Abstract: In actuality, the financial investment in Enterprises of the world is common. This investment is performed using internet platforms and value markets. This can generate a loss for many investors due to the uncertainty of the future financial health of the Enterprise. A comparison for the prediction of financial health based on algorithms of machine learning, particularly Support Vector Machine (SVM), Artificial Neural Networks (ANN), K-Nearest Neighbors (KNN), and Adaptive Neuro-Fuzzy Networks (ANFIS) is presented. The database of the Taiwan Economic Journal from 1999 to 2009 is used, with 95 financial ratios of enterprises financially healthiest and with bankruptcy problems. In the ANN, the epoch number, numbers of neurons, activation functions in each layer, loss function, and learning rate are tested; also, an architecture of a Convolutional Neural Network (CNN) is implemented. In SVM, the experiments are performed using different kernels, polynomial, RBF, linear. Besides, the size of C, size of gamma, and size of the polynomial are varied. In KNN, experiments with different numbers of neighbors, types of weight, and values of P are realized. In ANFIS, experiments with variants of the numbers of fuzzy rules, quantity and type of membership functions, number of epochs, and input dimensions are performed. Optimization using Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) of the three models with the best results are performed; the optimization is based on the search for the best hyperparameters that would provide a higher accuracy. The neural network models presented the best average for all the proposed tests.

Resumen Con la disponibilidad de datos de alta frecuencia y nuevas técnicas para la filtración de señales, es pertinente preguntarse una vez más ¿podemos predecir los precios de los activos financieros? El presente trabajo propone un algoritmo para la predicción de retorno logarítmico del siguiente periodo. Se usan datos en frecuencias de 1 a 15 minutos, para 25 activos de alta capitalización en el mercado accionario mexicano. El modelo consiste en la aplicación de una wavelet seguida de una red neuronal de tipo Long Short-Term Memory (LSTM). En la literatura comúnmente se encuentra el uso de wavelets o de redes neuronales en aplicaciones financieras, la novedad de nuestro trabajo radica en la arquitectura particular que proponemos. Los resultados muestran que, en promedio, el modelo de neuro-wavelet propuesto supera tanto a un modelo ARIMA como a un modelo de red neuronal densa de referencia. Podemos concluir que, aunque más investigación es necesaria, dada la creciente capacidad técnica actual de los participantes del mercado, la inclusión del modelo LSTM neuro - wavelet al abanico de herramientas disponibles es de mucho valor, pues podría representar una ventaja sobre las herramientas predictivas tradicionales.

Abstract With the availability of high frequency data and new techniques for the management of noise in signals, we revisit the question, can we predict financial asset prices? The present work proposes an algorithm for next-step log-return prediction. Data in frequencies from 1 to 15 minutes, for 25 high capitalization assets in the Mexican market were used. The model applied consists on a wavelet followed by a Long Short-Term Memory neural network (LSTM). Application of either wavelets or neural networks in finance are common, the novelty comes from the application of the particular architecture proposed. The results show that, on average, the proposed LSTM neuro-wavelet model outperforms both an ARIMA model and a benchmark dense neural network model. We conclude that, although further research (in other stock markets, at higher frequencies, etc.) is in order, given the ever increasing technical capacity of market participants, the inclusion of the LSTM neuro-wavelet model is a valuable addition to the market participant toolkit, and might pose an advantage to traditional predictive tools.