The customer baseline is required to assign rebates to participants in baseline-based demand response (DR) programs. The average baseline method has been widely accepted in practice due to its simplicity and reliability. However, the customer's baseline manipulation is little-known in the literature. We start from a customer's perspective and establish a Markov decision process to model the customer's payoff-maximizing problem. The behavior of a rational customer's underconsumption on DR days and overconsumption on non-DR days are revealed. Furthermore, we propose an approximated baseline method and show how the consumption distribution and program parameters affect the results. Due to the curse of dimensionality, a linear policy-based rollout algorithm is introduced to obtain a practical approximate solution. Finally, a case study is carried out to illustrate the baseline manipulation, where the simulation results confirm the effectiveness of the proposed methods and shed light on how to properly design baseline methods. Index Terms-Baseline manipulation, baseline method, demand response, dynamic programming, Markov decision process.