General Circulation Models (GCMs) have been widely used to produce long-term predictions of future climate change scenarios. The CMIP5 experiment provided an ensemble of historical and future climate simulations with a horizontal resolution of hundreds of kilometers. These spatial resolutions are too coarse to evaluate the climate impact on a local scale especially for small island which are often invisible for GCMs. The statistical downscaling technics allow to produce downscaled projections at a finer scale with relatively low computation cost. At EDF R&D, we use statistical downscaling technics such analogues and CDFt to produce finer climate information at local scale for enegy purposes. Assessing, the downscaling skills is a fundamental issue for our studies. In this work we use a Perfect model appraoch to evaluate the CDFt and the quantile-mapping skills to downscale climate projections (daily temperature and precipitaion) over CORSICA. The EURO-CORDEX simulations with a horizontal resolution of 11° were used as a psedo-reference. A set of experiment were conducted to answer a number of question such as the stationarity hypothesis. Result show that for temperature, CDFt performs very good in downscaling the cdf of the GCM temperature independently from the time interval between the calibration and validation periods. However, for quantile-mapping, the skill decreases as the time interval between the calibration period and the validation one increases.