結合基本面與類神經網路的最適投資組合-以台股為例

台灣在經濟環境不佳的情況下，單靠微薄的薪水已經無法滿足人們的生活的水準，想要改變這樣的現況最好的辦法就是學會如何投資。本文透過改良 Piotroski(2000)提出的 F-Score，將公司的基本面透過財報呈現出來，篩選出優良的公司使投資人能夠依據這個方向進行投資。除此之外，更透過類神經網路，來預測股價報酬率正負並進行交易策略模擬，目標是能夠打敗 ETF 台灣 50(0050)，使投資人能夠在辛苦的生活中找到一個財富自由的通道。實證結果顯示，本研究並提出的 C-Score 公司評分標準，透過 2007-2016 年共 10 年間的測試，這 10 年來的平均年報酬率達 21.63%，相對廣受一般投資人喜愛的 ETF 元大台灣 50(0050)能有更高的報酬率。評分在 12-13 分的個股 10 年帄均年報酬率更是高達 38.36%。在股價正負(持平)報酬率的預測上類神經網路 10 年平均正確率達 55.79%，提供投資人在操作上能有一個更好的方式，使用類神經網路的預測並進行機械式的操作更能避免一般投資人在操作上犯的錯誤。

關鍵字

F-Score ；基本面；台股；類神經網路；投資組合

並列摘要

Due to Taiwan's poor economic environment, people's salaries can no longer meet the living standards. The best way to change this situation is to learn how to invest. In this paper, by improving F-Score proposed by Piotroski (2000), the fundamentals of the company are presented through financial statements, screening out good companies so that investors can invest according to this direction. In addition, the neural network is used to predict the positive and negative returns of stock price and to simulate the trading strategy. The goal is to beat ETFs Taiwan 50 (0050), so that investors can find the way of wealth freedom in the hard life. The empirical results show that the average annual rate of return of the C-Score company proposed in this study has reached 21.63% over the 10 years from 2007 to 2016, and the ETFs 50 (0050), which is widely loved by ordinary investors, can achieve a higher rate of return. The 10-year average annual rate of return for companies with a score of 12-13 is 38.36%. The average 10-year correct rate of return on stock prices is 55.79%,providing a better way for investors to operate. The use of neural network-like prediction and mechanical operation can avoid the mistakes made by the general investors in the operation.