Accruals Quality, Shocks to Macro-uncertainty, and Investor Response to Earnings News

Abstract

Research shows that a firm’s prior accrual quality affects the market reaction to earnings news, as it allows investors to gauge the level of noise in the news. Investor response to earnings news tends to be weaker (stronger) when a firm’s accrual quality is lower (higher). We aim to examine how macroeconomic shocks affect investor reactions. Drawing on ambiguity aversion literature, we argue that if macroeconomic uncertainty spikes before the earnings announcement, investors will favor a ‘better safe than sorry’ attitude; that is, they will tend to react strongly (weakly) to all bad (good) news, taking less into account differences in accrual quality. We consistently find that macro-uncertainty shocks cause (i) a stronger reaction to bad news for low-quality (but not high-quality) firms and (ii) a weaker reaction to good news for high-quality (but not low-quality) firms. The results are robust to alternative model specifications and sensitivity tests. Additionally, we show that if macro-uncertainty resolves in the post-announcement weeks, investors correct their underreaction to high-quality good news, especially if the shock is not extreme.

Acknowledgement

We thank Beatriz García Osma (Editor) and two referees for their invaluable insights and suggestions. We also gratefully acknowledge the contributions of anonymous reviewers and workshop participants at the Department of Accounting of the Stockholm School of Economics, the 2021 Canadian Academic Accounting Association Annual Conference, the 2021 American Accounting Association Southwest Region Meeting, and the Fourth Israel Behavioral Finance Conference 2022. We also thank Bocconi University and Middle Tennessee State University for their support. This work was begun while Carlo D’Augusta was at Middle Tennessee State University.

Disclosure statement

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

Notes

1 Williams (2015) shows that increases (but not decreases) in the VIX, especially larger ones, affect investor response to good and bad news consistent with investors facing a situation of ambiguity. He interprets this finding to suggest that following decreases in the VIX (i.e., when uncertainty is resolving), investors behave as though the environment is not one of ambiguity (i.e., investors are confident about their knowledge of distribution parameters). In contrast, rising uncertainty shakes investors’ confidence and places them in an environment where ambiguity concerns significantly influence their processing of and reaction to earnings news.

2 For instance, Balakrishnan et al. (2016) and Francis et al. (2013) show a positive association between conditional conservatism and stock performance during the 2008 financial crisis, and Lang and Maffett (2011) show that financial reporting transparency has beneficial effects on a firm’s market liquidity, in particular during periods of crisis.

3 One such study is that by Caskey (2009). He shows that ambiguity-averse investors may prefer to trade based on aggregate signals that reduce ambiguity, even if it results in trading against investors with superior information. In subsequent studies, Hou (2015) adopts an ambiguity aversion framework to explain why the effect of discretionary accruals on the cost of capital is not diversified away. Ahmed et al. (2020) use such a framework to predict an association between past earnings volatility and stock returns during market crises. Hasan et al. (2018) find that individual investors increase their information demand during periods of greater market ambiguity.

4 For instance, Francis et al. (2007) hypothesize and find that the response coefficient is weaker for earnings announced by companies that exhibited lower accrual quality in the previous years, consistent with ‘fully Bayesian investors [who] place less weight on signals characterized by […] lower precision.’ Similarly, Ecker et al. (2006) build a proxy of investor perception of the risk that stems from low accrual quality and show that it is associated with a weaker response to earnings news. These results align with earlier research findings of a weaker reaction when the level of noise in the process generating the earnings signal makes current earnings a poor predictor of future realizations (Imhoff & Lobo, 1992; Lipe, 1990).

5 More in general, abundant literature has shown that firms develop a reputation for high-quality reporting and disclosure practices and that investors are aware of such reputation. For instance, Graham et al. (2005) interviewed more than 400 executives to determine the factors that drive reported earnings and disclosure decisions and find that they are ‘concerned about the company’s reputation for delivering reliable earnings and disclosing transparent information [believing that] a poor reputation for delivering transparent and reliable information can [hurt] stock performance.’ This finding is in line with earlier studies showing that companies’ reporting and disclosure reputations can vary across firms (Lang & Lundholm, 1993, 1996), and those good reputations are negatively associated with costs of capital (Healy et al., 1999; Welker, 1995). Similarly, in an experimental setting, Hodge et al. (2006) find that a good reporting reputation mitigates investor skepticism when the firm issues a signal that aligns with managers’ incentives.

6 For example, investors know they face a certain level of risk when they invest in a stock whose price is volatile, but they can precisely estimate the amount of volatility to expect. However, should unexpected events shake the investors’ confidence in the prior assessment of future volatility, the investors would now face a situation of ambiguity.

7 In Epstein and Schneider’s (2008, p. 198) words: ‘If an ambiguous signal conveys good (bad) news, the worst case is that the signal is unreliable (very reliable).’

8 This is the case discussed by Epstein and Schneider (2008).

9 In this study, we define accrual quality in a sense similar to Francis et al. (2007, p. 412), who argue that ‘a firm that sometimes overestimates, sometimes under-estimates – without any discernible pattern – will exhibit poor accounting quality’ and such poor quality will lead to low precision of the earnings signal. In other words, the low-AQ vs. high-AQ classification refers to the prior years’ noise level, which is ex-post observable. Therefore, high-AQ (low-AQ) identifies a situation where investors know past σ² to have been high (low). In contrast, ambiguity refers to investors’ confidence in their ability to use past noise to estimate current noise, which is ex-ante unobservable. In a low-ambiguity situation, investors are confident that current σ² is going to be in close proximity to past σ² levels. In a high-ambiguity situation, investors believe that current σ² could be anywhere within the [^Hσ², ^Lσ²] domain, regardless of past σ² levels.

10 We acknowledge that lower quality accruals could be driven either by management-specific factors (e.g., managers lacking expertise, the firm’s internal control system is deficient) or by fundamental factors (e.g., it is simply harder to avoid estimation errors when operating in a turbulent environment). Our hypotheses remain agnostic as to which one is the main driver. Indeed, investors could believe high-AQ’s expected noise E(σ²(ϵ)) to be smaller because the firm has superior managers, or it operates in an environment where accrual noise is lower or (likely) a combination of both. Whichever the case, macro-uncertainty shocks will reduce investors’ reliance on the conditional expectation E(σ²(ϵ) | AQ) when responding to the news. Nevertheless, we control for various factors correlated with fundamental volatility in all our models. Additionally, in an untabulated sensitivity test, we find that substituting the AQ proxy with two fundamental volatility proxies used by Williams (2015), namely cash flow volatility and daily returns volatility, does not replicate the results. These two facts suggest that our findings are not mechanically driven by fundamental volatility.

11 Consistent with Williams (2015), we expect β₃ to be insignificant in the absence of ambiguity concerns (i.e., when ΔVIX is negative and uncertainty is resolving).

12 Because control variables are standardized, in columns 2 and 4 the coefficient of UE represents investor response to earnings when ΔVIX is zero and all control variables are equal to their average values.

13 From this point onward, we focus on positive-ΔVIX observations.

14 The test UE^HIGH – UE^LOW > 0 at the bottom of Table 4 reports the statistical significance, with one-tailed p-values ranging between 11 percent and less than 1 percent.

15 Specifically, we obtain the residuals by running the following regression for each industry and fiscal year: TotalAccruals_it = δ₀ + δ₁1/Assets_it-1 + δ₂(ΔSales_it − ΔReceivables_it) + δ₃PPE_it + δ₄ROA_it + ϵ. For more details on the estimation of this model’s parameters, see Kothari et al. (2005).

16 This concern is already mitigated by the fact that we control for PriorRET and SDRET. The univariate correlation between our sentiment proxy and ΔVIX is less than 3 percent, which makes it unlikely that market sentiment is a significant driver of the results reported in Table 4.

17 This is consistent with Williams’ (2015) claim that such shocks are an ideal setting to test the effect of ambiguity concerns on the response to earnings because managers have not had the chance to observe such shocks when preparing financial statements.

18 In untabulated analysis, we also regress the absolute value of φ_it on a binary variable (Post-SOX) identifying earnings announced after 2002 and the control variables of Equation (3). The coefficient of Post-SOX is negative and significant at less than 1 percent level.

19 Therefore, we exclude from this test all announcements made before 2015 and those made in January or after April.

20 UE*COVID^HIGH (UE*COVID^LOW) is qualitatively similar (similar in magnitude but not significant) for good news (bad news) announcements. We note that the drop in sample size is a likely reason for the lower significance. Furthermore, we acknowledge the limitation of this approach, which, as noted by Williams (2015), fails to identify with precision the days when investors were exposed to surging ambiguity. Indeed, fully understanding investor reaction during a particular and eventful period, such as the COVID-19 pandemic, arguably requires additional and specific tests, which we leave to future research.