在台灣經濟新報(TEJ)財報重編事件資料庫下,本研究樣本空間包括1993~2011年,經扣除原資料庫資料部分遺失或未記錄的年度資料後,共計19年的460筆年度資料,其變數包括財報重編原因為會計錯誤者(其值以1代表)、財報重編前後的營收差異(%)、營業利益差異(%)、稅後淨利差異(%)、資產差異(%)、股東權益差異(%)等數據為樣本資料。經過本研究實證分析後,本研究結論如下: 一、 財務變數的解釋能力比較:表4-4、表4-6、表4-8、表4-10、表4-12等數據指出,財報重編前後的資產差異(%)解釋能力優於營收差異(%)、營業利益差異(%)、稅後淨利差異(%)、股東權益差異(%)等變數。 二、 經過學習期的求解與測試期的擇優結果,本研究得到表4-19的權重計算結果,以常數項的權重值高於營收差異(%)、營業利益差異(%)、稅後淨利差異(%)、資產差異(%)、股東權益差異(%)等變數來看,財報重編前後的營收差異(%)、營業利益差異(%)、稅後淨利差異(%)、資產差異(%)、股東權益差異(%)等變數對於財務重編事件發生機率的解釋模型仍需加入其他解釋因子。 三、 基因演算法與Logistic等演算方法比較: Logistic模型較具解釋能力。就分析方法上,Logistic模型可以搭配AIC(Akaike’s Information Criterion)與SC(Schwarz Criterion)、LR Test等統計檢定方法,會發生較高的解釋意義。而基因演算法可以排序方式,最佳化求解過程,解決非線性與複雜的最佳化問題,故應運用基因演算法處理公司財報品質所引起的市場反應預測之投資操作議題,則可以發揮其搜尋空間極大與較不容易陷入局部最佳解等優勢。
Based on the database of Taiwan Economic Journal(“TEJ”), the sample of this study include Financial statement restatement cases in the period from 1993 to 2011, with the financial variables such as percentage difference of sales, percentage difference of operating margin, percentage difference of net income, percentage difference of asset, percentage difference of equity. After the analysis in this study, this study concludes the following conclusions . Firstly, as far as the forecastingcapability of models are concerned, the outcomes of table 4-4, table 4-6, table 4-8, table 4-10, table 4-12, suggest that percentage difference of asset could deliver more forecasting capability than percentage difference of sales, percentage difference of operating margin ,percentage difference of net income, as well as percentage difference of equity. Secondly, after the optimal process of learning period and test period, the results in table 4-19 figure that weights of constant item in models are higher than percentage difference of sales, percentage difference of operating margin, percentage difference of net income, percentage difference of asset, percentage difference of equity, suggesting that future modeling should involve other factors. Thirdly, the logistic model could interpret more cases of financial restatement. As far as methodology selecting is concerned, logistic model could include the test models of AIC(Akaike’s Information Criterion),SC(Schwarz Criterion)、LR Test. Whereas, genetic algorithm (“GA”) could perform sorting procedure to optimal solutions of non-linear problems. Hence, GA is appropriate to deal with the market reactions resulting from the issue of financial statements, providing the suggestions for investing in the market.