基於深度學習的市場預測暨基金投資理財機器人

Abstract

前人研究發現，自金融海嘯後，民眾對傳統金融機構的信心衰減。隨著金融科技崛起，機器人理財顧問服務成為新趨勢。在本研究中，我們利用資產定價模型對於市場和宏觀經濟因素的解釋能力，去設計投資基金的投資組合最佳化策略，並基於variational autoencoder [29]和long-short term memory [25] 提出一個新的市場預測模型，使用宏觀經濟變量預測市場處於牛市或熊市。最後結合以上技術開發一執行市場預測、投資組合最佳化及自動化投資的理財機器人系統。我們以S&P500和Mutual Funds of U.S.的22年實際數據對本研究的貢獻進行驗證。在假設可以完美預測牛市和熊市的情況之下，我們所提出的策略平均年回報率為18.26%。應用我們的深度學習市場預測技術，準確率可達到84.3%，其平均年回報率可達13.87%。證實我們的模型比其他演算法更準確且能帶來更多收益。

Previous researches found that Robo-advisor (RA) has become a new trend after financial crisis, due to the rising of financial technologies and the public confidence in financial institutions is insufficient. The main contribution of this paper is to design a mutual fund portfolio optimization strategy empowered by the ability of Capital Asset Pricing Model (CAPM) of interpreting market and macroeconomic factors, and to propose a new market forecasting model based on variational autoencoder [29] and long-short term memory [25], which uses macroeconomic factors to identify whether the market is bull or bear. We combine the techniques above to develop an Robo-advisor that can predict future market, optimize portfolio and automate investment. Experiments use 22 years’ data of S&P500 and mutual funds of U.S. to validate our strategy. In the condition of predicting bull and bear perfectly, the proposed strategies achieve an average annual rate of return of 18.26%. The accuracy of our market prediction method can reach 84.3% and the rate-of-return of our RA is 13.87%. Our model is more accurate and profitable than other algorithms.