Journal of Econometrics

Volume 66, Issue 1-2, 01-December-1994

J. Econometrics Vol. 66 (1-2) pp. 251-287

Nonparametric estimation of structural models for high-frequency currency market data

a Ravi Bansal
b A. Ronald Gallant
c Robert Hussey
a George Tauchen

a Duke University, Durham, NC 27706, USA
b University of North Carolina, Chapel Hill, NC 27599, USA
c Georgetown University, Washington, DC 20007, USA

Received 1 July 1992; Accepted 1 January 1994


Empirical modeling of high-frequency currency market data reveals substantial evidence for nonnormality, stochastic volatility, and other nonlinearities. This paper investigates whether an equilibrium monetary model can account for nonlinearities in weekly data. The model incorporates time-nonseparable preferences and a transaction cost technology. Simulated sample paths are generated using Marcet's parameterized expectations procedure. The paper also develops a new method for estimation of structural economic models. The method forces the model to match (under a GMM criterion) the score function of a nonparametric estimate of the conditional density of observed data. The estimation uses weekly U.S.--German currency market data, 1975--90.

Keyword(s): Monetary model; Calibration; Simulation estimator; Exchange rates; Nonpara-metric

JEL Classification: C51