In out-of-sample exchange rate forecasting, purchasing power parity has higher predictive power over the longer term and Taylor rule models are predictive over the shorter term.
What’s Inside?
The author studies exchange rate forecasting using real-time data for a reasonably large number of countries over the post–Bretton Woods period. He uses an out-of-sample, constructed, real-time dataset for nine countries versus the US dollar over a long period (1973–2009). He evaluates both short- and long-horizon out-of-sample exchange rate forecasting performance of the linear exchange models using purchasing power parity (PPP) and Taylor rule fundamentals with constructed real-time price and inflation data and the consumer price index (CPI). He also estimates real-time output gaps using the industrial production index.
How Is This Research Useful to Practitioners?
Most available studies on exchange rate analysis use ex post revised data, which contain future information that was not available to policymakers and market participants at the time the forecasts were made. Hence, the same information cannot be used to evaluate the predictability of exchange rate models with an out-of-sample statistic. The author shows that panel estimation increases the predictability of the PPP model compared with the single-equation estimation.
He finds evidence of predictability with a panel estimation for seven out of nine countries with the Clark and West (Journal of Economics 2006) test (CW test) and five out of nine countries with the Diebold and Mariano (Journal of Business & Economic Statistics 1995) and West (Econometrica 1996) test (DMW test) against the driftless random walk. All of the countries exhibit predictability when tested against the random walk with drift regardless of which test statistic is used. One-quarter-ahead forecasts of the exchange rate model with PPP fundamentals are weaker than long-horizon forecasts.
The out-of-sample forecasting exercises also show that the predictability of the Taylor rule model is higher at the short horizon. The evidence with the single-equation Taylor rule model is found for one out of nine countries with both test statistics tested against the driftless random walk; it is found for three out of nine countries with the CW test and five out of nine countries with the DMW test against the random walk with drift. Because central banks target short-term nominal interest rates, these findings are reasonable.
How Did the Author Conduct This Research?
The author uses real-time quarterly data from 1973 to 2009 constructed from monthly data of the IMF’s International Financial Statistics (IFS) books for Australia, Canada, France, Germany, Italy, Japan, the Netherlands, Sweden, the United Kingdom, and the United States. He also uses the seasonally adjusted industry production index as a measure of each country’s income because of a lack of availability of consistent GDP data, a seasonally adjusted consumer price index, and an annual inflation rate. The author conducts econometric analysis based on panel estimation of the predictive regression with PPP and Taylor fundamental equations. He then calculates the forecasted out-of-sample test statistic using the CW, DMW, and minimum mean-squared prediction error (MSPE) tests followed by bootstrapping to evaluate the results.
The results indicate that the PPP model works best with the panel specification at the 16-quarter horizon (i.e., long-run PPP with panel methods works better than with univariate tests). Out-of-sample forecasts with Taylor rules suggest the existence of substantial heterogeneity in adjustments to equilibrium; hence, they cannot outperform time-series regression forecasts and work best in a single-equation framework over the short horizon.
Abstractor’s Viewpoint
The author reinforces the principles laid out by PPP and the Taylor model in exchange rate predictability.