An Empirical Study of Effectiveness of Detecting Business Earnings Management Using Cross-sectional Two-stage Regressions

　　本研究之目的在於運用統計分析、測試及比較三個橫斷面模式所檢測出企業盈餘管理之顯著水準與檢定力。本研究異於過去研究者，乃於過去研究係採用時間序列方式推估異常應計項目，而本研究係採取橫斷面方式推估異常應計項目。本研究除測試Modified-Jones模式（Dechow et al., 1995）與Margin模式（Peasnell et al.,2000）之外，亦併同測試本研究所發展的WHL模式。本研究在推估異常應計項目時，亦採Healy（1985）所主張之兩階段迴歸法而逐步推估出異常應計項目。研究結果顯示，三個模式在隨機抽樣下均有良好的顯著水準；另外，在偵測模式檢定力方面，WHL模式與Margin模式在費用操縱數額達到前期總資產的百分之三到百分之七之間時檢定力，其顯著高於Modified-Jones模式；而在銷貨操縱方面，除Margin模式外，WHL與Modified-Jones兩個模式在所操縱數額達前期總資產的百分之十時，兩模式檢定力皆能達到百分之百；在銷貨毛利操縱與壞帳操縱方面，除Modified-Jones模式外，WHL與Margin兩個模式在所操縱數額達前期總資產的百分之十時，兩模式檢定力皆未能達到百分之百；此外，本研究新增之存貨操縱假設發現，WHL模式在存貨操縱下模式的檢定力非常顯著的高於其他兩個模式，尤其在所操縱的數額達到前期總資產的百分之七時，WHL模式的檢定力已達百分之百。
　　This research examines specification and power of three detecting earnings manipulation cross-sectional models by using statistical analysis. Differing from prior research whose abnormal accruals were estimated by time-series model, we adopt cross-sectional model. In addition to Modified-Jones model (Dechow et al., 1995) and Margin model (Peasnell et al., 2000), we also test WHL model derived from this paper. Similar to Healy (1985), we estimate abnormal accruals by two-stage regression as well. The results indicate that these three models all have good specification under random sample. Further, with respect to compare the power of models, WHL and Margin model have significant higher power than Modified-Jones model while the amount of expense manipulation is between 3% to 7% of prior total asset. By the way of sales manipulation, the power of these models can reach 100% while the amount goes to 10% of prior total asset except for Margin model. However, the evidence of bad-debt and margin manipulation hypothesis show that the power of these models can not reach 100% while the amount goes to 10% of prior total asset except for Modified-Jones model. Besides, the evidence of our inventory manipulation hypothesis shows that the power of WHL model is extremely higher than other models. Especially, the power of WHL model has already up to 100% while the amount of inventory manipulation rises to 7% of prior total asset.