The Agricultural Resource Management Survey (ARMS) database contains information on production practice and financial information of farmers and farmland in United States. Unfortunately, the commodity-specific information is missing in ARMS dataset. By imputing these missing values, we can estimate the kernel density of net returns of the main commodities to make U. S. agricultural policy. In this thesis, we apply hot deck imputation, Bayesian approach and matrix completion to recover all the missing values and utilize the Kullback-Leibler divergence to determine the best fitted distributions for net returns of the six major commodities. Overall, these three approaches are effective to impute the missing values and adjust the agricultural policy in United States.