建構加值型營業稅選案查核模式之研究

A Study on Constructing Distinguishable Models of Value-Added Business Tax Revenue

洪梓玲(Zi-Ling Hong)

指導教授 ： 謝俊宏

國立臺中科技大學/管理學群/企業管理系事業經營碩士班/碩士(2011年)

摘要

營業稅稅收是國家重要財源之一，其違章逃漏金額卻高居首位，連帶侵蝕營利事業所得稅及綜合所得稅稅基，鑑於此，本研究應用區別分析、決策樹及類神經網路分析工具，判別逃漏稅及未逃漏稅案件，以建立加值型營業稅選案模型。研究樣本來自財政部臺灣省中區國稅局某稽徵所99年度選案查核案件，「逃漏稅」100筆、「未逃漏稅」200筆，共計300筆，預測變項結合營業稅及營利事業所得稅財務比率及非財務比率指標，並以分類矩陣評估各模型的分類效果。
實證結果顯示：（1）整體而言，類神經網路預測正確率達81.00％，優於決策樹之預測率80.00％及區別分析73.00％，故類神經網路對逃漏稅之區別能力，具有良好的成效，本研究模型可作為國稅局提高選案查核正確率之參考，（2）經區別分析及決策樹實驗結果發現，「存貨佔營業收入之比例」為偵測逃漏營業稅之顯著因子。另「毛利率佔同業利潤標準毛利率之比例」、「申報書類別」及「加值率佔同業利潤標準毛利率之比例」也是區別逃漏稅行為的因子。具有上述異常特徵之案件可優先列為選查對象，以提高查核實益。

關鍵字

加值型營業稅 ； 選案查核 ； 逃漏稅 ； 區別分析 ； 決策樹 ； 類神經網路

並列摘要

Business tax is one of the important sources of government finance, but business tax evasion is ranking in the first place among all types of taxes, eroding taxation bases of Profit-Seeking Enterprise Income Tax and Individual Income Tax. In this point of view, this study applies tools of discriminate analysis, decision tree and neural network to establish models of detecting business tax evasion. 300 samples selected in this research, including 200 honest cases and 300 tax-evasive cases, come from selective assessment of business tax case evasion in one of the tax offices under National Tax Administration of Central Taiwan Province, Ministry of Finance in 2010. Variables include financial ratios and non-financial indicators filed in Business Tax report and Profit-Seeking Enterprise Income Tax report. This study uses classification matrices to assess the distinguishing effect of each model.

The empirical results reveal that（1）Grossly speaking, the probability of distinguishing business tax evasion in the neural network model is 81.00%, in comparison with that of 80.00％ in the decision tree model and that of 73.00％ in the discriminate analysis model. The neural network model of this study can provide some suggestions for National Tax Administration to improve the accuracy of selective assessment of business tax.（2）From the discriminate analysis model and the decision tree model, “the ratio of inventory to sales revenue” is a significant character for detecting business tax evasion. Furthermore, “the ratio of gross profit to the average gross profit of the trade concerned”, “filing types” and “the ratio of value-added rate to the average gross profit rate of the trade concerned” are characters that can be use to discriminate the behavior of tax evasion. Any cases with abnormal characters listed above should be selected as first priority to enhance the assessment of business tax evasion.

並列關鍵字

Value-added business tax ； Tax case evasion ； Tax evasion ； Discriminate analysis model ； Decision tree ； Neural network

參考文獻

黃寶環（2009）。營利事業所得稅結算申報案件運用電腦選案查核之研究。中原大學會計學系碩士學位論文。

黃美祝（2008）。國稅局選案查核與企業租稅逃漏之探討。會計評論，第48期，P.35－66。

羅玉惠（2006）。整合財務比率與公司治理指標建構信用評等預測模型-區別分析與類神經網路之應用。國立台北大學企業管理學系。

邱家永（2008）。加值型營業稅冒退稅選案查核模式之研究。國立臺中技術學院事業經營研究所碩士班。

林月桃（2009）。影響逃漏稅之因素-以加值型營業稅為例。國立政治大學社會科學學院行政管理碩士學程第九屆碩士論文。