Non-technical electricity losses (NTL) are currently one of the main problems faced by electricity distribution utilities in Brazil. It is understood that part of these factors is not manageable by the utilities, and thus, they are passed on to consumers through electricity tariffs, following the limits defined by the Brazilian Electric Energy Agency (ANEEL). Despite the progress made by ANEEL, the methodology currently used by the Agency uses panel data to formulate these limits, however, some ad hoc measures are defined which end up resulting in loss targets that are not consistent with the reality of Brazilian distribution utilities. This dissertation aims to select models automatically for panel data, evaluating their applicability in setting non-technical loss targets. The proposed models were used in a panel data containing annual observations from 62 Brazilian electricity distribution utilities in the period from 2007 to 2017, thus generating 1,097,789 models with respect to the types of regressions in panel data, resulting in 3 potential models with NTL targets more adherent to the real possibilities of the Brazilian distribution utilities.