高報酬的投資工具往往也伴隨著高風險。在風險市場中如何保障投入的資產 價值並且能獲取不錯的利益是投資組合保險(Portfolio Insurance)研究的議題。 複製賣權策略(Synthetic Put Option)為投資組合保險常見的策略之一。此策略 透過連續調整資產分配在風險性資產與非風險性資產的部位，形成類似保護賣權 的資金結構，此結構能隨著市場上漲時獲取利潤，市場下跌時不跌過期初設定的 底值。然而一般複製賣權策略分配法則由Black-Scholes 選擇權評價模型推導。 此一方法為建構在假設下的模型驅動理論(model-driven approach)，無法配合環境 變動做適應性調整。 選擇權評價模型發展除了模型驅動理論外，另有一資料驅動理論(data-driven approach)。此發展方式不需嚴格假設，透過機器學習技術(machine learning)以及 大量資料的訓練與學習，使發展出的模型具有跟隨資料變動做出適應性調整的彈 性。本研究即嘗試使用機器學習技術之一的遺傳程式規劃(Genetic Programming) 發展複製賣權策略。 實證結果發現，本系統架構所發展出的複製賣權策略能比Black-Scholes 模 型為基礎的策略更為接近保護賣權的結構，以達到保險的效果。另外在績效上也 有不錯的表現。儘管如此，但發展出的策略也還有可能跌破保險底值。 There is lot of risk in the security market. How to protect one’s fortune from the risk is an important issue to investors. Portfolio insurance is one of the solutions to this question. It can help investors to gain the profit when the market is up and keep their portfolio equities when the market is down. The synthetic put option (SPO) is one kind of portfolio insurance strategies. It provides the insurance by using stocks and money to synthesize options. But it’s a dynamic portfolio insurance, it needs to adjust two position continuously in order to match the structure of options. Generally, most investors adjust the stock and money positions according to the Black-Scholes (B/S) option pricing model. But this way is model-driven approach. It has some defect. For example, model-driven approach can not adjust itself according to the change of environment. Data-driven approach is another way and it is more flexible. This research is want to rebuild the synthetic put option strategies by using genetic programming (GP) algorithm. GP is a kind of data-driven approach. After experiments in this research, GP really can find out synthetic program that is better than B/S in matching the structure of option and return.
