建構所得稅預測模式之研究

A Study on Constructing Forecasting Models for Income Tax Revenue

黃美玲(Mei-Ling Huang)

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)

https://doi.org/10.6826/NUTC.2011.00039

摘要

從我國賦稅收入結構可知，所得稅收入是國家重要的財政收入。精確的所得稅收入預測，可使財政主管機關更精確的擬編歲入預算，確實掌握政府未來可運用之財務資源，俾利政府施政計畫之推展，全面增進政府財務效能，發揮政府應有的功能及目標。本研究透過多元線性迴歸、時間數列預測及倒傳遞神經網路等方法，以國民所得、經濟成長率及工業生產指數等為預測變項，分別建立所得稅收入預測模式，並比較各模式的預測差異與預測效果，期望能提供更精確的預測模式供政府擬編歲入預算之參考。
本研究結果顯示，多元線性迴歸、時間數列及倒傳遞神經網路等預測模式之所得稅收入預測值經與實徵淨額比較後，其平均絕對誤差率最高僅為8.04％，顯示各預測模式的預測效果極為精準，且時間數列預測模式、多元線性回歸預測模式之預測效果較倒傳遞神經預測模式相對精準。再將前述結果與政府各年度編列之歲入預算數與實徵淨額間的平均絕對誤差率相比較，得出各預測模式之平均絕對誤差率遠低於所得稅收入預算數與實徵淨額之平均絕對誤差率，進一步證明本研究所建立之預測模式有極佳的預測效果，確能提供政府相關單位實務運用，對於歲入預算之擬編與施政計畫之推展有一定程度的助益。

關鍵字

所得稅 ； 預測 ； 多元線性迴歸 ； 時間數列 ； 倒傳遞神經網路 ； 類神經網路

並列摘要

According to the structure of tax revenues, Income Tax is a major source of government finance. The main goals of government shall improve public welfare, develop the national economy, enhance financial efficiency and strengthen administrative effectiveness by allocation of government resources. With respect to the management of government finance, the most important thing is to prepare the annual budget. This study intends to provide the government with accurate forecast in the most important portion of tax revenues, Income Tax revenue, for preparation of the government budget. In order to predict Income Tax revenue precisely, forecasting models are established separately through three methods, including Linear Multiple Regression Analysis Forecasting Method, Time Series Forecasting Method, and Back-Propagation Network. Each model uses three variables, nation income, economic growth rate and industrial production index.
We also compare the outcomes (forecasting Income Tax revenue) of these three methods with Income Tax revenue collected actually. The figures of Mean Absolute Percent Error (MAPE) between prediction made by three models in this study and Income Tax revenue actually collected are less than 8.04%. The result reveals that three models in this study give a fairly accurate forecast of Income Tax revenue. Furthermore, separate predictions from the model of Linear Multiple Regression Analysis Forecasting Method and the model of the Time Series Forecasting Method are more accurate than that from the model of Back-Propagation Network. In additions, the figures of MAPE stated above are lower than those between estimation used by the government in preparing the annual budget and Income Tax revenue actually collected. Thus, this study provides a better means for the government to predict Income Tax revenue while preparing the annual budget.

並列關鍵字

Income Tax ； Forecasting ； Linear Multiple Regression ； Time Series ； Back-Propagation Network ； Artificial Neural Networks

參考文獻

陳彥文（2003）。類神經網路之應用-以營業稅選案模型為例，銘傳大學資訊管理學系，碩士論文。

Mitchell, D., and R. Pavur “Using modular neural networks for business decisions,”Management Decision (40:1), 2002, Vol. 40 No. 1 pp. 58-63.

Moutinho, L., and Phillips, P. A. “The impact of strategic planning on the competitiveness, performance and effectiveness of bank branches: a neural network analysis,” International Journal of Bank Marketing (20:3), March 2002, pp.102-110.

Venugopal, V., and Baets, W. “Neural Networks and Statistical Techniques in Marketing Research: A Concept Comparison,” Marketing Intelligence & Planning(12:7), 1994, pp.30-38.

中文部分：