類神經網路預測台灣50股價指數之研究

The Study of Neural Network to Predict Taiwan ETF-50 Stock Index Price

黃華山 邱一薰

國立彰化師範大學資訊管理研究所

Hwa-Shan Huang, I-Hsun Chiu

Department of Information Management, National Changhua University of Education

huanghs@cc.ncue.edu.tw robert_chiu@msn.com

摘要

本論文是運用倒傳遞類神經網路做為預測台灣50股票指數的研究。台灣50 (ETF50)「指數股票型證券投資信託基金」，其指數之組成由台灣排名前50大之股票指數，依公司市值比例組成。因此，本研究利用其組成關係，找出台灣50排名前20名之成份股以及台灣50指數以往之技術指標資料，按成分股權重比例之影響，做為輸入類神經網路訓練之變數資料。本研究將收集到的樣本切割成兩部份，356筆日資料為訓練資料；100筆日資料為測試期資料。類神經網路經過訓練後、模擬ETF50股價指數，再選出 5 個平均誤差最小者，搭配本研究提出之交易策略，計算可獲取之投資報酬。經實驗發現：一、本研究提出之類神經網路模式具有良好的預測能力；二、類神經網路模式配合本研究提出之交易策略可獲得較高的報酬。

關鍵詞：倒傳遞類神經網路，台灣50

Abstract

This study is based on Back-propagation neural network theory to forecast Taiwan Top50 Exchange Tracker Fund (ETF50). Because ETF50 is composed of Taiwan Top 50 representative firms, this study applies the technical index of ETF50 and the top 20 of ETF50 stock indices as the inputs of the neural network. Our sample data is mainly separated into two parts, 356 records of training data and 100 records of testing data. After training the neural network, simulating the ETF50 price index and comparing with the mean square error, I choose the top 5 prediction modes. Besides, a speculative trading strategy is applied to evaluate the performance of the best return prediction model. Through my experiment, I found two points: 1. the neural network model proposed has good prediction capability; 2. neural network model applied with my speculative trading strategy can obtain higher return.

Keywords: Back-propagation neural network, ETF 50