Development of a Deep Learning-LSTM Trend Prediction Model of Stock Prices

Stocks prices follow the Brownian motion process that moves in a random and erratic state making it impossible to determine the future price based on the historical performance of the stock. However, the availability of the historical data of stock prices can't be ignored as it might hold hidden patterns that manifest the behavior of investors as well as the market sentiments and political conditions. Thus, investors have adapted mathematical models of regression and neural network analysis to process the stock's historical data along with the traditional fundamental analysis in forecasting the trend of stock prices. Neural network models have changed the landscape of formulating informed decisions on how investors can minimize their risks on their investments. However, deep learning models are dependent on the different hyper-parameters such as epochs, number of nodes in hidden layers, and number of hidden layers in order to produce the best trend prediction model as applied in stock investments. Unfortunately, the available literature and empirical studies are limited specifically on the number of hidden layers of deep learning models. This study attempts to explore the appropriate number of hidden layers to optimize the accuracy of a trend prediction model in stock trading.