Partners are Australia, Canada, Denmark, Hong Kong SAR, Japan, Korea, Norway, Singapore, Sweden, Switzerland, the United Kingdom, and the United States.
Price and wage data are from Eurostat and are not seasonally adjusted. The areawide measures reflect aggregates of the 11 participating countries through December 2000. Thereafter, the chained series include Greece as the 12th member country.
Manufactured goods include Sections 5–8 of the Standard International Trade Classification. The use of unit values (i.e., aggregrate, implicit deflators) is not ideal, but data limitations on direct trade prices necessitate their use.
For Canada, Japan, the United Kingdom, and the United States, all corresponding monthly series were drawn from the IMF’s International Financial Statistics, except wages, which were obtained from the Organization for Economic Cooperation and Development’s Analytical Database.
The differenced series for euro area consumer and import prices were borderline nonstationary (see Appendix). But, as is well known, unit root and stationarity tests have low power, making it difficult to distinguish between stationary and unit root processes in finite samples.
The identification scheme largely follows Choudhri, Faruqee, and Hakura (2005). That analysis also includes the interest rate in the VAR in order to further distinguish between the effects of interest rate and exchange rate shocks on prices. McCarthy (2000) and Hahn (2003) also use a Cholesky decomposition to examine pass-through based on a somewhat different model.
For the first variable (i.e., the exchange rate), the orthogonalized disturbance term is given by ε1t = μ1t. For the jth variable (j > 1) in the VAR, the corresponding shock term is given by εjt = μjt − cj,1εjt … − cj,j − 1εjt, where cj,i correspond to the entries of the Cholesky matrix (see Hamilton, 1994).
The exchange rate, however, helps predict (at least) trade prices (see Appendix). Restricted VAR estimates (not reported) excluding lagged prices from the exchange rate equation yield very similar results to those reported here.
Given the large number of parameters in the VAR, a parsimonious lag structure is sought to conserve degrees of freedom. The lag length p is chosen by starting with given maximum lag length and sequentially testing the incremental significance of dropping an additional lag based on the likelihood ratio test. Derived from bootstrapping methods, confidence intervals for the IRFs are shown in Figure 3.
Obstfeld and Rogoff (2000) provide evidence that the terms of trade decline with a currency depreciation—an observation that appears at odds with the implications of strict LCP models. See Lane (2001) for a review.
The exchange rate path is typically found to be less persistent (i.e., more mean-reverting) in the quarterly VAR when interest rates are included (and first in the ordering), but the (absolute and relative) pass-through effects are more similar to those reported here.
Gagnon and Ihrig (2001) find low degrees of pass-through (around 5 percent) in consumer prices for 20 industrial countries, and argue that pass-through has been declining.
These estimates should be taken as indicative, given the standard errors of the impulse-response functions. Based on bootstrapped standard errors, the 90 percent confidence interval for import prices is the widest, suggesting pass-through at 18 months in the range 0.6 to 1.4. The median value of this band suggests a slightly lower degree of import price pass-through (i.e., 0.95 at 18 months); otherwise, the median and estimated IRFs closely coincide.
The time pattern is similar, though, with most of the price adjustment transpiring by five quarters. Based on a quarterly VAR, Hahn (2003) reports similar findings on the degree and time pattern of pass-through in the euro area non-oil import deflator.
From the variance-covariance matrix, the correlations between residuals are less than 0.2, with the notable exceptions of the exchange rate and trade prices and between trade prices themselves.
This reduces the number of relevant reorderings considerably. The possible cases can be enumerated as follows:
The pass-through results are very similar despite the fact that the impulse-response path for the exchange rate itself differs across the two orderings, displaying less persistence in the latter case where the exchange rate shock depends recursively on the reduced-form residuals from all the price equations. The interpretation of the exchange rate shock becomes more difficult in this case, since it incorporates a mix of innovation terms.
The Leontief-type distribution process is needed only for moving intermediate goods across borders. For simplicity, the distribution cost for imports is the same whether they are used in the production of the intermediate or the final good. Note, however, that the consumption “technology” in equation (6) specifies a Cobb-Douglas process for transforming both home and foreign intermediates into a final consumer good.
Investigating extra-area import prices, Anderton (2003) finds that foreign suppliers attach a significant weight to the PTM strategy in efforts to maintain market share. Herzberg, Kapetanios, and Price (2003) find PTM to be the dominant consideration for U.K. import prices. Kieler (2001) provides comparative estimates for several industrial countries.
See also Kasa (1992) and Faruqee (1995) for analyses that introduce market-specific costs as a way to generate incomplete pass-through and PTM behavior. An alternative approach focusing on varying markups and demand elasticities through translog (i.e., non-CES) preferences can be found in Bergin and Feenstra (2001).
The home import price (in local currency terms) of a foreign variety, after and before distribution costs, is P̃Mt(y*) = PMt(y*) + δWt and thus depends partly on the domestic wage W.
Individual labor demands facing households depend on relative wages and aggregate labor demand:
Choudhri, Faruqee, and Hakura (2002) also examine the cases of wage and/or price flexibility (i.e., π = 0) and find that these model variants generally fall short in explaining the empirical impulse-responses.
The firms’ stochastic discount factor is consistent with that of households, based on the Euler condition in consumption:
Specifically, one can derive
In a noise-trader model, market participants can be subject to a stochastic bias in their expectations,
Reflecting the higher (monthly) frequency, the discount factor and interest rate are adjusted to produce consistent annualized rates with Choudhri, Faruqee, and Hakura (2005).
The exchange rate shocks from the VAR have permanent or highly persistent effects on the log level of the exchange rate. Consequently, the ξ shocks in the model are constructed to be permanent or highly persistent as well.
The standard errors for the accumulated impulse-response trajectories are similar, suggesting that minimizing the unweighted least squares should yield parameters broadly similar to the weighted least squares approach used in Smets and Wouters (2002). The weighted approach would give less weight to import prices in favor of consumer prices.
The elasticity of substitution σ between traded goods, broadly defined, was also chosen to minimize the distance between IRFs and found to be fairly low (σ = 2), in line with Choudhri, Faruqee, and Hakura (2002); Smets and Wouters (2002); and the references cited therein. The elasticity of substitution θ between specific varieties was also allowed to vary within a range (θ = 4 to 10), consistent with plausible markups.
Bekx (1998) reports the following 1995 shares of import and export prices that are respectively invoiced in domestic currency (figures in percent): Germany (52,75), Japan (23,36), the United Kingdom (43,62), and the United States (81,92). Note that the percentages for Germany refer to deutschemarks rather than all euro area legacy currencies.
See Dominguez (1999) for a related discussion of the limited role of the yen as an international currency, particularly as an invoicing currency. In principle, the choice of invoice currency is distinct from the issues of local versus producer currency pricing, but these issues share considerable overlap in practice.
The estimates for π typically range from 5 percent to 10 percent, suggesting an average duration between price changes of three to six quarters. The π estimates for the euro area (U.S.) are at the higher (lower) end of this range, suggesting shorter (longer) durations between price changes.
High estimates of implicit distribution costs in Japan may reflect aspects of its peculiar distribution system—including restrictive government regulations, predominance of small stores, complex marketing channels, and pervasive constraints on vertical integration. See Flath (2003).
The classic Marshall-Lerner-Robinson (MLR) condition requires that the sum of trade elasticities exceed unity for a depreciation to improve the trade balance under traditional pass-through assumptions (i.e., zero and full pass-through in export and import prices, respectively). For example, Isard and others (2001) use elasticity benchmarks of 0.7 and 0.9 for export and import volumes. When pass-through is incomplete, however, the MLR condition need not apply.