法人籌碼對台股未來走勢影響之研究

A Study of the Impact of Institutional Investors’ Holdings on TAIEX Index

查欣瑜(Hsin-Yu Cha)

指導教授 ： 陳安斌

國立交通大學/管理學院/財務金融研究所/碩士(2011年)

https://doi.org/10.6842/NCTU.2011.00178

摘要

本研究欲分析在台灣金融交易市場上占有重要地位的機構投資人，包含外資法人及主要以自營商和投信組成的本土法人，其在現貨市場、期貨市場與選擇權市場的進出布局，對台股未來走勢的影響。更進一步，希望藉由這些資料的分析，窺知股市未來的走勢，並以此作為投資的依據。從過去文獻及市場觀察可知，法人挾龐大資金於現貨、衍生商品市場作投機、避險、套利的佈局，以傳統的回歸方法單純探討現貨市場或期貨市場的籌碼，很難提供有說服力的投資建議。本研究利用倒傳遞類神經網路強大的非線性搜尋能力，學習外資法人和本土法人在現貨、期貨、選擇權的佈局的行為模式，並利用Chen and Hsu(2010)提出的金融工程物理學，將人工智慧演算法的輸入參數做一階與二階變量的處理，以更有效地學習法人籌碼佈局的物理力量。為探討法人籌碼對未來股價走勢的影響時間長短，本研究建立三個實驗組，分別學習、預測台指期貨未來一日、三日、五日的漲跌，期望藉由實驗的設計找出最適合的投資期間。為提供更多的投資策略與想法，本研究的實證交易策略，除了最單純的台指期買賣交易還有選擇權賣方的交易策略，提供不同風險偏好的投資人參考。
研究結果顯示，以法人籌碼資料預測台股走勢，準確率最高的區間是五天的漲跌。交易策略方面，以台指期為交易標的的獲利最大，若將獲利的波動考慮進去，選擇權賣方的策略則擁有較低的波動和較高的夏普比率。投資人可依照需求，選擇適合的交易策略。

關鍵字

三大法人 ； 台指期貨 ； 台指選擇權 ； 未平倉量 ； 倒傳遞類神經網路

並列摘要

This study intends to analyze how institutional investors’ holding arrangement on futures market, options market and the underlying stock market influences the trend of TAIFEX index. The study targets to get the insightful view of future index trends and profit from the investment with the suggestion of the research result. By reviewing literatures and observing from the market, it is known that institutional investors invest huge amounts of money on cash and futures markets with speculating, hedging, and arbitraging strategies. Therefore, analyzing their real intensions of the arrangements of positions by traditional regression methods is impossible. This study employs back-propagation neutral network (BPNN) with its tremendous non-linear searching ability to learn the behaviors of how institutional investors arrange their positions. Simultaneously, applying the idea of econophysics from Chen and Hsu (2010), extracting the changing strength of their holding positions by using the first-order and second-order derivatives of the input parameters as input variables of BPNN. To find the time range that the price trend is impacted by institutional investors’ positions most, the experiments are divided to three, learning and predicting whether TAIFEX futures price will rise or fall for the future one, three or five days respectively, expecting to determine the perfect investing period by this design of experiments. To offer more choices of investment strategies, two different empirical estimation methods are provided. The first strategy is trading using TAIFEX futures, and on the other hand, the second one is verified by selling options. The different characteristics of two strategies provide appropriate choices to investors with different risk preferences.
The research result indicates that the experiment predicting five day TAIFEX futures trend has the highest precision. For trading strategies, investing with TAIFEX futures earns highest profit. Taking profit volatility into account, the strategies of selling options are more stable and has higher Sharpe ratio. Investors can choose suitable strategies by their preference.

並列關鍵字

Institutional investors ； TAIEX futures ； TAIEX options ； Open Interest ； BPNN

參考文獻

[10] Chien-Cheng Lee, et al., “Federal Funds Rate Prediction: A Comparison Between the Robust RBF Neural Network and Economic Models”, Journal of Information Science and Engineering, Vol. 25, pp. 763-778, 2009.

[1] Albert S. Kyle, “Continuous Auctions and Insider Trading”, Econometrica, Vol. 53, No. 6, p. 1315, 1985.

[2] Ali F. Darrat, Shafiqur Rahman, “Has Futures Trading Activity Caused Stock Price Volitility?”, The Journal of Futures Markets, Vol. 15, No. 5, pp. 537-557, 1995.

[3] An-Pin Chen, Yu-Chia Hsu, “Dynamic Physical Behavior Analysis for Financial Trading Decision Support”, IEEE Computational Intelligence Magazine, Vol.5, Issue. 4, 2010.

[4] An-Sing Chen, et al., “Application of Neural Networks to an Emerging Financial Market: Forecasting and Trading the Taiwan Stock Index”, Computers & Operations Research, Vol. 30, pp. 901-923, 2003.