An analysis of the Bayesian classification of monetary assets in the United States

Using monthly time series data from 1959 to 2002, this paper attempts to cluster the Federal Reserve's 26 monetary assets according to mathematical properties, rather than by similarities regarding their liquidity. The new classes are formed using a Bayesian procedure entitled AutoClass, which groups the assets according to their various attributes such as asset quantities, user costs, velocity, and growth rates. This classification results in three distinct groups being formed. Thereafter, monetary aggregates are constructed using simple sum, divisia, and currency equivalent techniques. Using these new aggregates, a formal analysis of the interaction between money, prices, output and the interest rate follows, paying particular attention to the causal and cyclical relationships present.