由於台指選擇權市場的蓬勃發展,顯示台灣投資大眾漸漸能夠接受這種新興投資方式,因此本研究試圖使用人工智慧方法中的組合編碼基因演算法,建構由台指選擇權所組成的投資組合,藉由基因演算法強大的演化求解能力,求出投資組合的最適資金配置方式。 相較於證券等金融商品,選擇權的生命週期是相當短的,依到期日長短而定,最長不會超過8個月,而過去研究多未使用長天期的訓練資料,在演化的成效上是值得懷疑的,因此本研究也嘗試使用三種不同的資料處理方法串接選擇權歷史資料,理論上,可以由當日一直串接此選擇權的歷史資料至選擇權市場最初始的交易日,在交易期間上,我們選擇2003年至2004年台指選擇權交易資料做為實驗資料,在績效評估上,與平均分配資金策略與基因演算法所產生的資金分配策略比較。研究結果發現,平均而言基因演算法所產生的資金分配策略報酬率略優於平均分配資金策略,但在統計上並未有顯著優勢,另一方面,在比較各種不同歷史資料串接方法後,結果顯示,以選擇權的實際價格和理論價格之間的差距進行昇冪排序串接歷史資料,所獲得的績效最佳,同時也說明價值被低估的買權,其獲得正報酬率的機率較高。 Due to the growth of Taiwan Index Option (TXO) market, it is a new way to investment in Taiwan. This paper uses Combine Encode Genetic Algorithm to set up a portfolio which is composed of TXO options and to solve the fund allocation problem in TXO options portfolio. Compare to other securities, the life cycle of options is very short, depends on the expiration date, it varies from one month to eight month. The history data used before is too few to trust the result of GA. This paper uses three different methods to process option history data. In theory, the option history data can be expended to the first trading day of TXO market. Compare to original history, we can provide more data in GA evolution process. We select the TXO trading data from 2003 to 2004. We compare the portfolio performance between the GA’s strategy and the equally fund allocation strategy. The results show that the GA’s strategy can defect the equally fund allocation strategy, but in some cases, the statistic test result can not support this conclusion. In addition to the statistic test result, we also compare these different option history process methods, the result indicates that the best performance of all is the “ascend sort of the difference between option’s actual and theorical premium” method. It also indicates that the possibility of positive return of an option is high if that option’s value is underestimated.