本研究主要是運用人工智慧的方法改善過去純粹使用技術分析或經驗分析等過於主觀的缺點,並改善傳統開盤八法只考量價的缺點,取而代之的是三個取樣點之間的力道關係(變化率),並搭配不同的技術指標,再利用倒傳遞類神經網路來建構預測台灣加權股價指數期貨當日收盤價、以及隔日開收盤價漲跌方向的模型。研究期間從2011年1月3日至2011年11月29日。 由實驗結果我們得到以下幾個結論: 1.使用倒傳遞類神經網路所建構之模型於預測台灣加權股價指數期貨當日 收盤漲跌方向是可行的。 2.相同輸入變數之下,隔夜效應會使預測隔日開收盤漲跌模型方向準確率降 低。 3.預測當日收盤漲跌模型使用改良乖離率作為輸入變數,會比使用MA、 RSI或是KD指標,有更佳的預測方向準確率。
This study are base on manipulating the method of Artificial Intelligence to improve the flaw when people maneuvered technical or experience analysis merely, as well as modifying the shortcomings of Eight Indicators, which considered only price and quantity. Therefore, the traditional analyses are superseded by measuring the variation among three sampling points, variety technical indicators and inverse neutral network to establish a model; with that people could forecast the closing price on the day and the opening price on the next day of Taiwan weighted stock index future. Data collection is from third of January, 2011 to 29th of November, 2011. Some consequences from experiment are drawn as following, 1. Implement of inverse neutral network to project the tendency of Taiwan weighted stock index future is workable. 2. As inputting the same variable, the overnight effect would dilute the accurate rate. 3. Exercise modificatory BIAS as a input to predict the opening and closing market model would be more accurate than use MA, RSI and KD indicators.