雙動能策略與權重平滑效果之應用

Abstract

過去研究發現資產配置策略對投資組合的貢獻程度高達九成以上；而報酬預測是建構投資組合最核心的議題。本篇論文分析美國ETF市場資料，主要探討建構投資組合的策略及工具，創新動能投資策略並善用人工智慧科技的特性設計權重分配的規則，歸因四種投資組合的動能效果及權重效果，最後測試投資組合的穩健性及風險耐受性。戰略性資產配置(Strategic Asset Allocation)藉由動能的訊號產生的事件機率進行大類資產權重配置，能有效降低投資組合的整體風險，說明時間序列動能因子具有風險擇時的能力；戰術性資產配置(Tactical Asset Allocation)使用Black-Litterman模型結合長時間短期記憶神經網路轉換報酬分配來提高預測的準確度，其中長時間短期記憶神經網路預測準確率高達六成。研究結果發現，用以決定風險性資產權重的橫截面動能效果非常顯著，即便持有的資產屬於的投資組合類別(例如產業代表性ETF)，仍有機會透過動能效果增加額外的報酬，其中規避突發風險性衝擊的效果則來自於風險性資產池中納入避險性資產，說明同時具報酬與風險擇時的能力。因此本文建議投資人可以動態方式調整股債的權重來規避風險的衝擊，並搭配橫截面動能策略追求最大化目標報酬。

There is evidence that asset allocation strategies contribute more than 90% to investment portfolios, and return prediction is the core issue in portfolio construction. We conducted a data analysis in US ETF market, and focused on portfolio construction strategies and methods, including innovating dual momentum strategies, designing the rules of weight allocation by using the characteristics of artificial intelligence technology, and attributing the momentum effect and weight effect of investment portfolios. Finally, Robustness test and t-student test were used for statistical analysis. Strategic Asset Allocation is weighted based on the probability of events generated by momentum signals, which can reduce the overall risk of the investment portfolio effectively. It shows that time series momentum factor has the ability to market timing. On the other hand, Black-Litterman model combined with LSTM in Tactical Asset Allocation can be used to transform the distribution of returns, and improve the accuracy of prediction. Among them, prediction accuracy of LSTM is about 60%. The empirical results show that cross-sectional momentum effect used to determine the weight of risky assets is very significant. Even though the assets belong to an investment portfolio category (such as ETFs), there is still an opportunity to increase excess returns through the momentum. The principal conclusion was that investors can avoid the impact of risks through allocating the weight of stocks and bonds dynamically, and maximize the target returns by cross-sectional momentum strategy.