[Paper Review] ANN Model to Predict Stock Prices at Stock Exchange Markets

Barack Wanjawa, Lawrence Muchemi|arXiv (Cornell University)|Dec 17, 2014

Stock Market Forecasting MethodsDecision Sciences14 references16 citations

TL;DR

This paper proposes a feedforward multilayer perceptron artificial neural network (ANN) with error backpropagation for predicting stock prices on major exchanges. Trained on 80% of 2008–2012 data from the Nairobi and New York Stock Exchanges using a 5:21:21:1 architecture, the model achieved a mean absolute percentage error (MAPE) between 0.71% and 2.77%, demonstrating strong predictive accuracy across diverse markets.

ABSTRACT

Stock exchanges are considered major players in financial sectors of many countries. Most Stockbrokers, who execute stock trade, use technical, fundamental or time series analysis in trying to predict stock prices, so as to advise clients. However, these strategies do not usually guarantee good returns because they guide on trends and not the most likely price. It is therefore necessary to explore improved methods of prediction. The research proposes the use of Artificial Neural Network that is feedforward multi-layer perceptron with error backpropagation and develops a model of configuration 5:21:21:1 with 80% training data in 130,000 cycles. The research develops a prototype and tests it on 2008-2012 data from stock markets e.g. Nairobi Securities Exchange and New York Stock Exchange, where prediction results show MAPE of between 0.71% and 2.77%. Validation done with Encog and Neuroph realized comparable results. The model is thus capable of prediction on typical stock markets.

Motivation & Objective

• To develop a more accurate stock price prediction model than traditional technical, fundamental, or time series methods.
• To evaluate the performance of an artificial neural network (ANN) in forecasting stock prices on real-world stock exchange data.
• To validate the model’s predictive capability across different financial markets, including the Nairobi and New York Stock Exchanges.
• To compare the ANN model’s performance with established machine learning frameworks such as Encog and Neuroph.
• To demonstrate the feasibility of using feedforward multilayer perceptrons with backpropagation for high-accuracy stock price forecasting.

Proposed method

• The study employs a feedforward multilayer perceptron (MLP) with error backpropagation for training.
• The network architecture is configured as 5:21:21:1, representing input, two hidden layers, and one output neuron.
• The model is trained on 80% of the dataset, with 130,000 training cycles to optimize weight adjustments.
• Input features are derived from historical stock price data, with normalization applied to improve convergence.
• Model performance is evaluated using the mean absolute percentage error (MAPE) metric.
• Validation is conducted using two external frameworks—Encog and Neuroph—to ensure consistency and reliability of results.

Experimental results

Research questions

• RQ1 Can an artificial neural network model outperform traditional technical and fundamental analysis in predicting stock prices?
• RQ2 What is the predictive accuracy of a feedforward multilayer perceptron with backpropagation on real stock exchange data?
• RQ3 How does the model perform across different stock markets, such as the Nairobi and New York Stock Exchanges?
• RQ4 To what extent do external machine learning frameworks like Encog and Neuroph reproduce the same prediction results?
• RQ5 What is the optimal network configuration for minimizing prediction error in stock price forecasting?

Key findings

• The ANN model achieved a mean absolute percentage error (MAPE) ranging from 0.71% to 2.77% on test data from the Nairobi and New York Stock Exchanges.
• The model demonstrated consistent performance across both developed and emerging market exchanges, indicating robustness to market differences.
• Validation using the Encog and Neuroph frameworks produced comparable prediction results, confirming the model’s reliability and reproducibility.
• The 5:21:21:1 network configuration proved effective in capturing complex nonlinear patterns in stock price movements.
• The use of 80% of the data for training over 130,000 cycles led to stable weight convergence and improved generalization.
• The results suggest that ANNs with backpropagation can serve as a viable and accurate alternative to conventional stock prediction techniques.