Monetary disequilibria and the euro/dollar exchange rate

Abstract

Although stable money demand functions are crucial for the monetary model of the exchange rate, empirical research on exchange rates and money demand is more or less disconnected. This paper tries to fill the gap for the euro/dollar exchange rate. We investigate whether monetary disequilibria provided by the empirical literature on US and European money demand functions contain useful information about exchange rate movements. Our results suggest that the empirical performance of the monetary exchange rate model improves when insights from the money demand literature are explicitly taken into account.

Keywords:

• euro
• dollar exchange rate
• monetary model
• money demand functions

Acknowledgements

The research for this paper was carried out while Karsten Ruth was member of the Graduate Program Finance and Monetary Economics at the Goethe University Frankfurt. Financial support by the German Research Foundation (DFG) is gratefully acknowledged. An earlier version of this paper is published as part of Karsten Ruth's doctoral thesis and circulated as Deutsche Bundesbank Discussion Paper No. 18/2005. The opinion expressed in this paper does not necessarily reflect the viewpoint of the Deutsche Bundesbank.

Notes

Notice further that cointegration tests applied in the empirical exchange rate literature usually pay little attention to the fact that nominal money supply potentially follows an I (2) process (see Holtemöller 2004). In this paper, this is not an issue because we will deal with stationary monetary disequilibria derived from demand functions for real money (see Section 3).

Note that unit coefficients for relative money require that the underlying long-run money demand functions are correctly specified. In particular, if inflation is I (1) and has to be included in the long-run money demand as an additional variable (see, e.g. Coenen and Vega 2001), then the resulting fundamental exchange rate equation (6) has to be augmented by an inflation term.

Note that λ equals zero in case of a non-stationary interest rate differential.

Modeling deviations of the exchange rate from PPP in addition to deviations from economic fundamentals was also proposed by Mark and Sul (2001, p. 41).

All disequilibria are mean-adjusted, whereby the mean is calculated over the respective estimation sample (see Table 1).

We thank Claus Brand, Annick Bruggeman, Günter Coenen, Oliver Holtemöller, and Christian Müller for providing us with the original data used for their analyses. Data for Golinelli and Pastorello (2002) are available at http://spbo.unibo.it/pais/golinelli/macro.html. Funke (2001) essentially draws on the same data set as Coenen and Vega (2001). Kontolemis (2002) expands the data set of Brand and Cassola (2000) up to 2001:3. The US data can be downloaded at http://research.stlouisfed.org/fred2.

The own rate of M3 as used in Bruggeman, Donati, and Warne (2003) was computed for the period 2002:1–2002:4 by a dynamic simulation of a regression on (within a sample 1994:2–2001:4), where the Euribor 3-month rate obtained from the EAS. D(·) is the difference operator.

The (log) US GDP deflator and the (log) European GDP deflator enter the ec ^PPP term as foreign and domestic price levels, whereas ec ^ird includes the European and the US 10-year government bond yields. According to KPSS unit root tests, both regressors are stationary (see Appendix B). For the sake of completeness, please note that all equations contain a constant and an impulse dummy D924 capturing the effects of exchange rate market turmoils in fall 1992.

Estimation of Equation (9) involves generated regressors (the estimated monetary disequilibria) such that standard $t$-statistics have to be viewed with caution (compare with Pagan 1984). Note, however, that the inference problem due to generated regressors cannot be seen as a distinguishing disadvantage of the structural approach (see, e.g. MacDonald and Taylor 1994).

For all equations, a maximum lag length of k=6 was chosen, thereby removing potential autocorrelation from the residuals.

Employing the Johansen approach in order to efficiently estimate the parameters of the fundamental Equation (4) is one possible choice. Alternatively, single equation approaches such as FM-OLS or DOLS would also ensure asymptotic efficiency (see Rapach and Wohar 2002, p. 365).

It should be stressed that – strictly speaking – this forecast evaluation has to be classified as ‘pseudo’ ex-ante. That is because definitions of the ec ^m(*)-terms (monetary disequilibria) draw on parameter estimates obtained by others over a sample which includes parts of the forecast horizon. However, given stable cointegrating relations, parameter estimates should not change substantially over the forecast horizon.

Note that the results do not change qualitatively when lagged first differences of the exchange rate are included in Equations (11) and (12).