金融海嘯後，黃金市場持續熱絡且價格節節攀升，但近日愈發不穩定有急速下滑之趨勢，未來多空行情眾說紛紜。選擇權是一種多空皆有利可圖的商品，但其非線性的價格因素特徵，適合以倒傳遞類神經網路(Back-Propagation Neural Network, BPNN)找出較適價位。目前資料採礦技術被廣為應用在各領域，採礦工具也一直推陳出新，幾乎都有類神經網路的功能，本研究採用學術界常用的採礦工具WEKA，針對美國CME黃金期貨選擇權自1983年1月至2011年3月每日買權結算資料，以BPNN進行評價研究。實證結果發現，除BS (Black-Scholes Option Pricing Model)模型既有基本變數外，注入選擇權及標的商品的當日及前日結算未平倉量有助於提升模型準確度。震盪走勢時，多注入CME隱含波動率並不會使模型全然提升品質。
After the subprime financial crisis, gold market continues to heating up and the price is climbing. However, the gold market has turned downward in 2013 with a rapid price decline, and the outlook is divergent.
Options payoff could be expected in both bull and bear market. Its nonlinear pricing characteristic could apply back-propagation neural network (BPNN) to identify a suitable price level. Nowadays data mining technology is widely used in various fields. Data mining tools have been innovation, and most of them have included neural network function.
In this study, I use WEKA, a widely-used academic data mining tool, to conduct evaluation studies by BPNN. The data uses U.S. CME gold future call options daily settlement price with the period from January 1983 to March 2011.
Empirical results show that, in addition to BS model (Black-Scholes Option Pricing Model) basic variables, the additional variable inputs of current and previous open-interest volume from options and the underlying commodity could improve the model accuracy. In an up-down oscillation market, including sorely CME option implied volatility might not be necessary to improve the model accuracy.
