Large Devaluations and the Real Exchange Rate

The paper asks why large devaluations are typically followed by large, persistent falls in the real exchange rate accompanied by low domestic inflation, and argues that the primary force behind this pattern is the slow adjustment of nontradable-goods and distribution-service prices, not a failure of relative PPP for goods that are actually traded (Section 1, Introduction). The authors note that at both short and long horizons, the rate of exchange-rate passthrough to prices is much lower for nontradable goods than for goods that are actually traded, and that this finding casts doubt on the alternative view that large devaluations produce large deviations from relative PPP for imported and exported goods.

The conventional decomposition – classifying goods as tradable and services as nontradable – is misleading because the retail price of a nominally “tradable” good embeds two nontradable components: distribution costs and “local goods” (Section 3.1, “The Composition of the CPI Basket”). Distribution costs – wholesale and retail services, marketing and advertising, and local transportation – are estimated, following Burstein, Neves, and Rebelo (2003), to average roughly 50 percent of the retail price of consumption goods; since distribution services are intensive in local land and labor, this implies half of the retail price of a tradable good is actually a nontradable input. Separately, many goods conventionally classified as tradable (the paper’s example is yogurt in Argentina) are in fact “local goods” produced solely for domestic consumption and never actually cross the border. Combining these two adjustments, the authors estimate that the share of the CPI basket that is truly composed of pure-traded goods may be as low as 14 percent, compared with a traditional estimate that nontradables account for only about 50 percent of the CPI.

Measuring the price of tradable goods at the border (import and export prices “at the dock”) rather than at retail shows that relative PPP is a reasonable description of pure-traded-goods prices, even though retail tradable-goods prices adjust much more slowly – a sharp contrast the authors treat as their central empirical fact (Section 2, “Large Devaluations,” Table 1). For example, in the first three months after Brazil’s 1999 devaluation, import and export prices rose by 44 and 37 percent respectively, comparable to the 45 percent nominal depreciation, while the retail price of tradable goods rose by only 4 percent, and the price of nontradables and services rose by just 1 percent. The authors also check that import price indices are not dominated by capital and intermediate goods rather than consumption goods, and find the two move closely together, and that public-goods (government-controlled) prices behave similarly to other nontradables prices, ruling out administered pricing as the explanation for slow nontradables inflation.

A sequence of price-accounting exercises shows that assuming all goods are tradable, or that nontradables exist without distribution costs or local goods, greatly overstates actual post-devaluation inflation, while incorporating both distribution costs and local goods brings the model’s implied inflation close to the data (Section 3.2, Table 4). Under Assumption A (all goods traded), predicted first-year inflation in Brazil is roughly 40 percent against an actual rate of about 9 percent. Adding nontradables without distribution costs or local goods (Assumption B) still greatly overstates inflation (about 27 percent for Brazil). Adding a 50 percent distribution margin (Assumption C) brings predicted Brazilian inflation down to about 15 percent, still above the actual 9 percent; only once local goods are also incorporated (Assumption D) does predicted inflation closely match the actual rate across the sample countries, though in Argentina the fully adjusted prediction is somewhat below the actual rate. The authors interpret the true weight of pure-traded goods as lying between the Assumption C and D bounds.

Under the fully adjusted specification (nontradables, local goods, and distribution costs), the implied markup movements needed to exactly reconcile predicted and actual retail tradable prices are modest – generally below 10-20 percent – whereas under the narrower specification that excludes local goods and distribution costs, the implied markup swings are implausibly large (Section 3.2, “Markups and Measurement Error,” Figure 3). For example, if distribution costs and local goods are ignored (Assumption B), reconciling the model with the data would require the gross markup in Argentina to fall below 0.69 (implying retailers sold at a loss for extended periods) or, alternatively, would require implausibly high pre-devaluation markups above 1.85 in Argentina and 1.5 elsewhere. The authors treat the modest, plausible size of the markup adjustment needed under the full (distribution-costs-and-local-goods) specification as supporting evidence for that specification.

Using Economist Intelligence Unit (EIU) cost-of-living data that holds product quality roughly constant, the authors find that EIU-based inflation rates exceed official partial-CPI inflation rates in four of five countries, consistent with a flight from quality that biases official inflation downward (Section 3.3). For example, in Brazil, one-year partial CPI inflation is about 7 percent while the corresponding EIU inflation rates are around 15 percent. The mechanism is that official CPI price inspectors, when an item is discontinued and replaced, must split the price change between a pure-price effect and a quality-difference effect; after a large contractionary devaluation, when consumers switch to cheaper substitutes and product turnover rises, imperfect correction for these quality changes can impart a systematic downward bias to measured inflation. Mexico, where the authors have institutional evidence of very few unplanned CPI substitutions around the devaluation, shows EIU and official inflation tracking closely, which the authors treat as a validity check on the EIU-based approach.

Decomposing the real exchange rate, following Engel (1999), into the log-difference in pure-traded-goods prices across countries versus the log-difference in nontradables prices relative to pure-traded goods, the authors find that the pure-traded-goods component explains only a small fraction of the total RER movement in their large devaluation episodes – for example, at a six-month horizon, only 20, 9, 7, 2, and 3 percent of the RER movement in Argentina, Brazil, Korea, Mexico, and Thailand, respectively (Section 4, “Engel Decompositions”). This is in sharp contrast to the traditional decomposition using retail tradable prices, which attributes 88-99 percent of the RER movement in the Korean case, for instance, to the “tradable” component. The authors also show that allowing for time-varying markups in the distribution sector does not change the conclusion that the pure-traded-goods component is small.

The Argentina case study, using disaggregated CPI data, a purpose-built survey of prices in Buenos Aires, and supermarket scanner data, shows that inflation is higher for product categories with a larger market share of imported and exportable goods, that goods prices adjust far more frequently than services prices, and that there is direct evidence of flight from quality (Section 5). Between December 2001 and December 2002, as the peso/dollar rate fell by roughly 124 percent (log terms), import prices rose 111 percent, the retail price of imported goods rose only 83 percent, the retail price of tradable goods (imported plus local) rose just 52 percent, and nontradables rose only 13 percent. A weekly survey of 53 goods and 10 services in Buenos Aires supermarkets found a median weekly frequency of price change of 31 percent for goods versus zero percent for services (median time between price changes of 4.8 weeks versus effectively infinite). Scanner data show the median share of SKUs “destroyed” (disappearing from shelves) rose from 16 percent (year to June 2001) to 26 percent (year to June 2002), and the relative price of products losing market share to those gaining share rose from a median ratio of 1.1 to 1.3, both consistent with flight from quality; a companion AC Nielsen consumer survey found 88 percent of respondents reported changing what they purchased, with 85 percent of those switching to cheaper brands.

In the two large RER appreciation episodes studied (Argentina’s 1991-95 Convertibility Plan and Mexico’s 1988-94 exchange-rate-based stabilization), the same nontradables-relative-to-pure-traded-goods channel dominates RER movements, mirroring the devaluation results (Section 6). Measured with retail tradable prices, changes in the price of “tradables” account for 89 percent (Argentina) and 58 percent (Mexico) of the RER appreciation, but movements in pure-traded-goods prices (import/export prices at the dock) account for only 21 percent and -4 percent of the RER movement, respectively – consistent with relative PPP holding reasonably well for pure-traded goods, and with the real appreciation instead reflecting a rising relative price of nontradables during these expansions.

The nontradables-relative-to-pure-traded-goods channel remains important for medium-size devaluations and small business-cycle exchange-rate fluctuations, but it is less dominant than in large devaluation episodes, where fluctuations in pure-traded-goods prices instead account for the majority of small movements (Sections 7.1-7.2). For four 1992 medium-size devaluations (Finland, Italy, Sweden, UK), the fraction of the 12-month RER movement attributable to pure-traded-goods prices ranges from about -5 percent (UK) to 60-62 percent (Finland, Sweden), compared with the 2-20 percent range found in the large devaluation episodes – so relative PPP holds up somewhat worse for pure-traded goods than in the large-devaluation sample, consistent with findings in Campa and Goldberg (2004). For small, HP-filtered, business-cycle-frequency fluctuations across ten OECD countries, the authors estimate that movements in the pure-traded-goods component account for about 63 percent of the standard deviation of the RER, so that, in line with Engel (1999) and Chari, Kehoe, and McGrattan (2002), fluctuations in the price of pure-traded goods across countries account for the majority of small RER movements – the opposite ranking from the large-devaluation case.

The authors state that they do not address why the rate of inflation for nontradable goods is so much lower than the rate of devaluation, and identify developing quantitative general-equilibrium models that can account for this phenomenon as an important task for future research (Section 8, Conclusion). Their contribution is thus empirical and accounting-based – establishing which prices move slowly and quantifying their importance for the RER – rather than a structural explanation of the underlying nominal-rigidity mechanism in the nontradable and distribution sectors.