핵심성과지표를 이용한 주식투자전략에 관한 연구

Abstract

Rouen et al. (2021)은 미국 상장기업의 연차보고서(10-K)에 머신러닝을 적용하여 일회성 손익을 식별하고이를 당기손익에 가감하여 산출한 핵심손익(core earnings)의 지속성과 가치관련성에 대하여 논의하였다. 이에 본 연구는 핵심손익과 전통적인 성과지표들의 지속성 및 가치관련성을 상대적으로 평가해보고, 주식투자전략에의 시사점을 도출하고자 한다. 2012-2020 기간의 국내 상장법인의 자료를 분석한 결과, 다양한 성과지표중에서 영업손익과 핵심손익의 미래이익의 예측력이 가장 높았으며 당기 및 차기 주식수익률과도 양(+)의상관관계를 보였다. 이러한 결과는 국내 주식시장이 영업손익과 핵심손익의 가치관련성(혹은 지속성)를 충분히즉각적으로 이해하지 못한다는 증거로 해석된다. 또한 영업외손익(당기손익과 영업손익의 차이)과 비핵심손익(당기손익과 핵심손익의 차이)은 일부 소규모 주식군(small-cap stocks)에서 당기 및 차기 주식수익률과 양(+)의상관관계를 보였다. 상기 결과는 영업손익과 핵심손익이 함께 주요 성과지표로 활용될 수 있으며, 이를 이용한투자전략이 일부 유효함을 의미한다. 본 연구에서는 핵심손익을 구성할 때 머신러닝 과정을 생략하고 손익계산서에서 즉각적으로 이용가능한 정보만을 활용하였으므로, 만약 핵심손익이 머신러닝 과정을 거쳐서 정제된다면영업손익보다 더 높은 지속성과 가치관련성을 나타낼 가능성을 배제할 수 없다. 그러나 핵심손익이 정교하게도출되는 과정에서 발생하는 비용을 감안하면, 손익계산서에 보고되는 영업손익을 핵심성과지표의 대안으로사용하는 것도 실익이 있다.

Separating transitory components from recurring components of earnings is a primary task in financial statement analysis. A recent study by Rouen et al. (2021) uses a proprietary database compiled by New Constructs and evaluate whether their earnings measure after excluding transitory components from GAAP earnings has incremental explanatory power for future earnings. They show that their “core” earnings measure (as referred to in Rouen et al. (2021)) is more persistent than Compustat-defined operating income, suggesting that a core earnings measure effectively excludes transitory (i.e., less persistent) components of GAAP earnings. They also find that investors and financial analysts do not immediately incorporate the differential implications of the recurring and non-recurring components of earnings into stock prices, and that a trading strategy based on their non-core earnings components yields annual abnormal return of 8%. Given the findings of Rouen et al. (2021), it is an empirical question whether a core earnings measure distinguishes between recurring and transitory components of earnings more effectively than do other earnings measures in countries outside the U.S. In this study, we follow the procedure outlined in Rouen et al. (2021) and attempt to exclude the non-recurring components from net income of Korean listed firms. Specifically, we identify the non-recurring components of earnings from the DataGuide database and estimate a core earnings measure by excluding from net income (1) currency fluctuations, (2) discontinued operations, and (3) gains and losses labeled as “other” on the income statement. We then compare the persistence and return predictability of core earnings and various commonly used adjusted income measures such as operating income, gross margin, income before income taxes, and income from continuing operations of Korean firms listed in KOSPI and KOSDAQ from 2012 to 2020. The results suggest that operating income is most persistent, followed by core earnings, income from continuing operations, gross margin, and income before income taxes when predicting one-year ahead net income. We further report that both non-operating income (i.e., net income minus operating income) and non-core earnings (i.e., net income minus core earnings) have some information contents in that they are predictive of one-year ahead net income with smaller magnitudes compared to their counterparts. Portfolio analyses based on operating income and core earnings show that both measures are predictive of future stock returns, implying that investors in the Korean stock market act as if they fail to fully incorporate the information contained in the transitory and permanent components of current earnings into stock prices, similar to the findings of Rouen et al. (2021). Our results confirm that 1) operating income and core earnings are comparably superior to other adjusted earnings measures as key performance indicators of a firm and 2) a trading strategy based on these measures yields profitable returns in the Korean stock market. Our results should be interpreted with caution. While Rouen et al. (2021) indicate that New Constructs uses both fundamental analysis and machine learning to extract transitory components of earnings, the core earnings measure used in this study only utilizes the income statement items readily available from the DataGuide database for Korean listed firms. A procedure that closely replicates Rouen et al. (2021) (including machine learning) may yield a core earnings proxy with improved persistence and future earnings predictability by distinguishing between transitory and permanent components of earnings more effectively. Despite the differences in the measurement, our results suggest that operating income is comparable to core earnings in terms of earnings and return predictability without considering a machine-learning algorithm for estimating core earnings. Given that implementation of a machine-learning algorithm could be costly, operating income could be an effective alternative to core earnings. Future research is warranted to further establish a cost-effective measurement of permanent components of earnings.