First, we determine whether the SETAR specification is superior to a unit root process for each SRER using the Enders and Granger (1998) threshold unit root test.
A failure to reject the null hypothesis implies that the SRER is nonstationary and consequently prices in two locations are disconnected.
Tables 2A, 2B, and 2C show the estimated threshold bands for each SRER for the three country pairs.
Tables 6A, 6B, and 6C report the estimated thresholds for each SRER, allowing for a different mean for the real exchange rate during the Tequila Crisis.
Using a SETAR model, we find strong evidence of nonlinearities in SRER dynamics across Mexico, Canada, and the United States in the pre- and post-NAFTA periods.
We obtain these parameters for each SRER corresponding to the three country pairs for both periods.
Nevertheless, the impact of changes in the sustainable debt to GDP ratio on the SRER estimates is relatively small: a 10 percentage point change in the ratio generated a change in the equilibrium real exchange rate of 0.
Data consistency is crucial for the SRER calculations, given the endogenous relationship between various variables, such as domestic and foreign demand or trade and financial flows.
We interpret the difference between an estimate of the SRER and the observed real effective exchange rate in two ways.
For 1998-2007 we compare our SRER estimates with observed values of effective real exchange rates and our estimates suggest that, first, forerunners' currencies were mostly overvalued and, second, latecomers' currency misalignments were volatile, with a gradual overvaluation trend towards the end of the period (see figure 3).
For post-2007 series we compare our SRER estimates with a real effective exchange rate commensurate with a peg to the euro and inflation equal to the announced inflation targets (see the forecast portion of figure 3).
The estimates of equilibrium exchange rates obtained in this round are more favourable than those from the two earlier (normalised) SRER estimates in Bulff and Smidkowa (2005) and (2007), see figure 4.