We develop a multi-agent based model of stock market and simulate it numerically. The agents are heterogeneous and risk-averse, adopting trading strategies deriving from technical analysis. The fluctuations are driven by a stochastic component in agent´s investment. The model is able to capture the stylized facts observed empirically in the stock markets in the normal regime, such as long tails in the distribution of returns, short-memory in the return temporal series and long-range correlation in volatility. Our simulations show that these empirical phenomenons emerge endogenously from the collective behavior of the agents, despite the simplicity of the model. This suggests that the complex market dynamics do not arise from the complexity of the individual investors but rather from interactions between them trough the market price formation.