使用AdaBoost之臺股指數期貨當沖交易系統

Abstract

本論文應用AdaBoost於臺灣股價指數期貨當沖交易，我們實作了一個交易系統，可依據盤中即時資料進行實務交易，亦可根據臺灣期貨交易所提供之行情資訊，進行歷史資訊回測模擬交易，檢驗交易策略之可行性。 我們選定臺灣股價指數期貨(以下簡稱臺股期貨)作為研究標的，以2004－2007年每分鐘成交資訊為訓練資料，檢測期間則為2008年1－6月。AdaBoost演算法在物件辨識方面有良好的效果，本交易系統即是用此方法訓練看漲與看跌兩個分類器，藉此找出臺股期貨適當的交易買賣點，提供交易人一個低風險且穩定獲利之投資參考。

We use AdaBoost for the Taiwan stock index future intra-day trading system. We design a trading system which can trade futures contracts automatically according to real-time streaming quotes. In addition, it allows us to use historical data for back testing and then examines the performance of our trading strategies. The training data are 1 minute candlesticks from 2004 to 2007. The period of testing starts from January 2008 and ends in June 2008. AdaBoost is an excellent machine learning technique for solving pattern classification problems. We train bull and bear classifiers by AdaBoost. These two kinds of classifiers support our trading system to find the proper time to long or short futures contract. In this thesis we provide low risk and steady profit models for futures traders.