دانلود رایگان مقاله انگلیسی یادگیری عمیق با شبکه های حافظه کوتاه مدت برای پیش بینی بازارهای مالی - الزویر 2018

abstract

Long short-term memory (LSTM) networks are a state-of-the-art technique for sequence learning. They are less commonly applied to financial time series predictions, yet inherently suitable for this domain. We deploy LSTM networks for predicting out-of-sample directional movements for the constituent stocks of the S&P 500 from 1992 until 2015. With daily returns of 0.46 percent and a Sharpe ratio of 5.8 prior to transaction costs, we find LSTM networks to outperform memory-free classification methods, i.e., a random forest (RAF), a deep neural net (DNN), and a logistic regression classifier (LOG). The outperformance relative to the general market is very clear from 1992 to 2009, but as of 2010, excess returns seem to have been arbitraged away with LSTM profitability fluctuating around zero after transaction costs. We further unveil sources of profitability, thereby shedding light into the black box of artificial neural networks. Specifically, we find one common pattern among the stocks selected for trading – they exhibit high volatility and a short-term reversal return profile. Leveraging these findings, we are able to formalize a rules-based short-term reversal strategy that yields 0.23 percent prior to transaction costs. Further regression analysis unveils low exposure of the LSTM returns to common sources of systematic risk – also compared to the three benchmark models.

Conclusion

In this paper, we apply long-short term memory networks to a large-scale financial market prediction task on the S&P 500, from December 1992 until October 2015. With our work, we make three key contributions to the literature: the first contribution focuses on the large-scale empirical application of LSTM networks to financial time series prediction tasks. We provide an in-depth guide, closely following the entire data science value chain. Specifically, we frame a proper prediction task, derive sensible features in the form of 240-day return sequences, standardize the features during preprocessing to facilitate model training, discuss a suitable LSTM architecture and training algorithm, and derive a trading strategy based on the predictions, in line with the existing literature. We compare the results of the LSTM against a random forest, a standard deep net, as well as a simple logistic regression. We find the LSTM, a methodology inherently suitable for this domain, to beat the standard deep net and the logistic regression by a very clear margin. Most of the times – with the exception of the global financial crisis – the random forest is also outperformed. Our findings of statistically and economically significant returns of 0.46 percent per day prior to transaction costs posit a clear challenge to the semi-strong form of market efficiency, and show that deep learning could have been an effective predictive modeling technique in this domain up until 2010.