After reading this work, I finally understood what stock prices are and how stock market predictions can be built. The paper highlights a bridge existing between machine learning and economics. Basically, the authors introduce a way to analyze stock market predictions from the technical analysis point of view. The idea is simple: train classical machine learning classifiers (neural networks, decision trees, K-nearest neighbor algorithms) on specific training patterns, and use them to predict Dow Jones Industrial Average time series.

The paper illustrates, first, how to select meaningful training data, and, second, how to use the different classifiers. In particular, the authors show that the correct direction that should be taken in future studies is to use an ensemble of classifiers. The results obtained exhibit 60 to 65 percent accuracy.

This high-quality paper is really easy to read, and multiple and exhaustive experiments have been considered.