistics (Toyo Keizai) in February of the following year. The Toyo Keizai surveys used in this study are from 1990 to 2002. ¹ Nikkei and Toyo Keizai are the leading economic media outlets, and many of the firms considered in the two abovementioned surveys were included in the Nikkei Stock Average.

Forecast revisions between November and December can be measured as the difference between the forecasts in Nikkei and those in Toyo Keizai for those who submitted their forecasts in both surveys in a given year. We consider the government forecasts and professional forecasters’ consensus forecasts as news because those forecasts are released during the period between the Nikkei and Toyo Keizai surveys. Around the second week of December, professional forecasters release their forecasts, and the consensus forecasts are also reported in Nikkei around mid-December. After the consensus forecasts are published, the government releases its forecast by the 25^th of December. Firms are likely to revise their forecasts after the government forecast is released because they sometimes update previously submitted forecasts when new information becomes available (Inoue, Kato, and Yamasaki 2012). The difference between government and professional forecasters’ consensus forecasts is also considered macro news by taking advantage of the release timing of the two forecasts. This variable also allows examining the relative impact of those forecasts.²

We restrict the sample to firms with observations equal to or greater than two. This results in an unbalanced panel with 67 observations. There are 17 firms; thus, on average, there are five observations for each year.³ For actual outcomes, we used real-time (initially released) data following the previous studies. Table 1 shows that firms tend to overpredict actual growth rates.⁴ Although revision is likely to be slightly negative,⁵ 63% of the observations for revisions are different from zero, indicating that most firms revise their forecasts within one month. The professional forecasters slightly overpredict compared to the firms while the government overpredicts more than both professional forecasters and firms. This indicates that the difference between those forecasts is likely to be positive.

Table 1. Descriptive statistics.

	Mean	Median	St. dev.	Minimum	Maximum
Actual real GDP growth	1.09	0.60	2.07	−1.90	5.70
Firm forecast	1.59	1.50	1.33	−1.50	4.50
Firms forecast error	−0.65	−0.50	1.72	−3.50	2.70
Revision	−0.12	0.00	0.44	−2.20	0.60
Government forecast	2.25	2.40	1.25	0.00	4.00
Professional forecast	1.67	1.73	1.50	−0.67	4.43
Difference	0.58	0.60	0.57	−0.43	1.87

Notes: Firm forecast denotes firm’s real GDP growth forecasts. Government forecast ($f^G$) denotes the Japanese government’s forecasts. Professional forecast ($f^P$) denotes professional forecasters’ consensus forecasts. Difference ($f^G-f^P$) denotes the difference between government and professional forecasts. Revision is a forecast revision that denotes the difference between a firm’s forecasts and the corresponding forecast in Toyo Keizai. St. dev. denotes standard deviation.

3. Methods

This study uses the following model (Born, Enders, Menkhoff, et al. 2022):

(1) $x_{\mathit{t}+1}-f_{t+1,t}^{i,N}={\beta }_0+{\beta }_1\cdot \mathit{micro}\mathit{new}{s}_t+{\beta }_2\cdot \mathit{macro}\mathit{new}{s}_t+{\epsilon }_t^i$(1)

where $x_{\mathit{t}+1}-f_{t+1,t}^{i,N}$ is the error of the forecasts $f_{t+1,t}^{i,N}$ for the next fiscal year for firm i submitted to Nikkei, and the actual data, $x_{\mathit{t}+1}$. Micro news is denoted as $f_{t+1,t}^{i,N}-f_{t+1,t}^{i,T}$, which is the forecast revision for firm i, where $f_{t+1,t}^{i,T}$ is the forecast submitted to Toyo Keizai. For macro news, we consider three variables: the Japanese government forecast, $f_{t+1,t}^G$, professional forecasters’ consensus forecast, $f_{t+1,t}^P$, and the difference between the two forecasts, $f_{t+1,t}^G-f_{t+1,t}^P$.

The null hypothesis of the FIRE model indicates that these coefficients are zero because forecast errors cannot be predicted. If micro or macro news can predict a forecast error, the information available at the time of the forecast update is not fully incorporated and thus contradicts the FIRE model. ${\beta }_1$ and ${\beta }_2$ indicate whether firms overreact or underreact to new information that is embedded in micro and macro news, respectively. When positive (larger) news is likely to be followed by a negative (smaller) forecast error (${\beta }_i<0,\mathit{i}=1,2$), the forecast revision is too strong from an ex-post point of view, indicating that there is an overreaction to news. Similarly, positive coefficients (${\beta }_i>0,\mathit{i}=1,2$) imply underreaction.

We use a random-effects model that includes the firm’s consensus forecast from the previous year because it is observable in real time by all firms.⁶ We estimate Equation (1)(1) $x_{\mathit{t}+1}-f_{t+1,t}^{i,N}={\beta }_0+{\beta }_1\cdot \mathit{micro}\mathit{new}{s}_t+{\beta }_2\cdot \mathit{macro}\mathit{new}{s}_t+{\epsilon }_t^i$(1) using Driscoll and Kraay (1998), which is robust to cross-sectional and temporal dependence on the error term. Survey respondents may know and communicate with each other. We assume an autocorrelation of lag 1 in the error term because forecasts for the following fiscal year rely on those of the ongoing year, as the actual values for the ongoing year are unavailable (Coibion and Gorodnichenko 2015). Further, we examine a sample excluding the period from 1990 to 1993, which was marked by a significant negative shock due to the bursting of the bubble (Ito and Hoshi 2020).

4. Results and discussion

Table 2 shows that the coefficients for micro news are not significant at the standard significance levels, suggesting that there is insufficient information for firms to incorporate within about a month.^7,8 This result contrasts with the previous studies showing an overreaction in macroeconomic expectations (Bordalo et al. 2020; Broer and Kohlhas 2022; Dovern, Müller, and Wohlrabe 2023). For macro news, firms overreact to government forecasts throughout the entire sample period, which is consistent with the responsiveness of firms’ output growth expectations to new information (Kumar, Gorodnichenko, and Coibion 2023). However, this result contrasts with Born et al’.s (2022a) findings that firms’ expectations of their own production overreact to micro news but underreact to macro news. This indicates that the formation of firms’ expectations varies across different target variables and news types. Our results add to the literature by showing that there is also an overreaction of firms’ expectations to macro news.

Table 2. Individual forecast results.

	Full sample				Sample since 1994
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Revision	0.074	0.194	0.353	−0.026	−0.066	−0.039	0.228	0.098
	(0.446)	(0.398)	(0.413)	(0.462)	(0.348)	(0.297)	(0.316)	(0.501)
Government		−0.841**				−0.942**
		(0.287)				(0.357)
Professional			−0.428				−1.074*
			(0.363)				(0.486)
Difference				−0.221				0.711
				(0.730)				(1.109)
Lagged firm’s consensus	−0.020	0.502	0.258	−0.029	−0.052	0.226	−0.133	−0.324
	(0.351)	(0.297)	(0.338)	(0.333)	(0.615)	(0.409)	(0.513)	(0.942)
Constant	−0.603	0.272	−0.413	−0.461	−0.470	0.782	0.724	−0.619
	(0.776)	(0.826)	(0.841)	(0.936)	(1.005)	(0.935)	(0.957)	(0.987)
Observations	67	67	67	67	48	48	48	48
Groups	17	17	17	17	15	15	15	15
Within R^2	0.001	0.124	0.054	0.005	0.001	0.179	0.248	0.034

Notes: Firm fixed effects are included. Standard errors are in parentheses and are estimated with a lag of 1, following Driscoll and Kraay (1998). ***, **, and * indicate the 1%, 5%, and 10% significance levels, respectively.

Furthermore, Table 2 shows that, since 1994, firms have overreacted to both government and professional forecasts. This finding suggests that firms place undue reliance on government forecasts in turbulent times, possibly due to the notion that various economic policies that the government takes to achieve its forecasts are effective. This is in line with a finding of central bank forecasting that central bank forecasts can influence the expectations of other economic agents by revealing information about policies and preferences even if central banks do not have an informational advantage over those agents (Hubert 2015). Table 2 shows that the coefficients for the difference between government and professional forecasts and the lagged firm’s consensus are not significant at the standard significance levels regardless of the sample periods. The latter indicates that firms incorporate the information efficiently into their forecasts.

Our findings suggest that the government’s macroeconomic forecasts play an important role in firms’ expectations formation and the macroeconomy. Tanaka et al. (2020) showed that larger forecast errors in Japanese firms’ GDP forecasts, lower profitability and productivity. These findings suggest that an overreaction to macro news may be associated with lower firm activity performance because the GDP forecasts of the Japanese government are too optimistic (Tsuchiya 2016), and weak macroeconomic performance induced by overly optimistic forecasts of authorities (Beaudry and Willems 2022) may occur through firm expectations channels.

5. Concluding remarks

Future research agendas should consider expanding the scope beyond the current limitations, which include a limited number of observations and sample periods focused solely on macro news derived from the government and professional forecasts. To enhance the depth of analysis, it is essential to examine various types of target variables, along with micro and macro news with more recent observations, including a wider coverage of industries and firm sizes. This broader approach will provide a more comprehensive understanding of the dynamics involved and offer a more nuanced perspective for future studies.

Ethical approval

This article does not contain any studies with human participants or animals performed by any of the authors.

Acknowledgements

The authors are grateful to anonymous referees for useful comments. Yoichi Tsuchiya gratefully acknowledges the financial support from the Japan Society for the Promotion of Science (JSPS; grant number JP22K01510).

Disclosure statement

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

Data availability statement

The data that support the findings of this study are available from the author upon reasonable request.

Additional information

Funding

The work was supported by the Japan Society for the Promotion of Science [JP22K01510].

Notes

¹ Unfortunately, surveys by Toyo Keizai have not been conducted except for the period from 1990 to 2002.

² Including the two forecasts simultaneously into our model results in an issue of serious multicollinearity because the two forecasts are highly positively correlated with a correlation coefficient of about 0.9.

³ See Table A1 in the Supplementary material for details.

⁴ Firm forecast errors are negative and statistically significant at the 1% level.

⁵ See Figures A1 and A2 for distributions of revision and forecast errors. Figures A3 show that there is no correlation between the two variables.

⁶ The random effects model was preferred to the fixed effects model according to the Hausman test, although not reported..

⁷ See Table A2 for results of the consensus forecasts of firms for comparison. Models (1) through (3) shows that the coefficients of micro news are positive, which is in line with the finding that the consensus forecasts underreact to news (Coibion and Gorodnichenko 2015). However, it shows none of the coefficients are statistically significantly different from zero probably due to a small number of observations.

⁸ We checked the robustness with pooled OLS due to a limited number of observations, and the results align with our main findings. See Table A3.