Exchange rate regimes and national price levels

m/locate/econbase
Journal of International Economics 70 (2006) 52–81
www.elsevier.co

Exchange rate regimes and national price levels☆

Christian Broda

University of Chicago, Graduate School of Business, USA

Received 22 October 2004; received in revised form 16 September 2005; accepted 23 November 2005

Abstract

This paper explores the role of exchange rate regimes in explaining deviations from the classic theory ofpurchasing power parity. Examining a broad panel of countries, I find that developing countries with fixedexchange rate regimes have national price levels that are 20 percent higher than those with flexible regimes.For industrial countries, the relation between regimes and price levels is qualitatively similar but weaker. Iinvestigate several explanations for this pattern, and find that exchange-rate overshooting in floats, inflationinertia in pegs and expansionary policies can explain only 5 percentage points of the observed differences. Ialso show that even though the observed pattern could be the outcome of a class of open economy modelspioneered by Obstfeld and Rogoff, the data provides limited empirical support for the predictions of thismodel.© 2006 Elsevier B.V. All rights reserved.

JEL classification: E52; F33; F41

1. Introduction

A fundamental question in international economics is how does the price of the same basket ofgoods compare across countries when denominated in the same currency? This question aboutpurchasing power has given birth to the most influential theory of exchange rate determination.The purchasing power parity doctrine asserts that the exchange rate between two currencies isdetermined by the two countries' relative price levels, and therefore prices in a common currencyshould be identical across countries.

☆ I would especially like to thank my advisors Rudi Dornbusch and Jaume Ventura. I would also like to thank Eric vanWincoop and an anonymous referee for their helpful comments. I also benefited from discussions with Amartya Lahiri,Roberto Benelli, Guido Lorenzoni, Lucas Llach, Paolo Pesenti, Ken Rogo., John Romalis, Cedric Tille and Kei-Mu Yi.Rachel Polimeni provided excellent research assistance.

E-mail address: [email protected].

0022-1996/$ - see front matter © 2006 Elsevier B.V. All rights reserved.doi:10.1016/j.jinteco.2005.11.002

mailto:[email protected]

http://dx.doi.org/10.1016/j.jinteco.2005.11.002

53C. Broda / Journal of International Economics 70 (2006) 52–81

It has long been recognized that absolute purchasing power parity (PPP) does not hold inpractice. Departures from PPP were already recognized in the work of Ricardo, Keynes (1923)and Viner (1933). Qualifications to the theory took two forms: structural deviations from PPPrelated to lasting changes in equilibrium relative prices due to real factors (as in Harrod, 1939) ortransitory departures that arise as a result of the differential speed in adjustment of asset and goodsprices (as in Keynes, 1923).1

Despite the early theoretical work on disparties from PPP, it was not until the work of Balassa(1964) and Kravis and Lipsey (1983) that a systematic deviation from PPP was confirmedempirically. These authors found that real income per capita levels systematically influencerelative prices across countries. Balassa suggested that real income differences capture differencesin productivity across countries, while Kravis and Lipsey rely on different factor endowments toexplain the systematic correlation.2 Since then, the vast majority of empirical work related to PPPhas focused on relative PPP, i.e., whether the price of the same basket of goods moves togetherover time across countries when expressed in a common currency.3 Tests of relative PPP aretypically easier to perform as they only require data on real exchange rate indices rather thanactual cross-country price data.4 Recently, Imbs et al. (2005) and Crucini and Shintani (2004)have used micro data to evaluate the differences in persistence of individual goods relative toaggregate indices. Despite the dominance of relative PPP studies, a number of empiricalinvestigations have examined the extent of deviations from the law-of-one price using data onindividual goods across cities in Europe and U.S. (e.g., Rogers, 2001; Goldberg and Verboven,2005).

This paper explores the role of exchange rate regimes in explaining differences in the relativeprice of the same basket of goods across countries. Following Balassa (1964) and Kravis andLipsey (1983), national price levels are used to measure the extent of deviations from absolutePPP. National price levels are defined as the ratio between two countries' relative price and theirexchange rate. In particular, they are closely related to real exchange rates.5 The paper uncovers astrong empirical relationship between national price levels and exchange rate regimes that has notbeen previously studied. Specifically, developing countries with fixed exchange rates have highernational price levels than those with flexible rates. Using Penn World Table data for low andmiddle income countries, those with hard pegged regimes have national price levels that are

1 See Dornbusch (1988) and Rogoff (1996) for a more thorough discussion on departures from PPP. See Niehans(1990) for an explanation of Ricardo's view on international prices.2 Harrod (1939) was the first to suggest that productivity differences could explain systematic deviations from PPP.

Balassa (1964) and Samuelson (1964) argue that high-income countries are associated with a larger productivitydifference between tradable and non-tradable goods than low-income countries. Kravis and Lipsey (1983) and Bhagwati(1984) suggest that low-income countries have a higher labor intensity of non-tradable goods and relatively cheaper laborthan rich countries. Both explanations imply a strong (positive) relation between national price levels and income levels.3 A notable exception is Clague (1986). He shows how other structural characteristics (like trade balance and tourism)

influence national price levels. Bergstrand (1991) tries to distinguish between the Bhagwati (1984) and the Balassa(1964) explanation of why real per capita incomes and national price levels are correlated.4 As noted by Balassa (1964), even though Cassel's name has been associated with relative PPP he also formulated the

absolute version. See Lopez et al. (2005) for a recent survey on relative PPP studies.5 National price levels and real exchange rates are related but distinct. In theory, national price levels and real exchange

rates are identical if the shares in consumptions of goods are equal across countries. In practice, real exchange rates arecomputed using country-specific shares from government indices. By contrast, national price levels are computed usingthe same methodology across countries and actual price level data (see Section 2). This terminology is taken from thework by Kravis and Lipsey (1983), and continued by Heston and Summers (1991).

54 C. Broda / Journal of International Economics 70 (2006) 52–81

approximately 20 percent higher than those that fully float.6 The cross-sectional difference ishighly significant using all three available regime classifications (i.e., the IMF's officialclassification, Levy Yeyati and Sturzennegger's (2003) de-facto classification and Reinhart andRogoff's (2002) classification) during the period between 1980 and 1998.7 For industrialcountries, the association between regimes and price levels is qualitatively similar but weaker.

After documenting the facts, the paper examines three potential explanations of the sign andmagnitude of the observed pattern. Two of the explanations are taken from the existing literatureon currency crises and inflation stabilizations, and should be considered as transitory departuresfrom PPP. First, the tendency to run expansionary policies may lead to a real appreciation incountries with fixed exchange rate regimes but not in those with flexible regimes. This tendencymay be especially pronounced in developing countries. Second, sudden regime shifts can explainthe empirical association between prices and regimes. The exchange rate overshooting thataccompanies a currency crisis implies that low national price levels can be temporarily associatedwith flexible regimes (see Frankel and Rose, 1996). Similarly, countries that anchor theirexchange rates to stabilize inflation typically suffer from inflation inertia (see Cassel, 1928 andKiguel and Liviatan, 1991).8 This inertia implies that a country that adopts a fixed exchange rateto stabilize inflation should have a high national price level.

Unlike the first two explanations, the third one is based on a model of equilibrium nationalprice levels and constitutes what Dornbusch (1988) defines as a structural departure. In theirseminal work on stochastic general equilibrium models, Obstfeld and Rogoff (2000) showed thatthe second moments of economic disturbances could have significant impact on price andexchange rate levels.9 This framework can be used to show that countries that face more volatilityof nominal marginal costs will have higher national price levels.10 In particular, pegs have highernational price levels than floats if uncertainty is predominantly real or arises from foreignmonetary shocks. We show that given the relative variance of shocks observed in the data, themodel can explain the sign of the empirical link between prices and regimes.

Surprisingly, while we find some support for each of these three explanations they account fora small part of the strong negative relationship between regimes and national price levels. The twoexplanations that rely on transitory deviations from absolute PPP explain around 5 percentagepoints of the link between regimes and prices observed in the data. In particular, policy variablessuch as government expenditure and volatility of base money account for most of this reduction inthe link between regimes and prices. Moreover, we show that the predictions of the stochasticopen economy model receive only limited support in the data. As the volatility of real and foreignshocks rise, the national price levels in floats rise relative to pegs, opposite to the sign predicted bythe model. Furthermore, the model suggests no direct effect of regimes on national price levels

6 The results are also supported using World Development Indicators data. Both databases are based on data from theInternational Comparison Programme, but the way in which they aggregate and extrapolate the data differs.7 This pattern is also robust to the inclusion of the controls that are applied in the literature on the Balassa–Samuelson

hypothesis, and to different price definitions and aggregation methods.8 It was Cassel who first suggested that stabilization of inflation could lead to “overvaluation.” However, in Cassel's

view even small deviations from PPP would bring about large changes in trade flows that would correct any deviation.9 As argued in Obstfeld and Rogoff (2000), this insight has become a crucial result unveiled by the “new open economy

macroeconomics” framework. It has been extended by Devereux and Engel (1998, 2000) and Baccheta and van Wincoop(2000). Corsetti and Pesenti (2001b) were the first to formalize this idea in terms of CPIs. Recently a new literature ispursuing the same idea with models that cannot be solved in closed form, using second order approximations (seeBenigno and Benigno, 2003 and Kollman, 2003).10 Obstfeld and Rogoff (1998) apply the same approach to the level of nominal exchange rates. Baccheta and vanWincoop (2000) take a similar approach to assess the effect of exchange rate volatility on prices of traded goods.


once the interaction between shocks and regimes is considered, while there is a strong effect in thedata. The only prediction that finds some empirical support is related to the volatility of nominalshocks. As suggested by the model, when the volatility of nominal shocks rise, national pricelevels in floats rise relative to pegs. In sum, while the list of shocks considered in the empiricalanalysis is by no means exhaustive, these findings suggest that the model cannot adequatelyexplain the strong link uncovered in the data.

The link between national price levels and exchange rate regimes is important for severalreasons. First, the empirical pattern found in this paper suggests a fresh contribution to the long-standing empirical debate on exchange rate regimes. This debate has been dominated by thepredictions of the Mundell–Fleming–Dornbusch model which is silent about the determination ofnational price levels. Second, exploring the link between exchange rate regimes and national pricelevels can help explain the large difference that exists between national price levels of countrieswith similar income levels. For instance, the price level of Panama (a dollarized economy) in the1990s was 60 percent that of the US while the price level of neighbouring Colombia (a countrywith a flexible exchange rate) was less than 40 percent that of the US. This difference cannot beexplained by the conventional Harrod–Balassa–Samuelson effect or the Kravis–Lipsey–Bhagwati hypothesis because Panama and Colombia had similar levels of real income per capitain the 1990s.11 Third, since national price levels and real exchange rates are closely related (seefootnote 5), distinguishing between transitory and structural deviations form PPP has importantimplications for the measurement of real exchange rate misalignment.

The paper is organized as follows. Section 2 discusses the data used and descriptive statistics.Section 3 provides the benchmark regression analysis. Section 4 discusses the two explanationsbased on transitory deviation, develops a model of equilibrium national price levels and tests thedifferent hypotheses. Section 5 summarizes the main contributions of the paper.

2. Data and preliminary diagnostics

The price data used in this paper comes from the Penn World Tables (PWT) Mark 6.0 and theWorld Development Indicators (WDI).12 Both databases use the survey on prices provided by theInternational Comparison Programme (ICP, United Nations). In this survey, prices of differentgoods and services (standardized with respect to quality) are collected for a selected number ofcountries (beginning with 60 in 1975 and increasing to more than 115 in 1998) at five-yearintervals (the so-called benchmark years). Drawing on this sample of prices, both sourcesconstruct purchasing power parity (PPP) indexes for each country relative to the United States,defining a country's national price level as the PPP index of that country divided by it's foreignexchange rate. PWT calculates national price levels for country i as:

NPLPWTj ¼ 1

ej us

Pi pijqijPi piqij

ð1Þ

where i are all goods available in the ICP survey, pij are price parities relative to the U.S., πi is the“international price” for good i used in the PWT index to compare quantities across countries (inshort, a weighted average of relative prices of the same good for all the countries included), andej us is the exchange rate between the country j's currency and the US dollar. Data for more than

11 In US 1996 dollars, Panama's and Colombia's real GDP per capita were 5002 and 4940, respectively.12 An earlier version of this paper used data from PWT Mark 5.6.


110 countries during the period between 1980 and 1998 are used in the regression analysis below(see Table A-I in the Appendix). We restrict the analysis to non-oil countries.13

The PWTandWDI differ in the techniques they use to compute good weights across countries,and they rely on different methods to extrapolate the data for those countries and years notincluded in the benchmark surveys.14 Despite the explicit effort to make the two sets comparable,significant differences between the PWT and WDI data remain. Differences are especiallypronounced for low and medium income countries.15

I use three different exchange rate classifications in the empirical analysis that follows. Thestandard reference for exchange rate regime classification is the International Monetary Fund'sAnnual Report on Exchange Arrangements. The IMF's classification consists of ten categories(1 to 10), broadly grouped into pegs (1 to 4), arrangements with limited flexibility (5), and “moreflexible arrangements,” which include managed and pure floats (6 to 10). This classificationcaptures the formal commitment to a regime, but it fails to capture whether the actual policieswere consistent with this commitment. For example, de jure pegs can pursue policies inconsistentwith their stated regime and require frequent changes in the nominal exchange rate, making thedegree of commitment embedded in the peg in fact similar to a float. In the case of floats, fear offloating can induce a central bank to subordinate its monetary policy to eliminate fluctuations inthe exchange rate, rendering a de jure float equivalent to a de facto peg.

The problems that arise from a pure de jure classification have prompted researches to usedifferent criteria to classify regimes. Reinhart and Rogoff (2002, hereafter RR) classify exchangerate regimes using information about the existence of parallel markets combined with the actualexchange rate behavior in those markets. RR's classification consists of six categories (1 to 6),where 6 is the most flexible arrangement.16 Levy Yeyati and Sturzennegger (2003, hereafter LYS)analyze data on volatility of reserves and actual exchange rates. I use LYS's three-wayclassification that consists of peg, intermediate and flexible regimes (1 to 3). Such de factoclassifications, however, can fail to distinguish between stability that results from policycommitments and that which results from the absence of shocks. For these reasons, the threeclassifications are used in the empirical analysis. In all three classifications, a higher index isassociated with a higher degree of flexibility.

Table 1 provides a simple comparison of the three classifications during the benchmark years.The table shows that the IMF and LYS have roughly the same number of country–year pairsclassified under pegs as floats during this period. By contrast, RR has three times more country–year pairs classified as pegs than floats. Moreover, the table shows that a large share of IMF pegsare also classified as pegs by RR and LYS, while only a small share of IMF floats are classified as

13 The top twenty oil producers according to Oil and Gas Journal, World Oil, and EIA were excluded from our sample.The qualitative results of the paper are unaffected by the inclusion of the top oil producing countries. See Broda (2002)for these results.14 For a complete description of the ICP survey see the Handbook of the International Comparison Programme (2000).For detailed description of the method used by the World Bank see World Development Indicators (2000). See Summersand Heston et al. (2002) for the methodology used in the Penn World Tables.15 The correlation between the PWT and WDI price levels for low and middle income countries is 0.36 and 0.65,respectively. For high-income countries, the correlation is 0.92.16 More specifically, 1 includes no separate legal tender, pre-announced peg, narrow bands or currency boardarrangements, and de facto pegs. Category 2 includes de jure and de facto crawling pegs with narrow bands. Othercrawling pegs are included in 3, together with managed floats. 4, 5 and 6 includes respectively, freely floating, freelyfalling, and regimes with dual market in which parallel market data is missing.

Table 1Exchange rate regime classifications

Number of Countries classified as:

Peg Intermediate Float

IMF classification 181 93 180( – , 0.68, 0.74) ( – , 0.32, 0.37) ( – , 0.12, 0.14)

RR classification 281 93 91(0.40, – , 0.50) (0.26, – , 0.19) (0.79, – , 0.44)

LYS classification 129 71 104(0.58, 0.75, – ) (0.10, 0.12, – ) (0.84, 0.29, – )

Data for benchmark periods only (80, 85, 90, 95). In parentheses we present the share of IMF, RR, and LYS, respectivelythat have the same classification. For instance, 68 percent of the IMF pegs are also classified as pegs by RR. Definitions:IMF peg (1–4), intermediate (5), float (6–10); RR peg (1–2), intermediate (3), float (4–6); LYS peg (1), intermediate (2)and float (3).


floats in the other classifications. A high share of floats in RR are classified as floats in LYS, andvice versa.

Fig. 1 presents descriptive statistics of the raw data. Each bar-chart shows the mean of thenational price level for different income levels and exchange rate regimes. Each row shows theprice–regime relationships for a different classification, and each column presents comparisonsfor different benchmark years. For expositional purposes, a single dummy for fixed regimes isdefined for each classification.17 Most charts show that fixed regimes have higher price levelsthan flexible regimes. For instance, the bottom right chart shows that the average price level inmiddle-income countries with fixed regimes according to LYS classification is 59.9 percent of theU.S. level, while in more flexible regimes with the similar per-capita income the price level is49.3 percent of that of the U.S. price level. A test of the equality of means across regimes suggeststhat this difference is statistically significant at the 5 percent level.

3. Empirical method and main results

The basic cross-country analysis uses data from the benchmark years for the InternationalComparison Programme, for the period between 1980 and 1995, and 1998 (denoted by thesubscript t). Specifically, we estimate the following regression:

npljt ¼ dt þ byjt þ gERRjt þ KXjt þ ejt; ð2Þwhere nplj is the log of the national price level of country j relative to the U.S. (NPLUS=100), andERRj is country j's exchange rate regime. For comparison purposes, I normalize all threeclassifications to lie between 0, the most fixed regime classification, and 1, the most flexible one.In turn, yj is the log of the real per capita GDP of country j, and

Xj ¼ OPENþ

;OPEN� y−

;CAyþ

; SIZE−

;KAOPENþ=−

; FINDEVþ=−

;

� �

17 The more disaggregate classifications are used in the next section. High income countries are excluded from thecharts because the number of fixed regime dummies is very small. In the case of high income countries the regressionsexploit the thinner classification.

National Price Levels (WDI) in Low Income Countries by Exchange Rate Regime

National Price Levels (WDI) in Middle Income Countries by Exchange Rate Regime

199519901985

Float (10) Peg (20) Float (19) (**) Peg (18) Float (31) (*) Peg (15)

Float (16) Peg (19) Float (21) (*) Peg (18) Float (23) Peg (21)

LYS

RR

IMF

Float (5) Peg (8) Float (2)(*) Peg (6) Float (10) (**) Peg (15)

199519901985

Float (27)(**) Peg (17) Float (37) Peg (14) Float (52) (*) Peg (14)

Float (19) Peg (24) Float (31)(*) Peg (27) Float (27)(*) Peg (43)

LYS

RR

IMF

Float (8) Peg (10) Float(15)(*) Peg (13) Float (19) (**) Peg (14)

50

40

30

20

50

40

30

20

50

40

30

20

60

50

40

60

50

40

60

50

40

Fig. 1. National price levels (WDI) in low income countries by exchange rate regime. National price levels (WDI) inmiddle income countries by exchange rate regime. Number of Countries in parentheses. Stars next to Float correspond tothe level of Significance for a test of the equality of means across regimes. (⁎⁎⁎), (⁎⁎), and (⁎) stand for 1, 5, and 10 percentsignificance, respectively.



is a matrix of control variables suggested by previous theoretical work on national price levels(expected signs appear above the variables). A key control variable in the analysis is the real per-capita income level, yj.

18 Since Balassa (1964), Samuelson (1964), and Kravis and Lipsey (1983)(hereafter, BS–KL), a country's real per-capita income level has been recognized as an importantdeterminant of the national price level. Balassa (1964) argues that real per-capita income levelsare good proxies for productivity differentials between tradable and non-tradable goods. Kravisand Lipsey suggest that income levels are associated with factor endowments and factor rewards.Finally, Bergstrand (1991) suggests a demand-oriented explanation for why real per-capitaincome levels should determine national price levels. All these studies find that price levels tendto increase with the level of real per-capita income. This relationship also receives ample supportin the databases used in this paper.

Other potential determinants of the national price level have been suggested in the literature.Kravis and Lipsey (1987) argue that a high propensity to trade should pull a country's pricestowards the world average (OPEN×y) and should increase the prices of non-tradables for a givenincome level (OPEN).19 Krugman (1978) argues that in countries where spending is temporarilyhigher than income, real exchange rates should also be higher. Therefore, the current account as ashare of GDP (CAy) is included in Xj.

Moreover, since we are interested in capturing the effect that the exchange rate regime has onprices, determinants of exchange rate regimes that could potentially have a direct impact on nplsare also included. For instance, large countries are more prone to adopting flexible regimes, and asin Clague (1985), may also have lower price levels as they enjoy larger endowment of factors.Therefore, we include SIZE, measured separately as log GDP in dollars and land area, in thebenchmark regression to avoid attributing to the exchange rate regime an effect that comes from acountry's size. In addition to size, the degree of capital account openness (KAOPEN) and thelevel of financial development (FINDEV), proxied by broad money as a share of GDP, arecommon exchange rate regime determinants that might have a direct effect on prices.20 Inparticular, a capital account liberation can lead to disinflation as it raises the penalty for loosemonetary policy (see Bartolini and Drazen, 1997). By contrast, liberation can also lead to weakerdomestic financial markets which can eventually lead to inflationary pressures. The measure ofcapital account openness used is taken from Chinn and Ito (2005), which varies between −2 and 2.As will be clear in the next section, KAOPEN has a significant impact on prices, and fullycaptures the impact that FINDEV has on prices. We therefore drop FINDEV from the benchmarkregression as it has the additional benefit of increasing the number of observations by 10 percent.

19 They argue that trade tends to raise the prices of relatively abundant factors. If poor countries tend to have abundantlabor, and if non-tradables tend to be labor intensive, openness should result in higher prices once you control for theincome level.20 For an examination of the determination of exchange rate regimes see Barro and Tenreryo (2004) and Levy Yeyatiand Sturzennegger (2004).

18 The exact definitions and sources for the control variables are the following: OPEN is taken from the PWT, and isdefined as (Exports+ Imports) /CGDP. For SIZE, the log of real per-capita GDP from the PWT (WDI) is used in columnswith price data from PWT (WDI). CAy is the current account balance as a share of GDP, and FINDEV is broad money asa share of GDP. Both variables are constructed using IFS data. KAOPEN is the measure of capital account opennessdeveloped in Chinn and Ito (2005). This measure is based on the four binary dummy variables reported in the IMF'sAnnual Report on Exchange Arrangements and Exchange Restrictions (AREAER). The variables include information onthe existence of multiple exchange rates, restrictions on current account transactions, restrictions on capital accounttransactions, and the requirement to surrender of export proceeds. For details on the construction of this index, see Chinnand Ito (2005).


Table 2 shows the main results of the paper. It presents the results for Eq. (2) using twodifferent price databases and the three different exchange rate regime classifications (ERRIMF,ERRRR, and ERRLYS), all normalized to lie between 0 and 1. The exchange rate regimevariable η is negative (the more flexible the regime, the smaller the price) and highlysignificant in all specifications. The first three columns use data from the Penn World Tables.These regressions suggest that hard pegs have npls that are 21 percent higher than fullyfloats using the IMF classification, 17 percent using RR's classification, and 12 percent withLYS's classification. Furthermore, regressions (4) to (6) confirm a similar result using theWDI price data. On average, using WDI data, hard pegs have national price levels roughly20 percent larger than fully floats.

Most variables included in the control matrix are significant. Most notably, the BS–KLeffect (β) is positive and highly significant in all the regressions. The openness variables(OPEN, OPEN×y), although highly significant, have signs that are opposite to those foundby Kravis and Lipsey (1988). The table suggests that openness reduces a country's pricelevel, but the price level falls less with openness the higher the country's output. Highercurrent account deficits or larger land areas do not have a clear impact on prices. Larger

Table 2National price levels and exchange rate regimes

Dependant variable: log national price level

ERR classification IMF RR LY-S IMF RR LY-SPrice data PWT PWT PWT WDI WDI WDIYear dummies Yes Yes Yes Yes Yes Yes

(1) (2) (3) (4) (5) (6)

ER regime −0.211 −0.175 −0.126 −0.315 −0.224 −0.115− (4.81)⁎⁎ − (2.44)⁎ −(3.20)⁎⁎ − (7.70)⁎⁎ −(3.49)⁎⁎ − (2.51)⁎

Log per capita Y 0.247 0.263 0.230 0.300 0.336 0.312(10.04)⁎⁎ (9.84)⁎⁎ (7.14)⁎⁎ (8.90)⁎⁎ (9.92)⁎⁎ (7.63)⁎⁎

Open −0.002 −0.003 −0.003 −0.001 −0.001 −0.001− (4.18)⁎⁎ − (6.20)⁎⁎ −(4.72)⁎⁎ − (1.86) −(2.98)⁎⁎ − (2.35)⁎

Open⁎Y 0.004 0.005 0.005 0.000 0.000 0.000(3.98)⁎⁎ (4.87)⁎⁎ (3.84)⁎⁎ (4.36)⁎⁎ (4.70)⁎⁎ (3.61)⁎⁎

BCA (share of GDP) −0.210 −0.118 −0.423 −0.053 0.377 0.160− (1.14) − (0.47) −(1.46) − (1.55) (1.61) (0.60)

Land −0.002 −0.005 −0.021 −0.013 −0.020 −0.031− (0.14) − (0.41) −(1.53) − (1.23) −(1.81) − (1.94)

Log Y (in dollars) 0.032 0.004 0.045 0.061 0.031 0.052(2.23)⁎ (0.25) (2.37)⁎ (5.02)⁎⁎ (2.25)⁎ (2.84)⁎⁎

KA open 0.070 0.065 0.084 0.067 0.068 0.080(5.84)⁎⁎ (4.63)⁎⁎ (5.85)⁎⁎ (5.30)⁎⁎ (4.70)⁎⁎ (4.72)⁎⁎

Constant 1.164 1.709 1.181 0.096 0.470 0.328(4.70)⁎⁎ (6.48)⁎⁎ (3.96)⁎⁎ (0.35) (1.65) (0.89)

Observations 423 400 291 411 382 271R-squared 0.72 0.68 0.74 0.66 0.64 0.67

Acronyms can be found in the text. Robust t-statistics appear in parentheses (⁎significant at 5% level; ⁎⁎significant at 1%level). All regressions are estimated for benchmark years (1980, 1985, 1990, 1995) and 1998. The data from PWTcolumnare all from PWT but for CAY from the International Financial Statistics (IFS) database. The data for WDI columns arefromWDI except for CAY that comes from IFS. GC is expressed as a share of GDP. Openness equals imports plus exportexpressed as a share of GDP; Open⁎Y is openness times country's relative GDP to the U.S., which is normalized to 1BCAY is expressed as a share of GDP. KA OPEN comes from Chinn and Ito (2005).

s

s;


countries, as measured by log of dollar GDP, have higher npls. Moreover, countries that haveless capital account restrictions have significantly higher npls. The magnitude of this effect isimportant, as moving from a financially closed economy to a fully open capital account canincrease npls by around 20 percent.

Table 3 examines the relationship between regime and national price level in each of theincome groups. We modify Eq. (2) to allow for different δs, βs and ηs across income groups. Forlow and middle income countries, we find that the coefficient on the exchange rate variable, η issignificantly negative in all cases except for middle income countries with RR's classification.For low-income countries (and PWT data), the price level in countries with fully flexibleregimes is on average 23 percent lower than in those with fully fixed regimes (using IMF, RRand LYS the differences are 30, 30 and 10 percent respectively). For middle-income countries,the price level in fully floats is 17 percent lower than in hard pegs. The difference ranges from7 percent using the RR classification to 32 percent using the IMF classification. For high-incomecountries, the difference is, on average, negative but insignificant, except when using the IMFclassification.

Table 3Exchange rate regimes and income levels

Dependant variable: log national price level


(1) (2) (3) (4) (5) (6)

ERR⁎LI −0.305 −0.299 −0.099 −0.211 −0.355 −0.128−(4.59)⁎⁎ − (2.94)⁎⁎ − (2.10)⁎ − (3.20)⁎⁎ − (4.52)⁎⁎ − (2.54)⁎

ERR⁎MI −0.148 −0.103 −0.206 −0.324 −0.076 −0.147−(2.47)⁎ − (0.94) − (3.38)⁎⁎ − (6.07)⁎⁎ − (0.77) (2.39)⁎

ERR⁎HI −0.195 −0.098 −0.066 −0.224 −0.112 −0.038(2.91)⁎⁎ − (0.95) − (1.45) − (2.98)⁎⁎ − (1.04) − (0.70)

Log per capita Y⁎LI −0.054 0.183 −0.243 −0.053 −0.118 −0.282−(0.76) (3.56)⁎⁎ − (3.12)⁎⁎ − (0.92) − (1.95) − (4.00)⁎⁎

Log per capita Y⁎MI 0.352 0.202 0.355 0.613 0.618 0.51(8.09)⁎⁎ (4.75)⁎⁎ (6.57)⁎⁎ (7.87)⁎⁎ (7.22)⁎⁎ (3.60)⁎⁎

Log per capita Y⁎HI 0.585 0.237 0.44 0.294 0.3 0.281(7.52)⁎⁎ (5.86)⁎⁎ (3.65)⁎⁎ (6.65)⁎⁎ (6.63)⁎⁎ (5.48)⁎⁎

Open 0.000 −0.002 −0.001 −0.001 −0.001 −0.001(0.46) − (3.89)⁎⁎ − (1.53) − (0.85) − (0.17) − (0.38)

Open⁎Y −0.001 0.002 0.000 0.000 0.000 0.000−(0.60) (2.42)⁎ − (0.14) (0.94) (0.92) (0.75)

BCA (share of Y ) 0.042 0.014 −0.037 −0.083 0.101 −0.255−(0.16) (0.06) − (0.14) − (3.03)⁎⁎ − (0.47) − (1.07)

Land −0.017 −0.005 −0.031 0.003 −0.002 −0.018−(1.69) − (0.39) (2.32)⁎ (0.25) − (0.20) − (1.20)

Log of Y (in dollars) 0.027 0.002 0.047 0.032 0.012 0.03(2.12)⁎ (0.20) (3.16)⁎⁎ (2.85)⁎⁎ (1.08) (1.88)

KA open 0.039 0.028 0.051 0.044 0.033 0.051(3.42)⁎⁎ (1.90) (3.65)⁎⁎ (3.70)⁎⁎ (2.58)⁎ (3.75)⁎⁎


Same notes as in Table 1 apply. Robust t-statistics in parentheses (⁎significant at 5% level; ⁎⁎significant at 1% level). Alregressions include a low, middle and high income dummy.

l


To assess the importance of the exchange rate regime and income variables, the within-income group regressions were performed with and without the yj and ERRj variables (notreported). For middle income countries the regression without both variables (but includingyear dummies and the rest of the control variables) has an R-squared of 0.46. The inclusionof yj raises the R-squared to 0.50, while the inclusion of ERRj (using the IMF classification)increases it to 0.54. In the case of low (high) income countries the explanatory power risesfrom 0.26 (0.69) to 0.29 (0.81) with the inclusion of yj, and to 0.36 (0.82) by addingERRIMFj

.21

4. Testing alternative explanations

4.1. A expansionary policies, currency crises and exchange-rate based stabilizations

Froot and Rogoff (1995) and Bergstrand (1991) suggest the importance of “demand” factorssuch as the share of government spending in income (GC), in explaining temporarydeviations from absolute PPP. More generally, monetary and fiscal policy variables cancapture the tendency to run expansionary policies which lead to long-run inflation and realexchange rate appreciations in fixed regimes. This tendency may be especially pronounced indeveloping countries, where the empirical relationship between prices and exchange rateregimes is strongest. To test this hypothesis we include in the benchmark regression anumber of policy variables. To capture monetary policy indicators, I include the (lagged)consumer price inflation rate, Lπ, and the volatility of base money, as it is usuallypresupposed that countries with higher (and more volatile) inflation rates have lower nationalprice levels. Additional policy variables such as (lagged) domestic credit growth and fiscaldeficits were dropped since the existing variables, GC and Lπ, fully captured the impact ofpolicy on npls.

The second potential explanation for the strong relationship between national price levelsand exchange rate regimes also relies on transitory deviations from PPP. Dornbusch's (1976)seminal paper on the overshooting of the nominal exchange rate suggests that measuringnational price levels when a country is faced with a currency crisis may induce flexibleregimes to have a low national price level. The intuition for this result is that regimes maytemporarily have a low national price level once a country has undergone a majordevaluation but before prices have adjusted. In this interim period, a country would beclassified as having a floating regime and would have a low national price level.

The strong relationship can also respond to episodes of exchange-rate based stabilizations.A common feature of these episodes is the existence of inflation inertia in relatively fixedexchange rate regimes. This implies higher inflation rates relative to devaluation rates in theearly stages of the stabilization process which leads to high national price levels. This was firstnoted by Cassel (1928), recently emphasized by Kiguel and Liviatan (1992), and documentedby Calvo and Vegh (1999) and Goldfajn and Valdes (1999). Note that both explanations arebased on sticky prices and therefore respond to transitory considerations. As soon as domesticprices are fully flexible, both theories would suggest no difference in national price levelsacross exchange rate regimes.

21 A table with the results of the regressions with and without these variables using the different datasets and exchangerate classifications is available from the author upon request. In general, the explanatory power of the IMF's classificationis larger than that of LYS and smaller than RR's classification.


We identify episodes of currency crises and exchange-rate based stabilizations to investigatewhether the strong link between exchange rate regimes and national price levels in the previoussection responds to these factors. We use Frankel and Rose's (1996) list of currency crisisepisodes. They use annual data for more than 100 developing countries from 1971 to 1992, andclassify crises as episodes with a nominal depreciation of at least 25 percent that also exceed theprevious years' change in the exchange rate by at least 10 percent. This dummy was extended toinclude the European countries that suffered a sharp depreciation in 1992/3, and the countriesinvolved in the Asian crisis.22 I also constructed a dummy for the episodes that Calvo and Vegh(1999) classify as exchange-rate based stabilizations. The country–year pairs included in thesedummies appear in the Appendix Table A-I.

More generally, we want to identify whether the effect found in the data responds toshort-run effects that may arise from sudden regime shifts. For this purpose we construct adummy that takes the value of 1 if the country–year pair fails to satisfy the followingrestriction:

ERRjt−2 ¼ ERRjt−1 ¼ ERRjt ¼ ERRjtþ1 ¼ ERRjtþ2 ð3Þ

Thus, we identify whether the results from previous section are being driven by flexibleregimes that have recently become floats (e.g., crises not captured by Frankel and Rose'sdummy) or pegs that have recently adopted fixed regimes that are not associated with inflationstabilizations. Table A-I also shows the country–year pairs that satisfy this restriction between1980 and 1998.

Table 4 shows the same regressions as in Table 2 but including additional policy variables, andallowing for η to vary across different samples. In particular, I interact the exchange rateregime classification variable (ERR) with the Frankel and Rose dummy (FR), the Calvoand Vegh dummy (CV) and the changing–exchangerate–regime dummy (CER).Surprisingly, the strong relationship between regimes and prices is still present in mostcolumns when we control for policy and shortrun considerations. That is, the η coefficientfor all those country–year pairs that do not satisfy restriction (3), and those that are notincluded as currency crisis or exchange-rate stabilization episodes is still negative in allspecifications.23 In particular, after controlling for policy and short run considerations, thenational price level in hard peg is around 15 percent lower than in a fully float with thesame level of income.24 This implies that controlling for these two potential explanationsreduces the η coefficient by 5 percentage points on average. Indeed, both the governmentexpenditure and the monetary volatility have the expected signs and are highly significantin all specifications. These variables play a particularly important role in columns (2) and(5), where the η coefficient falls by around 10 percentage points. In terms of the short-run

22 Mexico is not in our sample because it is a major oil producer.23 In the working paper version of this paper we present robustness checks to reduce the effect of short-run factors evenfurther, restriction (3) is strengthened to include 10 years in which the regime is constrained to be constant. Results for thecountries that had constant exchange rate regimes for this longer period are still negative and significant.24 The estimated difference using the IMF, Reinhart–Rogoff and Levy Yeyati–Sturzenegger classifications is around 25,8.5 and 9 percent, respectively.

Table 4National price levels and exchange rate regimes: policy and short-run considerations

Dependent variable: log national price level


(2) (4) (6) (8) (10) (12)

ER regime −0.213 −0.073 −0.085 −0.295 −0.108 −0.098− (4.32)⁎⁎ − (0.78) − (2.00)⁎ − (5.72)⁎⁎ − (1.07) − (1.98)⁎

Log per capita Y 0.335 0.364 0.344 0.246 0.249 0.286(11.31)⁎⁎ (11.18)⁎⁎ (8.47)⁎⁎ (6.27)⁎⁎ (6.35)⁎⁎ (6.34)⁎⁎

Open −0.002 −0.003 −0.003 −0.001 −0.001 −0.001− (4.00)⁎⁎ − (5.77)⁎⁎ − (4.88)⁎⁎ − (1.49) − (2.28)⁎ − (1.98)⁎

Open⁎Y 0.003 0.004 0.004 0.001 0.001 0.001(3.45)⁎⁎ (4.31)⁎⁎ (3.61)⁎⁎ (2.96)⁎⁎ (2.93)⁎⁎ (2.47)⁎

BCA (share of Y ) −0.135 −0.159 −0.302 −0.053 0.365 0.097− (0.85) − (0.78) − (1.26) − (1.67) (1.64) (0.36)

Land 0.007 0.003 −0.001 −0.011 −0.022 −0.018(0.74) (0.27) − (0.05) − (1.00) − (1.94) − (1.01)

Log Y (in dollars) 0.026 −0.001 0.022 0.071 0.044 0.043(2.01)⁎ − (0.11) (1.30) (5.09)⁎⁎ (2.96)⁎⁎ (2.16)⁎

KA open 0.073 0.066 0.092 0.071 0.079 0.091(6.22)⁎⁎ (4.95)⁎⁎ (6.36)⁎⁎ (5.47)⁎⁎ (5.68)⁎⁎ (5.18)⁎⁎

Policy variablesGC (share of Y ) 0.076 0.089 0.093 0.008 0.014 0.005

(4.95)⁎⁎ (5.21)⁎⁎ (4.05)⁎⁎ (2.96)⁎⁎ (4.93)⁎⁎ − (1.33)CPI inflation (lagged) 0.000 0.000 0.000 0.000 0.000 0.000

(0.34) (0.21) (0.46) (1.45) (1.18) (0.79)CPI inflation volatility −0.000 −0.000 −0.000 −0.000 −0.000 −0.000

− (3.56)⁎⁎ − (3.60)⁎⁎ − (3.18)⁎⁎ − (2.43)⁎ − (2.08)⁎ − (2.07)⁎

Short-run considerationsERR⁎NCER 0.059 −0.045 −0.033 0.02 −0.05 −0.012

(1.24) − (0.46) − (0.71) (0.39) − (0.50) − (0.22)ERR⁎FR −0.208 −0.11 −0.778 −0.03 −0.063 −0.667

− (1.95) − (0.76) (10.87)⁎⁎ − (0.27) − (0.55) (5.56)⁎⁎

ERR⁎CV 0.198 0.197 0.44 0.056 −0.15 0.422(2.10)⁎ − (0.83) (9.28)⁎⁎ (0.55) − (1.07) (5.27)⁎⁎

Constant 0.372 0.803 0.416 0.214 0.727 0.549(1.25) (2.38)⁎ (1.09) (0.77) (2.64)⁎⁎ (1.42)


Same notes as in Table 1 apply. Robust t-statistics in parentheses (⁎significant at 5% level; ⁎⁎significant at 1% level).


considerations, the coefficients on ERR⁎CER are generally negative, suggesting that amongcountries that have changed regimes in a 5-year window around t the price differencesbetween pegs and floats are even larger. The coefficients on ERR⁎CV and ERR⁎FR aregenerally the correct sign, that is, countries that had exchange rate crises have lowernational price levels than the average float, and countries that had exchange-rate basedstabilizations had higher national price levels than the average peg. Despite their correct


sign, the interaction terms are only significant using the de-facto classification of LYS, andwith the PWT when using the IMF classification.25,26

4.2. Equilibrium national price levels

In light of the low explanatory power of the explanations based on transitory deviations fromPPPwe provide amodel of equilibrium national price levels across different exchange rate regimesthat can potentially explain the observed pattern. Themodel is based on a stochastic open economymodel with sticky prices as in Obstfeld and Rogoff (1998) and Corsetti and Pesenti (2001a).

4.2.1. The modelThe model is composed of households, firms and a government in each country, Home and

Foreign. There is a continuum of identical households with population normalized to one in eachcountry. Uncertainty arises from monetary and productivity shocks. A representative householdconsumes all types of goods, supplies labor, and holds money through a cash-in-advanceconstraint. It also owns a proportion of domestic firms and receives its profits. A representativehousehold maximizes expected utility

EUðc; lÞ ð4Þwhere l is leisure and c is a Cobb–Douglas composite of home traded goods (ch), foreign tradegoods (cf), and non-traded goods (cn), which are constant elasticity of substitution indexes:

c ¼ cann cahh c

aff

cj ¼Z 1

0cjðiÞðh−1Þ=h

� �h=ðh−1Þfor jaJ ¼ n; h; ff g

where θ is the elasticity of substitution within non-traded and within traded varieties, andΣj∈Jαj=1. Foreign agents have symmetric preferences (foreign variables appear starred).

Labor income is given by w(1− l), where w is the wage rate and 1 the time endowment. Firmsprofit earned by the household sector are denoted by π. For each realization of the shocks, thehouseholds' budget constraint is

XjaJ

Z 1

0pjðiÞcjðiÞdi ¼ wð1−lÞ þ puY ð5Þ

25 In the case of currency crises, this result can respond to the varying degrees of flexibility of the post-crisis regime.
Even though the post-crisis regime is usually assumed to be a float, for the events selected by Flood and Rose (1995) ascurrency crisis, only forty out of ninety-one of the events were classified as having flexible regimes one year after thecrisis (using IMF's classification). Therefore, the interaction term with FR can have an ambiguous effect on therelationship between prices and exchange rate regimes.26 In the case of exchange-rate based stabilizations, inflation inertia usually tends to increase the national price levelrelative to the initial post-stabilization level. However, the post-stabilization level also depends on the nominal exchangerate level at which the exchange rate is fixed. This implies that the short-run effect introduced by including these cases ispositive when the nominal exchange rate is initially set at a low value, and negative when it is set at a high value.Therefore the evidence presented here does not contradict the inflation inertia behavior or the evidence presented inGoldfajn and Valdes (1999) that suggests that real exchange rate appreciations are more common in fixed exchange rateregimes.


The first order conditions for consumption and leisure can be written as

ucwP¼ ul ð6Þ

cjðiÞ ¼ ajpjðiÞpj

� �−hYpj

for jaJ ð7Þ

where uc and ul are the marginal utilities of consumption and leisure, and Eq. (7) represent thedemand for type j good i. P is the overall price index, defined as

P ¼ dpann pahh p

aff ¼ d

YjaJ

Z 1

0pjðiÞ1−hdi

� �aj1−h

where δ=αnαnαh

αhαfαf.

Finally households need cash to purchase goods. Since the cash-in-advance constraints arebinding, the nominal value of output Y sold by Home firms is equal to the stock of domesticmoney supply,

Y ¼ M :

At the end of the period, the government imposes a tax of M.I now turn to the description of the behavior of firms. Each good is produced with 1 /a

units of labor, where a is a random productivity term. Profits of a Home firm i producinggood n and t are:

pn ¼ pnðiÞcnðiÞ−wacnðiÞ

ph ¼ phðiÞchðiÞ þ Sp⁎h c⁎h ðiÞ−

waðchðiÞ þ c⁎h ðiÞÞ

where Home demands are given by Eq. (7), and Foreign demands are defined similarly.Firms set the price of their product before observing the productivity and monetary shock.

After the uncertainty is revealed, they cannot change prices, and labor is determined by thedemand for goods at the preset prices. Firm i producing good j maximizes the value of profits, E(ucπj), subject to consumers' demands. Optimal prices are:

pnðiÞ ¼ phðiÞ ¼ /Euc wa M

EucMð8Þ

where / ¼ hh−1

. To briefly review known results, prices can be interpreted as a markup, ϕ, over thecertainty equivalent of costs divided by the certainty equivalent of sales (see Obstfeld and Rogoff,1995). Nominal costs (in marginal utility of consumption units) are proportional to w /a M whilesales are proportional to M.27 Note that Eq. (8) shows an explicit relationship between prices,

27 Corsetti and Pesenti (2001a) and Baccheta and van Wincoop (2000) present similar interpretations to their optimalpricing equations.


productivity shocks and monetary policy (or, more importantly for the purpose of this paper,exchange rate regimes). The intuition behind this relationship will be clear below. The optimalprices are analogous in the Foreign country.

Let S be the exchange rate and assume that the law of one price holds.28 Then,

p⁎h ðiÞ ¼1SphðiÞ: ð9Þ

For future reference we obtain the equilibrium level of employment. Using Eq. (9) and theresource constraint, cn+ch+ch*=a(1− l), we obtain:

ð1−lÞ ¼ a−1Mpn

: ð10Þ

Finally, the equilibrium exchange rate and exchange rate regimes have to be defined. Theexchange rate follows directly from the trade balance condition Sph⁎ch⁎=pf cf. After substitutingthe demand functions for domestic and foreign goods, we obtain

S ¼ MM⁎ ð11Þ

Since the model has 2 countries only, there is a single bilateral exchange rate, and there can onlybe one exchange rate regime in the world. By contrast, the empirical section examines manycountries with different exchange rate regimes. Therefore, to derive the predictions that can betested in the data, I compare two 2-country models, where in one the Home country has chosen tofix its exchange rate (relative to Foreign, the only other country in the world), while in the otherthe Home country allows its exchange rate to float.29 Without lost of generality assume that theForeign country's monetary supply is entirely driven by a random component that is independentof whether Home decides to peg or not to the Foreign currency, i.e.,M⁎=μ. Further assume that amonetary policy rule consistent with Eq. (11) defines an exchange rate regime. We denote thedegree of flexibility of an exchange rate regime by R∈ [0, 1], and the monetary policy rule thatcorresponds to each regime as:

M ¼ l1−RaReR: ð12ÞHome's monetary policy rule is comprised of three key variables, the foreign random component,μ, Home's productivity shocks a, and a domestic random component ε that is independent ofproductivity and the foreign monetary shock. Note that when R=0, Eq. (11) is constant, andHome has chosen to fix its exchange rate. In this case, Home has to fully accommodate foreignmoney supply shocks, cannot use monetary policy to respond to productivity shocks, and cannotgenerate monetary expansions that are independent of the foreign shock. As R increases, however,so does the degree of flexibility of the regime. When R=1, Home has a fully flexible monetary

28 In the calibration Section 1 comment on the effects of relaxing this assumption and allowing for the possibility of local-currency pricing (see Devereux and Engel, 1998 and Corsetti and Pesenti, 2001a for models with this characteristic).29 Alternatively, two countries with different regimes could be compared in a 4-country version of the model. In asymmetric world, where only 2 countries decide to peg their currencies together, the relationship between national pricelevels and regimes would be qualitatively identical to that from comparing two 2-country models. The 4-country versionwould also require further notation and space. Moreover, since I am interested in the positive implications of the 2-country model, I assume that the Foreign country just accepts the exchange rate regime that Home chooses.


policy that completely accommodates real shocks, that does not automatically react to foreignmoney supply shocks, but also has the capacity to generate domestic nominal disturbances. Thedomestic random component ε can be seen as a source of nominal instability that is usuallyassociated with flexible exchange rate regimes.

4.2.2. Equilibrium NPL under different exchange rate regimesThe national price level (NPL) of a country is defined as the ratio of the purchasing-power

parity of the country's currency to its foreign-exchange rate. In our model, the log of NPL (npl)can be expressed as

npl ¼ ppp−sppp ¼ anðlogpn−logp⁎n Þ þ ahðlogph−logp⁎h Þ þ af ðlogpf −logp⁎f Þ :

where ppp is a purchasing power index (in logs) analogous to the second term in Eq. (1).30 Giventhe assumption of the law of one price in the traded sector, we obtain that

npl ¼ anðlogpn−logp⁎n −sÞ: ð13ÞTo best understand the main intuition behind the relationship between prices and exchange rate

regimes I explicitly compute how npls vary with R. In common with most other stochastic openeconomy models (e.g., Obstfeld and Rogoff, 1995 and Corsetti and Pesenti, 2001a), only verysimple utility functions allow explicit solutions of the model. Preferences are assumed to be:

Uðc; lÞ ¼ logc−1tð1−lÞt; ð14Þ

where v≥1 is measures the curvature of the labor disutility.In this case, by combining Eq. (8) with Eq. (6) and using Eq. (10), we obtain the key pricing

equation of the “new stochastic open economy” literature:

pn ¼ /½Eða−1MÞt�t−1 : ð15ÞAs opposed to non-stochastic models, non-traded prices in this framework are a markup (ϕ) overexpected marginal cost. The impact of policy and shocks on expected marginal costs is acombination of two effects. The first effect is related to the covariance between policy and shocksand can be most easily described when t=1. In this case, expected marginal costs are a “weighted”average of productivity shocks with “weights” given by the strength of domestic demand, M. Inparticular, if policy and productivity shocks are positively correlated, then expected marginal costsare lower as firms expect low (high) productivity events to have a low (high) “weight” as they areassociated with lower (higher) domestic demand. Thus, when policy and shocks are uncorrelatedprices include a “price premium” relative to the positive-correlation case.31

The second effect is related to the case where υN1. From Eq. (15) we can see that the morevolatile is a− 1M (note that this implies more volatile equilibrium employment, see Eq. (10)) thehigher is the expected marginal cost that the firm faces. This is because firms have to compensate

30 Given the symmetry assumed in the model (i.e., αj=αj⁎), real exchange rates (RER) and national price levels areidentical. Formally, rer=αn log pn+αh log ph+αf log pf−αn⁎ log pn⁎−αh⁎ log ph⁎−αf⁎ log pf⁎− s=npl. However, if αj≠αj⁎,then npls differ from rer.31 This intuition is very similar to that found in Corsetti and Pesenti (2001a,b).


workers for the additional disutility that comes from the employment volatility that workers face.Intuitively, charging a higher price reduces the cost of uncertainty for workers as it reduces thelevel of expected labor effort.32 Higher expected marginal costs, in turn, lead to higher prices.Since this effect depends on the disutility from labor volatility we will refer to it as a “riskpremium” in prices.33

Using the monetary policy rule (12), we obtain a relationship between exchange rate regimesand prices:

pn ¼ /½EðaR−1l1−ReRÞt�t−1 : ð16ÞFor instance, if R=1 (fully flexible regime) monetary policy completely stabilizes ex-postfluctuations in marginal costs (and employment) that originate in productivity shocks and foreignmoney supply shocks. However, floats are subject to the marginal costs fluctuations that comefrom domestic money supply shocks that are independent of a and μ. As R falls, when theeconomy is hit by a productivity shock monetary policy does not fully accommodate thedisturbance and marginal costs are not fully stabilized. Moreover, with Rb1, domestic monetarypolicy has to react to foreign shocks generating additional volatility in marginal costs. Proposition1 follows immediately:

Proposition 1. If a, μ, and ε follow independent log normal distributions with zero mean andvariances given by σa

2, σμ2 and σε

2, then we obtain the following 3 predictions:

a) Exchange rate regimes impact npls solely through the uncertainty of shocks;b) The higher the variance of real and foreign shocks, σa

2 and σμ2, respectively, the lower are npls

in countries with flexible regimes relative to pegs;b) The higher the variance of domestic monetary shocks, σε

2, the higher are npls in countries withflexible regimes relative to pegs.

Proof of Proposition 1. Replacing Eqs. (16) and (11) in Eq. (13), and differentiating with respectto R we obtain the following expression:

dEðnplÞdR

¼ −antð1−RÞðr2a þ r2lÞ þ antRr2e :

The predictions follow directly from a)–c):

dEðnplÞdR

j r2a¼r2l¼r2e¼0 ¼ 0 aÞ

dEðnplÞ2dRdr2a

¼ dEðnplÞ2dRdr2l

¼ −antð1−RÞV0 bÞ

32 Formally, if vN1 then a higher price reduces the amount of money that workers are willing to pay to eliminateuncertainty (i.e., the certainty equivalent, CE). In general, for a stochastic z, then CE ¼ EðzpnÞt−ðEzpnÞt ¼1ptn½Ezt−ðEzÞt�N0 where the inequality follows from tN1 and the definition of convexity. As pn increases (i.e., average

employment falls), CE falls and so does the disutility that comes from employment volatility.33 There has been research related to temporary layoffs and compensating wage differentials in other fields. For instance,Abowd and Ashenfelter (1997) derive and test a model where the competitive wage includes a risk compensationproportional to the size of the employment variation in different industries.


dEðnplÞ2dRdr2e

¼ antRz0: cÞ

□

The intuition behind predictions b) and c) is the following. In the case of productivity shocksand υ=1, a passive monetary policy (i.e., low R as in pegs) implies marginal costs fluctuate withproductivity shocks, rising when productivity is low and falling when it is high (see Eq. (16)). Thehigher is R (more flexible regimes), the more policy and productivity shocks co-vary positively,which leads to lower expected marginal costs (as in Corsetti and Pesenti, 2001b). This implies thatmore flexible regimes have a lower “price premium.”Moreover, as policy dampens the impact ofproductivity shocks on marginal costs, more flexible regimes face lower volatility of employmentand with υN1 this implies a smaller “risk premium” on prices.34 As the uncertainty due to realshocks increases, the lower are national price levels in floats relative to pegs.

A similar effect is found in the case of foreign monetary shocks. In this case, pegs (low Rs)have to accommodate the domestic money supply to foreign money supply changes while floats(high Rs) can let the exchange rate absorb the changes in the foreign money supply withoutchanging the domestic money supply. This leads pegs to have a lower covariance betweenproductivity shocks and policy and higher volatility of marginal costs. Both channels tend todecrease national price levels in floats relative to pegs.

The impact of domestic nominal shocks on npls is similar to that of real or foreign shocks,since a higher σε

2 rises the volatility of marginal costs, and therefore rises the expected marginalcost. However, as opposed to real or foreign shocks, uncertainty related to domestic nominalshocks increases with the level of flexibility of the regime. In particular, the higher is σε

2, thehigher is the npl in floats relative to pegs.

Proposition 1 gives direct predictions on national price levels since in the model prices aresticky and firms set prices optimally. However, the same qualitative results are obtained in a setupwere wages, rather than prices, are preset (see Obstfeld and Rogoff, 2000). In the followingsubsection we examine whether these effects are also present on an international comparison ofwages.

Finally, given the relative variance of shocks observed in the data, the model can explain thesign of the empirical link between prices and regimes. However, even for large values of t (i.e.,the disutility from having employment volatility), it is hard to reconcile the magnitude of theempirical difference between regimes with that predicted by the model. Formally,

npljR¼0−npljR¼1 ¼1

2ant½ðr2a þ r2lÞjR¼0−r

2e jR¼1�:

Using the volatility of real GDP to proxy for σa2, the volatility of foreign monetary base to proxy

for σμ2 and the excess volatility in domestic base money over σa

2 +σμ2 to proxy for σε

2, we find thatfor the median country in the sample used in the empirical exercise, (σa

2 +σμ2)|R=0=0.75% and

σε2|R=1=0.25%.35 This suggests that for υ∈ [1, 6],36 the difference in npls between a hard peg and

a fully float range from 0.25 to 1.5 percent. Thus, given the relative volatility of shocks observed

34 For developing countries, there is empirical evidence that suggests that flexible exchange rate regimes have lessoutput variability. See Ghosh et al. (1997), Levy Yeyati and Sturzennegger (2003) and Broda (2001). For developedcountries, there is no such evidence.35 The median standard deviations of these variables in the sample used are: αa=7%, σμ=5% and σM=10%.36 Values for υ of 1 and 2 are standard in the real business cycle literature. I report results for higher values of υ to showthat the model has a hard time replicating the empirical differences even at implausibly high values υ.


in the data, the model suggests a higher npl in hard pegs than in floats, but the magnitude of thedifference between regimes is only a small fraction of the empirical difference that is notaccounted by sudden regime shifts or expansionary policies. We turn next to examining whetherthis effect, though small, is partly behind the empirical finding.

4.2.3. An empirical testIn light of the predictions derived in the previous section, I modify the benchmark regression to

include for interactive effects between the exchange rate regime and the volatility of shocks.Formally,

npljt ¼ dt þ byjt þ g0ERRjt þ g1ERRjt � r2a þ g2ERRjt � r2l þ g3ERRjt � r2eþ KXjt þ ejt; ð17Þ

where, according to predictions derived, the expected signs for the η coefficients are thefollowing: η0=0, η1=η2b0 and η3N0. I proxy the variance of domestic real shocks, σa

2,with the observed variance of per-capita GDP throughout the entire period between 1980and 1998. In the robustness section, I also present results using the variance of terms-of-trade and government expenditure shocks as proxy for the volatility of real shocks. σμ

2 isproxied using the volatility of base money in an anchor country that varies by country.37

Since it is hard to empirically disentangle what part of the domestic money supplyvariance is coming from each source, I use the difference between the volatility of moneyand the volatility of real and external shocks to proxy for σε

2. Results are qualitativelysimilar using different measures of monetary aggregates, but I only present results usingthe volatility of base-money, the variable that most closely resembles the theoreticalshock.38

Table 5 presents the results for the 3 different exchange rate regime variables used, and for bothPWTandWDI data. The main results of this table suggest that while the previous explanations areunchanged by the inclusion of additional variables, only one of the three predictions of the modelare supported in the data. In particular, η3 has the expected sign in most of the regressions, that is,the higher is the variance of nominal domestic shocks, the higher are the npls of floats relativepegs, and is significant in two of the regressions. However, predictions a) and b) are not borne inthe data. Among countries with higher real volatilities, floats have higher npls relative to pegs,while b) suggests that the opposite should be true. η1 is generally statistically significant at the10 percent level.

Moreover, as opposed to prediction a), the coefficient on the ERR, η0, is negative and highlysignificant. The magnitude of this coefficient needs some further interpretation. For instance, theinclusion of the volatility interactions in column 1 makes the coefficient jump from −0.19to −0.32 (column 2). Evaluated at the median of all variances, this coefficient implies thatthe difference between regimes is around 26 percent, almost 6 percentage points higherthan the original coefficient.39 A doubling of the real volatility reduces the difference

37 For countries in Asia and America, the anchor country used is the United States, for countries in Europe, I choseGermany to be the benchmark country, and finally France was the benchmark country used for the African countriesincluded in the sample.38 Data sources for all new variables except terms of trade used in this section are from IFS. For the terms-of-tradeseries, the source is WDI.39 This is computed as g0þ g1r2a þ g2r2l þ g3r2e .

Table 5National price levels and exchange rate regimes: model predictions

Dependent variable: log national price level

ERR classification IMF IMF RR RR LY-S LY-SPrice data PWT PWT PWT PWT PWT PWTYear dummies Yes Yes Yes Yes Yes Yes

(1) (2) (3) (4) (5) (6)

ER regime −0.190 −0.321 −0.013 −0.033 −0.095 −0.140− (3.44)⁎⁎⁎ − (3.65)⁎⁎⁎ − (0.12) − (0.19) − (1.89)⁎ − (1.86)⁎

Log per capita Y 0.327 0.332 0.334 0.338 0.318 0.310(10.63)⁎⁎⁎ (10.75)⁎⁎⁎ (9.86)⁎⁎⁎ (9.89)⁎⁎⁎ (7.65)⁎⁎⁎ (7.24)⁎⁎⁎

Open −0.002 −0.002 −0.003 −0.003 −0.003 −0.003− (3.45)⁎⁎⁎ − (3.82)⁎⁎⁎ − (5.22)⁎⁎⁎ − (5.06)⁎⁎⁎ − (4.20)⁎⁎⁎ − (3.59)⁎⁎⁎

Open⁎Y 0.003 0.003 0.004 0.004 0.004 0.004(2.78)⁎⁎⁎ (2.96)⁎⁎⁎ (4.21)⁎⁎⁎ (4.08)⁎⁎⁎ (2.99)⁎⁎⁎ (2.92)⁎⁎⁎

BCA (share of Y ) −0.107 −0.109 −0.105 −0.105 −0.183 −0.196− (0.55) − (0.56) − (0.49) − (0.48) − (0.72) − (0.78)

Land 0.007 0.006 0.007 0.008 −0.001 0.003(0.61) (0.49) (0.56) (0.62) − (0.09) (0.17)

Log Y (in dollars) 0.017 0.024 −0.016 −0.013 0.016 0.026(1.05) (1.41) − (0.95) − (0.78) (0.76) (1.18)

KA open 0.067 0.067 0.060 0.060 0.088 0.088(5.07)⁎⁎⁎ (5.08)⁎⁎⁎ (3.92)⁎⁎⁎ (3.99)⁎⁎⁎ (5.41)⁎⁎⁎ (5.32)⁎⁎⁎

Policy considerationsGC (share of Y ) 0.075 0.065 0.073 0.071 0.081 0.073

(4.54)⁎⁎⁎ (4.08)⁎⁎⁎ (3.90)⁎⁎⁎ (3.82)⁎⁎⁎ (3.29)⁎⁎⁎ (3.06)⁎⁎⁎

CPI inflation (lagged) 0.000 −0.000 0.000 0.000 0.000 0.000(0.43) − (0.16) (0.24) (0.50) (0.60) (0.11)

CPI inflation volatility −0.000 −0.000 −0.000 −0.000 −0.000 −0.000− (3.42)⁎⁎⁎ − (3.66)⁎⁎⁎ − (3.52)⁎⁎⁎ − (1.68) − (3.12)⁎⁎⁎ − (3.27)⁎⁎⁎

Short-run considerationsERR⁎NCER 0.026 0.005 −0.080 −0.077 −0.033 −0.029

(0.48) (0.09) − (0.76) − (0.75) − (0.61) − (0.55)ERR⁎FR −0.199 −0.148 −0.148 −0.114 −0.778 −0.720

− (1.40) − (1.07) − (1.02) − (0.75) − (9.56)⁎⁎⁎ − (6.37)⁎⁎⁎ERR⁎CV 0.244 0.208 0.298 0.313 0.502 0.471

(2.40)⁎⁎ (1.93) (1.20) (1.19) (8.67)⁎⁎⁎ (5.34)⁎⁎⁎

Model predictionsERR⁎VdY 0.109 0.003 0.002

(1.73) (1.55) (2.19)⁎⁎

ERR⁎Vu 0.004 −0.007 −0.004(0.93) − (0.63) − (0.70)

ERR⁎Ve 0.000 0.000 0.000(2.43)⁎⁎ (1.05) (1.89)⁎

Constant 0.640 0.509 1.328 1.209 0.788 0.581(1.85) (1.44) (3.30)⁎⁎⁎ (2.91)⁎⁎⁎ (1.86)⁎ (1.31)


Dependent variable: log national price levelRobust t-statistics in parentheses (* significant at 10% level; ** significant at 5% level; *** significant at the 10% level).


IMF IMF RR RR LY-S LY-SWDI WDI WDI WDI WDI WDIYes Yes Yes Yes Yes Yes(7) (8) (9) (10) (11) (12)

−0.249 −0.360 −0.112 −0.489 −0.114 −0.326− (4.28)⁎⁎⁎ −(3.62)⁎⁎⁎ − (0.97) − (2.63)⁎⁎⁎ − (1.96)⁎ − (2.98)⁎⁎⁎

0.256 0.258 0.251 0.269 0.282 0.275(6.28)⁎⁎⁎ (6.30)⁎⁎⁎ (5.92)⁎⁎⁎ (6.49)⁎⁎⁎ (5.85)⁎⁎⁎ (5.47)⁎⁎⁎

−0.001 −0.001 −0.001 −0.001 −0.001 −0.001− (2.13)⁎⁎ −(2.08)⁎⁎ − (1.98)⁎⁎ − (2.17)⁎⁎ − (2.00)⁎⁎ − (1.35)

0.000 0.000 0.000 0.000 0.000 0.000(2.93)⁎⁎⁎ (2.73)⁎⁎⁎ (2.60)⁎⁎⁎ (2.51)⁎⁎ (2.32)⁎⁎ (1.84)⁎

0.267 0.275 0.365 0.296 0.251 0.338(1.15) (1.24) (1.54) (1.28) (0.89) (1.10)−0.020 −0.022 −0.022 −0.022 −0.029 −0.020− (1.73) −(1.82) − (1.74) (−1.73) − (1.58) − (0.98)

0.055 0.061 0.041 0.049 0.038 0.053(3.25)⁎⁎⁎ (3.53)⁎⁎⁎ (2.24)⁎⁎ (2.68)⁎⁎⁎ (1.64) (2.03)⁎⁎

0.071 0.072 0.074 0.074 0.080 0.085(4.87)⁎⁎⁎ (4.81)⁎⁎⁎ (4.66)⁎⁎⁎ (4.73)⁎⁎⁎ (4.03)⁎⁎⁎ (3.91)⁎⁎⁎

Policy considerations0.009 0.008 0.012 0.012 0.003 0.005(3.20)⁎⁎⁎ (3.14)⁎⁎⁎ (4.14)⁎⁎⁎ (4.19)⁎⁎⁎ (0.92) (1.39)−0.000 −0.000 −0.000 −0.000 −0.000 −0.000− (1.40) −(0.94) − (1.10) − (0.14) − (0.59) − (1.23)−0.000 −0.000 −0.000 −0.000 −0.000 −0.000− (2.00)⁎⁎ −(0.89) − (1.85) − (0.72) − (1.63) − (2.06)⁎⁎

Short-run considerations0.004 −0.011 −0.054 −0.078 0.008 0.043(0.07) −(0.19) − (0.51) − (0.76) (0.13) (0.76)−0.002 0.022 0.043 0.168 −0.653 −0.685− (0.02) (0.20) (0.38) (1.44) − (4.90)⁎⁎⁎ − (4.91)⁎⁎⁎

0.029 0.070 −0.101 0.004 0.432 0.477(0.27) (0.60) (0.67) (0.02) (4.74)⁎⁎⁎ (5.05)⁎⁎⁎

Model predictions0.147 0.005 0.003(1.86)⁎ (2.76)⁎⁎⁎ (1.97)⁎⁎

0.004 0.009 0.003(0.84) (0.95) (0.61)−0.000 0.000 0.000−(1.03) (1.32) (0.91)

0.598 0.474 0.828 0.522 0.867 0.490(1.63) (1.27) (2.16)⁎⁎ (1.38) (1.86)⁎ (0.97)

335 335 326 326 225 2070.60 0.61 0.56 0.58 0.60 0.62



between regimes in about 5 percentage points to 21 percent.40 For the other 5 columns, theimplied difference between regimes evaluated at the median volatilities is 25 percent. Thisaverage is derived using coefficients that are not precisely estimated, and so should beinterpreted with caution.

4.2.4. Robustness and extensionsI have kept the model purposedly simple to focus on the sign of the relationship between

exchange rate regimes and national price levels. However, the benchmark model can beextended in several ways. First, other sources of shocks can be included. Broda (2002)examines the impact of including terms-of-trade shocks in the model presented in the previoussubsection. The terms-of-trade shock is modeled as changes in world preferences towards thehome traded goods (see Stockman and Tesar, 1995 and Ventura and Kraay, 2002). I find thatterms-of-trade shocks behave similarly to productivity shocks, and result in npls being higher inpegs than in floats.41 Similar results can be obtained by including government spending shocksas in the appendix to Obstfeld and Rogoff (1998). Table 6 includes the variance of these shocksas a robustness check in the benchmark regression. None of the interactive terms with thevariance of terms of trade or government expenditure shocks have a significant negativecoefficient.

The model in the previous subsection was derived in terms of sticky prices and optimal pricesetting behavior. Similar predictions can be obtained from a model where wages are sticky andagents set them in advance (as in Obstfeld and Rogoff, 2000). For this reason, in Table 6 (columns4–6) I examine whether dollar wages across countries are systematically linked to regimes(international wage data come from Freeman and Oostendorp, 2000). Indeed, despite the smallersample due to limited wage data, two out of three ERR variables have strong negativecoefficients. That is, pegs have higher wages in dollars relative to floats. Controlling for theinteraction effects, the difference in dollar wages across IMF and LYS regimes is around40 percent. Interestingly, in the RR case, several variables including FR and CV dummies, fiscaland inflation indicators and the interaction terms absorb most of the direct effect of the exchangerate regime.

The benchmark model can also be extended to include local-currency-pricing in thetradable sector. This can be particularly important for developed countries, as Frankel et al.(2004) show. By relaxing the assumption of the law of one price, the channels described inthe last subsection are unaffected, since they work through the price of non-traded goods.However, relaxing this assumption (as in Betts and Devereux, 2000) has implications fornpls. Under complete pricing to market and no non-traded goods P=P⁎, and therefore npl=−s. Thus different regimes have different equilibrium npls as long as the average exchangerate is affected. Obstfeld and Rogoff (1998) derive the risk premium in exchange ratesunder flexible regimes. Under certain circumstances, floats can be associated with a moreappreciated exchange rate than pegs, which under pricing to market implies a higher npl infloats. Broda (2002) compares npls for traded and non-traded goods separately and finds noevidence for this positive relationship between prices and exchange rate flexibility among

40 The median variance of real shocks is around 0.5 percent for the overall sample.41 More formally, I assume that the share of home goods in the traded good sector, αh, is stochastic and has a log normaldistribution; the monetary policy rule in Eq. (12) is extended to be M ¼ ð ah

EahÞ1−R.

Table 6National price levels and exchange rate regimes: terms-of-trade and wages

Dependant variable: Log national price level Log wage (in dollars)

ERR classification IMF RR LY-S IMF RR LY-SPrice data PWT PWT PWT – – –Year Dummies Yes Yes Yes Yes Yes Yes

(1) (2) (3) (5) (7) (9)

ER regime −0.298 −0.33 −0.207 −1.435 0.403 −1.314− (3.22)⁎⁎ − (1.68) −(1.94) − (3.55)⁎⁎ (0.50) − (2.75)⁎⁎

Log per capita Y 0.374 0.397 0.385 1.184 1.059 1.276(11.64)⁎⁎ (11.88)⁎⁎ (8.46)⁎⁎ (10.41)⁎⁎ (6.18)⁎⁎ (7.48)⁎⁎

Open −0.002 −0.003 −0.002 −0.004 −0.007 −0.004− (2.73)⁎⁎ − (4.51)⁎⁎ −(2.22)⁎ − (1.86) (3.43)⁎⁎ − (1.51)

Open⁎Y 0.002 0.003 0.001 0.004 0.009 0.005(1.47) (2.92)⁎⁎ (1.12) (1.41) (2.75)⁎⁎ (1.30)

BCA (share of Y ) −0.292 −0.363 −0.177 0.433 −1.082 0.512− (1.31) − (1.48) −(0.58) − (0.38) − (0.80) − (0.40)

Land −0.004 −0.003 −0.017 0.081 0.038 0.077− (0.34) − (0.27) −(1.11) (2.74)⁎⁎ (0.94) (1.94)

Log Y (in dollars) 0.012 −0.033 0.016 0.000 −0.097 −0.014− (0.63) − (1.90) −(0.73) (0.01) − (1.83) − (0.30)

KA open 0.079 0.079 0.085 0.065 0.098 0.037(5.86)⁎⁎ (5.85)⁎⁎ (4.83)⁎⁎ (1.32) (1.36) (0.64)

Policy considerationsGC (share of Y ) 0.067 0.063 0.083 0.319 0.291 0.356

(3.70)⁎⁎ (2.98)⁎⁎ (3.09)⁎⁎ (3.44)⁎⁎ (2.10)⁎ (2.89)⁎⁎

CPI inflation (lagged) 0.000 0.000 0.000 −0.001 −0.001 0.000(0.35) (0.75) (0.06) (1.30) − (2.02)⁎ (0.83)

CPI inflation volatility 0.000 0.000 0.000 −0.001 0.000 0.000(3.77)⁎⁎ (0.80) (2.92)⁎⁎ − (2.39)⁎ (1.69) (0.20)

Short-run considerationsERR⁎NCER 0.054 −0.002 −0.031 −0.298 −0.206 0.227

(0.95) − (0.02) −(0.53) − (1.63) − (0.51) (1.11)ERR⁎FR −0.085 0.052 0.93 0.798 −0.234

− (0.58) − (0.31) (2.48)⁎ (2.01)⁎ − (0.91)ERR⁎CV 0.271 0.632 0.382 −0.566 −1.137

(2.49)⁎ (2.29)⁎ (4.53)⁎⁎ − (1.23) − (2.99)⁎⁎

Model predictionsERR⁎VdY 0.337 1.217 0.741

(1.35) (1.70) (2.25)⁎

ERR⁎VdTT −0.37 −0.631 0.087− (1.30) − (1.24) (0.31)

ERR⁎VdG 0.298 5.131 2.805(0.18) (1.80) (1.34)

ERR⁎Vu 0.785 1.452 0.35 4.549 −4.903 3.016(1.65) (1.81) (0.62) (2.14)⁎ − (1.29) (1.15)

ERR⁎Ve 0.003 −0.003 0.001 0.013 −0.025 0.000(3.10)⁎⁎ − (1.89) (1.11) (2.61)⁎ − (3.47)⁎⁎ (0.08)

Constant 0.497 1.302 0.397 −4.652 −1.329 −5.293(1.24) (2.88)⁎⁎ (0.76) − (3.34)⁎⁎ − (0.62) − (2.82)⁎⁎


Robust t-statistics in parentheses (⁎significant at 5% level; ⁎⁎significant at 1% level). Terms of trade data is fromWDI, andinternational wage data from Freeman and Oostendorp, 2000.



traded −s.42 Moreover, that paper finds that the bulk of the differences in npls acrossregimes comes from non-traded goods.

The framework (14) can be enriched in several dimensions. First, a more general specificationfor consumption can be used. It can be shown that when 1

1−gc1−g is used, the magnitude of the

differences across exchange rate regimes slightly increase with γ. Second, under local currencypricing, the same utility as in Eq. (14), and monetary shocks, Baccheta and van Wincoop (2000)show that the price level is lower in floats than in pegs. The main difference between producer andlocal currency pricing comes from the direct link that exists between the equilibrium employmentand the exchange rate regime under local currency pricing. In particular, a higher correlation inmoney supplies (i.e., a more fixed regime) leads to a higher comovement between employmentand money supply under local currency pricing, and no change in this comovement underproducer currency pricing. This particular effect leads to lower relative national price levels infloats under local currency pricing than under producer currency pricing.

5. Final remarks

The main contribution of this paper is to uncover an empirical regularity between exchangerate regimes and national prices levels. The paper finds significant differences in national pricelevels across regimes. In developing countries with hard pegged regimes national price levels areroughly 20 percent higher than in those with fully flexible arrangements. The effect is presentusing both de facto and de jure exchange rate classifications.

The paper then examines the origins of the higher national price levels found in pegs. The stronglink between exchange rate regimes and price levels is only partly explained by transitoryconsiderations. The paper considers the tendency of some developing countries of runningexpansionary policies, and episodes associated with currency crises, exchange rate basedstabilizations and sudden switches in regimes. It finds that these considerations can account for5 percentage points of the observed difference in national price levels across regimes. Therefore,after accounting for policy and short-run considerations, the core of the empirical relationshipremains unexplained.

In light of the low explanatory power of the transitory departures from PPP the paper exploresthe possibility that the differences observed in the data respond to different equilibrium nationalprice levels across regimes. The paper builds on the new stochastic open economy modelspioneered by Obstfeld and Rogoff to derive equilibrium national price levels across differentexchange rate regimes. In theory, given the empirical variances of real and nominal shocks in thesample, the model can potentially account for the qualitative aspects of the data but fails to explainthe magnitude of the observed differences. In the data, however, the predictions of the model findonly limited support.

None of the explanations presented in the paper are fully satisfactory to explain the magnitudeof the empirical patterns observed. More work is needed to understand the reasons behind whycountries with fixed exchange rate regimes are associated with higher national price levels thanthose that have flexible regimes.

42 A possible interpretation of this result is that it is evidence against pricing to market and in favor of producer currencypricing. Another interpretation is that goods classified as traded are not purely traded and include a non-tradedcomponent.


Appendix A

Table A1Countries and Years used in Sections 3 and 4 (⁎) (⁎⁎⁎)

Country name
Code Year Country name Code Year Country name Code Year
Angola
AGO (1991) Georgia GEO (1996) Netherlands NLD(⁎⁎)
(1977–96)

Albania
ALB (1994–96) Ghana GHA (1980–82,1988–96)
Nepal
NPL (1977–80,1985–90)
Argentina
ARG (1984–86,1993–96);(1985–86x)
Guinea
GIN (1990,1991,1996)
NewZealand

NZL(⁎⁎)

(1981–82,1987–96)

Armenia
ARM (1995–96) Gambia, The GMB (1977–83,1988–96)
Pakistan
PAK(⁎⁎)
(1977–79,1984–96)

Australia
AUS(⁎⁎)
(1978–80,1985–96)

Guinea–Bissau

GNB(⁎⁎)

(1980,1985–94);(1991x)

Panama
PAN(⁎⁎)
(1977–96)

Austria
AUT (1977–91,1996)
Greece
GRC(⁎⁎)
(1978–96)
Peru PER (1987,1992–96)
Burundi
BDI (1977–80,1985–89,1994–96)
Guatemala
GTM (1977–86,1991–96);(1986x)
Philippines
PHL(⁎⁎)
(1977–79,1986–96)

Belgium
BEL(⁎⁎)
(1977–96)
Guyana GUY (1977–78,1993–96)
Papua NewGuinea

PNG
(1980–91,1996)
Benin
BEN(⁎⁎)
(1977–96);(1981x)

Honduras
HND (1977–87,1996)
Poland
POL (1988,1993–96)
Burkina Faso
BFA(⁎⁎)
(1986–96)
Croatia HRV (1996) Portugal PRT (1979–87,1994–96)
Bangladesh
BGD(⁎⁎)
(1981–96)
Haiti HTI (1977–88,1993–96)
Paraguay
PRY (1977–86,1991–95);(1984x)
Bahrain
BHR (1982–96) Hungary HUN (1984–92) Romania ROM (1994–95) Belarus BLR (1995–96) Ireland IRL
(⁎⁎)
(1981–96) Sudan SDN (1977–82,
1989);(1982x)

Belize
BLZ(⁎⁎)
(1984–96)
Iceland ISL (1977–85,1990–96)
Senegal
SEN(⁎⁎)
(1977–96)

Bolivia
BOL(⁎⁎)
(1985–96)
Israel ISR (1979–80,1993–96)
Singapore
SGP (1977–84,1989–96)
Brazil
BRA (1977–87,1996);(1983x–85x–87x)
Italy
ITA (1981–89) SolomonIslands
SLB
(1981–95)
Bhutan
BTN (1983–96);(1991x)
Jamaica
JAM (1992–96) Sierra Leone SLE (1992–96)
Botswana
BWA (1982–96);(1985x)
Jordan
JOR (1978–86,1991–95)
Sao Tomeand Principe

STP
(1996)
CentralAfricanRepublic

CAF(⁎⁎)

(1977–96);(1981x)

Japan
JPN(⁎⁎)
(1977–96)
Slovenia SVN (1995–96)
(continued on next page)

Table A1 (continued)

Country name Code Year Country name Code Year Country name Code Year


Switzerland
CHE (1994–96) Kenya KEN (1977–84,1989–90,1995)
Sweden
SWE (1979–89,1994–96)
Chile
CHL(⁎⁎)
(1984–94)
Cambodia KHM (1995–96) Swaziland SWZ (1977–96);(1984x)
Cote d'Ivoire
CIV(⁎⁎)
(1977–96);(1981x)

Kiribati
KIR (1988–96) Seychelles SYC(⁎⁎)
(1981–93)

Cameroon
CMR(⁎⁎)
(1977–96);(1981x)

St. Kitts andNevis

KNA(⁎⁎)

(1986–96)
Syrian ArabRepublic
SYR(⁎⁎)

(1977–96);(1988x)

Colombia
COL (1977–91,1996)
Korea, Rep.
KOR(⁎⁎)
(1977,1982–94)

Chad
TCD(⁎⁎)
(1977–96);(1981x)

Comoros
COM(⁎⁎)
(1984–96)
Lao PDR LAO (1991–92);(1980x, 1985x)
Togo
TGO(⁎⁎)
(1977–96);(1981x)

Cape Verde
CPV(⁎⁎)
(1988–95)
Lebanon LBN (1991–96);(1984x–1990x)
Thailand
THA(⁎⁎)
(1986–94)

Costa Rica
CRI (1977,1984–89)
Sri Lanka
LKA(⁎⁎)
(1979–96)
Tonga TON (1987–88,1993–96)
Cyprus
CYP(⁎⁎)
(1977–96)
Lesotho LSO(⁎⁎)
(1977–96);(1984x)

Trinidad andTobago

TTO
(1978–90,1995–96);(1986x)
CzechRepublic

CZE
(1992–94) Lithuania LTU (1996) Tunisia TUN (1977–83,1988–96)
Germany
D E U(⁎⁎)
(1993–96)
Moldova MDA (1995–96) Turkey T U R(⁎⁎)
(1977–96);(1978x,1984x,1988x)

Denmark
DNK(⁎⁎)
(1980–96)
Madagascar MDG (1977–79,1988–91, 1996)
Tanzania
TZA (1990,1995, 1996)
DominicanRepublic

DOM
(1977–82,1996)
Maldives
MDV (1993–96) Uganda UGA (1994–96)
Algeria
DZA (1977–91,1996);(1991x)
Macedonia,FYR

MKD
(1996) Uruguay URY (1984–89,1994–96)
Ecuador
ECU (1977–80,1985,1991);(1990x)
Mali
MLI(⁎⁎)
(1977–96);(1981x)

St. Vincentand the Gre.

VCT
(1981–96)
Spain
ESP (1978–85,1991–96)
Malta
MLT (1977–96) Vietnam VNM (1993–96)
Estonia
EST (1994–96) Mozambique MOZ (1994–96) Vanuatu VUT (1985,1990–96)
Ethiopia
ETH (1983–90,1995)
Mauritania
MRT (1977–84,1989)
Samoa
WSM (1984–85,1990–96)
Finland
FIN (1977–89) Malawi MWI (1977–81,1986–91)
Yemen,Rep.

YEM
(1992–93)
Fiji
FJI(⁎⁎)
(1977–96)
Malaysia MYS (1977–90,1995)
SouthAfrica

ZAF(⁎⁎)

(1981–96)

France
FRA(⁎⁎)
(1981–96)
Namibia NAM (1994–96);(1992x)
Zambia
ZMB (1978–80,1994–96);(1983x)

Table A1 (continued)

Country name Code Year Country name Code Year Country name Code Year


Gabon
GAB(⁎⁎)
(1977–96);(1981x)

Nicaragua
NIC (1977–87);(1985x)
Zimbabwe
ZWE (1982–91,1996)
Notes: (⁎) Country/Years pairs included if all 3 of the following conditions were met: (1) Country/Year had price data(either for the WDI or PWT database), (2) Country/Year had an IMF exchange rate regime classification and (3) theCountry had a constant exchange rate regime for Year(−2) to Year(+2). See Section 4 for a detailed explanation ofcondition 3. (⁎⁎) Countries included in Table 2, Column (5). (⁎⁎⁎) An “x” stands for country–year pairs that were excludedespite restriction (3) being met. This is because these country–year pairs appear in Frankel and Rose (1996) classificationof currency crisis or in Calvo and Vegh (1999) list of exchange-rate based stabilizations.

References

Abowd, J., Ashenfelter, O., 1997. Anticipated unemployment, temporary layoffs, and compensating wage differentials. In:Hallock, K. (Ed.), The Collected Essays of Orley Ashenfelter, vol. I, pp. 167–196.

Baccheta, P., vanWincoop, E., 2000. Does exchange rate stability increase trade and welfare? American Economic Review90, 1093–1109.

Balassa, B., 1964. The purchasing power parity doctrine: a reappraisal. Journal of Political Economy LXXII, 584–596(December).

Barro, R., Tenreryo, S., 2004. Economic effects of currency unions. NBER Working Paper 9435. revised.Bartolini, L., Drazen, A., 1997. Capital account liberalization as a signal. American Economic Review 87, 377–400

(March).Benigno, P., Benigno, G., 2003. Price stability in open economies. Review of Economic Studies 70 (4), 743–765 (no 245).Bergstrand, J., 1991. Structural determinants of real exchange rates and national price levels: some empirical evidence.

American Economic Review 81 (1), 325–334 (March).Betts, C., Devereux, M., 2000. Exchange rate dynamics in a model of pricing to market. Journal of International

Economics L (1), 215–244 (February).Bhagwati, Jagdish, 1984. Why services are cheaper in poor countries. Economic Journal XCIV, 279–286.Broda, C., 2001. Coping with terms-of-trade shocks: pegs versus floats. American Economic Review XCI (2), 376–380

(May).Broda, C., 2002. Uncertainty, exchange rate regimes and national price levels. Federal Reserve Bank of New York Sta.

Report 151. July.Calvo, G., Vegh, C., 1999. Inflation stabilization and BOP crisis in developing countries. In: Taylor, J., Woodford, M.

(Eds.), Handbook of Macroeconomics, vol. Ic. North-Holland Publishing, Amsterdam, pp. 1531–1614.Cassel, G., 1928. Post-war monetary stabilization. Columbia University Press.Chinn, M., Ito, H. “What Matters for Financial Development?,” mimeo University of Wisconsin, January 2005.Clague, C., 1985. A model of real national price levels. Southern Economic Journal 51 (4), 998–1017 (April).Clague, C., 1986. Determinants of the national price level: some empirical results. Review of Economics and Statistics

LXIII (2), 320–323.Corsetti, G., Pesenti, P., 2001a. Welfare and macroeconomic interdependence. Quarterly Journal of Economics CXVI (2),

421–445 (May).Corsetti, G., Pesenti, P., 2001b. International dimensions of optimal monetary policy. NBER Working Paper No. 8230.

April.Crucini, M., Shintani, M., 2004. Persistence in law-of-one-price deviations: evidence from micro-data. Vanderbilt

University Working Paper No. 02-W22R. July.Devereux, M., Engel, C., 1998. Fixed vs. floating exchange rates: how price setting affects the optimal choice of exchange-

rate regime. NBER Working Paper No. 6867. December.Devereux, M., Engel, C., 2000. Monetary policy in the open economy revisited: price setting and exchange rate flexibility.

NBER Working Paper No. 7665. April.Dornbusch, R., 1976. Expectations and exchange rate dynamics. Journal of Political Economy 84 (6), 1161–1176

(December).Dornbusch, R., 1988. Purchasing power parity. The New Palgrave: A Dictionary of Economics. Stockton Press, NewYork.


Flood, R., Rose, A., 1995. Fixing exchange rates: a virtual quest for fundamentals. The Journal of Monetary EconomicsXXXVI, 3–37.

Frankel, J., Rose, A., 1996. Currency crashes in emerging markets: an empirical treatment. Journal of InternationalEconomics 41 (3–4), 351–366 (November).

Frankel, J., Parsley, D. and Wei, S. “Slow Passthrough Around the World: A New Import for Developing Countries?mimeo Harvard University (2004).

Freeman, R., Oostendorp, R., 2000. Occupational wages around the world. NBER Working Paper 8058. December.Froot, K.A., Rogoff, K., 1995. Perspectives on PPP and long-run real exchange rates. In: Grossman, G.M., Rogoff, K.

(Eds.), Handbook of International Economics, vol. III. New Holland Publishing, Amsterdam.Ghosh, A., Gulde, Ostry, Wolf, 1997. Does the nominal exchange rate regime matter? NBERWorking Paper No. 5874. January.Goldberg, P., Verboven, F., 2005. Market integration and convergence to the law of one price: evidence from the

automobile industry. Journal of International Economics 65 (1), 49–73.Goldfajn, I., Valdes, R., 1999. The aftermath of appreciations. Quarterly Journal of Economics CXIV (1), 229–262

(February).Harrod, R., 1939. International Economics. University of Chicago 1957.Heston, Summers, 1991. The Penn World table (Mark 5): an expanded set of international comparisons, 1950–1988.

Quarterly Journal of Economics CVI (2), 327–368 (May).Imbs, J., Mumtaz, H., Ravn, M., Rey, H., 2005. PPP strikes back: aggregation and the real exchange rate. Quarterly Journal

of Economics 120 (1) (February).Keynes, J.M. (1923), A Tract on Monetary Reform. Macmillian and St.Martin’s Press for the Royal Economic SOciety,

1971.Kiguel, M., Liviatan, N., 1991. The inflation–stabilization cycles in Argentina and Brazil. In: Bruno, K., et al. (Ed.),

Lessons of Economic Stabilization and its Aftermath. The MIT Press, pp. 191–231.Kiguel, M., Liviatan, N., 1992. The business cycle associated with exchange-rate stabilizations. World Bank Economic

Review 6 (2), 279–305.Kollman, R., 2003. Monetary Policy Rules in an Interdependent World. CEPR DP 4012.Kravis, Irving B., Lipsey, Robert E., 1988. National price levels and the prices of tradables and nontradables. Papers and

Proceedings of the One-Hundredth Annual Meeting of the American Economic Association. American EconomicReview, vol. 78 (2), pp. 474–478 (May).

Kravis, Irving B., Lipsey, Robert E., 1983. Toward an explanation of national price levels. Princeton Studies inInternational Finance, vol. 52. Princeton University Press, p. 21. November.

Kravis, Irving B., Lipsey, Robert E., 1987. The assessment of national price levels. In: Arndt, Sven W., Richardson, DavidJ. (Eds.), Real Financial Linkages Among Open Economies. MIT Press, Cambridge.

Krugman, P., 1978. Purchasing power and exchange rates. Journal of International Economics 8 (3), 397–407 (August).Levy Yeyati, E., Sturzennegger, F., 2004. Optimal Exchange Rate Regimes. Available at http:// www.utdt.edu/~ely/papers.

html.Levy Yeyati, E., Sturzennegger, F., 2003. “To Float or to Fix: Evidence on the Impact of Exchange Rate Regimes on

Growth,”. American Economic Review 93 (4), 1173–1193.Lopez, C., Murray, C., Papell, D., 2005. State of the art unit root tests and purchasing power parity. Journal of Money,

Credit and Banking 361–369 (April).Niehans, J., 1990. A History of Economic Theory. The John Hopkins University Press, Baltimore and London.Obstfeld, M., Rogoff, K., 1995. Exchange rate dynamics redux. Journal of Political Economy 102, 624–660 (June).Obstfeld, M., Rogoff, K., 1998. Risk and exchange rates. NBER Working Paper No. w6694. August.Obstfeld, M., Rogoff, K., 2000. New directions for stochastic open economy models. Journal of International Economies

vol.L, 117–153 (February).Reinhart, C., Rogoff, K., 2002. The modern history of exchange rate regimes. NBER Working Paper No. w8963.

May.Rogers, J., “Price Level Convergence, Relative Prices and Inflation in Europe,” mimeo Federal Reserve Board, 2001.Rogoff, K., 1996. Purchasing power parity puzzle. Journal of Economic Literature 34 (2) (June).Samuelson, P., 1964. Theoretical notes on trade problems. Review of Economics and Statistics XLVI, 145–164.Stockman, A., Tesar, L., 1995. Tastes and technology in a two-country model of the business cycle: explaining

international comovements. American Economic Review LXXXV, 168–185 (March).Penn World Tables, Mark 5.6 available at http://pwt.econ.upenn.edu/.Ventura, Jaume, Kraay, Aart, 2002. Trade integration and risk sharing. European Economic Review 46, 1023–1048 (June).Viner, J., 1933. Studies in the Theory of International Trade. Allen and Unwin 1955.

http://20www.utdt.edu/~ely/papers.html

http://20www.utdt.edu/~ely/papers.html

http://pwt.econ.upenn.edu/

Further reading

Baxter, M., Stockman, A., 1989. Business cycles and the exchange rate regime. Journal of Monetary Economics XXIII,377–400.

Benigno, P., 2002. A simple approach to international monetary policy. Journal of International Economics LVII,177–196.

Heston, A., Summers, R., Aten, B., 2002. Penn World Table Version 6.1, Center for International Comparisons at theUniversity of Pennsylvania (CICUP). October.

International Monetary Fund Publication, 2001. Annual Report on Exchange Arrangements and Exchange Restrictions(AREAER).

International Monetary Fund Publication: Dornbusch's Overshooting Model After Twenty-Five Years, Kenneth Rogoffspeech (2002).

Lee, J. and M. Chinn, “Current Account and Real Exchange Rate Dynamics in the G-7 Countries” IMFWorking Paper No.02/130.

Mendoza, E., 1991. Real business cycles in a small open economy. American Economic Review 81 (4), 797–818(September).

Mussa, M., 1986. Nominal exchange rate regimes and the behavior of the real exchange rate. In: Brunner, Karl, Meltzer,A. (Eds.), Real Business Cycles, Real Exchange Rates and Actual Policies. North Holland Publishing, Amsterdam,pp. 117–213.

Ricardo, D., 1951–1955. In: Saffra, P. (Ed.), The Works and Correspondences of David Ricardo. Cambridge UniversityPress, Cambridge.

Williamson, J., 1994. Estimating Equilibrium Exchange Rates. Institute of International Economics.


Documents

Exchange rate regimes and national price levels