One branch of the banking literature in recent years has been devoted to developing dynamic theories of bank behavior. Goodfriend's (1983) model of the bank discount-window borrowing decision highlighted the deficiency of continuous-time static models that followed in the tradition of Klein (1971).(1) In the context of a banking model in which (implicit) intertemporal administrative costs imposed by the monetary authority were an important feature of the bank portfolio allocation problem, Goodfriend demonstrated how the inclusion of dynamic elements in a banking model could significantly influence the behavioral and policy implications that emerge from such models. Cosimano (1987, 1988), Cosimano and Van Huyck (1989), and Bundt, Cosimane, and Halloran (1992a, 1992b) have built on Goodfriend's approach and examine more fully the effects of the intertemporal bank decision process on the rational expectations properties of alternative monetary and banking policies.
A common feature of this dynamic banking literature is the implication that the asset and liability decisions of banks are jointly independent. Cosimano (1987) attributes this portfolio separation property solely to his assumption that dynamic adjustment costs are additively separable in the sense that these costs exist on either the deposit side or the loan side of the banks' balance sheet but not on both sides simultaneously. Yet, cost separability has been a common feature in static implicit- cost models that, nonetheless, imply portfolio interdependence.(2) The precise nature of the intertemporal adjustment-cost mechanism is, therefore, an important issue because policy conclusions ultimately rest on the assumed cost structure.
In this paper we develop a model of the banking firm that enables us to test for portfolio separation. Our theoretical model generalizes existing intertemporal adjustment-cost models by assuming that these costs coexist simultaneously on both sides of the bank's balance sheet. The optimal deposit supply and loan demand functions that emerge from our extended model show that each of these functions depends both on one- and two-period quantity lags and the lagged, contemporaneous, and expected future values of interest rates, including loan, deposit, and short-term borrowing (for example, federal funds) rates. Models exhibiting portfolio separation or asset-liability independence emerge as a special case of the extended model. This feature enables us to test conditions, paralleling those first highlighted by Sealey (1985), that are consistent with the portfolio separation hypothesis. When portfolio separation holds, deposit supply and loan demand depend only on a one-period quantity lag and contemporaneous and expected future differentials of the own rate and the federal funds rate. Thus, the property of portfolio separation can be examined empirically by determining whether the zero restrictions required by that property are supported in the econometric estimates of the banks' deposit supply and loan demand functions.
Our empirical analysis uses a panel-data technique to jointly estimate bank loan demand and deposit supply functions for four classes of banks, ranging in size from $1 billion in assets to the largest U.S. banking organizations. The results indicate that the portfolio separation feature of previous dynamic models does not accurately characterize bank behavior, at least for the banks examined in this study. Several of the key predictions of these models are at odds with the estimated forms of both the deposit and loan equations. This finding applies even to the largest class of banks that are the prime candidates for satisfying the portfolio separability conditions because they are more likely to be risk neutral and to have ready access to interbank lending markets. We conclude that recent dynamic models of the banking firm are based on overly restrictive assumptions that may limit the breadth of their …