One and a half decades after the seminal paper by Cooper et al. (2008) documenting that corporate investment is negatively related to subsequent stock returns, the so called asset growth anomaly, researchers still actively investigate the underlying origins of its occurrence. Early studies on the US stock market decomposed firms’ asset growth rates into subcomponents and linked those subcomponents to the two major competing theories of mispricing and market efficiency ( Titman et al., 2004 ; Cooper et al., 2008 ). On the other hand, early studies using international settings, outside the US, used country-level proxies to disentangle between these two theories ( Titman et al., 2013 ; Watanabe et al., 2013 ).

Recently published studies on the field provide new theories or empirical settings, directly or indirectly linked to the above-mentioned competing theories. Using a US setting, Goto et al. (2020) provide evidence that managerial equity incentives mitigate managers’ empire-building motives, leading to a less pronounced asset growth effect. Ma et al. (2023) argue that the anomaly is due, in part, to investors’ behavioral biases since high asset growth creates more challenging informational environments. Abdoh and Varela (2021) show that changes in total factor productivity serve as one of the drivers behind the asset growth effect. Studies outside the US market using firm-level data either relate their empirical findings to few and specific channels of risk or mispricing ( Cai et al., 2019 ; Artikis et al., 2023 ) or indirectly favor one of the existing theories without formally testing or addressing a specific channel ( Papanastasopoulos, 2017 ; Artikis et al., 2021 ).

However, we have to mention here that one study using international stock markets deviates from the pattern presented. Notably, Artikis et al. (2022) , using firm level data, for non-financial listed firms in European stock markets, decompose firms’ total assets into two subcomponents, drawing their inspiration from Richardson et al.’s (2006) decomposition on accruals. The authors link those subcomponents to the most cited channels under both the mispricing and the rationality camp. However, the authors also employ country-level proxies to address their developed hypotheses.

The present study aims to provide new insights into the ongoing debate on the origins of this prominent asset pricing puzzle in European stock markets by employing an empirical setting that encompasses a) firms’ earnings sign as a heuristic, b) asset growth decomposition proposed by Artikis et al. (2022) and c) country regions proposed by the OECD¹ and the European Commission. Unlike previous studies on the drivers of the asset growth anomaly in European equity markets using country level proxies ( Watanabe et al., 2013 ; Titman et al., 2013 ; Artikis et al., 2022 ), all variables employed in this study are measures at the firm-level and no country proxies are employed. In addition, we employ both Artikis et al. (2022) asset growth decomposition combined with firms’ earnings sign serving the role of a heuristic as a more well-rounded indicator of mispricing/rationality, in order to provide further insights on the rationality versus mispricing debate behind the occurrence of the asset growth anomaly.

The present paper’s contribution is as follows:

First, it expands upon recent research on the field in the EU by examining 24 countries including Bulgaria, Check Republic, Croatia, Cyprus, Hungary and Russia, countries neglected by recent studies on the EU field, although represented in worldwide organizations (e.g. OECD). We separate these countries into two country regions, namely, the Eastern-Southern and the Western-Northern country regions. Countries in Western Europe had economically liberalized far earlier than countries in Eastern European, which started liberalizing after the dissolution of the Soviet Union. Consequently, Eastern European members of the EU joined it much later than their western counterparts, which further gave Western Europe an advantage in free trade, access to the European market and financial support. Lastly, structural problems within these countries, like corruption and inefficient governance, have only exacerbated the problem. According to the 2020 CPI, several Eastern European countries have relatively high levels of corruption, including Bulgaria, Romania, and Belarus, which all scored below 50 on a scale of 0 (highly corrupt) to 100 (very clean).

Second, by combining firms’ earnings sign ( Papanastasopoulos, 2017 ) and Artikis et al. (2022) asset growth decomposition we can address the origins of the asset growth effect (mispricing or rationality) by using only firm level data since a) we re-enforce inferences provided by Papanastasaopoulos (2017) by including a subcomponent representing investment decisions and a subcomponent representing accounting distortions and b) at least indirectly addressed the linkage of asset growth subcomponents to the rational and mispricing camp without employing time invariant country level proxies.

Finally, our data ranges from 1996 to 2020 in terms of accounting data and from 1997 to 2022 in terms of firms’ return investigating the COVID-19 pandemic as a major macroeconomic event that tremendously affected the financial markets ( Okorie & Lin, 2021 ; Wang et al., 2021 ). Thus, the present paper fills important gaps in the literature by providing a more nuanced understanding of an international asset growth anomaly.

The remainder of this study is organized as follows. Section 2 presents the research design and the testable hypotheses proposed herein. In Section 3, we provide details about our data, sample formation, and variable measurement. In Section 4, we discuss the empirical results, and Section 5 offers concluding remarks.

Consensus has not yet been reached on what causes the anomaly. Two prominent theories have been proposed; one assumes some form of irrationality and the other is potentially consistent with rationality. Under the rational camp, two prominent channels are used to explain the negative effect of firms’ asset growth on subsequent stock returns. The real options theory ( Berk et al., 1999 ; Carlson et al., 2004 ) suggests that when managers invest, real options are exercised and converted into less risky assets on the balance sheet leading to lower expected returns. On the other hand, the q-theory of optimal investment ( Lyandres et al., 2008 ; Li et al., 2009 ; Li & Zhang, 2010 ) suggests that managers invest up to the point where the expected cost of capital equals the expected return on investment.

Under the mispricing camp, four theories have been proposed by existing literature. The first explanation relates to agency problems and notably, to managers’ empire building tendencies ( Titman et al., 2004 ) pursuing their own personal agenda even if this is not aligned with firm’s best interests. The second theory relates to firms’ market timing behavior in financing decisions ( Baker & Wurgler, 2002 ), where corporate insiders opportunistically raise equity financing when their stocks are undervalued, exploiting their inside information. The third explanation relates to earnings management suggesting that high balance sheet growth could be exploited as firm executives manipulate earnings upwards, especially when raising external capital and/or there is a forthcoming merger and acquisition ( Teoh et al., 1998a ; Teoh et al., 1998b ; Dai et al., 2017 ).

The fourth explanation relates to investors’ extrapolation bias ( Lakonishok et al., 1994 ; Cooper et al., 2008 ), where investors excessively extrapolate firms’ past growth, overvaluing higher asset growth firms leading to subsequent lower returns. All mispricing based explanations center around naïve investors who do not properly and/or timely evaluate publicly available information. Thus, high asset growth firms’ lower returns are a form of market correction to the initial misreaction.

Both Watanabe et al. (2013) and Titman et al. (2013) argue that if the strongest asset growth effect takes place is existent and robust in more developed economies, it can only be in line with a risk-based explanation of the overall puzzle since, they exhibit a greater degree of market efficiency, less room for managerial discretion over earnings, greater investors’ sophistication, less limits-to-arbitrage and stronger investors’ protection mechanisms. With respect to less developed economies, Titman et al. (2013) ² argue that under the optimal investment hypothesis, the asset growth anomaly is also expected to be more pronounced in less developed economies since they are of fewer market participants, fewer arbitrageurs and greater fluctuations in wealth leading to greater fluctuations in returns. However, Watanabe et al. (2013) relate a pronounced asset growth anomaly in less developed economies exclusively to mispricing.

Artikis et al. (2022) decomposed firms’ asset growth into two separate components: one that reflects real investment growth and another that captures accounting distortions and/or less efficient use of existing capital. Extrapolating the argumentation provided by Richardson et al. (2006) , the authors argue that if asset growth encapsulates real investment growth, higher sales should lead to an increase in assets associated with an increase in output. On the contrary, if there is no increase (or even a decrease) in sales, then a decline in efficiency, either because of accounting distortions or because of less efficient use of existing capital should have led to an increase in firm’s total assets.

A direct implication is that the effects of diminishing marginal returns to increased investment, favoring a risk-based explanation, should be limited to the growth component but, the very same component could also be consistent with the opportunistic use of managerial discretion to generate accounting distortions that temporarily inflate earnings, favoring a mispriced-based explanation ( Artikis et al., 2022 ). On the other side, the effects of temporary accounting distortions that manipulate earnings upwards and/or reduce efficiency should be captured mainly by the efficiency component of asset growth.

Studies in the field ( Ma et al., 2023 ) argue that heuristics may serve as the basis for some form of behavioral biases (that is, some form of mispricing). Papanastasopoulos (2017) provides evidence that the asset growth anomaly is more pronounced in loss firms. Ma et al. (2023) argue that high asset growth creates more challenging informational environments. High asset growth future returns and cash flows are less predictable and therefore create uncertainty regarding stock valuation level. Thus, naïve investors facing a high degree of uncertainty are more likely to rely on heuristics to arrive at valuation level estimates. Papanastasopoulos (2017) argues that since losses are more transitory and less informative about future prospects ( Hayn, 1995 ; Collins et al., 1999 ), traditional valuation models may lead to unreliable estimates of firm values for loss firms. Using losses as a heuristic may lead to systematic errors in translating financial information into stock returns. In addition, the author argues that investors are more likely to misinterpret asset growth of loss firms, favoring a mispricing based story³.

Adopting the notion that losses might serve as an additional indicator of mispricing, we expect that the combined information with a rational based indicator (e.g. asset growth and/or investment growth component) should lead to a weaker effect of those variables, since the disagreement between the signals implies that signal is more likely to be due to noise. But when both signals agree (either asset growth and/or accounting distortion component), the conditional probability that each signal is due to mispricing rather than noise is high.

All of the above being said, we form our hypotheses as follows:

H1: In Western-Northern sub-region combining firms’ profits (losses) with asset growth rates and/or the investment growth subcomponent should lead to a more (less) pronounced effect of both variables on subsequent stock returns, under rationality (mispricing).

H2a: In Eastern-Southern sub-region combining firms’ losses with asset growth rates, the investment growth subcomponent and/or the accounting distortion subcomponent should lead to a more pronounced effect of all variables on subsequent stock returns, under mispricing.

H2b: In Eastern-Southern sub-region combining firms’ profits with asset growth rate and/or the investment growth subcomponent should lead to a more pronounced effect of both variables on subsequent stock returns, whereas combining firms’ profits with the accounting distortion subcomponent should lead to its less pronounced effect on subsequent stock returns, under rationality.

The United Kingdom is by far the largest market in the sample. In addition, Russia is the second largest market in the Eastern-Southern country region and of these European countries arbitrarily excluded from previous studies on the field. Our sample also includes the COVID19 pandemic. The outbreak of COVID-19 pandemic has tremendously affected the financial markets, through a fractal contagion effect ( Okorie & Lin, 2021 ), leading to efficiency decreases ( Wang et al., 2021 ). Thus, in this section, we examine whether our initial results still hold if both the UK and Russia are excluded from the analysis, as well as the COVID-19 pandemic period.

Specifically, Τable 7 excludes the UK and Russia, while Table 8 excludes the COVID-19 pandemic. Initial patterns in coefficients are qualitatively the same. Still, the largest asset growth anomaly takes place in less developed economies and is mitigated in magnitude when more development countries along with firms’ losses are considered. Reported results suggest that there is a small decrease in AG’s coefficient estimate when loss firms are considered after excluding Russia from the sample (from −0.1701 to −0.1617), but a small increase when loss firms are considered after excluding the COVID-19 pandemic (from −0.1701 to −0.1826).

Table 9 & Table 10 present the results of asset growth’s decomposition. Table 9 excludes the UK and Russia, while Table 10 excludes the COVID-19 pandemic. Results are consistent with the ones provided in the initial sample. SG’s explanatory power is the strongest when more developed countries are considered and mitigated when loss firms are considered. ΔΑΤ’ s effect is more pronounced in less developed countries and mitigated when profit firms are considered.

Table 7 reports the results from cross-sectional regressions of annualized raw returns on asset growth, after controlling for size, book-to-market ratio and momentum. Annual cross-sectional regressions are estimated over the period 1996-2020 using OLS with clustered standard errors. From the initial sample, we have excluded for robustness purposes Russia and the United Kingdom. The regressions are estimated for the entire sample of firms, for the subsample of eastern-southern firms and for the subsample of western-northern firms. We report separate coefficients for each country region and coefficients for all countries combined. The all-countries row is formed with firms from all countries (i.e., pooled sample). Panel A reports the results from cross-sectional regressions of annualized raw returns on asset growth, after controlling for size, book-to-market ratio and momentum, conditional on firms’ classification into Loss firms, whereas Panel B the results from cross-sectional regressions of annualized raw returns on asset growth, after controlling for size, book-to-market ratio and momentum conditional on firms’ classification into Profit firms. Firms’ classification into Profit and Loss firms is based on firm’s year-end Net Income. If firm’s Net Income is positive (negative), the firm is classified as a Profit (Loss) firm. The classification takes place within each sub-sample and the polled sample as well). In the last row, we report the difference in AG coefficient from cross-sectional regressions between country region subgroups (i.e. Western-Northern/Eastern-Southern). Standard errors for coefficient’ differences estimates are given by: $se\left(b_H-b_L\right)=Var\left(b_H\right)+Var\left(b_L\right)-2Cov\left(b_H,b_L\right)$ where $Cov\left(b_H,b_L\right)$ is estimated by multiple-equation OLS. ***, **, * represent statistical significance at 1%, 5%, and 10% level, respectively (one-tailed). Regression estimated is: $RE{T}_{i,t+1}={\gamma }_{0,t}+{\gamma }_{1,t}A{G}_{i,t}+{\gamma }_{2,t}\ln {\left(SZ\right)}_{i,t}+{\gamma }_{3,t}\ln {\left(BM\right)}_{i,t}+{\gamma }_{4,t}MO{M}_{i,t}+u_{i,t+1}$.

Table 8 reports the results from cross-sectional regressions of annualized raw returns on asset growth, after controlling for size, book-to-market ratio and momentum. Annual cross-sectional regressions are estimated over the period 1996-2020 using OLS with clustered standard errors. From the initial sample, we have excluded for robustness purposes the COVID19 period. The regressions are estimated for the entire sample of firms, for the subsample of eastern-southern firms and for the subsample of western-northern firms. We report separate coefficients for each country region and coefficients for all countries combined. The all-countries row is formed with firms from all countries (i.e., pooled sample). Panel A reports the results from cross-sectional regressions of annualized raw returns on asset growth, after controlling for size, book-to-market ratio and momentum, conditional on firms’ classification into Loss firms, whereas Panel B the results from cross-sectional regressions of annualized raw returns on asset growth, after controlling for size, book-to-market ratio and momentum conditional on firms’ classification into Profit firms. Firms’ classification into Profit and Loss firms is based on firm’s year-end Net Income. If firm’s Net Income is positive (negative), the firm is classified as a Profit (Loss) firm. The classification takes place within each sub-sample and the polled sample as well). In the last row, we report the difference in AG coefficient from cross-sectional regressions between country region subgroups (i.e. Western-Northern/Eastern-Southern). Standard errors for coefficient’ differences estimates are given by: $se\left(b_H-b_L\right)=Var\left(b_H\right)+Var\left(b_L\right)-2Cov\left(b_H,b_L\right)$ where $Cov\left(b_H,b_L\right)$ is estimated by multiple-equation OLS. ***, **, * represent statistical significance at 1%, 5%, and 10% level, respectively (one-tailed). Regression estimated is: $RE{T}_{i,t+1}={\gamma }_{0,t}+{\gamma }_{1,t}A{G}_{i,t}+{\gamma }_{2,t}\ln {\left(SZ\right)}_{i,t}+{\gamma }_{3,t}\ln {\left(BM\right)}_{i,t}+{\gamma }_{4,t}MO{M}_{i,t}+u_{i,t+1}$.

Continued

Table 9 reports the results from cross-sectional regressions of annualized raw returns on asset growth’s decomposition, after controlling for size, book-to-market ratio and momentum. Annual cross-sectional regressions are estimated over the period 1996-2020 using OLS with clustered standard errors. From the initial sample, we have excluded for robustness purposes Russia and the United Kingdom. The regressions are estimated for the entire sample of firms, for the subsample of eastern-southern firms and for the subsample of western-northern firms. We report separate coefficients for each country region and coefficients for all countries combined. The all-countries row is formed with firms from all countries (i.e., pooled sample). Panel A reports the results from cross-sectional regressions of annualized raw returns on asset growth’ decomposition, after controlling for size, book-to-market ratio and momentum, conditional on firms’ classification into Loss firms, whereas Panel B the results from cross-sectional regressions of annualized raw returns on asset growth’ decomposition, after controlling for size, book-to-market ratio and momentum conditional on firms’ classification into Profit firms. The first classification into Profit and Loss firms is based on firm’s year-end Net Income. If firm’s Net Income is positive (negative), the firm is classified as a Profit (Loss) firm. The classification takes place within each sub-sample and the polled sample as well. In the last row, we report the difference in SG, ΔΑΤ and (SG * ΔΑ) coefficients from cross-sectional regressions between country region subgroups (i.e. Western-Northern/Eastern-Southern). Standard errors for coefficient’ differences estimates is given by: $se\left(b_H-b_L\right)=Var\left(b_H\right)+Var\left(b_L\right)-2Cov\left(b_H,b_L\right)$ where $Cov\left(b_H,b_L\right)$ is estimated by multiple-equation OLS. ***, **, * represent statistical significance at 1%, 5%, and 10% level, respectively (one-tailed). Regression estimated is: $RE{T}_{i,t+1}={\gamma }_{0,t}+{\gamma }_{1,t}S{G}_{i,t}+{\gamma }_{2,t}\Delta A{T}_{i,t}+{\gamma }_{3,t}\left(S{G}_{i,t}\ast \Delta A{T}_{i,t}\right)+{\gamma }_{4,t}\ln {\left(SZ\right)}_{i,t}+{\gamma }_{5,t}\ln {\left(BM\right)}_{i,t}+{\gamma }_{6,t}MO{M}_{i,t}+u_{i,t+1}$.

Table 10 reports the results from cross-sectional regressions of annualized raw returns on asset growth’s decomposition, after controlling for size, book-to-market ratio and momentum. Annual cross-sectional regressions are estimated over the period 1996-2020 using OLS with clustered standard errors. From the initial sample, we have excluded for robustness purposes the COVID-19 period. The regressions are estimated for the entire sample of firms, for the subsample of eastern-southern firms and for the subsample of western-northern firms. We report separate coefficients for each country region and coefficients for all countries combined. The all-countries row is formed with firms from all countries (i.e., pooled sample). Panel A reports the results from cross-sectional regressions of annualized raw returns on asset growth’ decomposition, after controlling for size, book-to-market ratio and momentum, conditional on firms’ classification into Loss firms, whereas Panel B the results from cross-sectional regressions of annualized raw returns on asset growth’ decomposition, after controlling for size, book-to-market ratio and momentum conditional on firms’ classification into Profit firms. The first classification into Profit and Loss firms is based on firm’s year-end Net Income. If firm’s Net Income is positive (negative), the firm is classified as a Profit (Loss) firm. The classification takes place within each sub-sample and the polled sample as well. In the last row, we report the difference in SG, ΔΑΤ and (SG * ΔΑ) coefficients from cross-sectional regressions between country region subgroups (i.e. Western-Northern/Eastern-Southern). Standard errors for coefficient’ differences estimates is given by: $se\left(b_H-b_L\right)=Var\left(b_H\right)+Var\left(b_L\right)-2Cov\left(b_H,b_L\right)$ where $Cov\left(b_H,b_L\right)$ is estimated by multiple-equation OLS. ***, **, * represent statistical significance at 1%, 5%, and 10% level, respectively (one-tailed). Regression estimated is: $RE{T}_{i,t+1}={\gamma }_{0,t}+{\gamma }_{1,t}S{G}_{i,t}+{\gamma }_{2,t}\Delta A{T}_{i,t}+{\gamma }_{3,t}\left(S{G}_{i,t}\ast \Delta A{T}_{i,t}\right)+{\gamma }_{4,t}\ln {\left(SZ\right)}_{i,t}+{\gamma }_{5,t}\ln {\left(BM\right)}_{i,t}+{\gamma }_{6,t}MO{M}_{i,t}+u_{i,t+1}$.

Tabulated results suggest that SG’s power to explain future returns is augmented (mitigated) by excluding either the UK or the COVID-19 period, when we consider both more developed (less developed) markets and profit (loss) firms. ΔΑΤ’s coefficient is augmented when we exclude Russia from the Eastern-Southern under profit firms. Its increase is even larger when we exclude the COVID-19 period under loss firms. On the other hand, although ΔΑΤ’s explanatory power remained almost unchanged when we excluded the UK from the Western-Northern country region, its power is mitigated in terms of magnitude and statistical significance when we excluded the COVID-19 period.

Compared to previous studies on a global asset growth effect, this research employed less aggregated data to provide new insights into its underlying origins. We employ Artikis et al. (2022) asset growth decomposition as well as losses as a heuristic to disentangle between a mispricing-bases and a risk-based explanation behind its occurrence. In addition, we examine both asset growth subcomponents, conditional on earnings’ sign, for more developed and less developed European economies.

The findings from portfolio and regression analysis suggest that both country regions exhibit a robust asset growth effect. Based on existing literature ( Watanabe et al., 2013 ; Titman et al., 2013 ), an asset growth effect in the Western-Northern country region is more aligned with a risk-based explanation. This finding is also supported by the fact that the asset growth effect is augmented when profit firms are considered. On the other hand, an existent asset growth anomaly in the Eastern-Southern country region rejects a combined q-theory and market discipline story. An asset growth effect in the Eastern-Southern region, which is augmented when loss firms are considered, is favoring a market inefficiency story.

In addition, our results are in line with Artikis et al. (2022) asset growth decomposition. Both components seem to account for the overall market puzzle in both country regions. The investment growth (accounting distortion) component is more pronounced in more developed (less developed) economies and mitigated (augmented) when loss firms are considered. Our initial results and inference remain robust when we exclude the United Kingdom and Russia or the COVID19 pandemic period from our analysis.

This work has been partly supported by the University of Piraeus Research Center.

^aNote: “W” denotes that the relevant data item comes from Worldscope.

Appendix B presents basic information of the sample by country. % of Participation is the percentage participation of each country in the overall sample. Time Period is the beginning and ending year of participation in the sample.

¹OCED focuses on Southeast Europe whereas the term Northwest Europe is adopted by the European Commission.

² Titman et al. (2013) argue that if the q-theory of investment and the behavioral/empire building theories are not mutually exclusive, but instead combined into one story, then one might also expect an offsetting positive relation between firms’ asset growth and subsequent stock returns.

³Moreover, loss firms are harder to arbitrage ( Baker & Wurgler, 2007 ) leading to a delayed correction of their initial misevaluation. Limits to arbitrage play an important role in explaining the asset growth anomaly ( Li & Zhang, 2010 ; Lam & Wei, 2011 ; Lipson et al., 2011 ). However, since proxies of investment frictions and limits to arbitrage are highly correlated, it is difficult to distinguish between prediction of the q-theory (with investment frictions) and the overinvestment theory (with limits to arbitrage). Watanabe et al. (2013) , however, concluded that in an international setting country characteristics representing limits-to-arbitrage theory had only limited power in explaining the asset growth anomaly.

⁴Since we form one year ahead forward-looking returns, monthly returns are collected for the period of 1988-2022.

⁵Note that firms with negative book value of equity are not considered in the final sample.

⁶Following the proposed methodology, if R_t or R_t−1 is greater than 300% and (1 + R_t)(1 + R_t−1)⁻¹ is less than 50%, both returns are discarded.

⁷The initial period of 1988-2020 was reduced due to lack of data within certain countries.

⁸According to Petersen (2009) , both OLS and the Fama-MacBeth standard errors are biased downward. Petersen (2009) reports evidence that only clustered standard errors are unbiased as they account for the residual dependence created by the firm effect. Thus, we estimate the OLS regression with clustered s.e. on one-dimensional clustering, i.e. separately for a time effect and a firm effect, as well as on two-dimensional clustering accounting for both a firm and a time effect. The results in all cases are qualitatively the same.

⁹The accounting distortions and/or efficiency component and the interaction term between the investment growth component and the accounting distortions and/or efficiency component enter the regressions with a negative sign since increases (decreases) in them lead to decreases (increases) in asset growth rate.