Heads I Win, Tails You Lose: New Evidence of Long-Run
Asymmetry in Exchange Rate Pass-Through by Exporting Firms∗
Raphael Brun-Aguerre†
J.P. Morgan, London
Ana-Maria Fuertes‡
Cass Business School, City University London
Matthew Greenwood-Nimmo§
Faculty of Business & Economics, University of Melbourne
December 9, 2013
Abstract
This paper studies the response of import prices to exchange rate shocks in a framework
that accommodates asymmetry both in the short-run dynamics and in the long-run equilibrium relationship. Estimation of asymmetric single equation error correction models for 33
countries over the period 1980 to 2010 reveals stronger pass-through of depreciations than
appreciations in the long-run, which is suggestive of exporter pricing power. In a panel context, the long-run asymmetry is significant and robust to different country grouping criteria
such as emerging versus developed. There is a strong positive nexus between the extent of
the asymmetry and import dependence, which suggests that exporters price-to-market in
the long-run. However, this link weakens as the importer gains greater freedom to trade
internationally. Our findings suggest that exporters on the whole are able to exploit weak
competition structures over the long-run. Since the prevalent form of pass-through asymmetry is welfare-reducing for consumers, import-dependent economies can benefit from trade
liberalization.
Keywords: Exchange Rate Pass-Through; Asymmetry; Nonlinear ARDL Model; Random Coefficients Panel Data Model; Emerging Markets.
JEL Classifications: F10; F14; F30; F31.
∗
Correspondence to: M.J. Greenwood-Nimmo, 3.12 Faculty of Business and Economics, University of Melboune, Carlton 3053, Australia. We would like to thank Matthieu Bussière, Charlie Cai, Jerry Coakley, Annina
Kaltenbrunner, Minjoo Kim, Donald MacLaren, Phil McCalman, Viet Nguyen, Adrian Pagan, Kate Phylaktis,
Kalvinder Shields, Yongcheol Shin and Ron Smith for their many helpful suggestions. The views expressed herein
do not reflect those of J.P. Morgan.
†
Email: [email protected]
‡
Email: [email protected]
§
Email: [email protected]. Tel: +61 3 8344 5354.
1
1
Introduction
In a perfectly competitive and frictionless environment, exchange rate fluctuations should be
rapidly, completely and symmetrically reflected in import prices. In practice, however, many
firms operate in imperfectly competitive markets subject to various frictions. As a result, exchange rate pass-through (ERPT) may be incomplete and may exhibit various complexities,
including sluggish adjustment and asymmetry with respect to depreciations and appreciations
(i.e. sign asymmetry). A thorough understanding of the nature of ERPT is central to the analysis of global trade imbalances and the study of shock propagation in the global economy. ERPT
is also, therefore, of direct concern to agencies charged with the optimal conduct of sovereign
macroeconomic policy as well as international bodies charged with global oversight.
This paper focuses on the narrowest notion of pass-through to the prices of goods observed
at the dock (i.e., when they first arrive in the destination country) which contrasts with broader
definitions widely employed in the literature, including pass-through to the price of imported
goods at the retail store counter or, more broadly still, to the general price level. By focusing
acutely on import prices, we aim to shed light on the aggregate pricing behaviour of exporting
firms conditional on various quantifiable characteristics of the export market without having to
control for additional confounding factors arising after the goods arrive at the dock, including
tariff structures, local transportation, distribution and retail costs.
ERPT elasticities can plausibly range between zero and one depending on exporters’ pricing
strategies. When export prices are set as a markup over marginal costs, the willingness of
exporting firms to hold the price of final goods constant in the local currency of the importing
market by absorbing exchange rate fluctuations into their markup – a strategy known as local
currency pricing (LCP) – results in incomplete ERPT. Effectively, incomplete or zero passthrough implies a deterioration of exporting firms’ margins in the case of depreciations and an
improvement in the case of appreciations.1 If LCP prevails, then the importing economy (i.e.,
the buyer) is insulated from terms-of-trade shocks and, in turn, from any expenditure-switching
effects arising from currency shocks. On the other hand, if exporters are reluctant to allow their
margins to fluctuate with the exchange rate then ERPT will be complete in accordance with the
Law of One Price (LOOP) – a strategy known as producer currency pricing (PCP). Under PCP,
imported goods will become more expensive (cheaper) following a depreciation (appreciation).
When the importing economy pursues inflation-targeting monetary policy, the impact of
1
Throughout this paper, the term appreciation (depreciation) will refer to an increase (decrease) in the value
of the importer’s local currency relative to the exporter’s currency.
2
exchange rate fluctuations on import prices will be relevant not just to producers and consumers
but also to policymakers and regulators. Under complete ERPT, a depreciation of the domestic
currency will lead to a commensurate increase in import prices, which will be reflected to some
degree in domestic consumer price inflation. However, much of the recent empirical literature
suggests that import price ERPT is incomplete. Weaker ERPT reduces the effect of exchange
rate fluctuations on consumer price inflation and, therefore, a smaller interest rate adjustment
will be required to maintain a targeted rate of inflation. An enriched understanding of the
nature and extent of ERPT will thus enhance the central bank’s ability to conduct monetary
policy in an optimal manner. The nature and extent of ERPT is also relevant for the choice
of the exchange rate regime as the fear of floating exhibited by many developing economies is
often linked to their apprehension about (near-) complete ERPT.
In the context of the 1970’s currency realignments, Kreinin (1977) documented varying
degrees of ERPT to different countries. He found relatively muted pass-through to US import
prices at 50 percent, stronger but still incomplete pass-through in both Germany and Japan (60
and 70 percent, respectively) and complete pass-through in Italy. Furthermore, the observation
that the various currency crises of the 1990s were not generally associated with high rates of
inflation provides ample anecdotal evidence of incomplete ERPT.2 The apparent resilience of
import prices to exchange rate fluctuations has generated a large literature concerned with
quantifying the extent of ERPT. The related issue of whether ERPT is endogenous to the
importing economy is receiving increasing attention, but the crucial issue of whether ERPT is
driven mainly by micro or macro factors remains unresolved (Brun-Aguerre et al., 2012; Bussière
and Peltonen, 2008; Choudhri and Hakura, 2006; Campa and Goldberg, 2005; Dornbusch, 1987).
A parallel theoretical literature has sought to explain sign asymmetric pass-through into
import prices whereby depreciations and appreciations need not be reflected to the same extent
in import prices. Intuitively, this phenomenon will arise if the relative incentives of exporters to
adopt LCP or PCP strategies differ between phases of appreciation and depreciation. According
to the capacity constraints theory, if exporting firms are operating at or near full capacity they
cannot easily accomodate the surge in demand that could result from an appreciation of the
currency in their destination market. In such a setting, exporters may rationally choose not to
pass on appreciations. This is consistent with short-run downward import price stickiness and
2
For example, although the Finnish banking crisis led to a cumulative depreciation of 29% between 1991 and
the beginning of 1993, the average rate of CPI inflation over the same period was just 3.5%. Similarly, despite a
50% depreciation of the Korean Won between 1996 and the beginning of 1998, the average rate of inflation was
just 5%.
3
the notion that prices rise faster than they fall (Peltzman, 2000).
The market share theory posits that foreign firms seeking to gain or defend market share
may be quite willing to pass appreciations into import prices in order to quote competitive
prices (Krugman, 1987; Marston, 1990). The level of competition faced by the exporter in the
local market may also influence its pricing policy and the degree of pass-through asymmetry. For
instance, an exporter with considerable market share in a given destination country may exercise
its pricing power by opting to only pass on depreciations so that its margins are not eroded by
exchange rate fluctuations (Bussière, 2007). By contrast, in a more competitive environment,
exporters have stronger incentives to pass on appreciations in order to preserve or increase their
market share. As the decision to absorb local currency depreciations entails a narrowing of the
exporter’s margin, such pricing policies are unlikely to be pursued systematically in the long-run.
By contrast, the decision of whether or not to pass on appreciations can be viewed equivalently
as a decision of whether or not an exporter should allow its margin to widen, and so exporters
may view pricing policies that exploit appreciations as a viable long-run strategy.
The technology switching theory advanced by Ware and Winter (1988) suggests that appreciations will be passed on more strongly than depreciations if exporters can strategically alter
the source of their production inputs (e.g., by switching from foreign to domestic sources and
vice versa) and the type of production technology. Since technology switching (even at no cost)
takes time to be implemented and contracts with input providers are likely to have a fixed term,
this mechanism can explain asymmetries predominantly in the long-run.
Despite these various theoretical explanations of asymmetric ERPT, the existing empirical
literature is surprisingly sparse and narrowly focused both in terms of the range of countries
considered and the methodology employed. The majority of existing studies focus on a few
countries and only consider asymmetries in the short-run. Examples include Herzberg et al.
(2003) for the UK, Marazzi et al. (2005) and Pollard and Coughlin (2004) for the U.S. at industry
level, Khundrakpam (2007) for India and Bussière (2007) for the G7 economies. Furthermore,
the findings of these studies are mixed and so a general consensus is yet to emerge.3
We are aware of only a few papers that consider long-run ERPT asymmetry in a coherent
manner. Webber (2000) uses partial sum decompositions to analyse asymmetric long-run passthrough of exchange rates into import prices for 8 Asian economies. His results strongly suggest
3
For example, Pollard and Coughlin (2004) document short-run sign asymmetry for about half of 30 industries
studied using data spanning the period 1978-2000 but the direction of the effect is ambiguous. By contrast, based
on country-level data from 1975 to 2001, Herzberg et al. (2003) cannot refute the hypothesis that the short-run
import ERPT mechanism is linear. Meanwhile, based on his investigation of short-run ERPT to both import and
export prices, Bussière (2007) stresses that asymmetries cannot be ignored.
4
that depreciations are passed through more powerfully than appreciations over the long-run.
More recently, Delatte and Lopez-Villavicencio (2012) employ the flexible nonlinear autoregressive distributed lag (NARDL) framework developed by Shin et al. (2013) to analyse exchange
rate pass-through asymmetry into the general price level of 4 developed economies – Germany,
Japan, the UK and the US – from 1980Q1 and 2009Q3. Their dynamic import price model
considers positive and negative partial sum processes of the exchange rate as driving factors
both over the short-run and the long-run. Their principal finding is that depreciations of the
local currency in Germany, Japan and the US are passed through to the general price level
more forcefully than appreciations in the long-run, a result which they associate with low levels
of competition and downward prices stickiness. Moreover, they note that their pass-through
coefficients are larger than those commonly reported on the basis of symmetric models.
Our contributions are threefold. First, we analyse a very rich dataset comprising 33 countries,
19 of which are developed markets (DMs) and 14 are emerging market (EMs). Despite the
growing importance of EMs in international trade, very few studies have as yet considered a
wide cross-section of both EMs and DMs.4 Furthermore, our analysis employs trade-weighted
foreign export prices to enhance the accuracy of our estimates. By contrast, existing studies
have typically proxied foreign export prices using consumer or producer price indices, or various
other cost measures for the exporting country.5 Secondly, at a methodological level, we provide
the first extension of the NARDL technique to the dynamic heterogenous panel data context in
order to simultaneously exploit the observed time and cross-section variation to achieve enhanced
inference. Finally, we investigate whether the characteristics of the importing market can explain
the extent of long-run ERPT asymmetry. We achieve this by both varying the composition of
groups within our panel models and also by estimating cross-sectional models where the level of
asymmetry as measured by the country-specific NARDL models is regressed on selected importer
characteristics.
Our results reveal that depreciations are passed-through into import prices more strongly
than appreciations in the long-run. Our panel exercises reveal that this phenomenon is robust
to different country groupings and that there is no significant difference between EMs and DMs.
This indirectly suggests that the fear of floating observed in many EMs may be unfounded.
Furthermore, in conjunction wit the fact of international trade in recent decades
Our cross-sectional analysis reveals a positive association between the extent of long-run
4
The only notable example of which we are aware is Webber (2000), who studies a small cross-section of 8
Asian economies including both emerging and developed markets.
5
For example, Bussière (2007) uses producer price indices and Marazzi et al. (2005) use consumer price indices.
5
ERPT asymmetry and the degree of import dependence, which indicates that the pricing decisions of exporting firms are influenced by their market power. However, this effect is less
pronounced when the importing economy enjoys greater freedom to trade internationally and
also when it is growing rapidly. The moderating effect of freedom to trade arises because greater
openness enhances competitions and subjects exporters to market discipline. Meanwhile, the
effect of the output gap is likely to reflect the opportunism of exporters that wish to gain market
share in a vibrant economy.
Our findings speak to policymakers as well as a broad literature on ERPT. Long-run ERPT
asymmetry raises a general concern about market power among exporters, where pricing-tomarket may be endemic in a setting of weak competition for many classes of traded goods. Our
finding that depreciations are more fully reflected than appreciations in import prices in the
long-run indicates that import prices exhibit downward rigidity. Where imports account for
a large share of the representative consumption basket, this downward stickiness will also be
reflected in the general price level, implying that ERPT may strongly influence both realised and
expected inflation. This may offer a partial explanation of the fact that inflation did not decline
substantially in most countries during and after the global financial crisis despite significant
contractions in aggregate demand.
The asymmetry arising from ERPT may complicate the conduct of monetary policy and
may impair the ability of exchange rate changes to correct trade imbalances. The absence
of exchange rate targeting in many developed economies means that imported inflation will
largely depend on the interest rate decisions of the central bank, at least in the absence of nonconventional monetary policies. Therefore, if ERPT exhibits asymmetry, then the impact of
monetary policy on inflation will also be asymmetric. Finally, our analysis of country variation
in the extent of asymmetry reveals a potentially important policy response. By enhancing trade
freedom, economies can subject exporters to greater market discipline, reducing the scope for
rent-seeking pricing behaviour in the form of ERPT asymmetry.
The paper proceeds as follows. Section 2 discusses the data used in estimation of the NARDL
models and the variables that we consider as drivers of asymmetric ERPT in the panel and crosssectional regressions. Section 3 introduces our estimation framework and Section 4 presents our
results. Section 5 concludes and draws out the policy implications of our research, while further
details of the dataset and bootstrapping techniques may be found in the Appendices.
6
2
Data Description
2.1
Countries and Key Variables
Our analysis proceeds in three stages. First, we obtain individual measures of pass-through
by estimating dynamic NARDL models on a country-by-country basis following the technique
developed by Shin et al. (2013). Second, we jointly exploit the time-series and cross-section
dimensions of our dataset to obtain panel pass-through measures for various country groupings
using the Mean Group estimator of Pesaran and Smith (1995). Finally, we conduct crosssectional regressions to identify the drivers of asymmetric ERPT.
We focus on the following 33 importing economies (14 EMs and 19 DMs): Argentina, Australia, Belgium/ Luxembourg, Brazil, Canada, Chile, China, Colombia, Czech Republic, Denmark, Finland, France, Germany, Greece, Hong Kong, Hungary, Ireland, Israel, Italy, Japan,
Korea, Mexico, Netherlands, Norway, New Zealand, Singapore, South Africa, Spain, Sweden,
Switzerland, Thailand, the UK, and the US. Collectively, these countries accounted for 69% of
world imports in 2010, with 48% attributable to the DMs and 21 percent to the EMs.
For each country, i = 1, 2, . . . , 33, we gather quarterly data on three variables that will be
used to estimate country-specific ERPT equations. Firstly, the exchange rate, si,t , is the local
(importer’s) currency price of a unit of the foreign (exporter’s) currency. This is computed as
si,t = 1/N EER, where N EER is the nominal effective exchange rate index of foreign currency
per unit of domestic currency. Next, the domestic import price, pi,t , is an index measure of the
domestic price of goods and services at the dock proxied by customs unit value indices. Finally,
the effective foreign export price, p∗i,t , is an index measure of the foreign price of goods and
services coming into country i. Specifically, for each importing country i = 1, ..., 33, we compute
PJ(i) j j∗
p∗i,t = j=1 wi,t
pt where j = 1, ..., J(i) denote the trading partners of importing country i, and
j
wi,t
are the corresponding import shares.6 Hence, p∗i,t measures the ‘rest-of-the-world’ foreign
export price faced by country i, and is the same measure used in Brun-Aguerre et al. (2012).
The panel dataset is unbalanced and the longest available time span is from 1980Q1 to 2010Q4.
The precise sample periods are detailed in Appendix A.
Table 1 summarises the distribution of logarithmic quarter-on-quarter changes in the exchange rate, import and export prices. As expected, we observe considerably more volatility
among the EMs than the DMs. For example, the standard deviation of exchange rate changes
6
Import shares were calculated using the IMF’s Direction of Trade Statistics. Where a trading partner’s export
prices were unavailable, they were replaced with aggregate IMF export unit value indices for developing, emerging
or oil exporting countries.
7
reaches its highest value at 30.48% for Argentina relative to its lowest value of just 1.43% for
Belgium. Similarly, the largest quarter-on-quarter depreciation across all countries stands at
19.54% for Brazil while the largest one among the DMs is 1% for Greece. A similar pattern can
be observed for quarter-on-quarter import price changes.
[Insert Table 1 around here]
The columns in Table 1 labelled ‘depr(+)’ and ‘appr(-)’ provide a count of the observations
in which the exchange rate appreciated and depreciated, respectively. On average, the two
counts are similar; for DMs, the proportion of depreciation quarters is 34–65% across countries
while the corresponding range for EMs is 31–72%. This is important, because our application of
the NARDL framework will identify regimes according to partial sum processes which separate
appreciations from depreciations. Therefore, to achieve reliable finite sample inference, the
respective regime probabilities should not deviate substantially from 50%. Finally, the table
reports two well-known unit root tests applied to the variables in log levels, the ADF test where
the null hypothesis is unit root behaviour against the alternative of stationarity, and the KPSS
test where the two hypotheses are reversed. In conjunction with (unreported) results of the two
tests applied to log first-differences, the table confirms that the three variables are difference
stationary.
2.2
Importer Characteristics as Drivers of Pass-Through Asymmetry
The country-specific ERPT estimates from the NARDL models are likely to reveal some elements
of commonality across countries as well as some heterogeneity in the extent of asymmetry. This
raises the question of which factors drive this cross-sectional variation. We consider the following
variables which are drawn from the existing literature on ERPT, and investigate whether they
have an impact on asymmetry:
(i) The binary variable Emerging (equal to 1 for EMs and 0 for DMs) allows us to distinguish
between EMs and DMs, where the resulting groups can be seen in Table 1.7 This will allow
us to test whether the fear of floating prevalent among EMs is substantiated by differences
in ERPT to EMs and DMs.
7
Our classification follows The Economist’s listing which we adopt because of its emphasis on the real economy.
For our sample, the lists by The Economist and the IMF’s World Economic Outlook (October 2008) coincide.
However, the classification of some of the countries is controversial: Hong Kong, Singapore and Israel are classified
as DMs by MSCI Barra and FTSE but as EMs by the IMF and J.P. Morgan; South Korea is listed as a DM by
the FTSE but as an EM by the MSCI and IMF.
8
(ii) Dornbusch’s 1987 theoretical model of price discrimination links the extent of ERPT to
a given destination market with the number of importing firms relative to the number of
local producers, and predicts stronger ERPT in small import-dependent markets. Import
dependence can be proxied as IDit ≡
Mi,t
GDPi,t −Xi,t ,
where Mi,t is the total value of imports,
GDP is nominal output, and Xi,t is the total value of country i’s exports.
(iii) Froot and Klemperer (1989) argue that exchange rate volatility may be negatively related
to ERPT in a competitive export environment, as exporters are prepared to absorb fluctuations in order to quote competitive prices to maintain or increase their market share. By
contrast, if exporters seek to stabilize their profit margins they will tend to engage in PCP,
resulting in a positive relationship (Engel, 2006). As noted by Gaulier et al. (2008), this
ambiguity can be linked to the exporter’s trade-off between stabilising marginal profits or
export volumes. Furthermore, exporters’ perceptions about whether exchange rate shocks
will be transitory or persistent will influence their pricing decisions, with greater perceived
persistence strengthening ERPT. We therefore compute a quarterly measure of FX volatility based on the realised standard deviation of daily foreign exchange returns such that
qP
N EERj
D
2
RVitF X ≡
j=1 [log( N EERj−1 )] , where D is the number of days in each quarter.
(iv) The output gap measures the stage of the country-specific business cycle, with a positive
gap implying that the economy is running above potential. Choudhri and Hakura (2006)
note that lower ERPT may be observed in this context if exporting firms try to ‘fill the
gap’ by absorbing exchange rate fluctuations in their profit margins in order to increase
sales. The output gap is computed as
∗
GDPi,t −GDPi,t
∗
GDPi,t
∗ , is
× 100 where trend output, GDPi,t
estimated using the HP filter.
(v) GDP per capita (denominated in thousands of nominal US$) measures the wealth of the
importing country. Wealth may influence ERPT if exporters price-to-market. The import
profile of wealthier economies is likely to differ from that of poorer economies, with nonessential goods whose demand is likely to be relatively price elastic accounting for a larger
proportion of imports. Furthermore, wealthier markets are likely to be more strongly contested and their participants better informed and more footloose than in poorer markets.
Overall, ERPT to wealthy markets may be weaker and exporters may be more willing to
absorb depreciations in order to defent their market share.
(vi) Demand for commodities is highly inelastic in the short-run due to habit formation, sunk
costs and the costs associated with technology-switching. This may promote rent-seeking
9
by exporters via asymmetric pass-through. In order to rank the countries in our sample
according to the influence of commodity prices on their exchange rate, we first regress
∆sit on a constant and quarterly logarithmic changes in a broad commodity index. We
run three regressions for each country in our sample using the following commodity spot
price indices from Datastream: (i) the Goldman Sachs Commodity Index (GSCI); (ii) the
Dow Jones-UBS Commodity Index (DJ-UBSCI); and (iii) the Thomson Reuters/Jefferies
CRB Index (TR/J CRB). We then average the slope coefficients across the three models to
compute our commodity importer measure. A negative coefficient signifies a ‘commodity
currency’, while a positive value indicates that a country is a net importer of commodities.
We are therefore able to rank the countries in ascending order of their slope coefficients.
(vii) Trade freedom captures the extent of frictions to international trade introduced by tariff
structures, trade quotas, inefficient and/or corrupt administration, capital controls etc.
We employ Component 4 of the Economic Freedom of the World index constructed by
Gwartney et al. (2012), which is an index is bounded between 0 and 10, where higher
values indicate greater freedom to trade. We expect that greater freedom will be linked
with stronger competition, thereby subjecting exporters to greater market discipline and
reducing the scope for opportunism.
(viii) Price level inflation serves as a link between ERPT and monetary policy. High levels
of inflation are associated with elevated uncertainty and also with stronger pass-through.
Importing economies whose monetary authority has lost credibility typically experience
high and volatile inflation and high levels of ERPT (Choudhri and Hakura, 2006; Taylor,
2000; Brun-Aguerre et al., 2012). We consider both the annual CPI inflation rate as well
as the volatility of inflation defined as its standard deviation over the sample period.
3
3.1
Pass-Through Parameter Estimation and Hypotheses Tests
Time-Series Analysis
Since the time dimension (T ) of our sample is large, we allow for full country (or unit-specific)
heterogeneity by estimating a separate empirical pass-through equation for each country. We
adopt the flexible NARDL modelling approach proposed by Shin et al. (2013). The NARDL
modelling framework involves the decomposition of the effective exchange rate into si,t ≡ si,0 +
10
−
s+
i,t + si,t where si,0 is an arbitrary initial value and
s+
i,t =
t
X
∆s+
i,j =
j=1
t
X
max (∆si,j , 0) , s−
i,t =
j=1
t
X
∆s−
i,j =
j=1
t
X
min (∆si,j , 0) ,
(3.1)
j=1
which are partial sum processes which accumulate positive and negative exchange rate changes,
thereby separating out periods of depreciation of the domestic currency (captured by s+
i,t ) from
periods of appreciation (captured by s−
i,t ). The initial value si,0 can be set to zero without loss
of generality.8 The NARDL model is built around the asymmetric long-run equilibrium relation
− −
∗
pi,t = βi+ s+
i,t + βi si,t + γi pi,t + ui,t ,
(3.2)
where (βi+ , βi− , γi )0 is an unknown vector of long-run parameters and ui,t is a stationary zeromean error process that represents deviations from the long-run equilibrium. Note that we
may write ui,t = pi,t − pei,t where pei,t is the equilibrium value of the import price for country i
conditional on the values taken by the explanatory variables. Substituting (3.2) into the errorcorrection term, ρi (pi,t − pei,t ), of a standard linear ARDL(p, q, r) model yields the following
NARDL(p, q, r) model for the import price
− −
∗
∆pi,t = αi + ρi pi,t−1 + θi+ s+
i,t−1 + θi si,t−1 + λi pi,t−1
+
p−1
X
j=1
ϕi,j ∆pi,t−j +
q−1 X
+
πi,j
∆s+
i,t−j
j=0
+
−
πi,j
∆s−
i,t−j
+
r−1
X
φi,j ∆p∗i,t−j + εi,t . (3.3)
j=0
where the long-run import ERPT parameters for the ith importing country are given by βi+ =
−θi+ /ρi , βi− = −θi− /ρi while γi = −λi /ρi captures the long-run relation between import and
export prices. As usual, εi,t ∼ i.i.d.(0, σi2 ).9 We estimate equation 3.3 using the lag structure
−
8
The construction of the two partial sum processes, s+
i,t and si,t , relies on the assumption of a zero threshold
defined in terms of ∆si,t . As discussed in Greenwood-Nimmo et al. (2013), this assumption can be relaxed to
accommodate one or more unknown threshold parameters which must be estimated. However, the generality
that could be gained in this way would come at the expense of the straightforward interpretation in terms of
appreciations and depreciations that is a central feature of our analysis.
9
Note that the derivation of the NARDL model detailed in Shin et al. (2013) employs a marginal DGP for
Pmax(q,r)−1
−
∗ 0
∆xi,t = (∆s+
Λi,j ∆xi,t−j + v i,t , and then conditions on v i,t to
i,t , ∆si,t , ∆pi,t ) of the form ∆xi,t =
j=1
control for any contemporaneous correlation between xi,t and the residuals of the unconditional NARDL model.
To conserve space, we do not repeat the derivation here but merely note that (3.3) is equivalent to the error
correction representation of the conditional model given in equation (2.10) of Shin et al. (2013). By virtue of its
conditional specification, the NARDL model provides valid estimation and inference in the presence of weakly
endogenous explanatory variables, with the caveat that it is not possible to identify contemporaneous causal
effects between the elements of ∆xi,t and ∆pi,t without making further assumptions. In the current context, it
is necessary to assume that exchange rate changes and export price changes can exert a contemporaneous effect
on import prices but not vice-versa. We do not consider this to be an unreasonable assumption. In any case, no
such assumption is required when working with the long-run parameters around which the large majority of our
11
p = q = r = 2 for all countries since it generally suffices to whiten the residuals.10 With this lag
structure in place, the relevant parameters capturing the short-run responses of import prices
+
+
−
− 0
to exchange rate fluctuations are (πi,0
, πi,1
, πi,0
, πi,1
) , while the remaining short-run parameters
are (ϕi,1 , φi,0 , φi,1 )0 .
To assess the significance of the long-run equilibrium relation, one can employ either the
FP SS bounds-test of Pesaran et al. (2001) or the tBDM test of Banerjee et al. (1998). The
former is a non-standard F -test of the joint restriction H0 : ρi = βi+ = βi− = γi = 0 in (3.3),
while the latter is a non-standard t-test of the single restriction H0 : ρi = 0 against the alternative
HA : ρi < 0. Pesaran et al. provide critical value bounds for both test statistics allowing for
combinations of I(0) and I(1) variables. Bounds-testing permits reliable inference regarding
the existence of a long-run levels relationship despite the variety of time-series properties that
may be observed when working with partial sum decompositions in practice. In light of the
−
various dependence structures that may exist between s+
i,t and si,t (including cointegration),
Shin et al. (2013) conclude that a conservative approach is to use the critical values tabulated in
Pesaran et al. (2001) by counting the number of stochastic regressors in the model prior to their
decomposition. Alternatively, one may compute empirical p-values for the PSS test by means
of a bootstrap.
The NARDL framework (3.3) is ideally suited to the empirical analysis of pass-through
equations because it is nonlinear-in-variables but linear-in-parameters and, as such, it is readily
estimable by OLS. Nevertheless, it can accommodate asymmetry both in the short- and long-run
responses of import prices to exchange rate shocks. Furthermore, (3.3) nests a number of simpler
pass-through models which facilitate statistical discrimination between the possible combinations
of short- and long-run asymmetry. One case arises under the restriction βi+ = βi− = βi , which
implies that long-run ERPT is symmetric. Meanwhile, cumulative short-run ERPT is symmetric
P
Pq−1 −
+
+
− 11
if q−1
j=0 πi,j =
j=0 πi,j , while ERPT is symmetric on impact if πi,0 = πi,0 .
For each importing economy in our sample, we evaluate an array of hypotheses concerning
the strength of ERPT into import prices (from zero to complete) and its nature (linear or
analysis revolves.
10
For the resulting NARDL(2,2,2) model, the Ljung-Box portmanteau test only suggests residual autocorrelation
for Hungary and Chile. Repeating the estimation for these two countries using Newey-West heteroskedasticity
and autocorrelation consistent (HAC) standard errors we obtained no change in the inferences on significance of
the estimated coefficients for Chile. For Hungary, using HAC errors the coefficient on s+
i,t−1 is now significant
at the 10% level whereas it was insignificant with OLS standard errors. In neither case are the results of the
cointegration tests materially affected. Therefore we report results based on OLS standard errors throughout.
+
−
11
In addition, Shin et al. consider a more restrictive form of short-run symmetry defined as πi,j
= πi,j
for
j = 0, 1, . . . q − 1, where this pairwise symmetry restriction implies additive symmetry but not vice versa. We
employ the additive restriction because we wish to focus on the cumulative adjustment to an exchange rate shock
as opposed to the per-period adjustment.
12
asymmetric) in both the long- and the short-run. Focusing initially on the long-run, we formulate
the following hypotheses:
Hypothesis 1 (Zero long-run ERPT) H01+ : βi+ = 0 for depreciations and H01− : βi− = 0 for
1+
1−
appreciations vs. alternatives HA
: βi+ > 0 and HA
: βi− > 0, respectively.
Hypothesis 2 (Complete long-run ERPT) H02+ : βi+ = 1 for depreciations and H02− : βi− =
2+
2−
1 for appreciations vs. alternatives HA
: βi+ < 1 and HA
: βi1− < 1, respectively.
3 : β + 6= β − .
Hypothesis 3 (Symmetric long-run ERPT) H03 : βi+ = βi− vs. HA
i
i
Similarly, we formulate the following hypotheses pertaining to impact ERPT (i.e. ERPT in the
same quarter as the exchange rate shock) and to cumulative short-run ERPT:
−
+
= 0 for
= 0 for depreciations and H04− : πi,0
Hypothesis 4 (Zero impact ERPT) H04+ : πi,0
4+
+
4−
−
appreciations vs. alternatives HA
: πi,0
> 0 and HA
: πi,0
> 0, respectively.
+
−
Hypothesis 5 (Complete impact ERPT) H05+ : πi,0
= 1 for depreciations and H05− : πi,0
=1
−
+
5−
5+
: πi,0
< 1, respectively.
: πi,0
< 1 and HA
for appreciations vs. alternatives HA
+
−
6 : π + 6= π − .
Hypothesis 6 (Symmetric impact ERPT) H06 : πi,0
= πi,0
vs. HA
i,0
i,0
Pq−1 +
7−
Hypothesis 7 (Zero short-run ERPT) H07+ :
j=0 πi,j = 0 for depreciations and H0 :
Pq−1 −
7− Pq−1 −
7+ Pq−1 +
j=0 πi,j > 0,
j=0 πi,j > 0 and HA :
j=0 πi,j = 0 for appreciations vs. alternatives HA :
respectively.
Pq−1 +
Hypothesis 8 (Complete short-run ERPT) H08+ :
j=1 πi,j = 1 for depreciations and
P
P
q−1 −
8+
+
8− Pq−1 −
: q−1
H08− : j=0
πi,j = 1 for appreciations vs. alternatives HA
j=0 πi,j < 1 and HA :
j=0 πi,j <
1, respectively.
Hypothesis 9 (Symmetric short-run ERPT) H09 :
Pq−1 −
j=0 πi,j .
Pq−1
j=0
+
πi,j
=
Pq−1
j=0
−
9:
πi,j
vs. HA
Pq−1
j=0
+
πi,j
6=
These hypotheses can be tested using Wald- and t-type test statistics which converge to
their standard asymptotic distributions. A further appealing feature of the NARDL framework
is that it is straightforward to compute asymmetric cumulative dynamic multipliers recursively
13
from the parameters of the NARDL-in-levels representation of (3.3) as follows (see Shin et al.
for details)
ms+
i,h
≡
h
X
∂pi,t+j
j=0
∂s+
i,t
h
−→
βi+ ,
and
ms−
i,h
≡
h
X
∂pi,t+j
j=0
∂s−
i,t
h
−→ βi− , h = 0, 1, 2 . . . H.
(3.4)
In the context of ERPT, the dynamic multipliers can be used to trace the evolution of the import
price over periods h = 0, 1, 2, . . . , H in response to a unit depreciation or appreciation of the
domestic currency in period h = 0. Dynamic multiplier analysis complements the asymptotic
hypothesis tests outlined above in two important ways. First, it provides a robustness check to
verify that the asymptotic inferences are not compromised too severely by the finite samples with
which we work.12 Second, by constructing bootstrap confidence bands around the differential
−
multiplier m+
i,h − mi,h , one may draw inferences about the presence of asymmetry over any
desired time horizon h. Details of the bootstrap procedure may be found in Appendix B.
3.2
Panel Analysis
By exploiting both the time and cross-section dimensions of the sample, panel estimation should
increase the signal-to-noise ratio relative to the time-series estimation outlined above, yielding
more reliable inference. To this end, we apply the Mean Group (MG) estimator of Pesaran
and Smith (1995), in which the coefficients of the panel data model are computed as an equallyweighted average of the coefficient estimates derived from the country-specific NARDL models.13
We verify that our panel is free from cross section dependence using the unbalanced panel
formulation of the CD test statistic proposed by Pesaran (2004), which returns a value of 0.823.
In this setting, MG estimation allows for a simple and flexible random-coefficients formulation
of the NARDL equation (3.3) that allows for full country heterogeneity in the parameters.
+
+
−
− 0
Let Θi = (βi+ , πi,0
, πi,1
, βi− , πi,0
, πi,1
) denote a vector that gathers the relevant pass-through
coefficients for country i ∈ [1, Ng ], where Ng ≤ N is the number of countries in the group for
which the MG estimates are to be computed (we will return to the issue of group selection
12
Shin et al. conduct a range of Monte Carlo experiments which reveal that the Wald tests for asymmetry
(Hypotheses 3, 6 and 9) generally exhibit acceptable size and power properties for sample sizes T ≥ 100. Bootstrap
inference is therefore particularly valuable when T < 100.
13
We also employed the Swamy (1970) random coefficients estimator which gives less weight to the countryspecific pass-through coefficients that are estimated with larger standard errors. However, for some groupings, the
covariance matrix of the Swamy estimator was not positive definite in which case we followed Swamy’s suggestion
and replaced it with the MG covariance matrix. Given that the MG estimates and the Swamy estimates were
qualitatively similar in all cases, we limit our attention to the simpler MG estimator. The Swamy estimates are
available on request.
14
MG
shortly). The MG estimator Θ̄
MG
Θ̄
Ng
1 X
=
Θ̂i ,
Ng
and its covariance matrix V (Θ̄
MG
) are defined as
Ng
and V (Θ̄
i=1
MG
X
1
MG
MG
)=
(Θ̂i − Θ̄ )(Θ̂j − Θ̄ )0 .
Ng (Ng − 1)
(3.5)
i=1
Having computed the MG estimates for a desired grouping of the countries in our sample, it
is straightforward to evaluate Hypotheses 1 to 9 at the group level.14 Furthermore, we can
compute group-level cumulative dynamic multipliers as in (3.4), for which empirical confidence
bands can be computed by bootstrapping as detailed in Appendix B.
To establish a baseline, the first grouping that we consider contains all 33 countries in the
sample. We then consider alternative groupings based on the importer characteristics outlined
in Section 2.2. For each country, we average the available observations on a given characteristic
over the longest available balanced sample and then rank the 33 countries accordingly to form
two groups of roughly equal size (17 and 16, respectively). Figure 1 shows the country rankings,
while Appendix C reports the pairwise correlations among the resulting cross-sectional values of
the importer characteristics. The rankings are generally uncontroversial and reflect stylised facts
in the global economy. For instance, Argentina and Brazil exhibit the highest inflation rates
and the greatest exchange rate volatility over our sample period. Meanwhile, Switzerland and
the Scandinavian countries lead the rankings for GDP per capita. Lastly, ranking the countries
according to their status as net commodity importers/exporters successfully separates those
countries known to have commodity currencies (Australia, Canada, Brazil, Chile, New Zealand,
Norway and South Africa) from the major net commodity importers (China, Hong Kong, Japan
and the U.S.).
[Insert Figure 1 about here]
4
4.1
Empirical Results
Individual Pass-Through Estimates and Hypotheses Tests
According to the LOOP for traded goods, the existence of a long-run equilibrium between the
import price pit , export price p∗it and nominal exchange rate sit will prevent them from drifting
too far apart over prolonged periods. Table 2 provides cointegration test results in favour of
14
Note that the MG estimator obtains the long-run pass-through measures by averaging βbi+ and βbi− across
all member of the group, i = 1, ..., Ng . Effectively, this presumes that the long-run parameters of interest are
E(−θ+ /ρ) and E(−θ− /ρ) instead of E(−θ+ )/E(ρ) and E(−θ− )/E(ρ). For further discussion, see Pesaran and
Smith (1995).
15
this proposition, thereby supporting the established practice of modelling ERPT in an error
correction framework (e.g. Delatte and Lopez-Villavicencio, 2012; Brun-Aguerre et al., 2012;
Kozluk et al., 2008; Campa et al., 2008). At least one of the tBDM and FP SS statistics reject the
null hypothesis of no cointegration for the large majority of countries. Interestingly, the evidence
of cointegration weakens when the restriction of symmetric long-run ERPT (i.e. βi+ = βi− ) is
imposed during estimation of the NARDL model. An important example is the U.S., where we
find strong evidence of an asymmetric long-run relationship but no evidence of a linear long-run
relationship. This is consistent with Shin et al. (2013)’s observation that imposing long-run
symmetry when the true DGP exhibits long-run asymmetry will lead to bias in estimation and
severely compromised inference.
[Insert Table 2 about here]
The time-series estimates of the NARDL model 3.3 with lag structure p = q = r = 2 are reported
in Table 3 alongside relevant diagnostics and Wald tests for Hypotheses 1 to 9 as detailed in
Section 3.1. The models perform well in the majority of cases, with little evidence of residual
autocorrelation and respectable values of the R̄2 . Indeed, it appears that the NARDL model is
particularly successful at capturing the behaviour of import prices in the EMs, where the R̄2
varies between 0.428 (Singapore) and 0.913 (Argentina), while the equivalent range for the DMs
is 0.274 (Spain) and 0.786 (Australia).
[Insert Table 3 about here]
There is strong evidence of asymmetric pass-through in the long-run. We find that 18/33
countries (almost 55%) experience asymmetric long-run pass-through, of which 7 are EMs and
11 are DMs. Furthermore, in all but one case, we find that the pass-through associated with
depreciations exceeds that of appreciations in the long-run (i.e. β̂i+ > β̂i− ). This is a remarkably
strong result which accords with the findings of Delatte and Lopez-Villavicencio (2012) and
Webber (2000). We cannot reject the null hypothesis of complete long-run pass-through following
a depreciation for 19/19 DMs and 9/14 EMs, while the corresponding proportions in the case
of an appreciation are 10/19 DMs and 8/14 EMs. It appears that while depreciations are often
fully reflected in import prices in the long-run, this is often not the case for appreciations.
Furthermore, this contrast is particularly marked when the destination market is developed.
The implication is that, in the aggregate, exporters may be either unwilling or unable to absorb
adverse exchange rate fluctuations into their operating margins in the long-run.
16
The form of long-run asymmetry that we observe is suggestive of imperfect competition
among exporting firms and of limited pricing power of importers. Indeed, the fact that we
observe asymmetry in the long-run indicates that exporters may be able to maintain pricing
power in the long-run. This would be feasible if exporters are able to consistently differentiate
their products over time by means of innovation, quality enhancements or technological progress,
all supported by appropriate marketing strategies. Alternatively, our results could arise from
either tacit or explicit price collusion among exporters regarding their response to exchange rate
fluctuations.
Turning our attention to impact ERPT, we observe asymmetry for 12/33 countries (36%),
with an even split between DMs and EMs. In this case, there is no clear direction of the asymmetry – 4/19 DMs (21%) and 3/14 EMs (21%) show stronger impact ERPT following depreciations,
while 2/19 DMs (11%) and 3/14 EMs (21%) show the reverse pattern. In accordance with the
results adduced in a number of previous studies (e.g. Campa and Goldberg, 2005, and Marazzi
et al., 2005), we are unable to reject the null hypothesis of zero impact ERPT to import prices
in the U.S. In most other countries, we observe significant and often considerable impact ERPT.
Indeed, for 6/19 DMs (32%) and 8/14 EMs (57%) we find that depreciations are passed through
fully on impact. Meanwhile, the equivalent figures for appreciations are 7/19 DMs (37%) and
6/14 EMs (43%). Note that while long-run pass-through is more complete for DMs than EMs,
the reverse pattern is true on impact. This is highly suggestive of price-smoothing on the part
of exporters selling to developed markets. The choice to smooth prices for DMs but not EMs
may reflect the greater importance of menu costs in DMs which generally exhibit lower and
more stable rates of inflation than their EM counterparts. Furthermore, exporters may perceive
a higher price elasticity of demand in DMs resulting from such factors as lower information and
switching costs, stronger institutions and higher levels of effective competition.
The null hypothesis of zero cumulative short-run ERPT can be rejected for 11/14 EMs and
14/19 DMs in the case of depreciations and 11/14 EMs and 16/19 DMs for appreciations. We
find evidence of cumulative short-run pass-through asymmetry for 10/33 countries (30%). Once
again, there is an even split between EMs and DMs and no clear pattern of asymmetry, with
2/19 DMs (11%) and 2/14 EMs (14%) showing stronger short-run ERPT for depreciation and
3/19 DMs (16%) and 3/14 EMs (21%) showing the opposite.
Figures 2 and 3 plot the cumulative dynamic multipliers associated with a unit depreciation
−
(m+
i,h ) and appreciation (mi,h ) of the domestic exchange rate for each country in our sample.
−
The differential m+
i,h − mi,h is plotted together with its 90% empirical confidence band in order
17
to provide a measure of the statistical significance of asymmetry at any desired horizon from
h = 0 . . . 24 quarters. This provides additional evidence the significance of short- and long-run
asymmetries to supplement the hypothesis testing conducted above.
[Insert Figures 2 and 3 about here]
The general tendency for depreciations to be passed-through more strongly than appreciations
in the long-run is clearly borne out by the multipliers. However, short-run asymmetry is far less
pervasive and, where present, the direction is rather mixed. In a number of cases, we observe a
switching pattern whereby appreciations are passed-through more strongly in the short-run after
which the pass-through of depreciations strengthens as the horizon increases. Notable examples
include Canada, Japan the US, Hong Kong and Singapore. This group of countries includes
some of the world’s most lucrative export markets, which are populated with well-informed and
affluent agents that enjoy considerable freedom to trade. Hong Kong and Singapore are notable
both as regional centres and for their rapid development and the emergence of a wealthy middle
class during our sample period. In such a setting, exporters may be more willing to defend
or expand their market share in the short-run by narrowing their margins to absorb adverse
exchange rate fluctuations.
A second group of countries experiences asymmetric pass-through only over the long-run.
Members include wealthy DMs such as Australia, Belgium, Denmark, Finland, Sweden, Switzerland and the U.K. as well as South Korea, which is among the most developed of the EMs. Most
of these economies are wealthy, relatively small and well established in the sense that they were
already affluent before the start date of our sample. In these markets, exporters are likely to
face sufficient competition to prevent rent-seeking price adjustments in the short-run. However,
the perceived gains from absorbing adverse exchange rate fluctuations in the short-run may be
too small for such a strategy to be enacted in practice.
Another group of economies is subject to asymmetry in both the short- and long-run whereby
depreciations are passed-through more strongly than appreciations in both cases. Members of
this group include Argentina, China, Greece, Israel and Thailand. A conspicuous aspect which
is common to these markets is that they are small or subject to restrictive trade regulations. As
such, exporters selling to these markets are likely to face a relatively low degree of competition
and will therefore be relatively unconstrained in their ability to engage in short-run rent-seeking
behaviour. A final group of countries exhibits roughly symmetric pass-through both in the shortand long-run. Its members are Chile, Colombia, Ireland, Italy, Mexico, the Netherlands, New
18
Zealand and South Africa, many of which are notable as either net commodity exporters or
re-export locations.
4.2
Panel Pass-Through Estimates and Hypotheses Tests
Panel pass-through coefficient estimates and test results are summarized in Table 4, firstly for a
panel composed of all 33 countries and subsequently for smaller panels formed by partitioning
our sample into ‘high’ and ‘low’ groups with reference to the importer characteristics outlined
in Section 2.2. The corresponding cumulative dynamic multipliers are plotted in Figure 4. A
result that resonates across country groupings is that the depreciation pass-through is generally
complete and is significantly stronger than that of appreciations in the long-run. Our results
suggest that, in the long-run, exporters tend to absorb appreciations by keeping the price roughly
constant for the importer as this will widen their operating margins. By contrast, exporters
generally pass depreciations through to import prices, making their products more expensive in
the destination market but leaving their operating margins intact.
[Table 4 and Figure 4 around here]
We find little evidence of statistically significant differences in the long-run pass-through estimates between country groups. For instance, the tests cannot distinguish between the long-run
ERPT coefficients for high vs. low inflation economies or for high vs. low per capita GDP
economies. This is reflected by the cumulative dynamic multipliers reported in Figure 4, which
generally a very similar pattern in the large majority of cases. In fact, the only case in which we
observe a significant difference in long-run ERPT between groups is the comparison of the high
vs. low commodity import cohorts, where the ‘high’ commodity import group contains the more
significant commodity importers in our sample. Long-run asymmetry is more pronounced in
the group which relies on commodity imports, and this effect derives from significantly stronger
long-run pass-through of depreciations in this group relative to the low commodity import group.
This finding suggests that the import price of commodities may react particularly asymmetrically with respect to exchange rate shocks, perhaps reflecting the relatively inelastic nature of
commodity demand and the secular trend of increasing global commodity demand throughout
our sample.
Table 4 reveals weak evidence of impact asymmetry but no evidence of cumulative short-run
asymmetry in any of the 17 country groupings that we consider. Where impact asymmetry (i.e.
π + − π − 6= 0) is significant, it is because the contemporaneous effect of a depreciation on import
19
prices exceeds that of an appreciation, a pattern which is again suggestive of weak competition.
This impact effect is significant in 6/17 (35%) groupings that we consider, including the high
commodity import group and the group which has experienced large exchange rate fluctuations.
The stage of development, GDP per capita and the inflationary environment also seem to be
important determinants of impact asymmetry. These findings are clearly reflected in Figure 4,
where a significant impact response of the import price to depreciations can be seen in each
case.
It is clear that both the time-series and panel NARDL models overwhelmingly reject the
hypothesis of long-run symmetry in ERPT in favour of the alternative that depreciations are
passed through more vigorously than appreciations. This is an important finding, not least
because the majority of published empirical research to date which has modelled the long-run
equilibrium relationship has assumed it to be linear. Furthermore, our results indicate that
once one allows for long-run non-linearity, then the evidence of short-run asymmetry weakens
dramatically. We therefore conclude that long-run asymmetry is a pervasive phenomenon that
cannot be ignored in future research.
4.3
Cross-Sectional Variation in Long-Run Asymmetric Pass-Through
Given the pervasive evidence of long-run pass-through asymmetry, we seek to explain the extent
of its variation across countries. To this end, we estimate cross-sectional regressions by OLS
where the dependent variable is the long-run asymmetry measure LRiasy ≡ βi+ − βi− which is
derived from the country-specific NARDL models and which is positive for the large majority
of countries in our sample. The explanatory variables that we consider are the time averages of
the importer characteristics introduced in Section 2.2 defined over the longest common sample
period. The results are recorded in Table 5.
[Table 5 around here]
While we are mindful of omitted-variable bias, we begin by estimating simple univariate crosssectional regressions to evaluate the effect of each variable in isolation. Import dependence
stands out as the most important driver of long-run asymmetry, with greater import dependence
corresponding to stronger long-run pass-through of depreciations than appreciations. Next, we
estimate a multiple regression including all of the importer characteristics. The results, reported
in Section B of the Table 5, confirm the central role of import dependence. To ensure that the
inclusion of the rate of inflation and/or its volatility among the explanatory variables does
20
not introduce significant endogeneity bias into estimation15 , we also report multiple regressions
excluding the two inflation variables. As can be seen in Table 5, the main results do not change
in a qualitatively important manner.
Building on these results, we explore the role of import dependence allowing for potential
interactions with other characteristics of the importing economy. Specifically, we consider a
general model in which import dependence enters linearly and also multiplicatively interacted
with all of the other explanatory variables, thereby allowing for non-constant marginal effects.
Significant coefficients on the interaction terms imply that the link between import dependence
and long-run pass-through asymmetry is attenuated/exacerbated by the economic characteristics
of the importing country. The results are recorded in Section C of Table 5.
Our results reveal that the positive association between import dependence and long-run
asymmetry is moderated by increased freedom to trade internationally. This is an intuitively
pleasing result, as an economy in which agents enjoy greater freedom to trade will be one in
which exporters face stronger competition and therefore have less opportunity to pass-through
depreciations more strongly/rapidly than appreciations without fear of losing market share.
Another significant interaction occurs between import dependence and the output gap, and
again exerts a moderating effect. This suggests that exporters pricing policies may take account
of the country-specific stage of the business cycle. In particular, where an destination market is
growing above potential, exporters may be seek to quote more competitive prices in the long-run
in order to gain market share over time. Lastly, it is interesting to note that we find no evidence
of significant differences in the extent of asymmetries for EMs and DMs, a result which supports
the view that the ‘fear of floating’ of many EMs has little economic justification.
Our cross-sectional analysis raises an important issue as the degree of trade freedom is
essentially a policy choice that has significant ramifications in our analysis. Therefore, to reduce
the welfare cost of asymmetric long-run ERPT, governments of import-dependent economies
may wish to pursue greater trade openness to promote effective competition and the market
discipline that it brings. In this way, they could reduce the scope for exporters to engage in
rent-seeking pricing strategies to the benefit of local consumers.
15
This may arise when depreciations are passed through more strongly than appreciations in the long-run, as
this would be expected to eventually feed into the general price level and therefore also into the volatility of price
level inflation.
21
5
Summary and Policy Implications
A thorough understanding of the nature and extent of ERPT into import prices is central to
the analysis of global trade imbalances, the conduct of monetary policy, and the appropriate
choice of exchange rate regime. Although the literature on the subject is ample, no existing
study has yet investigated sign-asymmetries in both the long-run equilibrium and the shortrun dynamics for a large sample of developed and emerging economies. This paper addresses
this lacuna by estimating nonlinear error-correction models that accommodate asymmetry in
both the short- and long-run behaviour of import prices for a large panel of 33 developed and
emerging economies. Careful analysis of the cumulative dynamic multipliers associated with
these models provides an illuminating summary of the traverse from initial equilibrium to the
new equilibrium following either a unit appreciation or depreciation of the domestic currency,
thereby clearly depicting asymmetries wherever they arise.
Our results overwhelmingly show that exporters engage in asymmetric pass-through in the
long-run. Specifically, the long-run pass-through associated with depreciations exceeds that of
appreciations. This effect is significant for the majority of countries in the time-series analysis
and is confirmed as a pervasive phenomena by panel estimation across a wide variety of country
groupings. Although our results reveal that long-run ERPT asymmetry is not confined to a
specific group of countries, further cross-sectional analysis reveals that the effect is stronger for
import dependent economies.
Overall, our results are indicative of weak competition structures in international trade, a
situation that allows exporting firms to extract rents from exchange rate fluctuations. Consequently, the response of import prices to currency changes is upwardly skewed, which may
translate into downward nominal rigidity, higher inflation rates over time and welfare losses for
consumers in the importing economy. This offers a partial explanation for the puzzling observation that, in many countries during the great recession, inflation did not fall as much as many
commentators expected it should despite anaemic global demand.
Our results raise a general concern about market concentration among exporters as well
as price discrimination between destination markets. These are important issues which will
be very difficult to regulate given their inherently multi-jurisdictional nature. Furthermore,
there are significant implications for the design and conduct of monetary policy, as domestic
policies aimed at controlling inflation and managing aggregate demand may be undermined by
asymmetric responses of import prices to exchange rate fluctuations, particularly where these
22
phenomena are poorly understood. Our analysis identifies an important policy response that
could help to reduce the extent of long-run ERPT asymmetry. Since the positive link between
the cross-sectional variation in import dependence and long-run asymmetry weakens with greater
trade openness, trade reform and liberalisation may be an effective mechanism for importing
countries to mitigate the effect of asymmetric ERPT. Finally, the similarity of our results for
both developed and emerging economies provides indirect evidence that the ‘fear of floating’ of
many emerging economies may be unwarranted on economic grounds. Therefore, by allowing
their exchange rates to float more freely, policymakers in emerging economies would gain greater
freedom to conduct policy in accordance with domestic stabilisation goals.
References
Banerjee, A., Dolado, J. and Mestre, R. (1998), “Error-Correction Mechanism Tests for Cointegration in a Single-Equation Framework”, Journal of Time Series Analysis, 19(3), 267-283.
Brun-Aguerre, R., Fuertes, A.-M, and Phylaktis, K. (2012), “Country and Time Variation
in Import Pass-Through: Macro versus Micro Factors”, Journal of International Money and
Finance, 31(4), 818-844.
Bussière, M. (2007), “Exchange Rate Pass-Through to Trade Prices: The Role of Nonlinearities
and Asymmetries”, Working Paper No. 822, European Central Bank.
Bussière, M. and Peltonen, T. (2008), “Exchange Rate Pass-Through in the Global Economy:
The Role of Emerging Market Economies”, Working Paper No. 951, European Central Bank.
Campa, J. M. and Goldberg, L. S. (2005), “Exchange Rate Pass-Through into Import Prices”,
The Review of Economics and Statistics, 87, 679-690.
Campa, J. M., Gonzalez Minguez, J. M. and Barriel, M.S. (2008), “Non-Linear Adjustment of
Import Price in the European Union”, Working Paper No. 347, Bank of England.
Choudhri, E.U., and Hakura, D.S. (2006), “Exchange Rate Pass-Through to Domestic Prices:
Does the Inflationary Environment Matter?” Journal of International Money and Finance,
25, 614-639.
Delatte, A.-L. and Lopez-Villavicencio, A. (2012), “Asymmetric Exchange Rate Pass-Through:
Evidence from Major Countries”, Journal of Macroeconomics, 34, 833-844.
Dornbusch, R. (1987), “Exchange Rates and Prices”, American Economic Review, 77, 93-106.
Engel, C. (2006), “Equivalence Results for Optimal Pass-Through, Optimal Indexing to Exchange Rates, and Optimal Choice of Currency for Export Pricing”, Journal of the European
Economic Association, 4, 1249-1260.
Frankel, J., Parsley, D. and Wei, S.J. (2005), “Slow Pass-Through Around the World: A New
Import for Developing Countries?” NBER Working Paper No. 11199.
23
Froot, K. and Klemperer, P.D. (1989), “Exchange Rate Pass-Through when Market Share
Matters”, American Economic Review, 79, 637-654.
Gaulier, G., Lahrèche-Révil, A. and Méjean, I. (2008), “Exchange Rate Pass-Through at the
Product Level”, Canadian Journal of Economics, 41, 425-449.
Goldfajn, I. and Werlang, R. (2000), “The Pass-Through from Depreciation to Inflation: A
Panel Study”, Working Paper No. 5, Banco Central do Brasil.
Greenwood-Nimmo, M.J., Shin, Y. and Van Treeck, T. (2013), “The Asymmetric ARDL Model
with Multiple Unknown Threshold Decompositions: An Application to the Phillips Curve in
Canada”, Mimeo: The University of Melbourne.
Gwartney, J.D., Hall, J.C. and Lawson, R. (2012), “Economic Freedom of the World: 2012
Annual Report”. The Fraser Institute, Vancouver, BC.
Herzberg, V., Kapetanios, G. and Price, S. (2003), “Import Prices and Exchange Rate Passthrough: Theory and Evidence from the United Kingdom”, Working Paper No. 182, Bank of
England.
Hsiao, C., Pesaran, M.H. and Tahmiscioglu, A.K. (1999), “Bayes Estimation of short-run
Coefficients in Dynamic Panel Data Models”. In C. Hsiao, K. Lahiri, L.-F. Lee and M.H.
Pesaran (Eds.) Analysis of Panels and Limited Dependent Variable Models: A Volume in
Honour of G.S. Maddala. Cambridge: Cambridge University Press.
Khundrakpam, J.K. (2007), “Economic Reforms and Exchange Rate Pass-Through to Domestic
Prices in India”, Working Paper No. 225, Bank for International Settlements.
Kozluk, T., Banerjee, A. and De Bandt, O. (2008), “Measuring long-run Exchange Rate PassThrough”, Economics, The Open-Access, Open-Assessment E-Journal, 2(6), 1-36.
Kreinin, M. E. (1977), “The Effect of Exchange Rate Changes on the Prices and Volume of
Foreign Trade”. IMF Staff Papers, 24, 297-329.
Krugman, P. (1987), “Pricing to Market When the Exchange Rate Changes”. In S.W. Arndt and
J.D. Richardson (Eds.) Real-Financial Linkages among Open Economies. Cambridge (MA):
MIT Press, 49-70.
Marazzi, M., Sheets, N., Vigfusson, R.J., Faust, J., Gagnon, J., Marquez, J., Martin, R.F.,
Reeve, T. and Rogers, J. (2005), “Exchange Rate Pass-Through to U.S. Import Prices: Some
New Evidence”, International Finance Discussion Paper No. 833, Board of Governors of the
Federal Reserve System.
Marston, R. (1990), “Pricing to Market in Japanese Manufacturing”, Journal of International
Economics, 29, 217-236.
Peltzman, S. (2000), “Prices Rise Faster than They Fall”, Journal of Political Economy, 108,
466-502.
Pesaran, M.H. and Shin, Y. (1998), “An Autoregressive Distributed Lag Modelling Approach
to Cointegration Analysis”. In S. Strom (Ed.) Econometrics and Economic Theory: The Ragnar Frisch Centennial Symposium. Econometric Society Monographs. Cambridge: Cambridge
University Press, 371-413.
24
Pesaran, M.H. and Smith, R. (1995), “Estimating Long-Run Relationships from Dynamic
Heterogeneous Panels”, Journal of Econometrics, 68, 79-113.
Pesaran, M.H., Shin, Y. and Smith, R.J. (2001), “Bounds Testing Approaches to the Analysis
of Level Relationships. Journal of Applied Econometrics, 16, 289-326.
Pesaran, M.H. (2004), “General Diagnostic Tests for Cross Section Dependence in Panels”,
Cambridge Working Papers in Economics No. 0435.
Pollard, P.S. and Coughlin, C.C. (2004). Size Matters: Asymmetric Exchange Rate Pass-Through
at the Industry Level”. Working Paper No. 2003-29C, Federal Reserve Bank of St. Louis.
Shin, Y., Yu, B. and Greenwood-Nimmo, M.J. (2013), “Modelling Asymmetric Cointegration
and Dynamic Multipliers in a Nonlinear ARDL Framework”. In W.C. Horrace and R.C. Sickles
(Eds.) Festschrift in Honor of Peter Schmidt. New York (NY): Springer Science & Business
Media. Forthcoming.
Swamy, P.A.V.B. (1970), “Efficient Inference in a Random Coefficient Regression Model”,
Econometrica, 38, 311-323.
Taylor, J. B. (2000), “Low Inflation, Pass-Through, and the Pricing Power of Firms”, European
Economic Review, 44 (7), 1389-1408.
Ware, R. and Winter, R. (1988), “Forward Markets, Currency Options and the Hedging of
Foreign Exchange Risks”, Journal of International Economics, 25, 291-302.
Webber, A. (2000), “Newton’s Gravity Law and Import Prices in the Asia Pacific”, Japan and
the World Economy, 12, 71-87.
25
26
Notes: To conserve space we have made use of the following abbreviations: G/D/T stands for GSCI/DJ-UBSCI/TR-J CRB; JPM stands for J.P. Morgan; NSO stands
for National Statistics Office; ME stands for Ministry of Economics; CM stands for Commerce Ministry; NCB stands for National Central Bank and Fraser stands for
The Fraser Institute. All remaining abreviations are widely known. Import and export prices are trade-weighted indices. NEER is the nominal effective exchange rate
of foreign currencies per unit of domestic currency.
APPENDIX A: Data Sources
APPENDIX B: Bootstrap Confidence Bands for Multipliers
The cumulative dynamic multipliers and associated non-parametric bootstrap intervals plotted
in Figures 2 and 3 are computed as follows:
1. Estimate the NARDL model (3.3) for country i by OLS to obtain the regression residubi =
als, ˆi,t , t = 1, ..., Ti and the vector of relevant pass-through parameter estimates Θ
−
+
+
+
+
−
−
0
(βbi , π
bi,0 , π
bi,1 , βbi , π
bi,0 , π
bi,1 ) from which the cumulative dynamic multipliers mi,h and m−
i,h
are computed. Note that the sample length for country i is denoted Ti to reflect the
differences in the estimation samples across countries detailed in Table 1.
2. Resample from ˆi,t with replacement and denote the vector of resampled residuals for
(b)
country i by ˆi,t , t = 1, ..., Ti .
(b)
3. Generate the bootstrap sample for country i, ∆pi,t , t = 1, ..., Ti , by recursion as follows,
taking the explanatory variables as given:
(b)
∆pi,t
(b)
b− −
b ∗
= α
bi + ρbi pi,t−1 + θbi+ s+
i,t−1 + θi si,t−1 + λi pi,t−1
+
p−1
X
j=1
(b)
ϕ
bi,j ∆pi,t−j
+
q−1 X
+
π
bi,j
∆s+
i,t−j
j=0
+
−
π
bi,j
∆s−
i,t−j
+
r−1
X
(b)
φbi,j ∆p∗i,t−j + εbi,t
j=0
(b)
4. Re-estimate the NARDL model for country i using the bootstrap sample ∆pi,t to obtain
+(b)
+(b)
−(b)
−(b)
−(b)
b (b) = (βb+(b) , π
bi,0 , π
bi,1 , βbi , π
bi,0 , π
bi,1 )0 .
the bootstrap parameter vector Θ
i
i
5. Compute the cumulative dynamic multipliers for country i using the bootstrap parameter
vector.
6. Repeat steps 2–5 B times and compute empirical confidence intervals of any desired width
around the cumulative dynamic multipliers obtained at step 1 in the usual way. Repeat
the process for all countries i = 1, 2, . . . , N .
The panel cumulative dynamic multipliers and the associated confidence bands plotted in Figure
4 can be computed as follows:16
7. Compute the mean of the N country-specific pass-through parameter estimates Θ̄ =
(β̄ + , π̄0+ , π̄1+ , β̄ − , π̄0− , π̄1− )0 from which the panel dynamic multipliers can be obtained.
8. Note that steps 2–4 have already been carried out B times for the full set of countries i =
1, 2, . . . , N . Now, for each bootstrap sample, b = 1, 2, . . . , B, compute the mean of the N
(b)
+(b)
+(b)
−(b)
−(b)
country-specific bootstrap coefficient vectors to obtain Θ̄ = (β̄ +(b) , π̄0 , π̄1 , β̄ −(b) , π̄0 , π̄1 )0 .
9. Compute the cumulative dynamic multipliers for each of the B bootstrap MG parameter
(b)
vectors Θ̄ and then compute empirical confidence intervals of any desired width around
the MG cumulative dynamic multipliers obtained at step 7 in the usual way.
16
Here we discuss the simple case in which the MG estimator averages over the NARDL parameter estimates
for all countries in the sample. Generalisation to the case where one averages over a subset of the countries follows
easily.
27
APPENDIX C: Pairwise Correlations between the Country Drivers
Notes: For each pair of drivers xA and xB , the figures reported represent the estimated correlation ρA,B =
corr(xA,i , xB,i ) for i = 1, 2, . . . , 33 where xA,i represents the time average of driver xA for the ith country.
p-values are reported in parentheses.
28
29
Notes: Depr(+) counts the number of quarters over the sample period during which the exchange rate change is positive; Appr(-) counts the number of quarters when
it is negative. ADF is the Augmented Dickey-Fuller test for the null that the variable is integrated of order one, I(1), against the alternative of I(0) behaviour. KPSS is
the Kwiatkowski-Phillips-Schmidt-Shin test for the I(1) null against the I(0) alternative The ADF and KPSS test equations include both a constant and a linear time
trend. The lag order of the ADF test is selected using the modified AIC criterion, and the results are based on the MacKinnon critical values. The bandwidth of the
KPSS test is based on the Newey-West estimator using the Bartlett kernel and the results are based on the critical values tabulated by KPSS. ∗ , ∗∗ and ∗∗∗ denote
rejection of the null at the 10%, 5% and 1% level, respectively.
Table 1: Descriptive Statistics
Table 2: Cointegration Test Results
Notes: PSS denotes the Pesaran et al. (2001) F -test of the null hypothesis ρ = β + = β − = θ = 0 against
the alternative of joint significance. BDM denotes the Banerjee et al. (1998) t-test of the null hypothesis ρ = 0
against the one-sided alternative ρ < 0. In both cases, the null hypothesis indicates the absence of a long-run
levels relationship. The relevant critical values tabulated by Pesaran et al. for the BDM t-test are -3.21 (10%),
-3.53 (5%) and -4.10 (1%). The equivalent values for the PSS F -test are 4.14 (10%), 4.85 (5%) and 6.36 (1%). ∗ ,
∗∗
and ∗∗∗ denote rejection of the null at the 10%, 5% and 1% level, respectively.
30
31
Notes: Data span is the effective sample used in estimation after adjusting for lags and first differences. β + (β − ) is the long-run pass-through associated with a
depreciation (appreciation). π0+ (π0− ) is the contemporaneous or impact pass-through reflected in the same quarter as the exchange rate shock. π1+ (π1− ) is the short-run
ERPT one quarter after the exchange rate shock. Where a coefficient is printed in bold face it is statistically greater than or equal to unity (indicating full pass-through)
at the 5% level. ∗ , ∗∗ and ∗∗∗ denote statistical significance at the 10%, 5% or 1% levels. Inferences are based on OLS standard errors.
Table 3: Individual Pass-Through Estimation Results
32
Notes: In all cases, the ‘low’ and ‘high’ cohorts include 17 and 16 countries, respectively. These cohorts are selected by ranking the countries in the sample according
to the corresponding economic criteria as shown in Figure 1. The panel estimates are obtained by applying the Mean Group estimator to the country-specific NARDL
models in 3.3 and inferences are based on the Mean Group covariance matrix in 3.5.
Table 4: Panel Pass-Through Estimation Results
33
Notes: The number of observations is N =33. The dependent variable is the degree of long-run asymmetry estimated as βbi+ − βbi− for country i, where the βi ’s are the
asymmetric long-run parameters from the NARDL models for countries i = 1, 2, . . . N . The regressors include a dummy variable (Emerging, which is equal to 1 for
EMs and 0 for DMs) and economic variables which are entered as time-averages over the longest common time period for all countries. Inferences are based on OLS
standard errors which are reported in parentheses since the Breusch-Pagan-Godfrey heteroskedasticity test was insignificant at all standard levels in all cases.
Table 5: Analysis of Country Variation in Long-Run Pass-Through Asymmetry
(a) Import Dependence
(d) GDP per capita (US$, thousands)
(g) Size FX Change
0.0
0.1
0.2
‐1
0
1
2
3
4
5
6
7
8
‐0.5
‐0.4
‐0.3
‐0.2
‐0.1
0
1
2
3
4
5
6
7
8
(b) FX Rate Volatility
(e) Commodity Importer
(h) Inflation Rate
0
1
2
3
4
5
6
7
8
0
1
2
3
4
5
6
7
8
9
10
‐0.6
‐0.5
‐0.4
‐0.3
‐0.2
‐0.1
0.0
0.1
0.2
(c) Output Gap
(f) Trade Freedom
(i) Inflation Volatility
Notes: Each figure ranks the countries in our sample according to the average value of the named driver (as defined in Section 2.2) over the largest common time
period for all countries, 1980Q1–2010Q4. In each case, the countries are partitioned into two groups which are identified by red/blue shading.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0
2
4
6
8
10
12
14
16
‐6
‐4
‐2
0
2
4
6
8
Figure 1: Country Rankings According to Selected Importer Characteristics (Average Values, 1980Q1–2010Q4)
HK
SG
BR
JP
US
CO
AR
AU
GR
UK
IT
NO
ES
FR
NZ
ZA
CN
MX
CL
DK
IL
DE
FI
CA
SE
CH
KR
IE
TH
CZ
NL
HU
BELU
CN
TH
CO
BR
AR
CL
MX
HU
CZ
KR
ZA
GR
IL
NZ
ES
HK
IT
SG
DE
CA
FR
BELU
AU
UK
JP
FI
NL
SE
US
IE
DK
CH
NO
BR
AR
MX
ZA
CN
KR
CL
JP
CO
AU
TH
NZ
HU
GR
IL
UK
SE
US
CZ
CA
FI
CH
NO
IT
IE
HK
ES
DE
SG
FR
DK
NL
BELU
HK
DK
SG
CN
US
BELU
DE
NL
ES
IT
FR
FI
DE
CH
IE
TH
CZ
UK
NO
SE
IL
CA
HU
MX
CL
KR
CO
JP
AU
NZ
AR
ZA
BR
BR
AU
KR
CL
MX
CO
ZA
CA
HU
SE
NZ
NO
UK
CZ
IL
IE
SG
IT
ES
GR
FR
NL
DE
BELU
FI
DK
TH
CH
AR
HK
CN
JP
US
HK
JP
CH
SE
SG
DE
FR
FI
CN
UK
NL
BELU
CA
NO
DK
IT
IL
TH
NZ
US
CZ
IE
ES
KR
AU
CL
GR
ZA
MX
CO
HU
BR
AR
AR
CO
CL
TH
HK
CN
SG
BR
JP
CZ
HU
ZA
NO
BELU
NZ
DE
SE
UK
DK
KR
IT
FI
AU
MX
IL
US
CH
GR
FR
CA
IE
ES
NL
CN
CO
ZA
AR
TH
KR
BR
MX
JP
GR
AU
NO
CZ
HU
IT
ES
FR
FI
CH
CA
SE
US
CL
DE
BELU
NZ
IE
DK
IL
NL
UK
SG
HK
DK
IT
DE
CH
FR
JP
CA
UK
NL
GR
KR
NZ
NO
SE
BELU
AU
ES
US
FI
CZ
SG
TH
HU
CL
CO
IL
CN
HK
BR
IE
ZA
MX
AR
34
35
-0.5
-1.0
-1.5
-0.5
-1.0
-1.5
-0.8
-0.4
0.0
0.4
0.8
1.2
(l) Netherlands
(c) Canada
(h) Greece
-1.0
-0.5
0.0
0.5
1.0
0.0
-.8
-.4
.0
.4
.8
-1.5
-1.0
-0.5
-1.5
-1.0
-0.5
0.0
0.5
1.0
(n) Norway
(i) Ireland
(d) Denmark
(r) United Kingdom
(m) New Zealand
(q) Switzerland
-1.0
-0.5
0.0
0.5
1.0
-0.5
0.0
0.5
0.5
1.0
1.5
-1.0
-0.5
0.0
0.5
(s) United States
-1
0
1
2
-3
-2
-1
0
1
-0.8
-0.4
0.0
0.4
0.8
1.2
(o) Spain
(j) Italy
(e) Finland
Notes: The solid (long-dashed) black line shows the cumulative dynamic multiplier effect of a one percent depreciation (appreciation) of the exchange rate on the import
price, measured in percentage points on the vertical axis. The heavy short-dashed blue line depicts the difference between these two cumulative dynamic multipliers
(i.e. it is computed as a linear combination of the solid and dashed black lines) while the shaded region reports its 95% bootstrap confidence interval. Tick marks on
the horizontal indicate quarterly intervals over a 24 quarter horizon.
-.8
-.4
.0
(p) Sweden
0.0
0.0
.4
0.5
-0.8
1.0
.8
0.0
-0.4
(g) Germany
1.5
0.8
1.0
2.0
0.4
2.5
-1.2
-0.8
1.2
(b) Belgium
0.0
-0.4
1.0
1.5
0.8
0.4
2.0
1.2
1.6
0.5
(k) Japan
(f) France
(a) Australia
-2
-1
0
1
2
3
1.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
-.8
-.4
.0
.4
.8
Figure 2: Cumulative Dynamic Multipliers from the Unrestricted NARDL(2,2,2) Model (Equation 3.3) – Developed Markets
36
-1.0
-0.4
-3
-2
-1
0
1
(k) Singapore
-1.0
-0.5
0.0
0.5
0.4
0.4
-0.8
-0.4
-0.8
-0.4
0.0
0.8
0.0
1.2
(h) Hungary
0.8
(l) South Africa
-1.5
-1.0
-0.5
0.0
0.5
1.0
(c) Chile
1.2
(g) Hong Kong
(b) Brazil
-1.0
-0.5
0.0
0.5
1.0
1.5
(m) South Korea
-1.0
-0.5
0.0
0.5
1.0
1.5
-1.0
-0.5
0.0
0.5
1.0
(i) Israel
(d) China
-1
0
1
2
(n) Thailand
-1.0
-0.5
0.0
0.5
1.0
-1.0
-0.5
0.0
0.5
1.0
(j) Mexico
(e) Colombia
Notes: The solid (long-dashed) black line shows the cumulative dynamic multiplier effect of a one percent depreciation (appreciation) of the exchange rate on the import
price, measured in percentage points on the vertical axis. The heavy short-dashed blue line depicts the difference between these two cumulative dynamic multipliers
(i.e. it is computed as a linear combination of the solid and dashed black lines) while the shaded region reports its 95% bootstrap confidence interval. Tick marks on
the horizontal indicate quarterly intervals over a 24 quarter horizon.
-1.2
-0.8
-0.4
0.0
0.4
(f) Czech Rep.
-0.5
0.0
1.0
0.0
0.4
0.8
0.5
0.8
(a) Argentina
1.0
1.2
Figure 3: Cumulative Dynamic Multipliers from the Unrestricted NARDL(2,2,2) Model (Equation 3.3) – Emerging Markets
37
(q) Comm. Importer
-0.8
-0.4
0.0
(r) Trade Freedom Low
-0.8
-0.4
0.0
0.4
0.4
(l) Output Gap Low
(f) Infl. Vol. Low
(b) DMs
(c) EMs
(s) Trade Freedom High
-0.8
-0.4
0.0
0.4
0.8
1.2
(m) Output Gap High
-1.0
-0.5
0.0
0.5
1.0
(g) Infl. Vol. High
-1.0
-0.5
0.0
0.5
1.0
-1.0
-0.5
0.0
0.5
1.0
(t) Size ∆FX Low
-0.8
-0.4
0.0
0.4
0.8
1.2
(n) GDP p/c Low
-1.0
-0.5
0.0
0.5
1.0
(h) Import Dep. Low
-0.8
-0.4
0.0
0.4
0.8
1.2
(u) Size ∆FX High
-1.0
-0.5
0.0
0.5
1.0
(o) GDP p/c High
-0.8
-0.4
0.0
0.4
0.8
1.2
(i) Import Dep. High
-0.8
-0.4
0.0
0.4
0.8
1.2
Notes: The solid (long-dashed) black line shows the cumulative dynamic multiplier effect of a one percent depreciation (appreciation) of the exchange rate on the import
price, measured in percentage points on the vertical axis. The heavy short-dashed blue line depicts the difference between these two cumulative dynamic multipliers
(i.e. it is computed as a linear combination of the solid and dashed black lines) while the shaded region reports its 95% bootstrap confidence interval. Tick marks on
the horizontal indicate quarterly intervals over a 24 quarter horizon.
(p) Comm. Exporter
-1.0
-0.5
0.0
0.5
(k) FX Vol. High
-1.0
0.8
-0.8
-0.8
1.2
-0.4
-0.4
-0.5
0.8
0.0
0.0
0.0
0.5
1.0
-0.8
-0.4
0.0
0.4
0.8
1.2
-0.8
-0.4
0.0
0.4
0.8
1.2
1.2
0.4
0.4
1.0
0.8
(j) FX Vol. Low
1.2
0.8
(e) Infl. Rate High
-1.0
-0.5
0.0
0.5
1.0
(a) All Countries
1.2
(d) Infl. Rate Low
-0.8
-0.4
0.0
0.4
0.8
1.2
-1.0
-0.5
0.0
0.5
1.0
Figure 4: Cumulative Dynamic Multipliers for Selected Country-Groups Computed by Mean-Group Estimation