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

Abstract

Global integration has increased the international linkages of financial markets for emerging market countries. A key channel for the international transmission of inflation and economic cycles is from exchange rate movements to domestic prices, known as exchange rate pass-through (ERPT). This article reviews the conceptual, methodological and policy issues connected with ERPT in emerging market and developing countries, and critically surveys selected empirical studies. A key contribution is to categorise and compare the heterogeneous methodologies used to extract ERPT measures in the empirical literature. Single equation models and systems methods are contrasted; frequent misspecifications that produce unreliable ERPT estimates are highlighted. The discerning policy-maker needs to ascertain by which methods ERPT measures were calculated, the controls and restrictions applied, and the time frame and stability of the estimates.

Acknowledgements

Janine Aron acknowledges support from the British Academy (British Academy Research Development Award). This research was also supported in part by grants from the Open Society Foundation and the Oxford Martin School. We are grateful to Gregory Farrell, Ekaterina Kortava and an anonymous referee for helpful comments.

Notes

^1. Emerging market and developing economies accounted for 40 per cent of world exports in 2011 compared with 21 per cent on average in the 1990s (IMF Direction of Trade Statistics).

^2. Menon comprehensively surveys and tabulates empirical studies on ERPT up the early 1990s; Goldberg and Knetter use a unifying framework to examine some key contributions on the law of one price, purchasing power parity, ERPT, and pricing-to-market, linking to the literature on industrial organization.

^3. Exchange rate changes are not always the same as exogenous exchange rate shocks. With feedback effects, the pass-through from exchange rate changes could be different from exchange rate shocks. This distinction is explained in this survey and applied in Aron, Farrell, et al. (2014).

^4. In the extreme case of PCP, the law of one price holds for all goods and purchasing power parity is satisfied in the aggregate (assuming markets are not segmented).

^5. ‘Switching costs’ provide another mechanism of this type, where incomplete pass-through arises because firms attract customers with lower prices in the current period, knowing that customers will remain attached to the firm in the future due to high switching costs.

^6. We assume here that marginal costs are constant for convenience of exposition.

^7. A linear demand function and flexible functional forms for demand as used in some of the micro-literature (Section 7) will generate variable mark-ups, unlike a constant elasticity demand function, where the mark-up is constant.

^8. Mishkin (2009) compares two versions of the Federal Reserve’s SIGMA DSGE model. In one, Dixit–Stiglitz constant elasticity demand functions result in complete long-run ERPT; but in the other, the assumption that the elasticity of demand increases as market share declines results in far lower levels of EPRT.

^9. This combines a linear demand curve, constant marginal costs and homogeneous goods (perfect substitutability between domestic and imported goods). The ERPT elasticity then depends on the relative market shares of domestic and foreign firms.

^10. The marginal cost curve here slopes upward; marginal costs fall as output contracts in response to an appreciation in the exporter’s currency, so ERPT is incomplete.

^11. Exporting firms building market share may squeeze mark-ups when the destination currency depreciates, to stabilise the local price (Channel 1), but leave mark-ups intact when appreciation lowers prices for importers. At the extreme, there could be full ERPT after appreciation and zero ERPT after depreciation.

^12. If firms face capacity constraints and cannot satisfy increased demand at the lower price, at least in the short term, then not all the appreciation would be passed through and local prices could rise (Knetter, 1994). After depreciation, downward price or wage rigidities (Channel 3) could limit a decline in the price so the ERPT could exceed zero. Lowering prices is usually more feasible than raising them, so a lower ERPT for depreciation than appreciation could arise again (Bussiere, 2007).

^13. This issue is by no means settled even in advanced countries, see (Dale, 2011) on the underestimation of ERPT to import prices in 2009 in the UK.

^14. A regime shift with the opposite effect is trade liberalization; even with tariff reduction leading to more open trade regimes, significant non-tariff barriers may remain which themselves limit ERPT (Menon, 1996).

^15. A survey for Asia is by Ghosh and Rajan (2007).

^16. Fendel, Frenkel, and Swonke (2008), surveying German exporters, contradicts both ‘pure’ LCP and ‘pure’ PCP hypotheses.

^17. By contrast, the pricing method chosen by foreign firms in most models is regarded as exogenous to domestic monetary policy (though see an exception in Devereux and Yetman [2010]).

^18. The constant elasticity demand function, though widely used in the literature, has the implausible implication that demand for a good only tends to zero when the price tends to infinity. For most traded goods, it seems more plausible that demand would become zero at finite price thresholds given the availability of partial substitutes. This guarantees a falling mark-up as that threshold is approached, for example through local currency depreciation, so resulting in incomplete ERPT.

^19. The term ‘super-elasticity’ (rate of change of the price elasticity of demand) has been used to describe how the mark-up varies as the price changes (see Nakamura and Zerom [2010] for empirical evidence).

^20. The constant, μ, is an average of industry-specific fixed effects (calculated when the log of the real exchange rate is zero). For simplicity, the relationship between the mark-up and the real exchange rate is log linear, and implies that the mark-up decreases (that is, the price elasticity of the underlying demand function increases) with the exporter’s price. A more general formulation could allow a flexible, non-linear demand function (see Section 7).

^21. A common simplifying assumption is that marginal cost is not affected by a change in the exchange rate. However, there may also be costs (for example, from imported inputs) that fluctuate with the exporter’s own trade-weighted exchange rate.

^22. Appreciation (a percentage rise) or depreciation (a percentage fall) in the exchange rate is expressed as a difference in the log of the exchange rate: .

^23. A non-stationary process can wander away from a constant or a linear trend without limit. An I(1) non-stationary process has to be differenced to become stationary or mean-reverting over time. An I(2) process needs to be differenced twice. In a random walk, a simple version of non-stationarity, the change in the exchange rate is equal to a random term, and its path over time is the cumulation of successive random terms, so that the exchange rate is not predictable, an uncomfortable result.

^24. This applies if the underlying series are non-stationary series integrated of order 1, or I(1). If they have higher orders of integration, say I(2), they will have to be differenced again to achieve a stationary series. Some prices are thought to have I(2) behaviour, though this can result from structural breaks in an I(1) process, so that inflation then behaves like an I(1) variable.

^25. The original Campa and Goldberg model is expressed in individual prices for industrial sub-sectors.

^26. Not included were commodity prices, demand in the exporting country and destination country costs.

^27. We define the exchange rate so that a rise is an appreciation, as above, which differs from the original Campa and Goldberg (2005) paper. Hence the pass-through measures are designated with a negative sign.

^28. If one lag of the dependent variable had been included as regressor, then would have to be to be divided by 1 minus the coefficient on the lagged dependent variable to calculate ‘long-run’ ERPT, but see online Appendix 1 for further discussion.

^29. Long-run price homogeneity (see Section 3) entails that .

^30. The Hodrick Prescott filter constructs a long-term trend over a sample, using the entire sample. Thus, current and later information on the exchange rate (and hence destination country CPI) is incorporated in the trend used to explain the CPI, implying an endogeneity bias.

^31. With both an ecm term and (one or more) lags in the dependent variable, computation could be done by dynamic simulation (see online Appendix 1).

^32. Chow tests evaluate parameter stability based on exogenously imposed breakpoint, whereas other tests use endogenously determined structural breakpoints (for example, Andrews, 1993). In both cases, the asymptotic tests entailed have limited statistical power in the data samples available for analysis. Rolling regressions can also be used to determine the location of the breaks.

^33. Several Canadian import prices are constructed by multiplying the foreign-currency price by the nominal exchange rate, so that the degree of ERPT is, by construction, equal to 1 for those prices, biasing upwards the average empirical estimates of ERPT (Bailliu & Bouakez, 2004).

^34. Goldberg and Knetter (1997, p. 13) furthermore argue these proxies may introduce measurement error that is correlated with exchange rates in a way that biases coefficients toward finding incomplete ERPT and excessive mark-ups, a problem that becomes more acute with foreign outsourcing (that is, as the share of costs incurred in the home currency then declines).

^35. Trade-weighted consumer prices are easily derived from real effective exchange rate indices, now more widely available for emerging markets from the mid-1990s (for example, from the BIS and IMF).

^36. The CPI in the EU has no owner-occupied housing, but in Australia and the United States it is included, for example.

^37. It makes a big difference to ERPT estimates if oil or commodity prices are included to help control for costs (Bussière & Peltonen, 2008). Hellerstein and others have objected to their inclusion for the United States where the NEER and oil prices appear to be correlated. This is unlikely to be the case for most emerging market countries.

^38. Advanced country benchmarks are given in Table 2.

^39. In Brun-Aguerre et al. (2012), for South Africa, one country in the panel, the error was to convert an import price already in rands (domestic currency) by multiplying it by the rand/dollar exchange rate. Details are available on request. This also biases their estimates for the short run: ERPT after one quarter measures up to 160 per cent (models a,b,c, see Table 1) and after five quarters at 250 per cent (model b).

^40. Calvo and Reinhart’s set of advanced countries consists solely of Australia and New Zealand, giving a short-run average ERPT rate of 7 per cent. This should be compared with Campa and Goldberg’s short-run average for 25 OECD countries of 46 per cent (though for a longer sample).

^41. Low, moderate and high inflation groups are defined by average inflation rates less than 10 per cent, between 10 and 30 per cent and more than 30 per cent, respectively. However, note there is a classic selection bias because the division into inflation groupings is done on the basis of the dependent variable (the inflation rate). Hence, group differences will be somewhat exaggerated.

^42. To facilitate comparison with single equation estimates, this study, like Bache (2007) and Ito and Sato (2008), divides the cumulative impulse response function (IRF) for domestic prices by the cumulated IRF for the exchange rate.

^43. The (pooled) Latin sample covers only 2000–2005, and, by avoiding the excesses of inflation of earlier years, shows lower ERPT than the SSA sample.

^44. For instance, Kohlscheen (2010) examines Spearman rank correlations between 6-month and 12-month ERPT coefficients for eight countries, including inflation (average and median), inflation variance, NEER volatility, trade openness, food and energy trade; and size (constant GDP). ERPT is mainly linked with the volatility of exchange rates and the composition of trade.

^45. In Table 1, Kohlscheen, Akofio-Sowah and Mihalek and Klau, could not find size effects. Bussière and Peltonen suggest that outliers induce poor empirical findings on size effects (for example, Japan is a large economy with high ERPT).

^46. The problem concerns the model used to analyse the time-varying ERPT elasticities from rolling window estimates. In this rolling exercise, the window start-point varies from 1980Q1 to 1985Q1and the end-point varies from 2004Q3 to 2009Q3, for the longest samples. The estimated ERPT elasticities will depend on information over the entire window used. Hence, sensitivity of these ERPT elasticities will be as great to structural breaks in the early part of the sample (for example, 1980Q1–1985Q1) as in 2004Q3–2009Q3. It is therefore incorrect to explain the time and cross-section variation of the ERPT elasticities with drivers dated only at the penultimate observation of each rolling window. The average information over the entire window is required for these drivers.

^47. This can generate spurious asymmetry findings which would not arise if heterogeneity had been allowed for.

^48. In practice, de jure regimes may not coincide with the actual regime; for example, due to the ‘fear of floating’.

^49. A peg is more vulnerable to rupture in its first year, but also more likely to endure the longer it has lasted, and if it breaks is then likely to re-form more quickly.

^50. Razafimahefa (2012) acknowledges that the findings need to be interpreted cautiously.

^51. Kohlscheen’s correlations for eight countries’ floats between measured ERPT and average inflation and inflation volatility, produce no clear pattern.

^52. This confirms the findings of Mishkin and Schmidt-Hebbel (2007) for emerging market targeters.

^53. Mihalek and Klau (2008) claim a fall in ERPT for most countries but this is unreliable (column 2 of Table 1).

^54. The South African study of Aron, Farrell, et al. (2014) also uses interaction effects.

^55. This bivariate VAR model lacks controls and may simply be picking up patterns in omitted variables.

^56. Menon (1995, 1996) emphasises, however, that under perfect competition a tariff will not reduce ERPT whereas a quota will. Quotas lower the ERPT for Australia.

^57. For example, imports plus exports to GDP or the import share of GDP; see Aron and Muellbauer (2007) for a critique and an alternative measure.

^58. Akofio-Sowah also interacts the change in trade flows with and finds it insignificant (unsurprisingly, as differencing renders the variable stationary). ‘Openness’ is not included separately.

^59. The Standard International Trade Classification (SITC) and Harmonized System (HS) are different trade classifications, enabling trade price comparisons amongst different countries and years, from the late 1980s. The UN-maintained SITC categorises by the materials used in production/processing stage to five digits, while the HS allows disaggregation by product up to at least six digits.

^60. See Aron, Creamer, Muellbauer, and Rankin (2013) for a brief survey of these studies.

^61. Aron and Muellbauer (2012) survey the international literature, also reporting results for South Africa.

^62. However, see comment by Sato in Campa and Goldberg (2008).

^63. The micro analyses for the beer and automobile industries use static equations; Nakamura and Zerom (2010) explore the coffee market with a dynamic model. However, all these models, and hence the role attributed to mark-up, rely heavily on the functional form assumptions employed for estimated demand.

^64. Barhoumi (2006), alone in this literature, tests and soundly rejects this restriction of homogeneous ERPT across countries.