Classification-based financial markets prediction using deep neural networks

Abstract

Deep neural networks (DNNs) are powerful types of artificial neural networks (ANNs) that use several hidden layers. They have recently gained considerable attention in the speech transcription and image recognition community (Krizhevsky et al., 2012) for their superior predictive properties including robustness to overfitting. However their application to algorithmic trading has not been previously researched, partly because of their computational complexity. This paper describes the application of DNNs to predicting financial market movement directions. In particular we describe the configuration and training approach and then demonstrate their application to backtesting a simple trading strategy over 43 different Commodity and FX future mid-prices at 5-minute intervals. All results in this paper are generated using a C++ implementation on the Intel Xeon Phi co-processor which is 11.4x faster than the serial version and a Python strategy backtesting environment both of which are available as open source code written by the authors.

Deep Learning for Forecasting Stock Returns in the Cross-Section

“…Conversely, they stated that careful hyper-parameter optimization may still yield advantageous results for the tuningintensive deep neural networks. Outside the stock market, Dixon et al [12] attempted to predict the direction of instrument movement for 5-min mid-prices for 43 CME listed commodity and FX futures. They showed 68% accuracy for the high ones.…”

Predictive intraday correlations in stable and volatile market environments: Evidence from deep learning

“…White [25] hypothesised early that artificial neural networks could be successfully used to deliver empirical evidence against all the three forms of the efficient market hypothesis, reporting an R 2 value of 0.175 when using a simple feed-forward network with five previous days of IBM stock prices as inputs for a regression task [see also 26]. More recently, Dixon et al [27] also implement an artificial neural network with five hidden layers for trinary classification, differing in an output that represents little or no change from the previously cited studies. Using data of CME-listed commodities and foreign exchange futures in five-minute intervals to generate a variety of engineered features like moving correlations, a single model is trained instead of a separate model for each target instrument, resulting in an average accuracy of 42.0% for the investigated three-class prediction task.…”

Deep Learning for Mortgage Risk

“…Khandani, Kim & Lo (2010) and Butaru, Chen, Clark, Das, Lo & Siddique (2016) examine other machine learning models of financial default. Recent applications of deep learning in financial economics include Sirignano (2016), who models limit order books and Dixon, Klabjan & Bang (2016), who model market movements. Heaton, Polson & Witte (2016) use deep learning for portfolio selection.…”