The Informativeness of Micro and Macro Information During Economic Crisis and Non-Crisis Periods: Evidence from Europe

Abstract

We investigate whether and how the information content of reported profitability and macroeconomic expectations changes when the state of the economy changes from non-crisis to crisis conditions. For this, we analyze data from 16 European countries over the period 2005–2015. We find macroeconomic expectations to be useful in predicting future profitability only during non-crisis periods and mainly for firms facing elastic demand for their products and services and firms without sequential losses. Current profitability as well as its cash flow and accruals components are much more informative predictors of future profitability than macroeconomic expectations in both non-crisis and crisis periods. Market pricing tests suggest that macroeconomic expectations are not informative and thus not priced by market participants during crisis periods and support efficient pricing of current profitability under both non-crisis and crisis periods. However, it is the cash flow component of profitability that is efficiently priced under both economic conditions, while the accrual component of profitability is mispriced during crisis periods. Overall, we provide evidence that reported accounting information is much more useful to stock market investors than macroeconomic expectations in both non-crisis and crisis economic periods.

Keywords:

• Economic crisis
• Micro and macro information
• Profitability prediction
• Stock market valuation

JEL codes:

• E00
• E01
• G14
• M41

Acknowledgements

We appreciate the helpful comments of Michel Magnan (editor) and two anonymous reviewers. We are grateful to Louis Chan, Giovanni-Battista Derchi, Dimosthenis Hevas, Alexis H. Kunz, Hunter Land, Karl Schuhmacher, Efthymios Tsionas, Oktay Urcan, Zach Wang, Georgios Zanias, seminar participants at Manchester Business School, Essex Business School and HEC Lausanne as well as to conference participants at the 39th EAA Annual Congress in Maastricht, the 13th Biennial Athenian Policy Forum Conference in Athens, the 2016 American Accounting Association Annual Meeting in New York and the 12th workshop on European Financial Reporting in Fribourg for helpful comments and suggestions. We also thank Stavros Panagopoulos for excellent research assistance.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 We refer to accounting information using the following terms interchangeably: ‘micro,’ ‘accounting profitability,’ and ‘reported profitability.’ We also use interchangeably the following terms: ‘macro,’ ‘macroeconomic information,’ and ‘macroeconomic expectations.’

2 Changes in interest rates are positively related to unexpected earnings. However, this effect only partially offsets the negative value effect in required returns.

3 A related stream of research examines earnings persistence as well as the relation between accounting information and stock returns conditional on macroeconomic information (Johnson, 1999; Gallo, Hann, & Li, 2013; Graham, King, & Bailes, 2000; Ho, Liu, & Sohn, 2001; Swanson, Rees, & Juarez-Valdes, 2003). Filip and Raffournier (2014) report a decrease in income smoothing and a decrease in accruals quality during the recent financial crisis in Europe.

4 We are thankful to an anonymous reviewer for suggesting this theoretical framework.

5 Ahir and Loungani (2014) support their conclusion using both OECD’s and Consensus Economics forecasts of GDP growth.

6 Specifically, while in a non-crisis period with a fiscal multiplier k, an x% increase in public expenditures will cause a k*x% increase in GDP, in a crisis period an x% decrease in public expenditures will cause a much larger decrease in GDP than k*x%.

7 Berkmen, Gelos, Rennback, and Walsh (2009) examine GDP growth revisions issued by Consensus Forecasts in 2008 for a sample of 43 emerging markets. They report growth forecast revisions ranging from −18% to −1.5%. Auerbach and Gorodnichenko (2011) provide evidence of large differences in the size of spending multipliers in recession and expansion periods.

8 The Li et al. (2014) model is as follows: $\begin{array}{cc}& RNO{A}_{t+1}=a_0+a_{1}MACR{O}_t+a_{2}RNO{A}_t\\ & +a_{3}BT{M}_t+a_{4}SIZ{E}_t+a_{5}DNO{A}_t+a_{6}DLOS{S}_t\\ & +a_{7}DDI{V}_t+a_{8}DIVYIEL{D}_t+u_{t+1}\end{array}$

9 As an example, we calculate the weighted average GDP growth forecast (MACRO_t) for a Greek company with international sales in France and Germany as follows: Weighted average GDP growth forecast = % sales in Greece * average GDP growth forecast of Greece + % sales in France * average GDP growth forecast of France + % sales in Germany * average GDP growth forecast of Germany (assuming that total sales are made in these three countries).

10 IMF publishes the GDP growth forecasts used to estimate the variable MACRO_t either late September or during October.

11 It is important to note that the returns windows in models (2) and (3) do not overlap.

12 We winsorize all variables at the 1% level.

13 The lack of a clear definition of net operating assets precludes us for including these firms in the analysis.

14 The estimated variance inflation factors (VIF) for all our models are below 4.53 suggesting that there is no significant multicollinearity issue.

15 As an additional analysis, we employ portfolio tests to explore the efficient pricing of the macroeconomic expectations during crisis and non-crisis periods. We sort stocks into five quintiles based on MACRO_t separately for crisis and non-crisis periods. We value weight the constituents of each quintile using the market value of equity and calculate portfolio returns for each month, 6 months ahead of the publication of the latest forecast of GDP growth. We then estimate regressions of portfolio returns on the four Fama and French (1992, 1993) factors. Similar to Li et al. (2014), the empirical results (untabulated but available upon request) confirm the mispricing of macroeconomic expectations during non-crisis periods (hedge alpha is positive and significant). However, there is no mispricing of the macroeconomic expectations during crisis periods (hedge alpha is insignificant), consistent with the results of Table 5, Panel A.

16 We also perform portfolio tests to explore the pricing of current profitability (RNOA_t) during crisis and non-crisis periods similar to those described in footnote 15 for (MACRO_t). The results support efficient pricing of current profitability in both non-crisis and crisis economic conditions.

17 We additionally perform our profitability prediction analysis by using aggregate data for both profitability (RNOA_t) and macroeconomic expectations (MACRO_t). We aggregate profitability variables following the methodology proposed by Kothari, Lewellen, and Warner (2006) for aggregate earnings. The aggregate RNOA_t series is simply the cross-sectional sum of operating income for all firms in the sample, subsequently scaled by the sum of lagged net operating assets. We use the same methodology for aggregate operating cash flows and aggregate accruals. The series are value-weighted using the market value of equity as a weight. We also aggregate MACRO_t on a country-level synthetic GDP measure. The results from the aggregated data are consistent with the results from firm-level data.

18 We further test the sensitivity of our results by using two alternative definitions of our CRISIS variable. First, we define year t as a crisis year if both the Debt-to-GDP ratio is higher than 60% and the Government Budget is negative (deficit) for two consecutive years (year t−1 and year t). Second, we define year t as a crisis year if both the GDP growth rate is negative and the Government Budget is negative (deficit) for two consecutive years (year t−1 and year t). Our empirical results (not tabulated but available upon request) remain robust to these alternative definitions of crisis.

19 The mean forecast error (FE) for GDP growth in the non-crisis and crisis periods is −0.171 and 1.834, respectively.

20 We are thankful to an anonymous reviewer for suggesting this analysis.

21 Using our 2005–2015 sample, we estimate RNOA_t+1 after multiplying the estimated coefficient values of model (1) with their respective variable values.

Additional information

Funding

The authors acknowledge financial support from the Research Center of the AUEB.