Neural Networks in Finance: Design and Performance

Abstract

Neural networks have piqued the interest of many financial modelers, but the concrete applications and implementation have remained elusive. This article discusses a step-by-step technique for building a potentially profitable financial neural network. The authors also demonstrate a successful application of the neural network to investing based on daily and monthly financial data. The article discusses various components of neural networks and compares popular neural network activation functions and their applicability to financial time series. Specifically, use of the tanh activation function is shown to closely mimic financial returns and produce the best results. Incorporating additional inputs, such as the S&P 500 prices, also helps improve neural networks’ forecasting performance. Longer training periods deliver strategies that closely mimic common technical analysis strategies, such as moving-average crossovers, whereas shorter training periods deliver significant forecasting power. The resulting neural network–based daily trading strategies on major US stocks significantly and consistently outperform the buy-and-hold positions in the same stocks.

TOPICS: Statistical methods, simulations, big data/machine learning

Key Findings

• • This article provides theoretical background and a step-by-step implementation of a neural network using financial data.

• • This article compares neural networks’ activation functions in the context of best fit to financial data.

• • This article demonstrates the implementation of monthly adjusted investment strategies based on neural networks.