Exchange Rate Pass-Through in Developing and Emerging Markets: A Survey of Conceptual, Methodological and Policy Issues, and Selected Empirical Findings

1. The Changing Focus of the Exchange Rate Pass-Through Literature

This article provides an overview of the literature on exchange rate pass-through (ERPT) to prices (the effect of exchange rate changes on domestic prices). The focus is the burgeoning empirical research on ERPT for emerging market and developing economies, and on the global as well as local monetary policy relevance of ERPT. For instance, the strategic pricing practices of emerging market exportersFootnote¹ breaking into new markets has been implicated in lowering global inflation (Figure 1).

By the late 1990s, there was a sizeable empirical literature on ERPT, mainly for the industrialised countries (Goldberg & Knetter, Citation1997; Menon, Citation1995).Footnote² Early research had a micro-economic focus from the industrial organisation literature, analysing the pricing strategies of monopolistic firms with industry-level and product-level trade price data, and ERPT to aggregate trade prices. The focus of the literature has since changed: volatility in exchange rates and persistent trade imbalances globally, have galvanised interest in the role of ERPT in monetary policy. The ‘New Open Economy Macroeconomics’ (for example, Obstfeld, Citation2002) provided a theoretical catalyst for empirical macroeconomic research on the ERPT, and the Taylor (Citation2000) staggered pricing model added a further impetus. These empirical models have investigated the delayed and incomplete ERPT, and the apparent decline in ERPT in some countries since the 1980s, for aggregate import prices and domestic consumer price indices. Taylor argues that the ERPT is itself influenced by the monetary policy regime (see Section 2).

At the research frontier, coming full circle from aggregate indices back to disaggregated data, is a new focus on price adjustment and heterogeneous ERPT, using the millions of prices that underlie the aggregate consumer and producer price indices. This approach has centred on industrialised countries, given the paucity of data sets (Klenow & Malin, Citation2011); the first emerging market study for ERPT is published in this special section (Aron, Creamer, Muellbauer, & Rankin, Citation2014).

Faced with a plethora of reported ERPT measures in the empirical literature, the discerning policy-maker needs to ascertain by which method these were calculated, which controls and restrictions were applied, and the time frame and stability of the estimates. Reported ERPT estimates from different methodologies are not directly comparable, as the underlying assumptions differ. A key contribution of this article is to categorise and compare the heterogeneous methodologies used to extract ERPT measures. Single equation models and systems methods are contrasted; and frequent misspecifications that produce unreliable ERPT estimates in empirical applications are highlighted.

The original definition of ERPT referred to the percentage change in import prices in domestic currency in response to a 1 per cent change in the exchange rateFootnote³ (now called Stage 1 ERPT). The definition has since been extended to address the effect of exchange rate movements on producer or consumer prices (Overall ERPT). The effect of a change in import prices on producer or consumer prices is known as Stage 2 ERPT. The time dimension is important: a distinction needs to be made between very long-run or ‘equilibrium’ measures of ERPT and measures over a shorter period of one or two years, considered most relevant for monetary policy.

Typically the sensitivity to the exchange rate will decline down the price distribution chain, from import prices ‘at the dock’ in the destination country, through wholesale and retail networks to final consumer prices. However, even for ‘at the dock’ import prices, ERPT may be incomplete if exchange rate changes elicit a less than equi-proportionate change in prices. It is helpful to distinguish between the pricing strategies considered in the theoretical models of the ‘new open-economy macroeconomics’. In producer currency pricing (PCP), prices are set and are sticky in the exporter’s currency.Footnote⁴ Then ERPT to import prices is complete: variations in the exchange rate leave the exporters’ prices in their own currency unchanged, while destination market prices vary closely with the exchange rate. However, considerable empirical evidence suggests the opposite: ERPT is incomplete, exporters’ prices vary closely with the exchange rate and the local (destination) prices remain fairly stable. This is known as local currency pricing (LCP). Goldberg and Hellerstein (Citation2008) suggest it is the convention in the literature that ‘incomplete pass-through’ refers to a single-destination market and pricing-to-market (PTM) to multiple-destination markets with market segmentation. Both generate deviations from the law of one price.

With complete or near complete ERPT to import prices and perhaps producer prices, a flexible exchange rate adjusts the trade balance through expenditure switching (for example, the ‘flight from quality’) towards locally produced substitutes, when imported goods become more expensive with destination currency depreciation. Provided that the ERPT is then incomplete to the retail or final goods prices, the gains in competitiveness will not be entirely inflated away (Obstfeld, Citation2002). Unsurprisingly, therefore, firms’ pricing behaviour and the ultimate ERPT to different prices are crucial in the policy debate on optimal monetary and exchange rate regimes (Section 2).

What can explain the pervasive findings of delayed and incomplete ERPT? Three possible channels are suggested by theory for generating incomplete ERPT, and much micro-empirical literature has been devoted to adjudicating amongst them (see Goldberg & Hellerstein, Citation2008). The destination country’s import price at the point of entry is defined as the exporter’s price divided by the (bilateral) exchange rate (Section 3). The exporter’s price is determined by the exporter’s marginal cost in own currency multiplied by a mark-up over the marginal cost. Thus, incomplete ERPT – where the exporter’s price varies closely with the variation in the exchange rate – could be due to a change in the mark-up (which we call Channel 1) and/or a change in the marginal cost (Channel 2). There is also a third factor: delayed (and hence incomplete in the short run) ERPT could be due to nominal rigidities that cause unresponsiveness in prices in the short run (Channel 3); for example, firms paying a ‘menu cost’ will adjust prices less frequently.Footnote⁵ In the very short run there is little change in the destination currency import price.

A common misconception noted by Goldberg and Hellerstein (Citation2008) is that incomplete ERPT reflects the degree of competition in the destination market. In perfectly competitive markets, where goods are homogeneous and the price equals marginal cost (there is no mark-up), the ERPT will be complete, or else firms will suffer a loss and have to exit the market. With imperfect competition, firms can charge a mark-up on costs, but there will be an effect on ERPT only if mark-ups are not constant but variable.Footnote⁶ Since, from the first order condition for the firm’s profit maximisation problem the firm’s mark-up is determined by the price elasticity of demand, the functional form of the demand curveFootnote⁷ matters for Channel 1 of ERPT.Footnote⁸ Also relevant for this channel (and related to the shape of demand curves) are two other factors that influence the firm’s mark-up: the ease of substitutability between similar domestic and foreign goods; and the degree of market segmentation. The lower is the substitutability (that is, the greater the product differentiation) and the greater the market segmentation (so arbitrage is limited even for the same good), the greater will be the market power of price-setting firms. These general points for the distribution of imports apply as much to an imperfectly competitive retail sector, where distributors can absorb part of the exchange rate fluctuations to maintain stable prices or expand market share, leading to what Hellerstein and Goldberg call a ‘double marginalisation’. Menon (Citation1995) further explores the implications for ERPT from the industrial organisation literature of different market structures and behavioural assumptions of firms, for example using the Cournot oligopoly model of Dornbusch (1987).Footnote⁹

Mark-up variation, however, cannot fully explain the extent of exporters’ price adjustment (see the evidence from micro-studies in Section 7). The exchange rate must therefore also impact marginal costs (Channel 2). Examples are where there are decreasing returns to scale in production of the exporter’s good (for example, due to short-run capacity constraints);Footnote¹⁰ imported inputs into the exporter’s good; and/or local non-traded costs in the destination market (costs of distribution and wholesale/retail such as transport, taxes, tariffs, storage, marketing, advertising, finance, insurance, real wages and rents). Goldberg and Hellerstein (Citation2008) observe that local non-traded costs generate different (incomplete) ERPT in different destination markets (that is, PTM), whereas the former two examples imply an incomplete ERPT that is the same over different markets. Local costs are unresponsive to exchange rate fluctuations: where (non-traded) local services comprise a high proportion of marginal costs, even large changes in imported costs may not affect local prices much. In practice, Channel 2 appears to be the biggest contributor to generating incomplete ERPT through imported inputs and/or local costs (Section 7).

Turning to Channel 3, price rigidity and other dynamic factors could contribute to incomplete ERPT, at least in the short run. The price stickiness found in micro-studies is insufficient to account for the slow adjustment of aggregate prices (Rogoff, 1996). Klenow and Malin (Citation2011) argue that a ‘contract multiplier’ is required to explain why real effects of nominal shocks appear to last several years, and potential multipliers include strategic complementarities, countercyclical mark-ups, sticky plans, and sticky information (including rational inattention, for example Reis [Citation2006]). The direct link between inflation and exchange rate changes is then obscured.

Other factors also blur the linkage between inflation and exchange rate changes. If consumers switch from imported goods to lower-quality, cheaper local brands, the ERPT will register as incomplete; these quality adjustments are often not properly measured in the consumer price index (Burstein, Neves, & Rebelo, Citation2003). The stage of the business cycle may influence exporter’s pricing decisions; for example, if there are production constraints in a boom, this could raise ERPT. A volatile macro-economic environment also affects ERPT, whether induced by external commodity price shocks, waning confidence in macro-management, or greater global risk aversion. These factors could all induce foreign exchange rate volatility, which in emerging markets with poor hedging opportunities is typically associated with a rise in ERPT as exporters price in their own more stable currency (though there are other possibilities, see Froot & Klemperer [Citation1989]). Finally, the central bank’s reaction function should be modelled in a system with the ERPT model, as we argue below, since the exchange rate is endogenous to policy with sometimes substantial feedback effects from monetary policy. For example, an inflation-targeting central bank would raise interest rates to head off inflation induced by an exchange rate shock, so correcting part of the initial exchange rate depreciation.

The standard models assume ERPT is linear (in logs of prices and the exchange rate), but recent research has investigated asymmetries in price responses to exchange rate fluctuations (for example, Bussière, Citation2007; Marazzi & Sheets, Citation2007; Pollard & Coughlin, Citation2003). There are potentially directional asymmetries, where depreciation may elicit a different proportionate price response than appreciation, and/or size asymmetries, where smaller changes may produce a different proportionate response to larger changes. The former may arise from strategic considerationsFootnote¹¹ and downward price rigidities.Footnote¹² Size asymmetries may result from ‘menu costs’ associated with changing prices: firms absorb small exchange rate changes in their margins and only pass through to prices those changes exceeding a size threshold (Channel 3).

Finally, much literature investigates a different non-linearity: apparent secular declines in ERPT with structural/regime changes.Footnote¹³ Possible micro-explanations include the metamorphosis of the trading basket from homogeneous raw materials (with rapid price adjustment) towards differentiated manufactured goods and services (Campa & Goldberg, Citation2005); or the changing geographical composition of trading partners (for example, Chinese imports lowering the United States ERPT, Marazzi and Sheets [Citation2007]). The key macro-explanation is the adoption of transparent and stable monetary regimes, a la Taylor (Citation2000) (see Section 2.2; for example, Gagnon and Ihrig [2004], for the United States consumer price index).Footnote¹⁴

Until comparatively recently, few analyses have tested the above hypotheses for trade prices and domestic prices in emerging marketFootnote¹⁵ and developing countries. These small, open and trade-dependent economies embody special features that can make it difficult to obtain reliable estimates of ERPT. Where the exchange rate has been actively targeted, for example in Asia, systems methods with feedback effects may be more reliable than single equation models. With extensive structural transformations, for example in South Africa (SA) and Central and Eastern European countries in the 1990s, shifts in ERPT should be tested for. The greater uncertainty facing economic agents also raises the relevance of non-linearities from threshold effects (menu, hedging and other transactions costs). Finally, hyperinflation and macro-volatility make for challenging empirical analysis.

The survey first summarises links between monetary policy and ERPT. Section 3 gives the theoretical underpinning of the partial equilibrium ERPT model that underlies most empirical studies. Section 4 critically compares different methodologies used to extract ERPT measures, and Section 5 raises important data issues. Section 6 examines and tabulates selected empirical evidence for emerging market countries using aggregate price data; and Section 7 reviews micro-based research. The survey concludes in Section 8.

2. Monetary Policy and Exchange Rate Pass-Through

Pinning down the ERPT mechanisms at different points in the price distribution chain, and hence deriving the ultimate implications for consumer price inflation and economic activity, is key to effective monetary policy-making. The degree and delay of ERPT also influences the effectiveness of trade balance adjustment through expenditure switching, and hence choice of the exchange rate regime. Two different theoretical contributions have spawned a host of empirical testing that links ERPT and monetary policy. In Section 2.1, the sometimes controversial expenditure-switching role of the exchange rate in the new open macroeconomic (NOE) models is explored with different pricing assumptions for firms. Increasing trade integration, shifts in market structure and in the weights of consumer price index (CPI) components might all be reasons for a decline in EPRT. In Section 2.2, the Taylor hypothesis is presented with its rationale for the decline in ERPT under credible monetary policy regimes.

3. The Theoretical Underpinnings of Empirical Pass-Through Analyses

We illustrate the partial equilibrium micro-founded mark-up equation that is the basis for the empirical analysis of ERPT to import prices using both single equations and some systems methods (which introduce some general equilibrium features, for example, the Johansen method, see below), at both aggregate and disaggregated price levels. We extend the exposition of Campa and Goldberg (Citation2005) for ERPT to import prices, to include a role for the domestic costs of the importing country and an expanded definition of the exporter country’s costs with commodity prices.

The import price index of a country measured at the point of entry can be defined as the price index of the exporter to that country, , converted to domestic currency using the exchange rate (expressed in foreign currency per unit of domestic currency of the destination country, so that a rise in the exchange rate is an appreciation):

(1)

Expressing the prices in logs, denoted by lower case letters, gives:

(2)

The exporter’s prices are expressed as a mark-up () over the exporter’s marginal costs ():

(3)

Using lowercase letters to reflect logarithms, and substituting Equation (3) into Equation (2), yields:

(4)

As noted in the Introduction, the mark-up, which is inversely related to the price elasticity of demand in the destination market, depends on the shape of the demand curve. The mark-up is unlikely to be a constantFootnote¹⁸ when the price and quantity vary, and it will also vary by sector or firm with the degree of product differentiation and market structure.Footnote¹⁹ The mark-up is a function of the real exchange rate and potentially of other macro-variables. Campa and Goldberg (Citation2005) formulate the mark-up in terms of the nominal exchange rate, but the real exchange rate is more relevant: it is more difficult to get a large mark-up if domestic prices in the destination market are low relative to the foreign prices expressed in domestic currency. The relationship between the mark-up and the (time-varying) real exchange rate, defined as the nominal rate adjusted by the price ratio of unit labour costs in the exporting country () and destination country (), and expressed in logs, can be simply approximated by:

(5)

The value of lies between 0 and 1: at φ = 0 there is PCP; and at φ = 1 there would be complete LCP (that is, the mark-up varies one-for-one with the exchange rate, if marginal costs were constant). In between, there is a degree of LCP, lending support to the theoretical emphasis on this type of pricing, even before distribution costs are added. The constant is given by μ.Footnote²⁰ For simplicity, Equation (5) omits other influences, such as demand conditions in the importing country. The exporter’s marginal costs are postulated to rise with a weighted average of market wages in the exporting country, , and commodity prices such as oil prices, , and with demand conditionsFootnote²¹ in the exporting country, , and in its destination market, . Expressed in logs:

(6)

Combining the above equations, import prices at the point of entry, before local taxes and distribution costs are added, can be specified as:

(7)

Equation (7) generalises the formulation of Campa and Goldberg by introducing the importing country’s relative domestic costs into the mark-up function, and exogenous commodity prices into the exporter’s marginal cost function. Note that this describes a long-run relationship and is not about temporary price stickiness.

The (long-run) ERPT coefficient is , capturing the exchange rate elasticity of import prices. If , so that , then there is producer currency pricing (PCP). The import price changes one-for-one with the exchange rate; that is, there is full ERPT. In this case, the mark-up is a constant, for with , it does not depend on the exchange rate (Equation 5). At the other extreme, if , so that , then there is no ERPT to prices. This is (complete) local currency pricing (LCP), where the exporters fully absorb any exchange rate fluctuations by squeezing their mark-ups, giving importers a stable price.

Equation (7) can be expressed as a long-run log linear regression specification, and this is conventionally used in the empirical literature to estimate ERPT to import prices (using several different methodologies, discussed further below):

(8)

where is the local currency import price, is a constant, is the (nominal) exchange rate,Footnote²² and are control variables representing exporter costs and domestic costs, captures a further element of exporter’s costs stemming specifically from commodity prices, such as oil prices, and and control for demand in the destination market and exporter’s market.

There are several plausible theoretical restrictions on Equation (8). Long-run lack of money illusion, when the exchange rate does not change, requires the price homogeneity restriction . This ensures that doubling foreign costs and destination country domestic costs, for instance, eventually doubles import prices. When the exchange rate can vary, long-run homogeneity in Equation (8) requires that foreign costs be translated at the exchange rate and hence that . Then doubling the exchange rate at given foreign prices, for instance, is equivalent to doubling foreign prices at a given exchange rate. Combining these two restrictions gives the long-run ERPT as . In the short run, such homogeneity restrictions might not hold; but in the long run they should hold. Analogous to the discussion below Equation (7), at the one extreme, if there is zero ERPT to prices (with complete LCP) then in the long run, . At the other extreme, with complete ERPT to prices under PCP, then in the long run, and .

Simpler versions of Equation (8) are generally used in the literature, omitting controls for domestic costs and commodity prices, and sometimes also omitting demand controls. Excluding control variables that are correlated with exchange rates could result in biased estimates of the ERPT coefficient, . Omitting domestic costs is potentially serious. Even at the dock, from the mark-up in Equation (5), domestic costs matter since they enter the real exchange rate. And further down the price distribution chain, the empirical literature suggests local distribution costs are important for incomplete ERPT (Section 7). Yet some authors impose PCP pricing in the long run (that is, they assume purchasing power parity (PPP) holds in the aggregate, see Table 1), without testing the restriction that unit labour costs are not part of the long-run relationship above; that is, they assume . This could neglect an important part of the transmission mechanism to prices from exchange rates. This issue is also important in differentiating the results from single equation and systems methodologies, discussed in Section 4.

Equation (8) captures Stage 1 ERPT. The Stage 2 ERPT, of import prices to consumer prices, can be thought of a weighted combination of import price ERPT to producer prices, and producer price ERPT to consumer prices. Substituting the determinants of import prices into an equation for producer prices, and then substituting the determinants of producer prices into an equation for consumer prices, would allow the elimination of both import prices and producer prices. The resulting reduced-form equation for Overall ERPT of exchange rates to consumer prices resembles Equation (8), though the coefficients will be different.

Specifically, for import prices measured further down the distribution chain – as inputs into the production of goods or retailed directly to the consumer – and for domestic wholesale and retail prices, local currency cost content would have to be added in Equation (4). For example, where D represents such prices down the distribution chain, the foreign and domestic costs enter as a weighted average, where is the weight on foreign marginal costs:

(9)

and is a function of destination country demand, , and domestic unit labour, , and other costs. As a result, the new Equation (8) for such prices down the distribution chain would have an increased coefficient (because of the increased weight on domestic costs) and a reduced coefficient (because of the lower weight on foreign costs).

Finally, most papers assume log linearity of price adjustment and do not test it against alternatives. One rationale behind non-linearities is transactions costs. These encompass not only linear proportional transactions costs, like ‘shipping costs’, but also fixed costs, like ‘menu costs’. Non-linearity is introduced because price adjustment may only begin once a particular threshold is reached; for example, for a threshold based on the size of exchange rate shocks, small shocks are absorbed into profit margins and ERPT occurs only for shocks exceeding the threshold. The role of transactions costs is closely linked to the concerns of the NOE models described in Section 2. The originators of these models call for a richer framework in theory and empirics, with proper articulation of costs (fixed and proportional), imperfect competition and wage–price rigidities, and a distinction between retail and wholesale pricing (for example, Obstfeld, Citation2002). A different type of non-linearity is presented by ‘structural breaks’ caused by regime changes. The ERPT parameter, , in Equation (8), might then not be constant but vary over the sample. Examples include a break in the monetary policy regime. The next section explains how non-linearities are tested for in log linear single equation models.

4. How is Pass-Through Estimated in Practice and Why Does it Matter?

This section contrasts the heterogeneous methodologies used to extract ERPT measures in the empirical literature. The resulting ERPT measures are not directly comparable due to differences in the underlying assumptions (though they often are compared, regardless). If, for any method, the models are misspecified, for example by omitting relevant determinants, the results will be biased. The time frames often differ for reported measures; they may encompass multiple trade, exchange rate or monetary policy regimes where the ERPT is not stable. The discerning policy-maker needs to know which method was used, how it was applied, whether the measure pertains to the long run or short run, and whether ERPT changed over the sample, inter alia.

4.1. Single Equation Methods where the Exchange Rate is Assumed Exogenous

The statistical time series properties of the underlying price and exchange rate series affect the validity of the different methodologies. A stylised fact for most floating exchange rates is that they are not only non-stationary, but are difficult to distinguish from a random walk over the sample.Footnote²³ Prices, too, are typically non-stationary. In some early studies of ERPT, statistical inference was invalid, as linear regressions using non-stationary series can produce spurious correlations (Granger, Citation1981).

If Equation (8) is estimated as a single equation using non-stationary data to capture the long-run equilibrium ERPT, , this relationship is valid only if the variables can be shown statistically to form a ‘co-integrated relationship’. This means that even though the series are individually non-stationary, a linear combination of them, with weights captured by the regression coefficients, is stationary (I(0)). If the series in Equation (8) are co-integrated, they can be thought of as having one or more equilibrium or long-run economic relationships.

The majority of studies have addressed the non-stationarity issue by differencing the data to form stationary series.Footnote²⁴ Then a first-differenced version of Equation (8) can be estimated with several lags on the different variables to allow a gradual adjustment to the exchange rate. Using differenced equations means the investigator does not wish to assume the existence of long-run relationships (or that they are absent when tested).

The seminal example of the differenced single equation approach is the multi-country import price study by Campa and Goldberg (Citation2005), who report that co-integration was not found when tested for. Their specification for a quarterly model, and expressed here in aggregate price index terms,Footnote²⁵ is:

(10)

where is a constant, is the real GDP of the destination (importing) country and the foreign costs are measured as a trade-weighted average of foreign unit labour costs.Footnote²⁶ The short-run ERPT (after one period) is given by , and they define as ‘long-run’ ERPT.Footnote²⁷ With quarterly data, this measure would give ‘ERPT after five quarters’, since it reflects the cumulative effect of five quarterly shocks.Footnote²⁸ This type of model is widely applied in the literature.

Such single equation methods are based on reduced-form regressions from a partial equilibrium model; they assume that exchange rates are exogenous, which implies that changes in the log of the exchange rate, , are exogenous shocks. They differ from the systems methods, discussed below, which allow feedback effects from endogenous exchange rates. If the other variables in Equation (10) were also exogenous, then would be a discrete approximation of the ‘impulse response function’ (that characterises the time path of the dependent variable in an equation in response to a shock from an explanatory variable) cumulated up to n periods after an exchange rate shock (see online Appendix 1 for discussion).

However, floating exchange rates and prices are determined simultaneously in a general equilibrium setting. Neglecting the channels through which the exchange rate is influenced by other economic variables may lead to biased and inconsistent estimates. For instance, if policy-makers raise the interest rate in anticipation of future inflation following exchange rate depreciation, this counters the original depreciation by appreciating the currency and the net shock is smaller. The assumption of exogenous exchange rates produces ERPT measures for a ‘gross’ exchange rate shock, without correcting for such negative feedback from monetary policy.

Single equation models expressed in differences omit possible long-run relationships. If co-integration between long-run level variables is indeed present, this is a misspecification and will bias the ERPT. Failure to find co-integration is attributed by de Bandt, Banerjee, and Kozluk (Citation2008) to taking insufficient account of structural change (assuming there has been no omission of relevant non-stationary variables). In the usual tests for co-integration, the weights in the co-integrating vector are assumed constant in the sample. If the long-run economic relationship (and hence the weights) between the variables alters through, for example, technological progress, regime change or institutional change, this might mistakenly find against co-integration.

It is desirable to combine the long-run co-integrating relationship between prices and their determinants in Equation (8) with short-run dynamic adjustment to deviations from equilibrium induced by shocks of various types. This can be accomplished in a single equation ‘equilibrium correction’ model (ecm), a typical formulation of which is as follows:

(11)

where the is the long-run cointegrating term with the foreign prices converted to the destination country’s currency. The speed of adjustment to equilibrium is given by in Equation (11).Footnote²⁹

A few studies (see Table 1) introduce a simple version of the ecm term which is the log of the real exchange rate: (. Generally is proxied by the trade-weighted average of foreign prices (usually the CPI, and then is proxied by the destination country’s CPI) and is the trade-weighted exchange rate. This formulation effectively imposes long-run homogeneity, by assuming without testing that and _; see Section 3. Long-run structural change, such as from productivity change (the Balassa–Samuelson effect), is addressed by creating the real exchange rate ‘gap’, subtracting a long-run trend from the real exchange rate. Use of the Hodrick Prescott filter technique to create this time-varying trend can strongly bias upwards the coefficient on the real exchange rate gap.Footnote³⁰

This equation includes the lagged import price in both the co-integrating term, and also in (one or more) lagged changes. This can make the calculation of the time-profile of ERPT complexFootnote³¹ and perhaps accounts for the rarity of the ecm approach in practice. An approximation by Aron, Farrell, Muellbauer, and Sinclair (Citation2014) is to include an ecm term and the lagged dependent variable only at lags pre-dating the period over which ERPT is calculated. The advantage is that ERPT estimates can be straightforwardly calculated from a summation of the coefficients, and the bias from neglecting long-run information and lagged import price changes is reduced (though not eliminated) by including ‘older’ information. This equation can be estimated using ordinary least squares.

A distinct advantage of the single equation methods is that asymmetries and other non-linearities, of growing importance in explaining incomplete ERPT, can be straightforwardly tested using split trends and interaction effects, or with explicitly non-linear versions. There is scarce but growing evidence of non-linearities at the first stage of the ERPT, to import prices, and overall, to consumer prices. The majority of papers follow Pollard and Coughlin (Citation2003) in testing for short- to medium-run asymmetries by distinguishing large from small exchange rate changes, appreciations from depreciations, or a combination of the above. Two dummy variables could be defined to identify months or quarters in which the absolute value of the exchange rate change exceeded a certain threshold; for example, 3 per cent (or a grid of threshold values), for large changes, , and small changes, . Interacting these dummy variables with provides separate estimates for ERPT under large and small exchange rate changes. In a similar manner, the effect of exchange rate volatility on ERPT could be measured by interacting a volatility measure with .

We have mentioned that the ERPT relationship may not be stable with regime changes. There are various ways of testing the stability of coefficients in single equation models. Several techniques are used. Some simply test for the presence of structural breaksFootnote³² in the ERPT regressions (for example, Bussière & Peltonen, Citation2008). A direct method analyses the effect of independent variables on time-varying estimated ERPT elasticities as the dependent variable in regressions (for example, Brun-Aguerre, Ana-Maria, & Phylaktis, Citation2012). The before-and-after approach splits samples at a chosen breakpoint, and runs separate regressions on the shorter samples, comparing the elasticities in each (Coulibaly & Kempf, Citation2010). A related approach, with the advantage that the sample is not curtailed, introduces an interaction term into the linear model. The coefficient of the interaction term (for example, between and a dummy that equals 1 during the inflation targeting era [for example] and 0 otherwise) reveals whether there is a change in pass-through in the regime period (Akofio-Sowah, Citation2009; Goldfajn & Werlang, Citation2000). Finally, explicitly non-linear estimation methods that allow varying parameters over time, such as the Kalman Filter model or Markov Switching models, could potentially be used.

5. Data Pitfalls for Emerging Market Countries

For emerging market countries, the available time series data on exchange rates, prices and especially domestic costs and demand proxy variables may be limited. This constrains the possible methodologies that can be applied; for example, co-integration techniques require long time series, even with high frequency sampling of data. Sample size becomes critical when investigating asymmetries (for example, in order to have a sufficient number of appreciation and depreciation episodes). Long samples, on the other hand, may span different trade, exchange rate or monetary regimes, and the ERPT relationship could be altered by structural breaks, so that variation over time of the ERPT elasticities needs to be tested for. Very short time series are typical of the micro-data on prices (see Section 7).

Even for industrialised countries, import price data may be less than adequate.Footnote³³ In the absence of comparable trade price data, there is continued wide-spread use of import (and export) unit values (that is, value divided by quantity for each item). These price proxies suffer from well-known deficiencies (Menon, Citation1995). Unit values alter with changes in the price, but also with the quantities of items shipped (even for identical imports that are ‘bundled’ in a different way). Measurement errors can be reduced by disaggregating to homogeneous commodity classifications; but the measurement bias remains for aggregated data. Import prices are also distorted by the prevalence of non-tariff barriers, even where tariffs have largely been reduced and made homogeneous (see Menon, Citation1995, p. 227). Domestic prices such as the CPI may be poorly constructed with inadequate adjustments for the introduction of new goods or those where prices and quality change rapidly (for example, mobile phones and computers; see Boskin Report, US [1996]), a neglected issue in CPI construction in emerging market countries.

Menon (Citation1995) advocates the use of a currency-contract weighted exchange rate to give a true representation of the currency fluctuation faced by the exporter. Most studies proxy this with (trade-weighted) effective exchange rate indices. The choice of exchange rate index affects estimated ERPT: Pollard and Coughlin (Citation2006) found sharp variation across United States manufacturing industries with eight constructed exchange rate indices (varying indices by countries and weighting arrangements). In this survey, all but one study (Table 1) uses nominal effective exchange rate indices (with the additional advantage of allowing some exchange rate variation in fixed bilateral regimes). Using a bilateral exchange rate arbitrarily assigns a weight of zero to all other trading partners, introducing biases.

Proxies are needed to control for the marginal cost variables which cannot be directly measured, and the cost indices typically used may be good proxies for average cost but not for marginal cost.Footnote³⁴ Yet Bussière and Peltonen (Citation2008) point to strong biases when excluding these controls. The bulk of earlier studies (and Brun-Aguerre et al. [2012] in Table 1) unfortunately use a ‘world price’ variable constructed as trade-weighted export prices (or export unit values) as a proxy for exporters’ costs (Menon, Citation1995). If unit value indices, these are subject to compositional changes. Perhaps more seriously, exporters’ pricing decisions are already partly incorporated in these country-wide export prices: unless there is no pricing-to-market or the same pricing-to-market to all markets, there will be distortions (see Menon, Citation1996). Recent studies, with easier recourse to real effective exchange rate data,Footnote³⁵ typically use trade-weighted consumer prices. The drawback of these is the preponderance of non-traded goods and services in the CPI.

Moreover, while considered the most comparable index across countries and hence often used in published REER measures, a little discussed issue is that countries differ in whether the CPI contains housing costs, which makes a considerable difference.Footnote³⁶ These problems can be limited using the trade-weighted foreign PPI: it measures costs at an earlier stage than the export price, often includes highly traded commodities, and does not depend on the export market targeted. The trade-weighted unit labour cost (used by Campa and Goldberg (Citation2005)), from an even earlier stage of the production process and excluding oil costs, is arguably less relevant than the PPI.

Sometimes commodity prices such as oil pricesFootnote³⁷ or an index of commodity prices are added to capture other foreign costs and supply side effects, helping to disentangle these from exogenous exchange rate effects. If the weighted PPI includes oil, this may make the oil price redundant, but this remains an empirical question. Some VAR studies include only an oil price, ignoring domestic sources of inflation as well as other foreign sources.

The vast majority of studies neglect the destination country cost measures (for example, unit labour costs). Excluding domestic cost controls is especially serious in CPI equations, particularly at monthly frequencies, as wage shocks are more likely in emerging/developing countries. Exclusion also usually entails the violation of long-run price homogeneity (Sections 2 and 4). Sometimes the PPI is used to proxy for domestic costs and competing prices (Coulibaly, Citation2010); the PPI adjusts more rapidly than wages and is a good measure of local costs. But an important caveat is that while it works well in a system, in a single equation the PPI is a poor proxy because it is highly endogenous. Unit labour costs, in contrast, respond with a long lag to inflation, as wage settlements are not frequently made. If unit labour cost data are unavailable, they could be substituted by nominal manufacturing wage rates; then productivity trends from output per worker data, or at the least, a time trend, should also be added to the equation to correct for productivity changes.

The availability of data may be dictated by the frequency chosen for the analysis: demand measures are typically unavailable for higher frequency data; for example, GDP data are usually published quarterly with considerable lags, and often heavily revised. The output gap and the rate of growth of real GDP are common demand proxies. The former suffers from the ‘endpoints problem’ in its construction (using the Hodrick Prescott filter). Another problem common to both measures is that they pick up negative supply shocks, relevant in developing countries vulnerable to supply shocks. If growth is negatively impacted by drought/floods (Africa, Thailand), hurricane (Caribbean) or Tsunami (Indonesia), the shock causes inflation not disinflation and reflects a rise, not a fall, in excess demand.

Discussion of the advantages and disadvantages of aggregate versus disaggregated data is postponed to Section 7.

6. Empirical Findings for Emerging Market Countries Using Aggregate Price Indices

Having covered the advantages and disadvantages of different methodological and data choices in Sections 4 and 5, we now take a critical look at the reliability of empirical evidence on ERPT. Are the findings from different methodologies on ERPT for industrial countries replicated for less advanced countries? We cover the following aspects in Sections 6.1 to 6.7: (1) the degree and dynamics of ERPT; (2) the diversity of estimates across a single country; (3) differences in ERPT across groups of countries; (4) evidence for asymmetries; (5) evidence for the ‘instability’ of ERPT with regime change; (6) sensitivity to exchange rate volatility; and (7) the macro and micro ‘drivers’ of ERPT.

Three tables summarise methodology and comparative findings. Table 1 is an empirical typology for multi-country studies of ERPT to import, export and consumer price indices in developing and emerging market countries. Table A1 in the online Appendix 2 presents the diverse findings for South Africa, from single-country and multi-country studies. Table 2 contrasts ERPT estimates across different country groups. The sample of countries classified as ‘emerging’ needs to be clarified, especially when comparing the findings across studies and with advanced economies (see Table 1 and online Appendix 2).

6.7. Instability in Pass-Through from Regime Change

A different kind of non-linearity is introduced if the ERPT parameter, , in Equation (8) is not constant but varies by regime over the sample. For linear models spanning different regimes in long samples robustness testing is necessary. There is a sizeable literature examining changing ERPT in the sample for advanced economies (Campa & Goldberg, Citation2005; Gagnon & Ihrig, 2004). For emerging markets, studies in Table 1 explore unstable ERPT in: fixed versus flexible exchange rate regimes; high, moderate and low inflation regimes; liberalised trade regimes; and monetary policy regimes (for example, adoption of inflation targeting). Several techniques are used (see Section 4.1).

6.7.1. Exchange rate regimes

The exchange rate regime is not irrelevant for macroeconomic performance, despite the ‘mirage of fixed rates’ and the ‘fear of floating’, Klein and Shambaugh (Citation2008) argue. They demonstrate that de factoFootnote⁴⁸ fixed exchange rates exhibit considerably greater bilateral exchange rate stability than flexible rates, controlling for country and year fixed effects, inflation behaviour and capital controls.Footnote⁴⁹ A less volatile regime is more likely to encourage invoicing in the destination country’s currency (Sections 4 and 5) and pricing-to-market from foreign exporters; hence the prediction is for lower ERPT under fixed regimes. On the other hand, a shift in an exchange rate peg is more likely to be perceived as a ‘permanent’ shock, so resulting in higher ERPT.

In Table 1, studies comparing fixed versus flexible regimes run up against problems defining the regimes for their countries. Barhoumi (Citation2006) sub-groups countries by exchange rate regime, 1980–2000, using the purely statistical method of Levy-Yeyati and Sturzenegger (Citation2005); Razafimahefa (Citation2012) classifies into ‘flexible’ and ‘fixed’ groups over 1985–2008 using the IMF’s ‘Annual Report on Exchange Arrangements and Exchange Restrictions’. Both classifications are sometimes at odds with the de facto institutional classification of Reinhart and Rogoff (Citation2004). Moreover, different regimes per country are spanned within the sample period. In several cases, part of the floating regime also coincides with the adoption of inflation targeting, when inflation stabilised and declined. The result from both studies, that fixed regimes have higher ERPT than flexible regimes (in the long-run for Barhoumi and after 1 year for Razafimahefa), may not be reliable, therefore.Footnote⁵⁰

Others isolate an exchange rate regime. Kohlscheen (Citation2010) confines attention to ‘floaters’, defining the regime-periods using Reinhart and Rogoff (Citation2004); Ito and Sato (Citation2008) largely confine analysis to the post-Asian crisis period for five countries; and Bussière and Peltonen (Citation2008) find dummies for exchange rate crises are important in their regressions, reducing differences in ERPT estimates between advanced and emerging market countries.

6.7.2. Monetary regimes

For advanced countries, the relationship between monetary policy regimes and ERPT has been widely examined. In Table 1, support for a positive and significant relationship between ERPT and average CPI inflation emerges strongly for emerging markets, and explains most of the ERPT differences between them and advanced economies.

The most-quoted study of the effect of inflation on ERPT is by Choudhri and Hakura (Citation2006). The ERPT estimates to CPI for 71 countries are segregated into low, moderate and high inflation regimes (see Section 6.3). Pearson and Spearman rank correlations show that, in cross-sections, ERPT in the different regimes (for horizons up to 20 quarters) is positively and significantly related to the average inflation rate and the variance of inflation. This holds across countries and within countries. At each horizon, average ERPT is lowest for the low inflation group and the highest for the high inflation group. When the average exchange rate and its variance are included in the regressions, the average inflation rate dominates.

Similar cross-sectional results for short-run ERPT, but without segregating inflation regimes, are found by Ca’Zorzi et al. (Citation2007) and Bussière and Peltonen (Citation2008).Footnote⁵¹ Ca’Zorzi et al. find for 10 emerging markets that four- and eight-quarter ERPT to CPI is inflation-dependent (a result robust to the identification in the VAR generating the ERPT estimates). Bussière and Peltonen (Citation2008), in bivariate regressions for 41 emerging countries, find the inflation average and volatility are positively associated both with a higher export price ERPT and a higher import price ERPT (one quarter horizon). For equilibrium ERPT for imports, Barhoumi (Citation2006) finds marginally higher elasticities for those amongst the 24 developing countries with high inflation regimes.

The before-and-after method is adopted by Coulibaly and Kempf (Citation2010), who separate monetary regimes by date of adoption of inflation targeting for 27 emerging markets (with some non-targeters). In a panel-VAR, the rate of ERPT falls for all three price indices (import, producer and CPI) after the institution of more credible monetary regimes for the 15 targeters.Footnote⁵² Moreover, the contribution of exchange rate shocks to price fluctuations declines after the adoption of inflation targeting, unlike in the control group of non-targeters.Footnote⁵³ Choudhri and Hakura (Citation2012) estimate heterogeneous (bi-variate) VARs for ERPT to import and export prices, but find no decline in their emerging market sample for data from 1998 to 2010, compared with 1985–1997.

Several studies use interaction methods, with similar findings.Footnote⁵⁴ Akofio-Sowah (Citation2009) interacts with currency group dummies in SSA in a regression for CPI inflation. Higher ERPT is found for currency groups where average inflation is highest (the COMESA group of countries) compared with other regional currency groups, suggesting the latter are more credible. For Latin America, he finds ERPT is lower for officially dollarised countries (or with moderate unofficial dollarization), than for unofficially dollarised economies. In Razafimahefa (Citation2012), a ‘general’ post-1997 dummy is interacted with in a panel regression for CPI inflation in SSA, and the ERPT falls in the lower inflation period beyond 1997 (Figure 1).Footnote⁵⁵

Figure 1. Decline in headline inflation for sets of countries.

Source: World Economic Outlook database (October, 2012), IMF.

Display full size

Finally, monetary aggregates and/or interest rates have been added in several VARs (for example, Ca’Zorzi et al., Citation2007; Ito & Sato, Citation2008). These extensions of the VAR tend to support the Taylor hypothesis, that an anti-inflationary monetary policy regime is associated with lower ERPT.

6.7.3. Trade policy liberalisation

The less advanced countries have experienced extensive trade policy liberalisation in recent decades. The ERPT will be unstable in long samples if it depends on trade policy. Theory supports a negative relation between ‘openness’ and inflation (Romer, Citation1993). The impact of trade liberalisation on ERPT is less clear. Tariffs and quotas restrict spatial price arbitrage, violating the law of one price (Channel 3, Section 1). Liberalising reduces the cost barriers to arbitrage, increasing the integration of markets. Prices become more sensitive to costs, which increases ERPT.Footnote⁵⁶ In the Cournot model of Dornbusch (1987) see Section 1, more competition increases ERPT, as does a greater foreign share of imports (that is, greater import penetration and a paucity of local substitutes). But if the trade composition evolves with liberalisation towards more differentiated goods (for example, from homogeneous commodities to manufactured goods) with lower ERPT, average ERPT may fall (see Campa and Goldberg [Citation2005] for the OECD). In emerging markets, the scale of past tariff and quota reductions could overwhelm such an effect, and China’s WTO entry in 2001 will have increased the share of mass-produced standardised imported products.

Trade liberalisation refers to lifting of tariff and/or non-tariff barriers. Proxying actual trade policy is fraught by measurement problems, especially for unobservables such as quotas, once widely used in SSA. Measures of average tariffs are sometimes used though the data are infrequent (and exclude quotas and non-tariff barriers). The ubiquitous empirical measure for ‘openness’ (import dependence or penetration) is trade flows to GDP in real or nominal terms;Footnote⁵⁷ but it is endogenous and influenced by many factors, including country size. The direct impact of trade policy or ‘openness’ on inflation is captured by including an empirical proxy in an ERPT regression; the impact on the ERPT itself is measured by interacting the proxy with in this regression, or more directly by cross-sectional or panel correlations or multi-variate regressions of ERPT estimates on the proxy and other variables.

The effect of reducing protectionism on the equilibrium ERPT to import prices is examined by Barhoumi (Citation2006). Countries are segregated by trade protection regime using tariffs. This author is one of the few who tests the assumption of homogeneity of long-run ERPT across countries assumed by pooling. He rejects the hypothesis. For pooled and non-pooled mean groups methods, lower tariff barrier countries indeed experience a higher equilibrium ERPT to import prices than higher tariff barrier countries. For short-run ERPT, cross-sectional correlations and regressions mainly fail to find the expected positive relationship between ERPT and ‘trade-openness’ using the trade flows proxy measures (Section 5). Kohlscheen (Citation2010) finds a strongly negative correlation of ERPT to CPI with trade-openness for ‘floaters’ (but does not control for inflation). Ca’Zorzi et al. (Citation2007) find negative correlations with trade-openness for four- and eight-quarter ERPT to CPI that are not statistically significant, but a positive link, also insignificant, after controlling for inflation. Using regressions, Choudhri and Hakura (Citation2006) find no significant link with an import to GDP ratio, without and with inflation, for current- and five- year ERPT elasticities. Bussière and Peltonen (Citation2008) find the same result, and similarly for the degree of product differentiation (proxied by the share of high-tech goods in total trade flows).

Ambivalent results for the CPI are achieved by Goldfajn and Werlang (2010) using the indirect method. Interacting trade flows and produces the expected positive sign and significance within a year (important for the Africa group), but the direct ‘openness’ term has the wrong sign, and results prove sensitive to the horizon and sample.Footnote⁵⁸ Frankel et al. (Citation2012) interact two variables with : trade flows and a proxy for tariff barriers (using tariff data, available for two years in the sample). Higher barriers apparently reduce ERPT to the CPI for all 71 countries (as in Barhoumi [Citation2006]); the effect disappears on adding the interaction with average inflation. For the group of developing and emerging market countries, the barrier effect reverses and is significant (when controlling also for interaction effects with trade flows, average inflation and exchange rate variability).

In a nutshell, the different studies suggest: (1) lowering the trade barrier (measured by tariffs) to price arbitrage raises ERPT in the short run and also at equilibrium for some prices; (2) in more ‘open’ economies (alternatively, more import-dependent/with greater import penetration/with a greater number of importers relative to domestic producers), the ERPT to both import prices and the CPI is raised when controlling for average inflation, but this effect is not always significant. The results are sensitive to the sample used.

7. Micro-Price Analyses with Heterogeneous Pass-Through

The studies surveyed in Section 6 concentrated on aggregate price indices. Aggregate data offer two distinct advantages. Underlying the aggregate indices are millions of micro-prices, and the heterogeneous weights applied to them, being derived from consumer price surveys (for example, for the CPI) and regularly updated, reflect their economic relevance. By contrast, the disaggregated price data studies usually aggregate with equal weights for simplicity, and often for a subset of prices that may not be the most intensively used. The resulting ‘aggregate’ is thus not closely connected with the key concerns of monetary policy-makers. A second important advantage is that with time series of aggregate price indices, a systems approach can help disentangle exogenous shocks from changes in the exchange rate, and hence address the monetary policy feedback effects in ERPT relationships. This is missing in micro-analyses (see below). Against these advantages, there is likely to be an aggregation bias in ERPT given the consistent evidence for large and persistent heterogeneity in sectoral and goods levels price adjustments: service prices are stickier than goods prices, and raw goods prices are more flexible than processed goods (Klenow & Malin, Citation2011). Imbs, Mumtaz, Ravn, and Rey (Citation2005) expect heterogeneity from the differing tradability of goods, degree of competition and transportation costs in different sectors, and adjusting for the observed bias helps solve the PPP puzzle (Rogoff, Citation1996), at least in their dataset.

ERPT has been intensively studied using disaggregated trade prices at industry, firm and product levels, mainly for industrialised countries. Even in Menon’s Citation1995 survey, almost half the studies on import price pass-through used disaggregated industry data, and a handful at the product level. More recently this relative prevalence has been facilitated by standardised international definitions of traded goods,Footnote⁵⁹ and this literature is surveyed by Goldberg and Knetter (Citation1997). As noted by Goldberg and Campa (Citation2010), these studies encompass: more aggregated cross-sectional industry studies; subsets of narrowly defined industries; and firm/product micro-studies that tend to focus on a product or industry. A disaggregated analysis to the product level using annual unit value trade data for the 1990s, and including emerging and developing countries, is by Frankel et al. (Citation2012). The study pools 8 specialised commodities (details in Table 1), 12 years and 76 countries, using product and country dummies (pooled using equal weights). The downward trend in ERPT to the pooled goods prices may be partly related to the monetary regime (interacting long-term inflation and exchange rate variability with ). Real wages in interaction effects are also relevant, reflecting the role of distribution and retail costs in the decline in the ERPT.

A new, burgeoning strand of research examines pricing behaviour with highly disaggregated consumer price, producer price and trade price data. Most of the small body of research pertains to the United States and euro area (surveyed in Klenow & Malin, Citation2011; Melick & Galati, Citation2006). Data sets underlying the official monthly CPIs and PPIs have become available; as well as weekly ‘scanner’ (barcode) data and daily data from web-based retailers. In the United States, micro-data underlying monthly import and export price indices are now provided by the Bureau of Labor Statistics (originating from surveys of firms). However, only a small subset of this research examines ERPT; for example, using US micro-trade prices (summarised in Gopinath, Citation2012), United States retail and wholesale coffee prices (Nakamura, Citation2008; Nakamura & Zerom, Citation2010) and beer prices (Hellerstein, Citation2008).

To our best knowledge, no published studies for emerging market countries examine ERPT to micro-price data underlying the CPI (the first is Aron, Creamer, et al. [Citation2014]). Some studies examine ERPT to the main sub-components of aggregate price indices such as the PPI or CPI covering brief time periods (for example, 31 components in Soffer [Citation2006]).Footnote⁶⁰ Parsley examines ERPT with partly disaggregated retail price data from the Economist Intelligence Unit (EIU) for emerging markets (for example, Parsley, Citation2012). Finally, a recent strand of the forecasting literature forecasts sectoral inflation for sub-components of the CPI, giving insights into ERPT.Footnote⁶¹

The key findings from the micro-data studies for industrial countries on ERPT are:

• Heterogeneous ERPT estimates are typically reported at the sectoral and goods levels, and ERPT is delayed and incomplete for imports and both retail and wholesale domestic prices.

• Those adjusting import and export trade prices frequently have a far higher medium-run ERPT than low-frequency adjusters (Gopinath & Itskhoki, Citation2010). United States sectors producing homogeneous goods (for example, raw products) mainly price in dollars and adjust prices more frequently than product-differentiated sectors, which are mainly non-dollar-priced (emphasising that choice of the invoicing currency is endogenous; see Gopinath and Itskhoki [Citation2010]). The macro-level implications are clear if the trade composition should adjust towards differentiated goods.

• For destination country retail and wholesale prices, together non-traded local costs and imported inputs are the most important source of incomplete ERPT, estimated to contribute 50–78 per cent to incomplete ERPT across industries as diverse as beer, coffee and automobiles. It is difficult to distinguish empirically between these factors using micro-data approaches (Goldberg & Hellerstein, Citation2008). Studies using alternative methodologies and more aggregated data reach similar conclusions (Burstein et al., Citation2003; Goldberg & Campa, Citation2010). The latter suggest that the greater use of imported inputs in traded and non-traded goods across countries and industries is the key contributor to changing ERPT, not changes in distribution margins.Footnote⁶²

• Mark-up adjustment accounts for most of the rest of incomplete ERPT (for example, reduces ERPT by 33 per cent [after six quarters] in Nakamura and Zerom [Citation2010] relative to a constant elasticity benchmark; and by manufacturers and retailers, by about half in the long-run in Hellerstein [Citation2008]).

• The role of nominal price rigidities in incomplete ERPT appears small in the longer run but is important in the delayed response of prices to cost in the short run (Nakamura & Zerom, Citation2010). Rigidity seems to occur mainly at the wholesale and not retail level; strategic complementarities in pricing are small for their model (see a similar view in Klenow and Malin [Citation2011]).

A caveat is needed as restricted structural models were used for these studies. The ERPT may be biased for the usual reasons; for example, the static single equation approachFootnote⁶³ misses dynamic price-adjustment and may overestimate the role of local costs; more generally, assuming exogeneity of the exchange rate neglects monetary policy and other feedback effects.

Studies at the level of industries, firms, products or retail goods potentially lend insight into underlying structural price adjustments and the sources of incomplete and changing ERPT. As indicated by Bailliu and Murray (Citation2010), this promising research area could assist in forecasting future patterns of ERPT of interest to monetary policy-makers (and should include emerging markets).

8. Conclusions for the Wary Policy-Maker

We surveyed recent literature on ERPT, focusing on developing and emerging market (DEM) countries. It is salutary to reassess the major areas of concern in Menon’s comprehensive survey of 1995. The stark imbalance in country coverage has been redressed with a growing literature on DEM economies in multi- and individual-country studies. The profession now largely uses methodologies that address the non-stationary nature of the underlying time series, the changing coefficients of ERPT in time and, increasingly, asymmetries in ERPT. There is continued widespread use of often unsatisfactory price proxies such as unit values (see data discussion in Section 5); the trend to studying aggregated price indices has increased in the last decade, given the monetary policy interest, but a new focus on EPRT with underlying micro-prices is set to grow as datasets become available (Section 7). Menon’s concern that choice of data and methodology significantly affects the reliability of published ERPT estimates is heavily reinforced by our survey. Lessons are presented to improve future research and guide policy-makers.

The key finding of this survey is the generally poor state of the empirical literature for DEM countries, leading mainly to inconclusive and often incomparable results on ERPT (illustrated for a single country, South Africa, in Section 6.2). In general there are five common mis-specifications contributing to the malaise. First, many multi-country studies ignore the country heterogeneity (except in country fixed effects) that arises from different levels of development and trade regimes, different monetary policies and histories of inflation, and differences between commodity exporters and commodity importers. Thus, before restrictions are imposed that ERPT coefficients are the same across all countries in the sample, these restrictions should be tested.Footnote⁶⁴ A second common error, particularly in multi-country studies, is to specify a highly restricted bivariate relationship between, for instance, import prices and the effective exchange rate when the relationship is multivariate, for example involving domestic costs, such as unit labour costs, and foreign prices, possibly including oil prices. A third common error is to assume very restrictive lag structures without careful testing (often when using information criteria in VARs; see further discussion in online Appendix 1). For example, the assumption of short lags and the omission of domestic unit labour costs (which are slow to adjust to shocks but eventually respond to imported inflation) could result in long-run ERPT being substantially under-estimated. However, in countries where powerful unions create domestic wage-push shocks which depreciate the exchange rate, there is a risk of over-estimating the impact of the exchange rate on domestic inflation. This example takes us to the fourth frequent mis-specification, which is to ignore the potential endogeneity of the exchange rate, and of domestic cost variables, when these are included. Particularly when active anti-inflationary monetary policy is operating, it seems likely that an initial exchange rate shock will be partly offset by an interest rate adjustment, leading to negative feedback of exchange rate shocks, and so reducing the longer term inflationary impact of the initial shock. Notably, these issues are almost invariably ignored in ERPT studies on micro-data. The important implication is that equation systems are needed to develop a deeper understanding of ERPT.

The shift in monetary policy regimes also illustrates a fifth potential mis-specification linked with the importance of non-linearities in DEM economies: the assumption of constant parameters, when structural breaks from regime changes shift the coefficients underlying the time profile of ERPT. To date, we are not aware of studies for emerging markets combining a systems approach with tests for structural breaks that potentially could occur anywhere in the equation system. The possibility of structural breaks is part of the reason why many researchers work with VARs or single equations specified for data in quarterly or monthly differences. For estimation of short-run ERPT or short-run forecasting, differencing the data is likely to improve robustness; but, to estimate medium-term ERPT robustly, which is arguably the time frame of most interest to monetary policy-makers, model selection strategies that test and control for breaks are needed (see Castle, Doornik, and Hendry, Citation2012).

Where does this leave the policy-maker? Given the pervasive empirical mis-specifications in the literature on ERPT in DEM countries, one can ask what stylised empirical facts emerge, apart from the obvious ones that short-run ERPT is smaller than long-run ERPT, that ERPT diminishes down the distribution chain from import prices at entry to consumer prices, and that there is considerable heterogeneity among countries. One important stylised fact is that there appears to be little systematic difference between advanced countries and DEM economies as a whole, once inflation history is controlled for. This is consistent with the view that ERPT has declined with more stable, anti-inflationary monetary policy regimes (see Taylor, Citation2000). However, though not investigated by our studies, technical change, changes in trade composition, the China effect on inflation in manufactures and the increasing CPI weight of services could also account for some of this effect.

There are some indications that exchange rate volatility may be associated with higher ERPT in the DEM countries with floating currencies, possibly because exchange rate changes are viewed as more permanent or because invoicing in DEM currencies is discouraged by volatility. However, in countries with occasionally shifting currency pegs, which tend to have lower exchange rate volatility, ERPT appears to be higher, probably again because exchange rate changes are viewed as more permanent. At this stage, we view the evidence on asymmetries in the response of domestic prices to positive/negative or large/small exchange rate shocks as too inconclusive and too sensitive to the specification errors in the underlying models for broad conclusions to be reached. Similarly, evidence on trade openness (import penetration) analysed using (endogenous) trade flows is ambiguous, though lower trade barriers measured using tariff measures appear to increase ERPT.

Fresh research work should concentrate on addressing the empirical deficiencies highlighted in this survey: comparing single equation and systems methods given their comparative advantages and disadvantages (Section 4.3); checking robustness to different empirical proxies and lag lengths in better-specified equations using general-to-specific approaches, and to variable ordering (in VARs); taking account of the non-linearities and regime changes that prove pervasive in DEM economies such as monetary and trade regime changes (implying instability in ERPT) and asymmetries of various types. A high priority for research on aggregate time-series data for DEM countries is to incorporate domestic costs in the form of unit labour costs. Such data for the aggregate economy are often unavailable, but for many countries there are data on wage rates, at least for some sectors of the economy and on productivity growth, even if only annual data. Unit labour costs could then be replaced by hourly or per person wage or earnings data and the productivity trend, without enforcing the restriction: But this needs to be studied in at least a three-equation system, in which domestic prices, wages and the exchange rate are treated as endogenous, and where, for small economies, foreign prices including oil prices are taken as exogenous. Where highly disaggregated micro-data are becoming available, studies of ERPT and of the frequency of price changes are a high priority. Such studies illuminate the effects of market structure and recent price history on current price setting and hence on the economics of inflation.

For the wary policy-maker, ‘forewarned is forearmed’ when it comes to the interpretation of ERPT results, but the message is stronger: policy-makers should insist on the misspecifications discussed above being addressed.