三大法人選擇權與期貨未平倉量之研究

Abstract

本研究運用人工智慧中的倒傳遞類神經網路進行學習及預測，分析三大法人、外資、投信與自營商於期貨與選擇權未平倉量的變化。將未平倉量以及其他籌碼面參數的變化行為轉換為物理力量，透過人工智慧強大的非線性搜尋能力找出未平倉量變化行為與台股加權指數結算日價格之間的隱含行為。進而分別預測三大法人、外資、投信與自營商於結算前1至5日、6至10日、11至15日相對於結算日台股指數收盤價的漲跌趨勢，並對法人之間的結果做比較。最後依據三大法人在期貨與選擇權佈局成本變化與台股指數的行為，提出動態輔助投資決策的建議。 研究結果顯示三大法人於結算日前1至5天對台股走勢預測準確率及獲利點數皆優於隨機交易模型；外資法人於結算日前6至10天對台股走勢預測準確率及獲利點數皆優於隨機交易模型，因此證實三大法人於結算日前1至5天及外資於結算日前6至10天對於台股結算日漲跌趨勢具有較高的預測能力，故能提供投資人作為投資決策之參考。

This study applies Back-Propagation Neural Network(BPNN) for investigating the relationship between TAIEX and Taiwan’s three major institutional investors’ open interest of futures and options. Daily values of three institutional investors, foreign investor, investment trust and security dealer’s open interest in TAIFEX futures and options were used to train the BPNN, and the output is to estimate the trend of TAIFEX on settlement day. The estimation result is grouped into 9 groups, including the three institutional investors, foreign investor, and the investment trust and security dealer’s estimation for 1 to 5 days, 6 to 10 days, and 11 to 15 days before the settlement day.

The empirical results indicate that the three major institutional investors’ estimation at 1 to 5 days before the settlement day achieved better prediction than random trading model. Furthermore, the foreign investor’s estimation during 6 to 10 days before the settlement day also attains more accurate prediction than random trading model.