The paper proposes a continuous time model of an FX market organized as a multiple dealership. The model reflects a number of salient features of the Czech koruna spot market. The dealers have costly access to the best available quotes. They interpret signals from the joint dealer-customer order flow and decide upon their own quotes and trades in the inter-dealer market. Each dealer uses the observed order flow to improve the subjective estimates of the relevant aggregate variables, which are the sources of uncertainty. One of the risk factors is the size of the cross-border dealer transactions in the FX market. These uncertainties have diffusion form and are dealt with according to the principles of portfolio optimization in continuous time. The model is used to explain the country, or risk, premium in the uncovered national return parity equation for the koruna/euro exchange rate. The two country premium terms that I identify in excess of the usual covariance term (a consequence of the 'Jensen inequality effect') are the dealer heterogeneity-induced inter-dealer market order flow component and the dealer Bayesian learning component. As a result, a 'dealer-based total return parity' formula links the exchange rate to both the 'fundamental' factors represented by the differential of the national asset returns, and the microstructural factors represented by heterogeneous dealer knowledge of the aggregate order flow and the fundamentals. Evidence on the cross-border order flow dependence of the Czech koruna risk premium, in accordance with the model prediction, is documented.
Keywords: Bayesian learning, FX microstructure, optimizing dealer, uncovered parity
Issued: July 2003
Download CNB WP No 4(2003 (pdf, 958 kB)