
- This event has passed.
Deep learning-based portfolio optimization with transaction costs
January 17 @ 9:00 am - 10:00 am CST

Speaker: Prof. Aihua (Eva) Zhang, College of Science, Math & Tech., Wenzhou-Kean University, Wenzhou China
Abstract: In order to obtain the optimal portfolio strategy maximizing the accumulated terminal wealth with transaction costs, in this paper, we propose a new prediction-based portfolio method combining with a long short-term memory (in short, LSTM) network which is an extended type of recurrent neural networks in deep learning. Our proposed method, named as LSTM-Prediction-based Portfolio (LSTM-PbP) with transaction costs, consists of two technical steps: finding the optimal portfolio strategy and predicting the future relative prices. For the price prediction, we use multi-layer LSTM; while for the optimal portfolio strategy, we solve the constraint maximization problem via relative entropy. We then update the future portfolio weights using the predicted prices and past portfolio weights. We iterate the process until the final investment period. Numerical experiments are also provided to show the accumulated wealth by following the obtained optimal portfolio strategy in comparison with the accumulated wealth under buy-and-hold strategy. The numerical results show that our model consistently outperforms the buy-and-hold strategy.