This paper proposes the application of nonlinear autoregressive neural networks for the forecast of the stock market index of the Ecuadorian stock market, Ecuindex. Forty-five NAR network structures are tested; modifying the number of lags in the index time series and the number of neurons in the hidden layer. In the test period, the best network has a MAPE error of less than 0.25% and a percent of success in change direction greater than 68%.