Black-Litterman模型結合長時間短期記憶神經網路

Black-Litterman Portfolios with LSTM Derived Views

林承緯(Lin Cheng-Wei)

指導教授 ： 廖四郎

國立政治大學/商學院/金融學系/碩士(2020年)

https://doi.org/10.6814/NCCU202100714

摘要

本研究嘗試將長時間短期記憶(LSTM)應用於金融資產價格走勢的預測，並結合Black-Litterman模型建構風險分散的全球性多元化投資組合。本研究以金融資產的價量相關資料及技術指標預測資產價格漲跌及漲跌幅度，將預測結果作為Black-Litterman模型的投資者觀點並進行資產配置，比較投資組合在不同條件限制、不同風險趨避係數及不同共變異數估計方法下的績效表現。本研究實證發現：LSTM在預測資產價格漲跌幅度的方面擁有相對較佳的預測準確率，但是對於漲跌的預測表現並不突出；在投資組合績效回測方面，本研究建構之投資組合的績效表現皆優於市值加權投資組合及iShares Russell 1000 ETF。其中，使用Ledoit-Wolf Shrinkage方法估計共變異數矩陣能夠使投資組合獲得較高的夏普比率。

關鍵字

深度學習 ； 長時間短期記憶神經網路 ； Black-Litterman模型 ； 投資組合 ； 時間序列預測 ； 演算法交易

並列摘要

In this thesis, we try to apply long short-term memory (LSTM) to forecasting price trends of financial assets. We combine the forecasts with the Black-Litterman model and construct various globally diversified portfolios. Historical price and volume related data and technical indicators are used as input data to predict following week’s excess return, and we use the forecasts to arrive at the investor views in the Black-Litterman model. Finally, we test the out-of-sample performance of various Black-Litterman portfolios, where market capitalization weighted portfolio, equally weighted portfolio and iShares Russell 1000 ETF are used as benchmarks. The empirical results show that LSTM has a fine predictive accuracy for the sign of excess return; however, a mediocre performance on forecasting the magnitude of excess return. We also find that the Black-Litterman portfolios we construct outperform the benchmark portfolios, and the portfolios with Ledoit-Wolf Shrinkage covariance estimates generate higher Sharpe ratios.

並列關鍵字

Deep Learning ； Long Short-Term Memory ； Black-Litterman Model ； Portfolio ； Prediction of Time Series Data ； Algorithm Trading

參考文獻

[1]Beach, L., & Orlov, G. (2007). An application of the Black-Litterman model with EGARCH-M derived views for international portfolio management. Financial Market and Portfolio Management, 21(2), 147-166.

[2]Black, F., & Litterman, R. (1991). Asset allocation: combining investor views with market equilibrium. The Journal of Fixed Income, 1(2), 7-18.

[3]Black, F., & Litterman, R. (1992). Global portfolio optimization. Financial Analysts Journal, 48(5), 28-43.

[4]Cao, L. J., & Tay, F. E. H. (2003). Support vector machine with adaptive parameters in financial time series forecasting. IEEE Transactions on Neural Networks, 14(6), 1506-1518.

[5]Donthireddy, P. (2018, July 19). Black-Litterman portfolios with machine learning derived views. ResearchGate. Retrieved September 22, 2019, from https://www.researchgate.net/publication/326489143_Black-Litterman_Portfolios_with_Machine_Learning_derived_Views