| Abstract: | 全球市場緊密相連，各商品間存在不同關係，如何有效率尋找多商品間之相關性，並分析大量商品配對組合於樣本內、外之獲利穩定度，為一大關鍵問題，目前尚未有完整系統能達成。
為有效尋找跨市場多商品之相關性，本研究以Ruggiero（2012）策略為範例開發與設計自動化計算、回測與分析多商品間相關性之平臺，其允許使用者客製化回測指定之商品、時間分割與參數設定，並以風險平價方式標準化各種不同商品，以進行排序與分析獲利之穩健度。在分析過程中為避免樣本內過度最佳化造成過度適配問題，使用「高原搜尋演算法」尋找績效高原，並透過機器學習DBSCAN方法依據高原搜尋結果發展「高原排序指標」評估不同高原間之穩健性，樣本外透過每日累積盈虧檔案分析參數組相關性之存續性，並進行排序與驗證。除Ruggiero（2012）提出之策略外，其他多商品之交易策略亦可套用本研究之架構，以驗證其配對組相關性之有效性。
研究使用2002年01月01日至2019年08月31日計122檔美國市場商品進行分析與驗證本平臺，其中47檔商品可用於交易商品，配對組合產生5,687組，以進行不同時間分割、高原點數門檻值、高原平均報酬與排序方法配對組分析。由於多商品分析將相當耗時，需大量運算資源產生績效檔案，研究使用「多節點偵測任務分散演算法」，整合多電腦分析提升運算效率。藉由資料視覺化技術呈現分析結果，以快速分析各商品之相關性與獲利穩定度，理解不同市場間之相關性程度，用以協助使用者選擇投資標的。;Global financial markets are highly connected. There exist different relationships between different markets. It is a big issue to find out lots of multiple financial products correlation efficiently and analyze lots of financial product pairs on profit robustness within in-sample and out-of-sample. However, there is no a well-built system that could solve this problem and achieve this goal.
The purpose of this research is to find out the Intermarket correlation of multiple products efficiently. This study is based on the strategy which Ruggiero（2012）proposes to develop, design, and implement a platform that could automate calculation, backtest, and analysis multiple financial product correlation. This high flexibility platform allows users to customize their own backtest tasks including financial products, periods splits, parameters, and the other configurations. To rank and analysis different pairs the robustness of profit, using the risk parity concept to standardize different kinds of financial products. It could be divided into two parts in-sample and out-of-sample at the stage of the analysis. In-sample applies the “Plateau Searching Algorithm” to find out the sound and robust optimization result surface and to prevent the optimization from overfitting. Based on the prior results, using the DBSCAN to analyze and develop “Plateau Indicator” to evaluate the robustness between different optimization surfaces. Out-of-sample uses the daily profit data to judge the existence parameters correlation to rank the pairs and verification the effectiveness of pairs. Besides the Ruggiero（2012） strategy, other strategies could also practice the architecture of this study to analyze and verification the effectiveness of pairs of correlation.
In this study, using 122 financial products in the American market, from January 1, 2002, to August 31, 2019, are used to analyze and verify this platform. Among these financial products, 47 financial products could be used for trading, the others 122 could be used to confirm the relationship, in summary, produces 5,687 combinations. All of the combinations would apply to different time splits, optimization points threshold, demand of average return, and rank methods to analyze the different pairs performance and the level of robustness on profit. In order to analysis on multiple financial products would be consumed lots of time and computing resources to generate the performance files and needed data, using the “Host-Node-Message Detection Distribution Algorithm” to integrate multiple computers to accelerate the efficiency of computing process.
Finally, the result of this platform analysis would be presented by the charts and tables of the data visualization. This could help the user to perceive the correlation and robustness between different financial products and realize the level of different market correlations quickly. It would be help users to determine the investment targets. |
