Rui Neves – författare

This book presents a system that combines the expertise of four algorithms, namely Gradient Tree Boosting, Logistic Regression, Random Forest and Support Vector Classifier to trade with several cryptocurrencies. A new method for resampling financial data is presented as alternative to the classical time sampled data commonly used in financial market trading. The new resampling method uses a closing value threshold to resample the data creating a signal better suited for financial trading, thus achieving higher returns without increased risk. The performance of the algorithm with the new resampling method and the classical time sampled data are compared and the advantages of using the system developed in this work are highlighted.

This book presents a genetic algorithm that optimizes a grid template pattern detector to find the best point to trade in the SP 500. The pattern detector is based on a template using a grid of weights with a fixed size. The template takes in consideration not only the closing price but also the open, high, and low values of the price during the period under testing in contrast to the traditional methods of analysing only the closing price. Each cell of the grid encompasses a score, and these are optimized by an evolutionary genetic algorithm that takes genetic diversity into consideration through a speciation routine, giving time for each individual of the population to be optimized within its own niche. With this method, the system is able to present better results and improves the results compared with other template approaches. The tests considered real data from the stock market and against state-of-the-art solutions, namely the ones using a grid of weights which does not havea fixed size and non-speciated approaches. During the testing period, the presented solution had a return of 21.3% compared to 10.9% of the existing approaches. The use of speciation was able to increase the returns of some results as genetic diversity was taken into consideration.

This book develops a quantitative stock market investment methodology using financial indicators that beats the benchmark of S&P500 index. To achieve this goal, an ensemble of machine learning models is meticulously constructed, incorporating four distinct algorithms: support vector machine, k-nearest neighbors, random forest, and logistic regression. These models all make use of financial ratios extracted from company financial statements for the purposes of predictive forecasting. The ensemble classifier is subject to a strict testing of precision which compares it to the performance of its constituent models separately. Rolling window and cross-validation tests are used in this evaluation in order to provide a comprehensive assessment framework. A risk-off filter is developed to limit risk during uncertain market periods, and consequently to improve the Sharpe ratio of the model. The risk adjusted performance of the final model, supported by the risk-off filter, achieves a Sharpe ratio of 1.63 which surpasses both the model’s performance without the filter that delivers Sharpe ratio of 1.41 and the one from the S&P500 index of 0.80. The substantial increase in risk-adjusted returns is accomplished by reducing the model’s volatility from an annual standard of deviation of 15.75% to 11.22%, which represents an almost 30% decrease in volatility.