在傳統上，在研究銀行績效方法大多以財務比例分析的方式，作為探討銀行經營效率，近年來不少的研究文獻採用了資料包絡分析法(Data envelopment analysis , DEA)進行績效的評估並加以改良成各種形式的分析，本研究以資料分析與統計為強項的R語言當作工具，並對各家銀行之經營績效狀況進行各角度的分析。研究資料的部分，本研究以台灣本土35家上市母銀行，蒐集2013至2018年間為研究的資料基礎，將研究分析方式分成三個面向做討論，除了以傳統DEA模式下的數據分析結果外，並且使用類神經網路的倒傳遞模式(Back Propagation Neural Network)整合了資料包絡分析法，利用其步驟式的反覆修正建構模型，將效率情行以最精準的方式呈現，其在效率前緣線上的真實情形，並進一步分析銀行的經營效率；同時也利用DEA模式下，以差額為基準的未定向差額變數分析(Non-Oriented Slack-Based Measure , non-oriented SBM)進一步分析比較各家銀行在投入與產出狀況調整狀況。另外，在效率上，DEA與SBM均未能完全顯現其真實效率，其模式下，所得出的最具技術效率(效率值為 1)的DMU不只一家，因此本研究另外以DEA超級效率模式(Super-Efficiency Model)及未定向SBM超效率模式(Super-Efficiency Model )分析並且劃分出各家銀行的相對的真實效率排名狀況。以上述的效率評估方式，多以相對的概念加以評估，本研究以另一個角度，利用交叉效率(Cross Efficiency Measure, CEM)探討其銀行以自評角度的角度探討不一樣的效率方式。以及採用麥氏生產力指數(Malmquist Index)分析數據的資料範圍，自2013-2018年間，台灣本土上市銀行的總要素生產力，以及逐年間的變化情形進行探討上的分析，並且利用差額分析法(Slack Variable Analysis)探討個別銀行狀況，自2008金融海嘯至今，正好屆滿10年左右的經濟蕭條的敏感期限，且在台灣眾多銀行的環境競爭激烈狀況下各家銀行是否健壯，透過本研究的各角度分析，現今整體銀行的經營績效逐年狀況，並針對個別以及整體績效狀況，給予在個別投入向以及產出項的探討以及實質的改善分析建議。

According to traditional research, most of the methods used to research bank performance are in the form of financial proportional analysis. In recent years, a number of research papers have adopted the Data Envelope Analysis(DEA) method, conduct performance evaluation and improve it into various forms of analysis. This paper uses the R language as a tool, which has a strong advantage in data analysis and statistics, and analyze the operating performance of local banks in Taiwan from multiple perspectives. The research data is based on 35 major banks in Taiwan, collection and collation of data from 2013 to 2018 as a basis for research, Integrating the Data Envelope Analysis(DEA) method using the Back Propagation Neural Network(BPN) method, using a step-by-step construct model to further analyze the operational efficiency of banks. In addition, in DEA model, this paper also uses Non-Oriented Slack-Based Measure (non-oriented SBM) to analyze and compare the input and output status of each bank for adjustment. As stated above in terms of efficiency, neither DEA nor SBM models can fully present their true efficiency. The results show that there is more than one bank with the most technically efficient DMU (with an efficiency value of 1). Therefore, this paper uses Super-DEA (Super-Efficiency Model) and Super-SBM (Super-Efficiency Model) analysis and separates the relative true efficiency ranking of each bank. As stated above, efficiency is mostly assessed in a relative manner. This paper takes a different perspective, Using the Cross Efficiency Measure (CEM) to explore the ways in which their banks are efficient from a self-assessment perspective. Next, using the Malmquist Index was used to analyze the change and source of total factor productivity for each bank from 2013-2018. Since the financial tsunami in 2008, nearly 10 years or so of dangerous period. With the competition from many banks in Taiwan, the overall operating performance of the banks is now year by year. Through the analysis of various perspectives in this paper, we provide suggestions and analysis.

第1章 緒論 1
1.1 研究背景 1
1.1.1 研究背景 1
1.1.2 銀行金融近況概述 3
1.2 研究目的與研究流程 9
第2章 文獻探討 13
2.1 資料包絡分析模型 13
2.1.1 資料包絡分析之概述 13
2.1.2 CCR模式 16
2.1.3 BCC模式 18
2.2 資料包絡分析超效率模式 20
2.3 差額變數分析法 21
2.4 交叉效率衡量法 22
2.5 麥氏生產力指數 23
2.6 神經網路 27
2.6.1神經網路概述 27
2.6.2 神經網路運作原理 28
2.6.3 神經網路的學習(訓練)過程 30
2.6.4 影響神經網路效能因素 31
2.7 倒傳遞類神經網路模式 32
2.7.1 倒傳遞類神經網路概述 32
2.7.2 梯度陡降法 33
2.7.3 倒傳遞類神經網路演算過程 35
2.8 未定向差額變量分析模型 40
2.8.1 未定向差額變量分析模型 40
2.8.2 未定向差額變量分析超級效率模式 43
第3章 方法實證與分析 45
3.1 研究樣本與資料來源 45
3.1.1 資料來源與選取範圍 45
3.1.2　變數選取定義 46
3.2 研究分析 51
3.2.1 DEA資料分析 51
3.2.2 差額變數分析 61
3.2.3 交叉效率衡量法 66
3.2.4 類神經網路建構 70
3.2.5 未定向差額變數分析模型 74
3.2.6 麥氏生產力指數 77
第4章 總結分析討論 83
4.1 個別銀行分析與討論 83
4.2 整體銀行趨勢討論 91
第5章 結論與建議 97
5.1 結論 97
5.2 建議 98
參考文獻 100
附錄 104

[1] Farrell, M.J. (1957). “The Measurement of Productive Efficiency”. Journal of the Royal Statistical Society, 120, 253-290.
[2] Charnes A, WW Cooper and EL Rhodes (1978). “Measuring the Efficiency of Decision Making Units.” EJOR 2: 429-444.
[3] Banker RD, A Charnes and WW Cooper (1984). “Some Models for Estimating Technical and Scale Inefficiencies in Data Envelopment Analysis.” Management Science 30(9): 1078-1092.
[4] Ali AI and LM Seiford (1993). “The Mathematical Programming Approach to Efficiency Analysis.” in Fried HO and SS Schmidt (eds.) the Measurement of Productive Efficiency: Techniques and Applications, Oxford U.K.: 120-159.
[5] Banker RD and RC Morey (1986). “Use of Categorical Variables in Data Envelopment Analysis.” Management Science 32(12): 1613-1627.
[6] Coelli et al. (2005) ,“An Introduction to Efficiency and Productivity Analysis” 2nd ed. New York, NY, United States.
[7] Andersen P and NC Petersen (1993). “A Procedure for Ranking Efficient Units in Data Envelopment Analysis.” Management Science 39(10): 1261-1264.
[8] Sexton TR, Silkman RH, Hogan AJ (1986). “Data envelopment analysis: Critique and extensions. In Measuring Efficiency: An Assessment of Data Envelopment Analysis (Ed.: Silkman RH), Jossey-Bass,”, San Francisco, pp. 73-105.
[9] Byungho Jeong, Jae-won Ha & Chang-Soo(2010), “Ranking DMUs in the DEA Context Using Super and Cross Efficiency”. WKDD : 212-215”
[10] F. Rosenblatt(1957), “The Perceptron: A Perceiving and Recognizing Automaton.” Cornell Aeronautical Laboratory, Buffalo, New York, Report 85-60-1, 1957.
[11] Gluck MA, Myers C & Meeter M(2005), “Cortico-hippocampal interaction and adaptive stimulus representation: a neurocomputational theory of associative learning and memory”, Neural Netw. 2005 Nov;18(9):1265-79. Epub 2005 Nov 4.
[12] Lippmann, R. P., Gold, B., & Malpass, M. L. (1987). “A comparison of Hamming and Hopfield neural nets for pattern classification (Tech. Rep. TR-769). MIT Lincoln Laboratory. Cambridge, MA: MIT Press
[13] Hecht-Nielsen R. (1990) Neurocomputing, Addison-Wesley Publishing Company, Reading, Massachusetts.
[14] B. Freisleben, M. Hoof & R. Patsch(1992), “Using counterpropagation neural networks for partial discharge diagnosis”, Neural Computing & Applications volume 7, pages318–333(1998)
[15] E. Michael Azoff(1994), Neural Network Time Series Forecasting of Financial Markets (A Wiley finance edition) by E. Michael Azoff (1994-06-23)
[16] Refenes,Zapranis & Francis(1994), “ Stock performance modeling using neural networks: A comparative study with regression models”, Volume 7, Issue 2, 1994, Pages 375-388
[17] Rumelhart and McClelland(1986), “ Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations”, MIT Press Cambridge, MA, USA
[18] 葉怡成(1999)。類神經網路模式應用與實作第六版,儒林圖書有限公司
[19] Cooper KF, et al. (2000) Ama1p is a meiosis-specific regulator of the anaphase promoting complex/cyclosome in yeast. Proc Natl Acad Sci U S A 97(26):14548-53
[20] Kaoru Tone (2001), “A slacks-based measure of efficiency in data envelopment analysis”,in European Journal of Operational Research 130(3):498-509 · May 2001 with 2,330 Reads
[21] Tone K (2002). “A Strange Case of the Cost and Allocative Efficiencies in DEA.” JORS 53(11): 1225-1231.
[22] Caves, D. W., Christensen, L. R., Diewert, W. E., (1982). “The economic theory of index numbers and the measurement of input, output, and productivity”. Econometrica 50 1393-1414.
[23] Caves, D. W., Christensen, L. R., Diewert,W. E.,(1982). “Multilateral comparisons of out-put, input, and productivity using superlative index numbers”. The Economic Journal 92 73-86.
[24] Färe, R, Grosskopf, S, Lindgren, B and Roos, P (1992) “Productivity changes in Swedish pharmacies 1980-1989: a non-parametric Malmquist approach”, Journal of productivity Analysis, 3 (1/2), 85-101.
[25] Berg et al (1993), “ButcherL-selectin-mediated lymphocyte rolling in MAdCAM-1Nature”, 366 (1993), pp. 695-698
[26] Miller, S. and Noulas, A. (1996). “The Technical Efficiency of Large Banks Production”. Journal of Banking and Finance, 20, 495-509.
[27] Noulas, A.G. (1997). “ Productivity Growth in the Hellenic Banking Industry: State Versus Private Banks ”, Applied Financial Economics, 7(3): 223–228
[28] R.D. Banker, J.L. Chan, J.M. Patton (Eds.), (1989). ”Econometric estimation and data envelopment analysis.” ,Research in Governmental and Nonprofit Accounting, Vol. 5, JAI Press Inc, Greenwich, CT, pp. 231-243
[29] R. G. Dyson, R. Allen, A. S. Camanho, V. V. Podinovski, C. S. Sarrico and E. A. Shale,(2001), “Pitfalls and protocols in DEA” ,European Journal of Operational Research, 2001, vol. 132, issue 2, 245-259
[30] Lehman, Ann (2005). “Jmp For Basic Univariate And Multivariate Statistics: A Step-by-step Guide . ” , Cary, NC: SAS Press. p. 123.
[31] 蔡素琴(2001) “台灣金融機構經營績效分析-以上市上櫃銀行為例”,義守大學管理科學研究所
[32] 多層感知機PowerPoint PPT Presentation
(https://www.slideserve.com/loe/4765123)。
[33] 李宏毅-梯度下降Gradient Descent
(https://www.pianshen.com/article/7256969712/)
[34] Byungho Jeong, Jae-won Ha, Chang-Soo Ok. (2010). “Ranking DMUs in the DEA Context Using Super and Cross Efficiency”. KDD, pp.212-215.