Price and Quote Index Prediction with Deep Learning

Estimating the values of stock indices, financial instruments, or cryptocurrencies is a complex task due to the involvement of economic, social, and political factors that are generally unpredictable or unquantifiable. The commonly used methodologies for this task include heteroscedastic models such as ARCH and GARCH, which predict volatility or the intensity of price changes through returns, allowing for short-term predictions of up to 1, 3, or 5 days, or even longer. Recently, Machine Learning models have shown promising results in this field. This research explores the use of Deep Learning techniques by constructing an artificial neural network of the Multilayer Perceptron (MLP) or Feed-Forward type to approximate the value of the Mexican Stock Exchange Index, which measures the performance of the Mexican stock market. A comparison of these models was made using standard metrics such as Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE), concluding that the MLP provides improved results. However, the limitations of the proposed learning method highlight the need for alternative models in stock index prediction, and the Power Law model is suggested as one such alternative.