A Multilayer Feedforward Perceptron Model in Neural Networks for Predicting Stock Market Short-term Trends

Stock market prediction is important for investors seeking a return on the capital invested, though this prediction is a challenging task, due to the complexity of stock price time-series. This task can be performed by conducting two primary analyses: fundamental and technical. In this paper, we examine the predictability of these two analyses using a multilayer feedforward perceptron neural network (MLP) and determine whether MLP is capable of accurately predicting stock market short-term trends. We utilize stock prices (2013 Mar – 2018 Jun) and twelve financial ratios of technology companies selected through a feature selection preprocess. Our model uses self-organizing maps (SOMs) for clustering the historical prices and produces a low-dimensional discretized representation of the input space. The best results are obtained through hyper-parameter optimizations using a three-hidden layer MLP. The models are integrated using a nonlinear autoregressive structure with exogenous input (NARX). We find that the hybrid model successfully predicts the short-term stock trends. The hybrid model yields the greatest directional accuracy (70.36%) as compared to fundamental and technical analyses (64.38% and 62.85%) and state-of-the-art models. The results indicate that the market is not fully efficient. Our model will be useful to practitioners seeking investing and trading opportunities and others interested in the study of financial markets.