In recent years, the importance of environment is widely recognized and various kinds of environmental policies have been implemented. However, when one establishes a policy, there are few cases that the contents and the period of the policy are evaluated by using prediction method for its effect. In this study, we aim to predict the effects of a policy in micro-level by using agent simulations and estimate the optimal period of the policy implementation by totally evaluating its effect. For sugarcane farmer of an area in Japan, we performed 200 thousand patterns of simulations in order to optimize the subsidy of green manure and its implementation period. In particular, when the values of the parameters in the simulation model are not clearly defined, we are able to show various futures caused by the policy implementation of interest based on large-scale simulations with a huge variety of the parameter values.