The Kyoto Protocol adopted in 1997 has three flexible mechanisms in order to prevent the earth from global warming. One of them is emissions trading. Saijo and Kusakawa (2002) designed some trading systems and conducted emissions trading experiments in order to evaluate them. Such human-based experiments, however, have several basic problems. First, any single experimental result might depend on a particular set of subjects. Second, a considerable number of repetitions are needed to have statistically robust results. In order to avoid these drawbacks, we construct an agent-based emissions trading simulation model that does not use human subjects, but artificial computer agents. It is relatively easy to repeat many simulations with many different parameters when we use the simulation method. Our focus is teaming through the trading and we employ the Genetic Algorithm (GA). We evaluate two trading designs: the penalty against non-compliance and the Commitment Period Reserve (CPR) adopted in the Marrakesh accords. The former has a question how much penalty should be imposed on countries with non-compliance. We have two cases. Case I is that the penalty is fixed, and Case 2 is that the penalty depends on the average contract price of emissions permits. We found that in Case I all countries tried to comply with their target, while in Case 2 demand countries had some incentive not to comply the protocol on purpose. The latter has a question whether we should adopt the CPR or not to the emissions trading system. We found that the CPR dared not be adopted because we confirmed that it had no significant effect.
