本研究目的在透過分散式運算分析多商品與主、副策略間適配性，並以Python建置自動化平台，改善分析效率。 研究中，交易策略使用布林通道為主策略，六道指標濾網為副策略，並加入分類型變數設計，將不同的副策略指定相對應的分類型參數，讓濾網條件修正主策略，並分析不同組合下的策略績效。 但台股上市、上櫃商品共有1,690檔，進行多商品分析相當耗時，若再加上主、副策略變化，更是曠日費時。因此，研究中設計出的系統，將「資料探勘」結合了「多電腦式分散運算架構」，整合多台電腦進行分析，加速整體運算過程。並以高原尋找法分析策略穿透性，在多商品情況下，找出績效相對穩定的策略及其參數，避免交易軟體著重單一商品之最佳化帶來過適(overfitting)問題。 當分析完成後，以資料視覺化技術圖表，呈現分析結果，讓使用者可以透過此平台，更快速分析多商品與主、副策略的適配性。 ;The purpose of this research is to analyze the suitability of portfolio in different sub-strategies through Distributed System. And build an automation platform in Python to improve the efficiency of portfolio analysis. In this study, we use Bollinger Bands as the main trading strategy, six indicator filters as sub-strategies. Then using classification parameters to present sub-strategies. So the sub-strategies can work as filter conditions to modify the main strategy under different combinations when we optimize the strategy. However, the number of stocks in Taiwan is 1,690, it will take lots of time to analysis all symbol. So this research combines Data Mining and Distributed System to integrate multiple computers for analysis in order to accelerate the overall computing process. It also uses the method of region growing which can find the plateau in profit report to analyze the penetrability of the strategy in portfolio. After that, we can get a stable strategy and its parameters, so as to avoid the overfitting of the optimization. Finally, the results will be presented by the charts of the data visualization. This will allow users to quickly analyze the suitability of portfolio in different sub-strategy through this platform.
