The paper considers stochastic optimization of the electricity procurement in the day-ahead power market. The novelty is in addressing the random errors of time series forecasting of electrical power loads and prices in the procurement. This problem is currently important because of the increased random variability in the power grid that is caused by growing integration of renewable generation. This paper presents a methodology for stochastic optimization using data-driven models. We consider non-parametric models of multivariate distributions based on multiple quantile regressions, built from historical data sets. The statistics, such as cost expectation, required for the stochastic optimization are computed numerically using these models. Applying the methodology to utility data shows that 2% improvement of the costs is feasible.