A key element in the implementation of an effective result-based mechanism for Reducing Emissions from Deforestation and forest Degradation (REDD+) is the reference level (RL). Setting RLs requires modeling and predicting deforestation trajectories for a business-asusual (BAU) scenario. This thesis looks into two aspects of the design and implementation of a Payment for Environmental Service (PES) scheme for REDD+. First, we apply spatial econometric panel data analysis to explore the drivers of deforestation in Indonesian districts. Spatial models come in many forms, and we test and identify the most suitable spatial model, the Spatial Autoregressive (SAR) model. Incorporating a spatial lag of the dependent variable does not only help us measure neighborhood effects but also improves the accuracy of estimates of other predictor variables that drive deforestation. We found a strong inter-district dependence, which implies that there could be synergistic gains in the implementation of forest conservation policies. Deforestation is contagious, but conservation efforts may have positive leakage (spillover), much like the effect of vaccination on those not treated.