본 연구는 이익조정연구에서 필수적으로 사용되고 있는 재량발생액 추정모형에 대한 근본적인 검토를 목적으로 한다. 이를 위해 이익조정에 관한 선행연구에서 가장 많이 사용된 수정존스모형(Dechow et al. 1995)의 논리적․실증적 결함을 조명하고, 대안이 될 수 있는 모형을 제안하며 이를 수정존스모형과 비교분석한다.
수정존스모형은 모형설정과정에서 표준화와 통계적 모형으로의 전환 순서가 바뀜에 따라 의도치 않게 기업규모의 역수가 모형에 포함되는 구조적 문제점을 안고 있다. 즉, 표준화된 변수들로 구성된 모형에 표준화되지 않은 기업규모의 역수가 변수로 포함되고 상수가 생략됨에 따라 회귀모형의 잔차인 재량발생액의 평균이 0이 아닌 문제점을 갖는다.
실증분석결과, 수정존스모형의 유동발생액 대리변수가 제 역할을 못하는 것으로 나타나 모형이 낮은 설명력을 갖는 주된 원인이라는 것을 확인하였다. 또한, 수정존스모형은 비재량발생액을 과소 추정하여, 비재량적인 부분을 재량적인 부분으로 오도하는 경향이 있는 것으로 나타났다. 본 연구에서는 이러한 점을 고려하여 추정오류를 제거시키되 모형의 설명력을 향상시킬 수 있는 두 개의 신모형을 제안한다.
한국자료를 대상으로 연구한 결과 본 연구가 제안한 두 개의 신모형은 수정존스모형에 비하여 실증적 우수성을 갖는다. 모형의 실증적 타당성을 세 가지 차원에서 검토한 결과를 요약하면 다음과 같다. 첫째, 모형의 종합적인 적합도를 나타내는 수정R2로 판단할 때 수정존스모형은 평균 0.13에 그친 반면 본 연구에서 제안한 두 개의 신모형은 각각 0.24와 0.26으로 두 배 정도 향상된 모형적합도를 보였다. 둘째, 모형의 부적합성(R2가 0.05미만인 경우) 차원에서 판단할 경우 수정존스모형은 부적합 비율이 38%에 달하지만 본 연구가 제안한 신모형은 각각 15%와 12%에 그쳐서 훨씬 양호한 것으로 나타났다. 셋째, 회귀계수의 일관성을 검토한 결과, 수정존스모형에서 유동발생액의 대리변수로 사용한 현금매출액변동은 불규칙적으로 양과 음의 부호를 나타내어 일관성이 없었지만, 신모형에서 유동발생액의 대리변수로 사용한 순매출채권변동은 대부분의 경우에 아주 유의한 양의 값을 갖고 상대적으로 일관성이 높았다. 한편 다국자료로 표본을 확장하여 추가분석을 실시한 결과 한국자료와 유사한 결과를 보였으며, 존스모형(Jones 1991) 역시 수정존스모형과 동일한 문제점이 있다는 것을 확인하였다.
본 연구는 재량발생액 추정모형에 대한 근본적인 검토를 통하여, 선행연구에서 가장 많이 사용되고 있지만 검증력이 약하다고 지적되어온 수정존스모형과 존스모형의 문제점을 논리적․실증적으로 검토하였다는 점에서 시사점을 갖는다. 더불어 단순한 문제 지적에 그치지 않고 대안을 마련하여 새로운 모형의 상대적인 우수성을 확인하였다는 점에서 학술적 기여도를 찾을 수 있겠다.
The objectives of the study are threefold. The first is to fundamentally analyze the models for estimating discretionary accruals which are used as vehicles for analyzing various earnings management studies. To this end, the study logically and empirically analyzes the so-called modified Jones model (Dechow et al. 1995) which is most frequently used in various earnings management and other related studies. The second objective of the study is to propose new models as alternatives and shows empirically that they are better than the modified Jones model. The third objective is to examine the non-discretionary accruals, which are the fitted values of the accrual models, as a way to identify their relationships with the discretionary accruals. The statistical theory posits that the non-discretionary accruals and discretionary accruals should be independent of each other. However, if the models are misspecified, the independency may not be satisfied. Therefore, the analyses of the non-discretionary accruals will help us confirm the possible misspecification issues of the modified Jones model.
The modified Jones model can be briefly described as follows:TA/At-1=β0(1/At-1)+β1∆CREV/At-1+β2PPEt-1/At-1+ε (1)Here, TA represents total accruals, At-1 lagged total assets, △CREV changes in cash revenue, and PPE property, plant and equipment. The modified Jones model has appealing characteristics in two aspects. First, it decomposes accruals into current accruals and non-current accruals. Second, the selection of the proxy variables is based on the assumption that current accruals will be proportional to the change in revenues, and that non-current accruals will be related to the level of the property, plant and equipment which captures the most important noncurrent accruals, namely depreciation expense. These two appealing characteristics explain why many of the earnings management studies use the model widely.
We believe that the modified Jones model suffers from some misspecification problems.
The first misspecification comes from the fact that it wrongly switches its order of standardization stage and transformation stage into a statistical model. As a result of the reversal of the two stages, the modified Jones model inadvertently introduces the inverse of firm size which is a non-standardized variable into the model and unnecessarily suppresses the constant term. This will lower the goodness of fit of the model since the inverse of firm size should be independent of total accruals by construction. Furthermore, by suppressing the constant term, the regression line should go through the origin, which results in non-zero residuals. The second misspecification is less clear than the first one, but may have more pronounced impact on the separation of discretionary accruals from total accruals. The use of △CREV as a proxy for current accruals may not be as good as it was intended since there are a variety of other elements of current accruals. Finally, even though the use of PPE as a proxy for non-current accruals is quite appealing, it may fall short of sufficiently reflecting the characteristics of non-current accruals. One additional factor we need to consider is the level of intangible assets since the significance of intangible assets ever increases in the information and knowledge-prone modern business environments.
Taking into accounts these drawbacks of the modified Jones model, the current study proposes new models which have sound logical validity and show much higher explanatory powers than the modified Jones model. The two new models are briefly described below in equations (2) and (3).
<New model 1>TA/At-1=β0 + β1∆REV/At-1+β2∆NREC/At-1+β3PPEt-1/At-1+ε (2)Here, TA, At-1, △REV, △NREC and PPEt-1 represents total accruals, lagged total assets, changes in revenue, changes in net receivables, and lagged property, plant and equipment, respectively. New model 1 does away with the inverse of firm size and uses a free-floating constant term. △REV is used as a variable to control for changes in operating activity levels. △NREC is used as a proxy for current accruals. However, it uses the same proxy for non-current accruals as the modified Jones model.
<New model 2>TA/At-1=β0 + β1∆REV/At-1+β2∆NREC/At-1+β3PPEt-1/At-1+β4INTGt-1/At-1+ε(3)New model 2 adds lagged intangible assets (INTGt-1) as an additional proxy for non-current accruals.
Using 2,881 firm-year observations from firms listed on the Korea Stock Exchange, the study documents that the modified Jones model fails to properly separate out non-discretionary accruals from total accruals by overshooting negative non- discretionary accruals. Furthermore, the study documents that the modified Jones model fails to satisfy the statistical requirement that the residuals of regression models should have a zero mean value because of unwarranted inclusion of the inverse of firm size in the model instead of free-floating constant. In contrast, the new models proposed in the study have much more power than the modified Jones model in separating out non-discretionary accruals from total accruals.
By running regressions for 26 industries, the study documents that the average adjusted R-squares of the new models (0.24 or 0.26) are about twice that of the modified Jones model (0.13). The study also reveals that the explanatory powers of the variables included in the modified Jones model are generally very weak and not robust, while those of the variables in the new models are much more powerful in terms of explanatory powers and consistency across industries. Using correlation analyses as a means to check whether discretionary accruals can really be interpreted as discretionary, the study shows that discretionary accruals from the modified Jones model are not much different from total accruals. The results indicate that the validity of the modified Jones model is questioned in performing its function of properly separating out discretionary accruals from total accruals. In related analyses of mean difference tests and regression analyses which attempt to identify the major determinants of accruals, the study also documents that the modified Jones model fails to function properly in estimating discretionary accruals by showing that the determinants of total accruals and discretionary accruals are not distinguishable from each other. In contrast, the new models proposed in the study outperform the modified Jones model in terms of goodness of fit, consistency and robustness of individual variables, and the separation of discretionary accruals from total accruals.
The study did additional analyses using an augmented multinational data set of 13,216 firm-year observations from 44 different countries. In the additional analysis, the Jones model (Jones 1991) was also included in addition to the modified Jones model (Dechow et al. 1995) for comparison with the two new models proposed in this study. The results from the additional analyses confirmed the evidence found from the main study using the Korean dataset.
Through more fundamental analyses of the models for the estimation of discretionary accruals, the current study contributes to the literature by showing that the modified Jones model suffers not only from theoretical misspecification problems but also from poor empirical performance. The current study merits the contribution to the literature because it proposes new models which are theoretically sounder than the modified Jones model and shows that the new models outperform the modified Jones model and its variation.