Thousands of BEERs: Take your pick

• Author: Christian Grisse,Konrad Adler
• Published date: 01 November 2017
• DOI: http://doi.org/10.1111/roie.12296
• Date: 01 November 2017

ORIGINAL ARTICLE

ThousandsofBEERs:Takeyourpick

Konrad Adler

1

|

Christian Grisse

2

1

Toulouse School of Economics, France

2

Swiss National Bank, Z€

urich,

Switzerland

Correspondence

Christian Grisse, Swiss National Bank,

B€

orsenstrasse 15, 8022 Z€

urich,

Switzerland.

E-mail: christian.grisse@snb.ch

Abstract

This paper explores the robustness of behavioral equilibrium

exchange rate (BEER) models. We highlight the importance

of model uncertainty, and employ real exchange rates com-

puted from price-level data to explore robustness to the

inclusion of country fixed effects. The estimated coefficients

—and therefore also the implied equilibrium values—are

sensitive to the combination of variables included in the

model, and to the inclusion of fixed effects. We identify sev-

eral variables that exhibit a robust link with real exchange

rates across specifications. Our findings can help policy-

makers in understanding the uncertainty associated with

estimates of equilibrium exchange rates.

1

|

INTRODUCTION

The assessment of equilibrium exchange rates plays an important role in international policy discus-

sions, in particular at the regular IMF article IV missions. Estimates of exchange rate misalignments

are a key input for the evaluation of global imbalances and the assessment of equilibrium exchange

rates helps policymakers to determine whether currency movements reflect temporary factors, such as

safe-haven effects that are likely to recede once market turbulence subsides, or whether they are driven

by more permanent changes in fundamental macroeconomic variables. Several approaches have been

proposed to estimate equilibrium exchange rates.

1

The problem for policymakers, however, is that

these approaches often produce very different results. Furthermore, even within any particular

approach, the estimates will still depend on a number of modeling assumptions.

One popular method for estimating equilibrium rates is the behavioral equilibrium exchange rate

(BEER) approach. This is also a key component in IMF exchange rate assessments.

2

In BEER models

a long-run relationship between real exchange rates and a set of fundamental macroeconomic variables

is estimated, typically in a panel regression across currencies. The predicted values from this regression

are then interpreted as equilibrium exchange rates.

BEER models have two weaknesses. The first weakness is that structural macroeconomic models

suggest many variables as potential drivers of real exchange rates, but with a limited dataset it is often

not be practicable or optimal to include all of these variables (some of which may have little explana-

tory power for real exchange rates) in the regression. This gives rise to the issue of model uncertainty,

as coefficients and equilibrium rates depend on the particular specification that the researcher is

1078

|

V

C2017 JohnWiley & Sons Ltd wileyonlinelibrary.com/journal/roie Rev IntEcon. 2017;25:1078–1104.

Received: 21 March 2016

|

Revised: 22 March 2017

|

Accepted: 3 April 2017

DOI: 10.1111/roie.12296

estimating. The second weakness is the need to include country-specific fixed effects. Real exchange

rates are typically computed using consumer prices as deflator, and are therefore index numbers that

do not have any meaningful cross-sectional variation. This requires the use of country-specific fixed

effects in panel regressions, but with fixed effects the predicted real exchange rates are by construction

on average equal to the observed real exchange rate average. This makes the estimated equilibrium

rates hard to interpret, especially when the sample is short. The estimation of BEER models based on

real exchange rates levels is therefore a focus of current research.

3

This paper assesses the robustness of BEER models, focusing on the role of (1) model uncertainty

and (2) country-specific fixed effects. To address the issue of model uncertainty we estimate real

exchange rate regressions for all possible combinations of the fundamental variables, and report the

distribution of coefficients and implied equilibrium exchange rates across this range of models. We

also use Bayesian model averaging to weigh these models in a statistically optimal way.

4

To assess the

role of country-specific fixed effects we use real exchange rates computed from price-level data, and

compare the distributions of coefficients and equilibrium exchange rates in panel models with and

without fixed effects.

We find that alternative specifications imply a range of values for the model coefficients, but some

variables—including the terms of trade, the real interest rate, credit, gross domestic product (GDP) per

capita, government consumption and net foreign assets—are robustly linked to real exchange rates

across these different models. Alternative specifications also imply a distribution of estimates for equi-

librium exchange rates. For the fixed effects’regressions, however, this distribution is remarkably tight

as alternative models tend to explain the time-series variation of real exchange rates quite well. We

conclude that when using BEER models for policy analysis it is important to emphasize the sensitivity

of the results to alternative model specifications. Understanding the uncertainty associated with esti-

mates of equilibrium exchange rates can help policymakers in interpreting the results.

Our framework, which is based on a cointegrating relationship between real exchange rates and

fundamentals, implies that estimated exchange rate misalignments should have predictive power for

exchange rate movements, with overvaluation (undervaluation) followed by real depreciation (appreci-

ation). In a simple in-sample forecasting exercise we find that this is indeed the case.

The empirical literature on the relationship between real exchange rates and fundamentals in gen-

eral and BEER models in particular is vast. A survey of papers published since 2000 (see Table 1

below) shows that the recent empirical literature has explored at least 13 different macroeconomic vari-

ables as drivers of real exchange rates—and several more when variables that are relevant only for

emerging market countries are also included (such as the share of administered prices and dummy vari-

ables capturing balance of payment crises). Furthermore, papers often use different proxies to measure

factors such as demography and the Balassa–Samuelson effect. Many papers only include few varia-

bles, motivated by a particular theoretical model. Bussière, Ca’Zorzi, Chudik and Dieppe (2010) and

Phillips et al. (2013) are two papers that explore a large number of fundamental variables. Almost all

papers—with the notable exceptions of Bussière et al. (2010) and Berka and Devereux (2013)—use

real exchange rate indices, and therefore include country-specific fixed effects.

This paper is most closely related to Bussière et al. (2010). That comprehensive study also explores

the role of model uncertainty in equilibrium exchange rate models, focusing on three alternative esti-

mation approaches (fundamental equilibrium exchange rates or FEERs, external sustainability

approach, and BEERs). We follow Bussière et al. in using Bayesian model averaging (BACE) and

computing price-level based real exchange rates, but concentrate on the BEER approach, which is the

IMF’s main tool for exchange rate assessments. Our contribution differs from and adds to Bussière

et al. in the following ways. First, we discuss the implications of model uncertainty not only for con-

clusions about the importance of alternative variables as drivers of real exchange rates, but also for the

ADLER AND GRISSE

|

1079