Aggregate earnings informativeness and economic shocks: international evidence

ABSTRACT Our study proposes the usage of aggregate earnings to forecast future GDP growth. Using empirical analyses with global quarterly data, we investigate whether aggregate-level profitability drivers, which are components of aggregate earnings, are relevant for forecasting GDP growth. After confirming that aggregate-level profitability drivers are useful for forecasting future GDP growth worldwide, we show that considering the effects of crises improves the forecast model of GDP growth. In addition, we suggest that predicting GDP growth using aggregate-level profitability drivers is relevant for stock valuation in developed countries, but not in emerging countries.


Introduction
The aim of our study is to propose the proper usage of aggregate-level earnings information for forecasting future gross domestic product (GDP) growth. A large body of empirical accounting research has investigated the usefulness of accounting information to investors. Financial statement analyses at the firm level have been a recurring topic of accounting research. Researchers have investigated the link between earnings and stock returns, forecasted future earnings, or estimated the intrinsic value of equity at the firm level (e.g. Ball andBrown 1968, 2019;Ohlson 1995;Courteau et al. 2006;Penman 2012). These studies contribute to efficient decision making on investment.
On the other hand, accounting information is not only practically useful for stock investment. It is also used to capture the macroeconomic conditions. For example, economic newspapers and governments often aggregate accounting information in their articles. To investigate the usefulness of aggregate-level accounting information, a new accounting research area has emerged recently. Konchitchki (2016) calls it as 'macro-accounting. ' Konchitchki (2016, 27) states that 'this new research area focuses on addressing real-life world problems using the value added that accounting can bring to various macro-level topics that are at the forefront of the academic and professional discussions.' 1 There are two streams in the macro-accounting literature. One focuses on the earnings-returns relation at the aggregate level (e.g. Kothari, Lewellen, and Warner 2006;Sadka and Sadka 2009;Kang 2019), and the other investigates the link between aggregate-level accounting information and the macroeconomy (Shivakumar 2007;Konchitchki andPatatoukas 2014a, 2014b;Gallo, Hann, and Li 2016). This study belongs to the latter stream.
Our studies are mainly based on Konchitchki and Patatoukas (2014b), which is one of the pioneering studies on macro-accounting. They investigate the information content of aggregate earnings based on the DuPont analysis framework. This framework decomposes the profitability (return on net operating assets: RNOA) into their drivers (return on sales: ROS, and asset turnover: ATO). Since these drivers are shown to be useful for forecasting future performance at the firm level (Soliman 2008;Penman 2012), Konchitchki and Patatoukas (2014b) apply this framework to the aggregate level. They empirically indicate that aggregate-level profitability drivers, which are components of aggregate earnings, are useful for improving the forecast accuracy of GDP growth in the United States (US). They also show the significantly positive relation between future stock market returns and the portion of future real GDP growth that is predictable based on aggregatelevel profitability drivers but that is not anticipated by stock market investors. Konchitchki and Patatoukas (2014b) conclude that this finding suggests that predicting GDP growth with aggregatelevel profitability drivers is relevant for stock valuation.
However, Konchitchki and Patatoukas (2014b) and other prior studies have two limitations. First, they do not consider the effects of financial crises. Although several crises have strongly damaged the world economy, prior studies do not examine the potential effects of the crises on the informativeness of aggregate earnings. Recent studies show that accounting quality strengthens during crisis periods (Filip and Raffournier 2014;Arthur, Tang, and Lin 2015). Doukakis et al. (2020) suggest that financial crises can change the relation between the macroeconomy and the profitability of listed firms. Based on these studies, financial crises will affect GDP growth forecasts using aggregate-level profitability drivers. Thus, we investigate whether the forecast accuracy improves by considering the crises' effects.
The second limitation concerns external validity, which arises from the empirical analyses based on only US data. We should exercise caution in extending empirical findings based on the data of one country to other countries without checking their suitability. The empirical results of several prior non-US studies are different from those of US research (e.g. Gabrielsen, Gramlich, and Plenborg 2002;Rahman, Yammeesri, and Perera 2010;Gajewski and Quéré 2013). Our sample covers 21 countries to explore the external validity of the usefulness and implications of aggregatelevel profitability drivers for predicting GDP growth.
We obtain three findings. First, we confirm that aggregate-level profitability drivers are useful for forecasting future GDP growth using global data. Second, we show that considering the effects of crises improves the forecast accuracy of GDP growth. Third, we suggest that predicting GDP growth using aggregate-level profitability drivers is relevant for stock valuation in developed countries, but not in emerging countries.
Our study contributes to macro-accounting research in three ways. First, it extends the external validity of the usefulness of aggregate-level profitability drivers for predicting GDP growth. Unlike prior US macro-accounting research (Shivakumar 2007;Konchitchki andPatatoukas 2014a, 2014b;Gallo, Hann, and Li 2016), we use global data to show that aggregate-level profitability drivers significantly improve the forecast accuracy of GDP growth. Second, we improve the forecast model of GDP growth by considering the effects of financial crises. Konchitchki and Patatoukas (2014b) construct forecast models of GDP growth including aggregate-level profitability drivers. We upgrade their models by including a crisis dummy and its interaction terms. Recently, the real economy has been suffering from the recessionary damage of the COVID-19 pandemic. Our evidence may make a timely contribution to current macroeconomic analyses. The third contribution concerns the implications for investors. Konchitchki and Patatoukas (2014b) report that predicting GDP growth by aggregate-level profitability drivers is relevant for stock valuation in the US. We extend their research, as our results suggest that their relevance for stock valuation is restricted to developed countries.
We organize the remainder of this paper as follows. Section 2 discusses literature review and research questions. Then, we present our research design in Section 3. Variable definitions and sample selection follow in Section 4. Our empirical results are reported in Section 5. Section 6 includes the robustness checks and additional analyses. Finally, Section 7 summarizes our study and provides concluding remarks.

Literature review and research questions
We review the related literature and present our three research questions in this section. Empirical accounting research has focused on firm-level accounting information. For example, Beaver (1968), Ball and Brown (1968), and their subsequent studies investigate the usefulness of firm-level earnings information for investors. 2 By contrast, several recent studies inspect accounting information from a different perspective. They calculate the cross-sectional average or sum of firm-level accounting information of listed firms in a country and focus on such aggregate-level accounting information (e.g. Ball, Sadka, and Sadka 2009;Sadka and Sadka 2009). These studies are classified as macroaccounting (Konchitchki 2016). Prior macro-accounting studies empirically support the usefulness of aggregate earnings for forecasting macroeconomic indicators, such as inflation (Shivakumar 2007), and monetary policy (Gallo, Hann, and Li 2016). Furthermore, it has also been examined whether aggregate earnings can predict GDP growth which captures the overall economy (Konchitchki andPatatoukas 2014a, 2014b).
There are two reasons that prior macro-accounting studies focus on the usefulness of aggregate earnings for GDP forecast. First, aggregate earnings are expected to have the potential to forecast future GDP growth. Since corporate profits are components of GDP and are correlated with other components, their growth is assumed to be a main driver of economic growth (Konchitchki and Patatoukas 2014a). Second, accounting earnings of listed firms are timely information. Listed firms must report quarterly financial statements in some countries, and those reporting dates are usually earlier than those of macroeconomic indicators. Prior US studies point out that listed firms release quarterly accounting information before some macroeconomic indicators (e.g. Konchitchki andPatatoukas 2014a, 2014b;Nallareddy and Ogneva 2017).
Therefore, Konchitchki andPatatoukas (2014a, 2014b) investigate the usefulness of aggregate earnings for GDP forecasts in the US. Konchitchki and Patatoukas (2014a) find that aggregate earnings changes have a significantly positive relation with future GDP growth. Konchitchki and Patatoukas (2014b) decompose aggregate earnings changes into aggregate-level profitability drivers based on the DuPont analysis method. They report that aggregate-level profitability drivers are incrementally useful for forecasting GDP growth. However, the external validity of their evidence is not examined around the world. Prior studies report that empirical results can be different from US evidence in several cases when they use non-US data (e.g. Gabrielsen, Gramlich, and Plenborg 2002;Rahman, Yammeesri, and Perera 2010;Gajewski and Quéré 2013). Thus, our first research question is whether aggregate-level profitability drivers are useful for predicting GDP growth worldwide.
Furthermore, considering the state of the economy will improve the forecast model of GDP growth using aggregate-level profitability drivers. Filip and Raffournier (2014) report that earnings management behavior significantly decreases during the global financial crisis. Arthur, Tang, and Lin (2015) show that earnings quality improves during the crisis. Related to the state of the macroeconomy, Loh and Stulz (2018) find that the stock-price impact of earnings forecast revisions is greater in bad times. These studies suggest that the quality and the importance of accounting information improve during economic shocks. In addition, Doukakis et al. (2020) find that macroeconomic expectations are useful in predicting firm-level future profitability only during non-crisis periods. Their findings suggest that the relation between the macroeconomy and the profitability of individual firms can differ depending on the state of the economy. 3 Based on these studies, the effects of aggregate earnings on future GDP growth can differ between crisis periods and non-crisis periods. Therefore, our second research question is whether considering the crises' effects improves the forecast accuracy of GDP growth. 4 Moreover, Konchitchki and Patatoukas (2014b, 689-691) engage in empirical analyses to obtain implications for stock valuation. Using US data, they examine the association between future stock market returns and the portion of future real GDP growth that is predictable based on aggregatelevel profitability drivers but that is not anticipated by stock market investors. As a result, they observe a significantly positive relation between them. This finding suggests that the link between aggregate-level accounting profitability drivers and the subsequent real GDP growth is relevant for stock valuation (Konchitchki and Patatoukas 2014b). They explain this positive relation based on the effects of aggregate-level profitability drivers on the market reaction. 5 Since investors' reactions can differ depending on the development levels of the stock markets, we investigate whether the relation can be observed in developed and emerging countries as our third research question.

Improved forecast accuracy of GDP growth with aggregate-level profitability drivers
We first test whether aggregate-level profitability drivers improve forecasting future GDP growth using global data. We investigate it using Equations (1) and (2), which are based on Konchitchki and Patatoukas (2014b) ΔGDP c;qþh denotes future GDP growth of country c in quarter q + h (h = 1, 2, 3, 4). ΔRNOA c;q represents the changes in RNOA. In Equation (2), we decompose ΔRNOA c;q into ΔROS c;q and ΔATO c;q , which represent the changes in ROS and those in ATO, respectively. MR c;q is market returns. Since ΔGDP c;q and MR c;q are added as control variables in prior studies (Konchitchki andPatatoukas 2014a, 2014b), we include them in our models. We focus on the incremental usefulness of aggregate-level profitability (ΔRNOA c;q ) and its drivers (ΔROS c;q , and ΔATO c;q ) for GDP forecasts. Thus, we compare the forecast accuracy of Equations (1) and (2) to that of the following base model: Equations (1) and (2) add aggregate-level profitability or its drivers to Equation (3). If they are incrementally useful, the adjusted R squares of Equations (1) and (2) should be significantly higher than those of Equation (3). We examine whether the differences in the adjusted R squares between the models are significant using Chi-Squared statistics estimated by likelihood-ratio tests. We additionally check the significance of the estimated coefficients on aggregate-level profitability and its drivers. If they have significant effects on future GDP growth in most countries, the estimated coefficients on ΔRNOA c;q ; ΔROS c;q , and ΔATO c;q should be significant in Equations (1) and (2).

Improved forecast accuracy by considering crises' effects
Second, we test whether considering the effects of crises improves the forecast accuracy. Equation (4) adds a crisis dummy (Crisis q ) and its interaction terms to Equation (2). We test whether the differences in the adjusted R squares between these models are significant. 7 If considering the effects of financial crises increases the forecast accuracy, the adjusted R squares of Equation (4) should be significantly higher than those of Equation (2).
However, even if the adjusted R squares of Equation (4) are higher than those of Equation (2), aggregate-level profitability drivers and their interaction terms might not contribute to the improved forecast accuracy. The reason is that only the crisis dummy and its interaction terms on GDP growth or market returns (Crisis q , Crisis q � ΔGDP c;q , and Crisis q � MR c;q ) might improve the adjusted R squares. If this might be the case, we need not consider aggregate-level profitability drivers to forecast GDP growth.
To reduce this concern, we compare the adjusted R squares of Equation (5) with those of Equation (4). Equation (5) excludes accounting variables from Equation (4). If aggregate-level profitability drivers and their interaction terms improve the forecast accuracy of GDP growth, the adjusted R squares of Equation (4) should be significantly higher than those of Equation (5).

Implications for stock valuation
The third part of our analyses focuses on the implications of predicting GDP growth with aggregate-level profitability drivers for stock valuation. Figure 1 illustrates the outline of this part of the analyses. We estimate the portion of future real GDP growth that is predictable based on aggregate-level profitability drivers but that is not anticipated by stock market investors (ΔGDP ACC c;qþ1 ), following Konchitchki and Patatoukas (2014b). ΔGDP ACC c;qþ1 is estimated by two-stage regression models. The first stage regresses future GDP growth on aggregate-level profitability drivers to estimate fitted values. In the second regression, the fitted values are regressed on the contemporaneous market returns to estimate the residuals, or ΔGDP ACC c;qþ1 . We investigate the relation between ΔGDP ACC c;qþ1 and future market returns (MR 3m c;qþ1 ) using Equations (6) and (7), following Konchitchki and Patatoukas (2014b).
ΔROS res c;q and ΔATO res c;q are the residuals estimated by the regressions of ΔROS c;q and ΔATO c;q on ΔGDP ACC c;qþ1 , respectively. Instead of ΔROS c;q and ΔATO c;q , these residuals are used to reduce the multicollinearity concern as well as Konchitchki and Patatoukas (2014b). They are included in Equation (7) as control variables. Our interest is whether the relation between ΔGDP ACC c;qþ1 and MR 3m c;qþ1 is significantly positive, as Konchitchki and Patatoukas (2014b) observe. Thus, we focus on the coefficients on ΔGDP ACC c;qþ1 in Equations (6) and (7). Since the development level of stock markets could affect the relation, we employ the same regressions splitting our sample into developed and emerging countries based on the MSCI market classification. 8 We employ heteroskedasticity-consistent standard errors proposed by White (1980) and control country fixed effects when running the above regressions. We also check that the maximum of the variance inflation factor (VIF) in each regression is lower than 10, indicating that multicollinearity does not bias our results.

Variable definitions
We collect quarterly data. We use firm-level variables to calculate aggregate-level accounting variables. We define aggregate-level RNOA (RNOA c;q ) as the sum of operating income divided by the sum of the average net operating assets. Following Konchitchki and Patatoukas (2014b), we define net operating assets as operating assets (total assets minus cash and short-term investments) minus operating liabilities (total liabilities minus long-and short-term debt). RNOA c;q is decomposed into ROS (ROS c;q ) and net operating ATO (ATO c;q ) as Equation (8). ROS c;q denotes the sum of operating income divided by the sum of sales. ATO c;q represents the sum of sales divided by the sum of average net operating assets. We calculate the seasonal (year-over-year) differences of these variables for the regression models to reduce the seasonal effects (ΔRNOA c;q ; ΔROS c;q , and ΔATO c;q ).

RNOA ¼ Operating Income Net Operating Assets
MR c;q is the log return of the stock market index. We measure market returns over the 12 months leading to the end of quarter q, because Konchitchki and Patatoukas (2014b, Table 3) show that 12month returns are the most useful for predicting GDP growth. ΔGDP c;q is seasonal (year-over-year) log growth of real GDP.
Crisis q is a crisis dummy. It takes the value of one when the country/quarter is classified as in a crisis period, and zero otherwise. Our sample covers three crises: the burst of the dot-com bubble in 2001, the global financial crisis from 2007 to 2009, and the ongoing recession induced by the COVID-19 pandemic in 2020. Thus, if our regression models include at least one of the variables using data during these periods, we classify the country/quarter as being in a 'crisis period.' ΔGDP ACC c;qþ1 is the portion of future real GDP growth that is predictable based on aggregate-level profitability drivers but that is not anticipated by stock market investors. Following Konchitchki and Patatoukas (2014b), we adopt two-stage regressions to estimate this variable. The first-stage regression model is Equation (9).
In the first stage, we estimate predictable information on GDP growth at quarter q + 1 with aggregate-level profitability drivers (ΔROS c;q , and ΔATO c;q ) controlling country fixed effects. Considering crises' effects on the forecast of GDP growth, we include a crisis dummy (Crisis q ) in Equation (9)

Sample selection
Our main data source is the S&P Capital IQ. We obtain the financial data of listed firms, stock market index, and real GDP from the database. The financial data is converted into dominant currencies at the historical rate. The dominant currency is defined as the currency that the most firm/quarter observations adopt as the financial reporting currency in the country. After collecting financial data, we impose the following requirements on the firm/quarter observations.
(1) Quarterly accounting information is available within 3 months after the end of the fiscal quarter.
(2) The main SIC codes are not 6000-6499 or 6700-6999. We impose Requirement (1), because we assume that accounting information collected within 3 months after the fiscal quarter-end is used to forecast GDP growth. We impose Requirement (2) to exclude financial firms, Requirement (3) to match fiscal quarters, and Requirement (4) to remove observations with missing variables. Requirement (5) dispels concerns about negative denominators.
After imposing these data requirements, we select country/quarters that contain 50 or more firm/quarters, because aggregating a few firms cannot diversify firm-specific information. We delete observations missing MR c;q , and ΔGDP c;q . Finally, we choose countries that have at least 20 country/ quarter observations. Our final sample contains 1,239 country/quarters covering 21 countries (Argentina, Brazil, Canada, Denmark, Finland, Germany, Greece, Hong Kong, Italy, Japan, Mexico, Peru, the Philippines, Singapore, South Korea, Spain, Sweden, Taiwan, Thailand, Turkey, and the US). Table 1 presents the descriptive statistics (Panel A) and correlation matrix for the variables (Panel B). Table 2 provides the descriptive statistics by country.

Improved forecast accuracy of GDP growth with aggregate-level profitability drivers
We examine the usefulness of aggregate-level profitability drivers to forecast future GDP growth in the global context. Table 3 compares the regression results with and without ΔRNOA c;q , while Table 4 presents the results with and without ΔROS c;q and ΔATO c;q . Our focus is whether aggregatelevel profitability and its drivers are useful for predicting GDP growth. Therefore, we focus on the differences in the adjusted R squares to test their incremental usefulness.
In Table 3, the differences in the adjusted R squares are not significant when dependent variables are GDP growth at quarters q + 1 and q + 2. Meanwhile, all the adjusted R squares significantly increase after adding aggregate-level profitability drivers in Table 4. When predicting GDP growth h quarters ahead, including ΔROS c;q and ΔATO c;q in the models significantly raises the adjusted R squares by 0.3% (h = 1), 0.8% (h = 2), 1.4% (h = 3), and 0.8% (h = 4). These results mean that considering aggregate-level profitability drivers significantly improves the forecast accuracy of GDP growth, which is consistent with Konchitchki and Patatoukas (2014b). In addition, we find significantly positive coefficients on ΔROS c;q regressing GDP growth one quarter ahead, which is consistent with the findings of Konchitchki andPatatoukas (2014a, 2014b). Overall, aggregate-level profitability drivers are useful for forecasting future GDP growth worldwide.

Improved forecast accuracy by considering crises' effects
In this subsection, we focus on whether considering crises' effects improves the forecast accuracy of GDP growth. Table 5 indicates the results of the regression models with and without the crisis dummy and its interaction terms. This table shows that the adjusted R squares significantly increase after including the crisis dummy and its interaction terms. When predicting GDP growth h quarters ahead, including these variables increases the adjusted R squares by 1.7% (h = 1), 3.8% (h = 2), 3.3% (h = 3), and 1.7% (h= 4). 10 These results indicate that considering the crises' effects significantly improves the forecast accuracy of the model. However, a concern remains that the significant increases in adjusted R squares in Table 5 might not be caused by the interaction terms on the aggregate-level profitability drivers (Crisis q � ΔROS c;q , and Crisis q � ΔATO c;q ). Instead, the crisis dummy and its interaction terms on GDP growth or market returns (Crisis q , Crisis q � ΔGDP c;q , and Crisis q � MR c;q ) might lead the increases. To dispel this concern, Table 6 displays the differences in the adjusted R squares of the regression models with and without interaction terms on the aggregate-level profitability drivers. We observe the significant increases in the adjusted R squares after including aggregate-level profitability drivers and their interaction terms. The drivers significantly raise the adjusted R squares by 0.7% (h = 1), 1.0% (h = 2), 1.6% (h = 3), and 0.5% (h = 4). These results show that considering both aggregate-level profitability drivers and crises' effects improves the forecast accuracy of GDP growth. Table 7 shows the results on the implications for stock valuation of predicting future real GDP growth using aggregate-level profitability drivers. 11 Using our full sample or sub-sample of emerging countries, we observe that the coefficients on ΔGDP ACC c;qþ1 are insignificant. Meanwhile, the coefficients on ΔGDP ACC c;qþ1 are significantly positive when using the sub-sample of developed countries, which is consistent with US evidence (Konchitchki and Patatoukas 2014b). These results suggest that predicting GDP growth using aggregate-level profitability drivers is relevant for stock valuation in developed countries, but not in emerging countries.  Notes: The table reports the results from country fixed-effect regressions. We report t-statistics using robust standard errors in parentheses proposed by White (1980). N is the number of observations. AR 2 is the adjusted R square of the regressions. Diff is the differences in the adjusted R squares between the two models whose dependent variables are the same. Chi 2 is Chi-Squared statistics on Diff estimated by the likelihood-ratio tests. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Notes: The table reports the results from country fixed-effect regressions. We report t-statistics using robust standard errors in parentheses proposed by White (1980). N is the number of observations. AR 2 is the adjusted R square of the regressions. Diff is the differences in the adjusted R squares between the two models whose dependent variables are the same. Chi 2 is Chi-Squared statistics on Diff estimated by the likelihood-ratio tests. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Notes: The table reports the results from country fixed-effect regressions. We report t-statistics using robust standard errors in parentheses proposed by White (1980). N is the number of observations. AR 2 is the adjusted R square of the regressions. Diff is the differences in the adjusted R squares between the two models whose dependent variables are the same. Chi 2 is Chi-Squared statistics on Diff estimated by the likelihood-ratio tests. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

Robustness checks
We conduct three robustness checks (untabulated). First, we examine whether different results are observed when we use nominal GDP, instead of real GDP. Although the adjusted R squares of the regression using nominal GDP are generally higher, we confirm that using nominal GDP does not affect our results much. Second, we check whether the significance of the coefficients is stable, using other standard errors. In the abovementioned tests, we use heteroskedasticity-consistent standard errors proposed by White (1980). We confirm that the significance does not change much when using heteroskedasticity-and autocorrelation-consistent standard errors proposed by Newey and West (1987) with lag length varying from one to four.
Third, we investigate whether aggregate-level profitability drivers and their interaction terms with the crises' dummy improve the forecast accuracy of GDP growth in both developed and emerging countries. After splitting our sample into a sub-sample of developed countries and that of emerging countries, we confirm significant increases in adjusted R squares after including aggregate-level profitability drivers and their interaction terms with the crises' dummy. Therefore, we conclude that aggregate-level profitability drivers are useful for predicting future GDP growth in both types of countries.

Additional analyses
To deepen our understanding of the crises' effects, we test whether considering each crisis improves the forecast accuracy of future GDP growth similarly (untabulated). In the abovementioned Notes: The table reports the results from country fixed-effect regressions. We report t-statistics using robust standard errors in parentheses proposed by White (1980). N is the number of observations. AR 2 is the adjusted R square of the regressions. Diff is the differences in the adjusted R squares between the two models whose dependent variables are the same. Chi 2 is Chi-Squared statistics on Diff estimated by the likelihood-ratio tests. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
analyses, we use a dummy variable representing the three crises (Crisis q ): the burst of the dot-com bubble in 2001, the global financial crisis from 2007 to 2009, and the recession induced by the COVID-19 pandemic in 2020. We replace Crisis q with Crisis 01 q , Crisis 07À 09 q , or Crisis 20 q to ascertain the similarity between these crises. Crisis 01 q , Crisis 07À 09 q , and Crisis 20 q are crisis dummies that take the value of one when our regression models include at least one of the variables using data during the burst of the dot-com bubble, the global financial crisis, and the recession induced by the COVID-19 pandemic, respectively. We investigate whether including each crisis dummy and its interaction terms with aggregate-level profitability drivers significantly increases the adjusted R squares. In these tests, we exclude country/quarter observations during the other crises to reduce their effects.
The results with Crisis 07À 09 q and Crisis 20 q are both similar to our main results. We confirm that the adjusted R squares significantly increase after including these crisis dummies and their interaction terms. Although data availability limits the number of observations about the ongoing recession induced by the COVID-19 pandemic, these findings suggest that considering the global financial crisis and the recession induced by the COVID-19 pandemic improves the forecast accuracy of future GDP growth.
Meanwhile, the increases in the adjusted R squares by including Crisis 01 q and its interaction terms range from 0.0% to 0.4%. Their significance levels are 10% at best. This could be due to the weaker impact of the dot-com bubble on the real economy. The dot-com bubble mainly affects the stock market and its impact on the real economy is more limited than that of other crises in our study. 12 These results suggest that crises that severely damage the real economy should be considered in constructing the forecast model of GDP growth.

Conclusion
The recent accounting research stream known as macro-accounting addresses the potential of accounting information to support economists. Konchitchki and Patatoukas (2014b) report that We report t-statistics using robust standard errors in parentheses proposed by White (1980). ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively, using two-tailed tests. N is the number of observations. AR 2 is the adjusted R square of the regressions.
aggregate-level profitability drivers are incrementally useful for forecasting future GDP growth in the US. To extend this research, we test whether these drivers are useful for predicting future GDP growth using global data. We also investigate whether considering crises' effects improves their forecast model of GDP growth with aggregate-level profitability drivers. To obtain implications for stock valuation in various markets, we examine whether predicting GDP growth with aggregatelevel profitability drivers is relevant for stock valuation. We obtain three results. First, aggregate-level profitability drivers possess incremental information on future GDP growth around the world. Second, we show that considering crises' effects further improves the forecast model of future GDP growth. Third, we suggest that predicting GDP growth with aggregate-level profitability drivers is relevant for stock valuation in developed countries, but not in emerging countries.
We contribute to the macro-accounting research in three ways. First, our results extend the external validity of the usefulness of aggregate-level profitability drivers for predicting GDP growth by using global data of 21 countries. We show that aggregate-level profitability drivers significantly improve the forecast accuracy of GDP growth around the world. To the best of our knowledge, our study is the first to investigate the usefulness of aggregate earnings in emerging countries. Our findings suggest that economists of emerging countries could also incorporate aggregate earnings into their GDP forecasts. Second, we extend the research of Konchitchki andPatatoukas (2014a, 2014b) by proposing the better usage of aggregate earnings information for forecasting GDP growth. We find that considering the crises' effects improves their forecast model of future GDP growth. Since the world economy is currently suffering from recession induced by the COVID-19 pandemic, our findings may make a timely contribution to current GDP forecasts. Third, our results suggest that predicting GDP growth with aggregate-level profitability drivers is relevant for investors in developed countries, but not in emerging countries. Although Konchitchki and Patatoukas (2014b) report the relevance in the US, it should be restricted in other developed countries based on our results.
However, the current study has the following limitation. We do not empirically investigate the mechanisms underlying the links between aggregate-level profitability drivers and future GDP growth. Shivakumar and Urcan (2017) propose two explanations for the relation between aggregate earnings growth and future inflation: firms' investment channel and consumers' consumption channel. Future research could investigate the mechanisms through which aggregate earnings growth affects future GDP growth following Shivakumar and Urcan (2017). Such research would broaden our knowledge of the link between accounting and the macroeconomy.
Nonetheless, our findings support the view that accounting information can assist macroeconomists and macro policy setters around the world. Although accounting standard setters have traditionally overlooked these groups as users, macro-accounting research could extend the potential users of accounting information. We expect that the field of accounting further contributes to macroeconomic discussion.
Notes expectations about discount rates and investors' expectations about growth as other prior studies show (e.g. Patatoukas 2014; Yoshinaga 2016). 6. Konchitchki and Patatoukas (2014b) additionally decompose ROS into the ratio of operating income before depreciation to sales and the ratio of depreciation to sales. However, unlike sales and operating profit, the depreciation is not timeously released by quarter in many countries. When we exclude firm/quarter observations without the depreciation data, the number of firm/quarter observations becomes less than one-third of the original (untabulated). If we calculate aggregate-level variables with these reduced observations, they do not reflect the general performance of listed firms in the country. Therefore, we do not use depreciation in our empirical analyses. 7. Subsection 3.2. contains only the decomposed models, because Konchitchki and Patatoukas (2014b) show that the adjusted R squares of the decomposed model is higher than those of the model using changes in RNOA. 8. Our sample includes 11 developed countries (Canada, Denmark, Finland, Germany, Hong Kong, Italy, Japan, Singapore, Spain, Sweden, and the US) and 10 emerging countries (Argentina, Brazil, Greece, Mexico, Peru, the Philippines, South Korea, Taiwan, Thailand, and Turkey), based on the MSCI definition. See https://www. msci.com/market-classification (accessed on 10 August 2020). 9. We exclude the crisis dummy and its interaction terms in Equation (10) to reduce the multicollinearity concern. We confirm that the VIF of regressions using Equations (6) and (7) gets over 200 and that the estimated coefficients are not stable when including them in Equation (10) (untabulated). 10. We find that Crisis q � ΔROS c;q has significantly positive coefficients but ΔROS c;q has significantly negative coefficients, although these coefficients are not our focus. These results suggests that the positive effects of aggregate-level changes in ROS on GDP growth reported in Konchitchki andPatatoukas (2014a, 2014b) can be observed during crisis periods. 11. We do not create a dummy variable to classify observations as in developed or emerging countries, since our models already include another dummy variable (Crisis q ) and its interaction terms. If we create a new dummy variable (e.g. a developed country dummy), triple interaction terms will be included in our model, which will cause a multicollinearity concern and difficulties in the interpretation of the results. Thus, we split our sample into developed and emerging countries. 12. We additionally investigate the differences in the average GDP growth between each of the crisis periods and non-crisis periods using unequal variance t-tests (untabulated). As a result, GDP growth during the global financial crisis and during the recession induced by the COVID-19 pandemic is significantly lower than during non-crisis periods at the 1% level, respectively. However, the difference in GDP growth is not significant between during the 2001 crisis periods and non-crisis periods.