GTC Silicon Valley-2019 ID:S9618:Deep Learning to Predict Regime Changes in Financial Markets Using Constrained Time Delay and Recurrent Neural Networks
Yigal Jhirad(Cohen & Steers),Blay Tarnoff(Cohen & Steers)
We'll discuss how applying deep learning to identifying market regimes can be valuable in helping anticipate and position a portfolio for significant structural shifts in the market. We will explain how we develop deep neural networks, including time delay and recurrent neural networks, and train them to identify and target intervals that delineate market state changes such as factor-based trends (e.g. growth vs. value), volatility regimes, and economic cycles. We'll also cover how the optimization algorithm used to drive the training relies on CUDA for high-performance computations on the GPU.