The objective of the medium and long-term hydrothermal scheduling problem is to define operational target for each power plant in order to meet the load at the lowest expected cost and respecting the operational constraints. Stochastic Dynamic Programming (SDP) and Stochastic Dual Dynamic Programming (SDDP) algorithms have been widely applied to determine the optimal operating policy for the hydrothermal dispatch. In both approaches, the stochasticity of the inflows is usually produced by periodic auto-regressive models - PAR (p), whose parameters are estimated based on available historical data. As the estimators are a function of random phenomena, besides the inflows uncertainty there is statistical parameter uncertainty, which is not captured in the standard PAR (p) model. The existence of uncertainty in the parameters means that there is a risk that the hydrothermal operating policy will not be optimal. This thesis presents a methodology to incorporate the PAR(p) parameter uncertainty into stochastic hydrothermal scheduling and to assess the resulting impact on the computation of a hydro operations policy. Case studies are presented illustrating the impact of parameter uncertainty in the system operating costs and how an operating policy that incorporates this uncertainty can reduce this impact.