(1983). He questions the seemingly high level of confidence in the announced currency
band of Mexico -as inferred from the relatively low interest differential- prior to the Mexican
crisis of 1994. After adjusting the interest differential by such a time-varying risk premium,
Werner reaches a better ex ante measurement for the realized devaluation. By employing a
similar framework for the East Asian crisis of 1997, Bratsiotis and Robinson (2005) proxy the
risk premium term using relative share of foreign currency denominated private debt. They
document that once the UIP condition is corrected for such a time-varying risk premium,
it can then be used to forecast exchange rate depreciation prior to crisis.
Similarly, Poghosyan et al. (2007) relate the risk premium to two factors, i.e. the ratio
of deposits in domestic to foreign currencies, and foreign exchange market interventions
of central bank. Using the Armenian data for the period 1997-2005, they test whether
the risk premium can account for the forward premium bias, and whether there exists a
maturity effect. The results show that risk premium rises as maturity increases. Moreover,
regressing excess currency return on its own lags within a GARCH-in-mean framework,
where the conditional variance in the mean equation reflects time-varying risk premium,
they reject both risk neutrality and rational expectations hypotheses.
Another strand of literature examines the time-series properties of risk premium, ρt ,
to shed light on market integration of the countries in question. Among others, Holte-
moller (2005) examines the monetary integration of the European Union accession coun-
tries through investigating the time path of ρt as defined in equation (11).24 Intuitively, the
magnitude and the volatility of the risk premia in these accession countries are expected to
decay gradually along the path of monetary integration. Accordingly, Holtemoller investi-
gates time series properties of ρt for each country and questions whether it has a declining
trend both in magnitude and volatility. ρt can be stationary only when domestic and the
foreign interest rates have the cointegrating vector (1,-1), and exchange rate changes are sta-
tionary. Holtemoller assumes the latter and tests whether the domestic and the Euro-zone
interest rates are cointegrated in this particular way. Within the context of regression-
based UIP analysis, a non-stationary risk premium term implies that domestic and foreign
22