Fundamental analysis, low accruals, and the accrual anomaly: Korean evidence

ABSTRACT

Prior studies in Korea document that low accrual firms yield extremely low returns, driving away abnormal returns of an accrual-based trading strategy. We examine whether the performance of an accrual-based trading strategy can be improved using fundamental analysis to distinguish financially strong firms (‘winners’) from financially weak firms (‘losers’) within low accrual firms. Using Korean data from 1994 to 2018, our findings are summarised as follows. First, applying FSCORE in Piotroski (2000) [Journal of Accounting Research, 38(supplement), 1–41] to distinguish winners from losers within low accrual firms, we find that winners yield much higher future returns than losers. Second, after excluding losers in the low accrual group, the accruals-based hedge portfolio exhibits higher abnormal returns. Lastly, we find that, among low accrual firms, higher FSCORE is associated with less negative accruals, higher future probability, and lower probability of delisting. Overall, our findings imply that the extremely negative accruals (i.e., low accruals) do not signal good fundamentals, although Piotroski (2000) treats the negative sign of accruals as a universally positive signal of future performance. It also implies that investors do not fully incorporate the implications of low accruals for future performance.

KEYWORDS:

• accrual anomaly
• fundamental analysis
• financial statement analysis
• low accruals
• mispricing

JEL CLASSIFICATION:

• G14
• M41

Acknowledgements

This work was supported by Hankuk University of Foreign Studies Research Fund.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Please refer to Appendix A for a detailed definition of the winners and losers used in this study.

2 Strictly speaking, we mean that low accruals indicate large negative accruals. For simplification, we use the term ‘low accruals’ instead of ‘large negative accruals.’

3 When the lowest accrual quintile is grouped by decile or quintile portfolios based on the FSCORE, each group has too few observations to derive meaningful results

4 Desai et al. (2004) examine whether the accrual anomaly is a manifestation of the value-growth anomaly. Using several proxies for value and growth, they find somewhat mixed results, depending on the proxy for value and growth. However, in Table 4, they report that accruals and B/M, a proxy for value and growth firms, capture different types of mispricing.

5 Beaver et al. (2007) also examine the effect of delisting firms on the book-to-market effect. In contrast to the accrual effect, when delisting firms are included in the sample, the magnitude of hedge returns based on the book-to-market ratio increases. This is due to lower returns of the short position of the low book-to-market stocks portfolio.

6 Another important finding by Kim et al. (2015) is that the accrual anomaly does not exist when asset-deflated accruals are used, whereas the accrual anomaly exists when percent accruals (i.e., accruals deflated by the absolute value of net income) are used.

7 Other than using a summary measure, another line of literature uses statistical methodologies to extract information from financial ratios and predict stock returns. References are herein (Bartram & Grinblatt, 2018, 2021; Ogneva et al., 2020; Yan & Zheng, 2017). A different line of literature uses fundamental statement analysis to efficiently construct portfolios (Lyle & Yohn, 2021).

8 Another extension of Piotroski (2000) is to develop an industry specific fundamental summary measure that aggregates various industry specific financial ratios in order to identify mispriced stocks in a specific industry. Mohanram et al. (2018) develops a summary measure of fourteen bank specific variables (BSCORE in their terms) and use it to predict future earnings and stock returns.

9 To be clearer, we use accounting data from 1994 to 2018 and stock return data from April 1995 to March 2020. The timing difference between accounting data and stock returns data is due to the three-month lag of accounting data. All listed Korean firms are required to disclose their financial statements within three months after fiscal year-end (e.g., by end of March for firms with a December fiscal year-end). Accrual portfolios based on fiscal year t accounting data are constructed from the first day of April of year t+1 and held to the end of March of year t+2. This portfolio construction approach ensures a feasible trading strategy to every investor.

10 Most listed Korean firms have a fiscal year-end month of December.

11 We mainly use quintile grouping instead of decile grouping to sort based on accruals because quintile grouping includes more observations in the lowest accrual group. When accruals-based decile grouping is used for the lowest accrual portfolio, the results remain similar to our main findings.

12 Many prior studies also compute accruals in the same way (Dechow & Ge, 2006; Kim et al., 2015). For readers interested in various measurements of accruals, please refer to Larson et al. (2018).

13 When total assets are used to deflate accruals, many prior studies on accrual anomaly use the average of lagged and current period total assets as a deflator. To maintain consistency with the deflators of firms’ fundamentals used in FSCORE, we use lagged total assets to deflate accruals. When we deflate accruals by average of lagged and current period total assets, our results remain qualitative similar (untabulated).

14 This result is consistent with prior studies (Nam, 2009; Kim et al., 2015) that find insignificant accrual-based hedge returns based on accrual deciles.

15 For example, if a firm’s FSCORE is 0 or 1, then it is assigned to FSCORE quintile 1, and if a firm’s FSCORE is 9 or 10, then it is assigned to FSCORE quintile 5.

16 The exact medium value of FSCORE is 4.5. Thus, when the median value of FSCORE is used for a threshold, firms with FSCORE of value of 5 through 9 are classified into winners and more firms are included in winners than when 5 is used for a threshold. We also use 5 as a threshold for classification of winners and losers, and the results do not change.