In this paper we present the design of a system for stock market trading assistance. The system will provide buy/sell signals. The aim is to maximize the accuracy of such signals, even with the cost of missing some of them. The architecture of the network will be determined using a genetic algorithm and for training differential evolution will be used. The support vector machine will perform a three-class classification (buy/sell/no action). For this, two support vector machines will be used, one classifying buy against don’t buy, and the other one classifying sell against don’t sell.