«A Structural Approach to Explaining Incomplete Exchange-Rate Pass-through and Pricing-To-Market By Pinelopi Koujianou Goldberg and Rebecca ...»
A Structural Approach to Explaining Incomplete Exchange-Rate Pass-through and
By Pinelopi Koujianou Goldberg and Rebecca Hellerstein∗
The continuing depreciation of the dollar against other major currencies, coupled with
concerns about the impact of China’s exchange-rate policy on domestic prices, has spurred
new interest in the exchange-rate pass-through literature. A recent study by economists at
the Board of Governors attracted wide attention by documenting a steady decline over the past decade in the pass-through of exchange rates into U.S. import prices. This ﬁnding was later challenged by a study published by the Federal Reserve Bank of New York (see Mario Marazzi et al (2005) versus Rebecca Hellerstein, Deirdre Daly, and Christina Marsh (2006)), which demonstrated that the ﬁnding of such a decline depends crucially on the speciﬁcation of the pass-through regression, and in particular the inclusion of commodity prices. This exchange highlights the need to understand the structural determinants of exchange rate pass-through, not only because such understanding is important when trying to forecast future pass-through patterns, but also because it provides guidance regarding the speciﬁcation of the appropriate reduced-form regression, and more generally, measurement of pass-through.
The increased availability of micro data on prices and quantities means that research uncov- ering these determinants is more promising than ever. This paper lays out a structural approach that can be used to identify the determinants of incomplete exchange rate pass-through. We argue that despite the use of parametric assumptions, our identiﬁcation method that relies on exploiting exchange rate variability is general, and can be applied to a variety of markets and data. We show that existing micro studies yield surprisingly robust results regarding the sources of incomplete pass-through, with non-traded local costs emerging as the primary cause, in spite of diﬀerences in industries and countries investigated, modelling assumptions and data. We conclude by noting limitations of this approach and suggesting directions for future research.
∗ Goldberg: Department of Economics, Fisher Hall, Princeton University, Princeton, NJ 08544- 1021, BREAD, and NBER, (e-mail: email@example.com); Hellerstein: International Research Group, Federal Reserve Bank of New York, 33 Liberty Street, New York, NY 10045, (e-mail: re- firstname.lastname@example.org). The views expressed in this paper are those of the authors, and do not necessarily reﬂect the position of the Federal Reserve Bank of New York or the Federal Reserve System.
1 1 The Empirical Framework
1.1 The Problem Consider the example of a German ﬁrm exporting to the U.S. To keep things simple, suppose it is a single-product ﬁrm. The price of theproduct, expressed in U.S. dollars ($) and euros (€), respectively, is given by the following two identities:
e Pjt denotes the price this ﬁrm charges in the U.S. market, when it is converted to euros, and not the price of this product in the domestic German market. Mjt denotes the proportional e markup, MCjt is the marginal cost, denominated in the producer currency (€), and Et is the bilateral exchange rate ($/€). When exchange rate pass-through is complete, changes in $ Et lead to proportional changes in the local currency (€) price of the good Pjt. Hence the $ variability in Pjt tracks the variability in Et. In contrast, the product’s price expressed in e the producer currency (euro) Pjt should remain constant. However, there is overwhelming empirical evidence pointing to precisely the opposite pattern: The variability in exchange rates e closely tracks the variability in Pjt, while the product’s local-currency price remains fairly stable over time. Hence, the data imply incomplete, and in fact very low, pass-through. This e is the phenomenon we are trying to understand. As evident from (2), for Pjt to co-vary with exchange rates, it must be true that exchange rates lead to either a change in markups Mjt, e or a change in euro-denominated marginal cost MCjt, or both. An additional explanation is that because of nominal price rigidities (e.g., menu costs), prices do not respond to changes in $ the economic environment at all, so that Pjt remains literally ﬁxed in the short run. In the data this would show as a “no change” in the local-currency price of the product, as opposed to a small, incomplete change implied by the markup or marginal-cost change channels. In the following we investigate how micro data can help us identify the relative contributions of the markup, marginal cost, and nominal-rigidity channels respectively, in generating incomplete pass-through. For expositional purposes we focus mainly on the case of a single destination market (in our example above, the U.S.), but the framework can easily accommodate extensions to a multi-destination framework. In fact, we argue that the multi-destination framework allows one to obtain sharper results and further reﬁne the hypotheses investigated. Following the 2 literature, we use the term “incomplete pass-through” to refer to a single-destination market and “pricing-to-market” to refer to multiple destination markets, where the incomplete passthrough also generates deviations from the Law of One Price.
1.2 A Reduced-Form Approach Before describing the structural approach, it is instructive to consider what can be learned from a reduced-form approach to the problem. To this end, let us focus on the case of a (let’s say German) ﬁrm that exports to multiple destinations. Consider the following reduced form regression relating the price the ﬁrm charges in each destination (converted to euros) to the
bilateral exchange rate:
e(3) ln Pmt = θt + λm + β m ln Emt + umt
where the subscript m denotes destination market, θt is a set of time eﬀects, λm is a set of destination market ﬁxed eﬀects, and Emt is the bilateral exchange rate between the destination market and Germany. This is the standard pricing-to-market (PTM) equation estimated in the literature, where β m denotes the PTM coeﬃcient. The market ﬁxed eﬀects λm proxy for both quality and markup diﬀerences across markets that do not vary over time, while the time ﬁxed eﬀects θt capture the common-across-markets changes in marginal costs and markups.
Given this setup, the ﬁnding of a β m that is signiﬁcantly diﬀerent from zero convincingly establishes that exchange-rate changes are associated with markup variation and that this variation is speciﬁc to each destination market (a more detailed discussion is provided in Pinelopi K. Goldberg and Michael Knetter (1997), pp. 1254-55). The reduced-form approach has several advantages: it is easy to implement, it does not rest on any functional-form assumptions, and the data requirements are modest. Apart from exchange rates, one needs only data on prices to estimate (3), though for the approach to be informative, the data must cover multiple destinations. On the negative side, the inferences one can draw from price data alone are limited.
Speciﬁcally, while the reduced-form approach reveals that the “variable-markup" channel is at work, it cannot tell us how large the markup variation is, or how this mechanism compares to the other potential channels. More importantly, an implicit assumption underlying this approach is that the marginal cost of selling to each destination is not aﬀected by exchange rates. This assumption is plausible when the data used are f.o.b. export prices, as in Michael Knetter’s (1989) original work; however, it is much more controversial if one employs consumer-level data with a large portion of local, destination-market-speciﬁc, non-traded costs.
31.3 A Structural Approach
Consider equation (2) again. Let us
from nominal rigidities for now. Our objective e is to decompose the variability in Pjt into the variability in marginal costs, and the variability in markups. The challenge is that neither marginal costs nor markups are observable. The problem looks remarkably similar to the one faced in empirical work in Industrial Organization.
Accordingly, the approach we propose is informed by recent advances in that ﬁeld. While the techniques are not new, we note that the application of Industrial Organization methods to the question of exchange rate pass-through has a particular appeal that is rarely found in domestic market applications: exchange rates provide a source of large and plausibly exogenous price variation in the data. It is this variation that we exploit to identify marginal costs and markups.
Hence, exchange rates play a double role in this framework: they are the object of investigation;
at the same time, they help us identify the parameters of interest in the underlying structural model. We next explain how the data can help us identify the relative contributions of markups, marginal costs, and nominal rigidities respectively. We base most of our discussion on a static model, but consider dynamic extensions at the end.
Variable Markups: A common misconception is that incomplete pass-through reﬂects the intensity of competition in the destination market. This is only partially true. The intensity of competition is relevant, but only to the extent that it produces variable markups. To see that, consider the case of a monopolist who faces a CES demand. If demand is elastic, the markup will be low. Nevertheless, given the CES structure, the markup will not change in response to an exchange rate change, so that - absent any changes in the marginal costs - pass-through will be complete! For a model to be able to generate incomplete pass-through, it is thus important to allow for functional forms that do not - by construction - imply constant markups. In the static, period-by-period proﬁt maximization problem, a ﬁrm’s markup Mjt is linked to the price
elasticity facing the ﬁrm by the ﬁrm’s ﬁrst order condition:
ηjt (Pjt, Pt, Zt ) (4) Mjt = η jt (Pjt, Pt, Zt ) − 1 where η jt denotes the price elasticity facing the ﬁrm. The notation is chosen to reﬂect the fact the price elasticity facing the ﬁrm will in general be a function of the ﬁrm’s price Pjt (and hence not constant), as well as a function of a vector of competitor prices Pt, and a set of exogenous variables Zt. The markup can be inferred from the ﬁrst order conditions once the price elasticity facing the ﬁrm has been estimated. This suggests the following approach for estimating markups: 1. Assume a particular utility (or demand) function and estimate the 4 relevant parameters using data on prices and market shares; 2. Assume a particular market structure and behavior; 3. Infer the markups implied by 1.and 2. using the ﬁrst order condition (4). As we noted above, the advantage of international data is that exchange rate variation provides a plausible source of identiﬁcation of the demand side parameters - at least in the partial equilibrium setting. Hence, while this approach is parametric, identiﬁcation is not driven by functional form assumptions alone. Perhaps the most unsettling aspect of this approach is the dependence of the derived markups on the functional form assumptions inherent in the ﬁrst step. In practice, a variety of approaches have been adopted in the literature to estimate the price elasticity facing the ﬁrm. One common approach, especially in the macro literature, is to assume a CES market demand together with Cournot competition. The Cournot assumption leads to a price elasticity facing the ﬁrm that depends on the market share: the larger the market share, the lower the elasticity of demand facing the ﬁrm. One can easily see how this generates incomplete pass-through; as the local currency price starts rising in response to an exchange rate depreciation, the market share of the ﬁrm under consideration declines, raising the price elasticity of demand facing the ﬁrm. While this approach can generate incomplete pass-through, the assumptions of CES demand and Cournot are hard to defend, especially when applied to a wide set of sectors across the economy. At the other end of the spectrum, one can assume a linear demand function, which generates - by construction - a pass-through of less than 50%. This particular functional form assumption can generate a signiﬁcant degree of price inertia that matches what is observed in the data; however, the linear demand structure does not match other features of the data (e.g., does not produce plausible demand elasticities).
An alternative approach that has been pursued in a series of recent papers is to adopt ﬂexible functional forms that have been shown to match cross-sectional patterns in the data well and produced plausible substitution patterns, and examine what these functional forms imply for pass-through. Examples of this approach include the estimation of nested logit (Pinelopi K.