This work analyze time series models for inflow generation. The generated series are used in the SDDP (Stochastic Dual Dynamic Programming) algorithm, that makes the optimization of the electrical planing of a country or region. The SDDP solves linear optimization problems, and it needs convexity of the problems involved. For that, a crucial exigency is that the forecast function of the inflow time series model be concave, and this is a common characteristic between the classical models commonly used to generate inflow series. Two of these will be analyzed and compared to the GAS (Generalized Autoregressive Score). The last one isn t concave, but can still be used in a phase of the SDDP algorithm (the forward phase). In conclusions, GAS had a better performance in the diagnosis tests even if it has only one model for the whole series, while the other two have a speciric model (with different orders) for each month. Otherwise, the monthly basis models had an better result in the scenarios generation test, mostly because it could represent better the probability distribution of each month, while GAS represent better the total probability distribution.