BANK PURCHASES OF EARNING ASSETS A DECISION UNIT MODEL*

Abstract

TRADITIONALLY, THE ATTENTION of monetary theorists has been directed toward explaining such aggregate behavior as consumption, saving, and investment. To this end, there has usually been an interest in such aggregate variables as interest rates, the money supply, velocity, and the price level. The advent of powerful computational and data-processing equipment makes feasible research efforts directed toward integrating models of micro behavior with larger systems of models relating to aggregates. This study consists of the specification and testing of several models predicting an individual bank's net purchases of earning assets. The broad hypothesis of the study is that certain underlying environmental, legal, and status variables have systematic influences on the behavior of banks. Therefore, the models are specified largely in terms of variables which differ among banks (e.g., size, population of city in which located, deposit mix, etc.). In addition, however, several tests are performed with models which include such timeseries variables as interest rates, certain broad measures of economic conditions, and variables which seek to account for seasonal movements. The data with which the models are tested stem from statements of condition from member and non-member banks located in the Eighth Federal Reserve District. These statements, or call reports, provide successive cross-sections at six-month intervals during the period from June, 1950, to December, 1958. The empirical portion of the study consists of estimating the parameters relating the inputs and status variables of individual commercial banks (the independent variables) to their output, net purchases of earning assets (the dependent variable). A by-product of this process consists of testing hypotheses relative to the influence of such variables as size, deposit experience, deposit mix, and other environmental variables. The parameters of the models are estimated by multiple linear-regression techniques. A majority of the parameter estimates attain acceptable levels of significance, but the variables