Accepted Manuscript

Title: On the effectiveness of exchange rate interventions in emerging

markets

Author: Christian Daude, Eduardo Levy Yeyati, Arne Nagengast

PII: S0261-5606(16)00013-9
DOI: http://dx.doi.org/doi: 10.1016/j.jimonfin.2016.01.004
Reference: JIMF 1635

To appear in: Journal of International Money and Finance

Please cite this article as: Christian Daude, Eduardo Levy Yeyati, Arne Nagengast, On the
effectiveness of exchange rate interventions in emerging markets, Journal of International
Money and Finance (2016), http://dx.doi.org/doi: 10.1016/j.jimonfin.2016.01.004.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will
undergo copyediting, typesetting, and review of the resulting proof before it is published in its
final form. Please note that during the production process errors may be discovered which could
affect the content, and all legal disclaimers that apply to the journal pertain.
 On the effectiveness of exchange rate interventions in
emerging markets
Christian Daude
Organisation for Economic Co-operation and Development

Eduardo Levy Yeyati1

Harvard University

Arne Nagengast
Deutsche Bundesbank

Highlights
 On average foreign exchange interventions are effective in moving the real exchange rate
in the desired direction.
 We find little evidence of asymmetries in the effect of sales and purchases.
 There is some evidence of more effective interventions for large deviations from the
equilibrium.

Abstract

We analyze the effectiveness of exchange rate interventions for a panel of 18 emerging market
economies during the period 2003-2011. Using an error-correction model approach we find that
on average intervention is effective in moving the real exchange rate in the desired direction,
controlling for deviations from the equilibrium and short-term changes in fundamentals and global
financial variables. Our results are robust to different samples and estimation methods. We find
little evidence of asymmetries in the effect of sales and purchases, but some evidence of more
effective interventions for large deviations from the equilibrium. We also explore differences
across countries according to the possible transmission channels and nature of some global
shocks.

JEL-classification: F31, F37

Keywords: exchange rate; FX intervention; equilibrium exchange rate


We would like to thank participants at the 2014 LACEA meeting in Sao Paulo and the 2014 Annual Meeting of the
Sociedad de Economistas del Uruguay for helpful comments and suggestions. The views expressed here are our own
and should not be attributed to the OECD, its member countries or the Deutsche Bundesbank.
1 Email: Eduardo_Levy_Yeyati@hks.harvard.edu Tel: +1 617 650 6459.
1

Page 1 of 40
 1. INTRODUCTION

Few macroeconomic policy topics have been as hotly debated as the exchange rate policy of
emerging economies. From the varieties of pegs of the 70s and 80s to the bipolar (float or fix)
view of the 90s, from the floating with inflation targeting paradigm of the early 2000s to the
“leaning against the wind” intervention of recent years, exchange rate policy have tended to
follow perceived lessons from crisis and respond to the ongoing macroeconomic juncture; hence,
its apparent mercurial nature.1 Nowhere have this debate been as predominant as in Latin
America, a laboratory for all sorts of creative solutions to the classic exchange rate dilemma,
namely, how to reconcile flexibility, on the one hand, and external competitiveness and financial
and macroeconomic stability, on the other. In this paper, we document that leaning-against-the-
wind exchange rate intervention in emerging economies has a significant impact on the real
exchange rate.

There are in principle two main reasons behind exchange rate intervention: the perception that the
exchange rate is moving away from a given target and the perception that it is moving away from
its equilibrium (or too fast towards a potentially new equilibrium). The difference between the
two cases should be obvious. The first one implies a degree of explicit (or implicit) rigidity in the
exchange rate that the intervention is intended to preserve. In the second case, by contrast,
intervention is “corrective”: it attempts to smooth out deviations from equilibrium or volatility
deemed to be potentially damaging (possibly in both direction although not necessarily in a
symmetric way), relative to an exchange rate “comfort zone” that reflects the behavior of a set of
moving fundamentals.

In other words, whereas in the first case the exchange rate is a predetermined target (to work as a
nominal anchor or as a tool to enhance international competitiveness and import protection), in
the second it is a flexible relative price that, to the extent that it is prone to cyclical deviations and

1
See Sarno and Taylor (2001) for an early survey with a focus on advanced countries, and Levy Yeyati and
Sturzenegger (2010) for a recent one with a focus on developing economies.
2

Page 2 of 40
misalignments (as well as sudden changes and overshooting), needs to be contained by active
policy. Given that many emerging markets (and nearly all economies in the broader sample used
in the empirical exploration of this paper) run a more or less flexible exchange rate regime and
could be grouped under the second case, in what follows we will focus on countercyclical
intervention of the second kind.

After so many ups and downs in the global financial cycle (the reflection of the not-always-
correlated swings in global liquidity, financial risk and risk appetite), the role of exchange rate
policy as a standard countercyclical tool has become increasingly apparent. While exchange rate
flexibility helps mitigate real (terms of trade) shocks in the textbook Mundell-Flemming
framework, the questions remains whether full flexibility should be allowed vis-à-vis more
transient financial shocks in a financially integrated emerging world, so that exchange rates
appreciate during the rallies of risky assets during the risk-on phase of the global financial cycle
and depreciate with the sharp selloffs that typically come at the time of the risk-off reversal. The
“leaning against the wind” nature of most exchange rate policy in the emerging world (and in
some advanced economies) seems to indicate that, at least from a policy-making perspective, this
textbook type of flexibility is often seen as more harmful that beneficial.

However, the debate at the economic mainstream has typically downplayed the exchange rate-
smoothing nature of intervention, attributing it to the precautionary or prudential motives of
reserve accumulation/decumulation, and grouping the direct sales and purchases of foreign
exchange with other related measures such as taxes or restrictions on capital mobility or
differential reserve requirements.2 However, there is plenty of evidence, both anecdotal and hard,
indicating that intervention is primarily geared to limiting what policy makers tend to see as
unwarranted (and possibly harmful) deviations from equilibrium levels: intervention correlates
negatively with exchange rate pressure and is often complemented with capital restrictions and
taxes that could only make any precautionary reserve build-up more costly.3 This contrasts with
the longer debate in the academic literature, which has concentrated on intervention strategies to
postpone or limit a devaluation (as in the “fear of floating” view) or, on the other extreme, to

2
For a discussion of the precautionary and prudential motives see Aizenman and Lee (2007) or Obstfeld et al. (2010).
The taxes and capital mobility explanations of the sales and purchases of foreign currency are discussed in a recent
IMF External Sector Report (see http://www.imf.org/external/np/pp/eng/2013/062013.pdf, pages 17- 20).
3
See, i.a., Levy Yeyati (2010) and Adler and Tovar (2011).
3

Page 3 of 40
preserve depreciated real exchange rates to foster growth and prevent “Dutch disease” effects
(which we could label the “development” view of exchange rate policies).4

However, a case can be made (as we attempt to do here) that most, if not all, exchange rate
intervention should be regarded, more generally, as a countercyclical macroeconomic tool aimed
at smoothing out short-run currency swings and to limit volatility. Indeed, fears of deviations in
either direction could be reconciled with a broader fear of exchange rate instability due to
transient or cyclical (and, at any rate, reversible) factors—a fear that, in turn, relates to the
potential impact of exchange rate misalignments on economic growth and financial stability.
Thus, for example, the reluctance of commodity exporters to let the real exchange rate appreciate
to reflect the commodity price boom in the early and mid-2000s would be driven not so much to
keeping the currency undervalued in a disguised beggar-thy-neighbour strategy as suggested by
the “development view” as to minimizing the size (and the economic costs) of an exchange rate
correction should the commodity cycle revert –as it actually did in 2008-2009 during the global
recession. On the other extreme, “fear of floating” interventions to slow down short-lived
depreciations may be seen as geared towards mitigating avoidable inflation pass-through or, in
financially dollarized economies, balance sheet effects on currency-imbalanced firms and banks.
Naturally, the exchange rate dynamics and the nature of the concerns differ according to direction
of the correction: it follows that both the reaction function of the monetary authority and its
impact on the exchange rate should, in principle, be asymmetric.

Whatever view behind central bank interventions we are willing to endorse, perhaps the most
critical, and possibly the most controversial aspect of the debate is whether interventions are
effective by exerting a significant and lasting influence on the exchange rate behavior, both on its
level and on its volatility.

A series of surveys by the BIS (2005, 2013) and the World Bank (de la Torre et al, 2013)
exploring the motives and effectiveness of intervention from the perspective of the central banks
of emerging economies, support the view that Central banks intervene actively, especially in the

4
The postponement of depreciations is discussed in Hausmann et al. (2000) and in Calvo and Reinhart (2002). The
growth motive of keeping a depreciated currency is discussed in Rodrik (2008), Hausmann et al. (2005) and Johnson
et al (2010), or Glüzmann et al. (2012). For “Dutch disease” type concerns and evidence see Rajan and Subramanian
(2011), or Cárdenas et al. (2011) for the case of LAC.
4

Page 4 of 40
spot market, to prevent excessive swings of the exchange rate (the “leaning against the wind”
motive) and that interventions are seen as an effective tool to achieve this objective (Table 1).

[INSERT Table 1 here]

However, because of the two-way causality between intervention and exchange rate variations:
central banks purchase (sell) dollars to partially offset appreciations (depreciations), the literature
has not been successful at documenting a significant and systematic link between interventions
and their desired effect on the exchange rate, which typically prove to be not significant and,
sometimes, show up with the “wrong” sign. One way of addressing these problems has been to
use high frequency data. This literature in general finds a statistically significant effect on the
exchange that goes in the “correct” direction. However, high-frequency studies do not provide
answers to how persistent the effects are over time or if they are cancelled out in the short-run.

Moreover, not all interventions are exchange rate driven: as a large body of research on
developing economies has shown that, at least in some cases and to some extent, intervention may
be due to reserve restocking after a currency crisis (Aizenmann and Lee, 2007), or to reserve
precautionary stock building (and un-building) so as to ensure a given reserve coverage of
monetary aggregates to dissuade potential speculative attacks on the national currency. In
particular, Obstfeld et al. (2010) argue that the relevant intervention objective is not to keep
reserves constant relative to GDP, exports or dollarized obligations but rather to keep it stable
relative to the broad monetary aggregate, the one that typically runs against reserves in a currency
attack. It follows that, for our measure of central bank intervention to capture exchange rate rather
than precautionary motives, it is essential to filter out precautionary reserve accumulation.

In this paper we test the effect of interventions on the real exchange rate using an intervention
measure that filters out possible precautionary causes, within an error correction model that takes
into account jointly RER dynamics, controlling for the fundamental determinants governing its
long-term (equilibrium) trend as well as and short-term non-fundamental financial drivers. We
also test for asymmetry (differential responses to dollar sales and dollar purchases) and
nonlinearities relative to the size of the shock (to explore, in particular, whether episodes of heavy
5

Page 5 of 40
dollar sales during crises are effective). Unlike in the recent related literature, we find that
intervention is significantly associated with the real exchange rate in the expected way, whereby
dollar purchases reduce appreciation pressures currency whereas dollar sales tends to dampen
depreciation pressures. Interestingly, despite the skewness of the distribution of exchange rate
changes (where depreciations are sharper but less frequent than appreciation), we find little
evidence of asymmetry.5 Our results survive several robustness checks.

The paper is organized as follows: section 2 describes data definitions and sources, section 3
reports the results and the robustness tests, and section 4 concludes.

2. Data, definitions and methodology

This section presents the main data and methodology used in our econometric tests of the
effectiveness of central bank intervention on exchange rate behaviour. We adopt a two-stage,
error correction model: we estimate first the equilibrium real exchange rate as a long run
relationship between the latter and a set of standard macroeconomic fundamentals to obtain a
measure of exchange rate misalignment, and next we run a short-run model of real exchange rate
changes on the exchange rate misalignment and the change in the fundamentals, augmented by
additional non fundamental financial variables that may introduce short-term deviations, and our
measure of intervention. In what follows we describe each stage in more detail.

a. Exchange rate deviations: Estimating the equilibrium RER

In order to know whether the exchange rate has gone too far away from equilibrium (to identify a
deviation) we need to define an equilibrium (or “normal”) level. This is, of course, a nontrivial
task given that some of the natural drivers of the real effective exchange rate (RER), including

5
See Brunnermeier Nagel and Pederson (2008) on the role of skewness in carry trade and currency crashes.
6

Page 6 of 40
many of the fundamental ones, tend to display volatile cyclical patterns, making any simple
estimate of the equilibrium zone a moving target.

A recent paper by the IMF (2013) illustrates both the complexity and the drawbacks of any
exercise geared to pin down equilibrium RER (ERER). The paper lists three types of arguments
behind the RER: i) traditional fundamentals: productivity, population growth, commodity terms of
trade, trade openness, share of administered prices to CPI; ii) short-term (financial and cyclical)
factors: VIX, interest rate differentials (which the IMF groups under the “policy” group), financial
home bias, expected GDP growth; iii) policy variables: capital controls, FX intervention, health
expenditure to GDP, private credit to GDP (a proxy for macro prudential policies).6

The estimation of a panel of RER against this set of variables open the question about what level
of the policy variables should be considered in order to evaluate the equilibrium level of the RER.
The IMF addresses this question by specifying a policy benchmark for each of these variables,
according to what “the country desks suggest would be desirable for the future”—which adds a
degree of subjectivity and entails some normative judgment. Similarly, there is the question about
the extent to which cyclical and short-term financial variables should be reflected in the estimated
equilibrium RER, rather than as explanatory factors behind its deviation from its fundamental
level.

Note, however, that a more traditional approach (as the one taken, e.g., in IMF (2006) or Bello et
al. (2010) that restricts attention to traditional fundamentals and leave non fundamental and policy
aspects aside) faces a problem that is the flipside of the previous one: if policies (or, to a lesser
extent, financial variables) drive the RER persistently away from the equilibrium, ignoring those
variables in the estimation may bias the results. For example, we may see equilibrium exchange
rates more undervalued than they actually are if they have been systematically influenced by
central bank foreign exchange purchases uncontrolled for in the model.7

6
The net foreign asset position (a fundamental), the output gap (cyclical) and the fiscal balance (a policy variable) all
fail to deliver significant results.
7
The bias should be smaller, however, if the omitted variables are not correlated with the rest of the fundamentals (or
tend to cancel each other over time).
7

Page 7 of 40
At any rate, given that there seems to be no uncontroversial way to approach the problem, here we
opt for one that takes country-specific fundamentals into account and we conduct a series of
robustness tests using a wide range of alternative definitions of the equilibrium RER. Thus, for
our baseline specification we estimate the long-run RER as a function of standard fundamentals:
productivity (prod), commodity terms of trade (ctot), net foreign assets (NFA), trade openness
(open) and government consumption to GDP (gov), using a dynamic ordinary least squares
(DOLS) estimation8 with quarterly data (interpolated using cubic spline when only annual
frequencies are available).9 In particular, we include the contemporaneous levels of the
fundamentals – our main coefficients of interest – as well as a one-quarter lag and lead terms, in
addition to the contemporaneous term, of the first difference of all fundamentals.10 We use a
balanced sample of 18 emerging and 11 advanced economies that excludes observations
corresponding to fixed exchange rate regimes and international financial centers. In order to
address the idiosyncratic nature of the fundamental determinants of ERER, in addition to pooling
all countries we also estimated the model by region as well as by grouping all emerging markets.11

The summary statistics of the main variables used in the paper are presented in Table 2.

[Insert Table 2 here]

8
DOLS provides direct estimates of the cointegration vector for a set of variables and can also be used in a panel
context (Stock and Watson, 1993; Mark and Sul, 2003). We also formally tested for cointegration between the
variables using the Westlund (2007) and Pedroni (1999, 2001) tests and rejected the null hypothesis of no
cointegration for most specifications.
9
All variables are taken in logs, with the exception of net foreign assets, which are measured as a share of GDP. The
real effective exchange rate (RER) comes from the BIS; net foreign assets (NFA) over GDP from the update of Lane
and Milesi-Ferretti (2007)’s wealth of nations dataset; commodity terms of trade (CTOT) from “The history of booms
and busts” dataset; productivity (PROD), proxied by ratio of GDP per capita to the US in PPP terms, from the Penn
World tables; and government consumption over GDP (GC) and openness (OPEN), computed as exports plus imports
over GDP, are from the IMF’s IFS. GC, PROD and OPEN are computed relative to the country's trade partners. We
prefer to estimate the model using the realized values of the variables (rather than their trends as in Bello et al. (2010)
to capture the elasticities of the RER more precisely.
10
In Section 3b we present the results of a number of robustness tests which show that our findings are not sensitive
to the number of leads and lags included in the estimation of the equilibrium RER.
11
Note that this simplified approach implicitly assumes that the relationship between RER and fundamentals are
similar across economies and that economies in the simple are, on average, in line with fundamentals (in other words,
there is no systematic deviation from the equilibrium in the simple as a whole). Hence, the trade-off between a
broader simple (which forces the coefficients to be the same across regions and levels of development) and a
narrower simple (which may assume persistent ERER misalignments as normal).
8

Page 8 of 40
The results of estimating the equilibrium RER reported in Table 3 are mostly significant and
consistent with the priors. The equilibrium level of the real exchange rate increases (appreciates)
with productivity, terms of trade, net financial assets (NFA), and government consumption; and
decreases (depreciates) with trade openness.

[Insert Table 3 here]

In turn, using the coefficients from the model reported in the table, we construct two “versions” of
the ERER: one associated with the values of fundamentals at any given point in time (the fitted
values of the regression), which we label as “ERER based on current fundamentals”, and another
one based on the medium-term values of the same variables, which we estimate as the HP-filtered
trend and label “ERER based on trend fundamentals”.

[Insert Figures 1 and 2 here]

A few aspects stand out from our estimates of equilibrium real exchange rates. First, real
exchange rates in EM display a larger volatility than in advanced economies (particularly, in
LAC).12 For example, the deviation with respect to the equilibrium exchange at current
fundamentals is on average more than twice more volatile in EM than in advanced economies.
Second, despite the presence of heavy intervention, the boom and bust cycle around the 2008
financial collapse was associated with important deviations from the predicted equilibrium values,
based either on realized or trend fundamentals—a finding that a priori justifies the leaning against
the wind policy displayed by many countries in the region. Third, fundamental values of the RER
may differ significantly depending on whether they are estimated based on current or trend (that
is, medium run) fundamentals, which may motivate intervention to offset, for example, capital
flows or a commodity bonanza that are perceived to be temporary. This last fact holds especially
for EM compared to advanced economies (Figures 1 & 2).

12
Hausmann, Panizza and Rigobon (2006) also document the significant difference in the volatility of RER in
developing countries compared to developed economies. They argue that this difference in the volatility is only
partially explained by larger shocks in fundamentals or nominal shocks.
9

Page 9 of 40
 b. Measuring intervention

A normative assessment of the optimal degree of intervention exceeds the scope of this paper. But
a positive question about its effectiveness is a good starting point: Can intervention depress the
value of the currency? There is surprisingly little consensus about the capacity of intervention to
fend off appreciation pressures; on the other hand, policymakers seem to prefer intervention to
benign neglect, despite the skepticism often voiced in academic and policy circles.

Quantifying this effect is not simple, because it entails not only a good account of other factors
that may be pressing on both the exchange rate and the level of reserves but also an accurate
measure of intervention itself. Intervention data is scarce and often substituted for reserve
changes, which may occur for reasons arguably unrelated to central bank intervention, including
the accrual of interest, the liquidation of foreign currency receipts by public companies, or the
rebuilding or reduction of the reserve buffer driven by prudential considerations. 13 And many
alternative or complementary vehicles of intervention described above (changes in the currency
composition of government debt or in the local interest rate, or intervention in futures markets),
while typically marginal compared with spot intervention as noted above, are difficult to trace
systematically and often left aside in empirical tests.

Moreover, interventions usually take place when the exchange rate is moving or expected to move
in the opposite direction to the expected effect of the intervention. This ‘endogeneity’ problem
usually results in intervention showing with the wrong sign in exchange rate regression equations,
with purchases (sales) of dollars associated with an appreciation (depreciation) of the local
currency, and no simple way to estimate the counterfactual exchange rate under no intervention.

To filter out the prudential motive for intervention and minimize the endogeneity bias, Levy
Yeyati et al. (2013) proposed to use, as a “strict” proxy for intervention, the change in the
reserves-to-M2 ratio (where reserves are computed as the central bank’s net foreign asset position

13
Adler and Tovar (2011) highlight this point, and use confidential data compiled by the IMF. Unfortunately, these
data are publicly available for only five countries in our sample. Fortunately, as the authors document, differences
between changes in reserves and actual intervention at lower-than-weekly frequencies tend to be minor and
uncorrelated with exchange rate changes. In turn, Obstfeld et al. (2010) argue that prudential reserves tend to move
together (ultimately, to cover) broad money aggregates such as M2 to avoid double bank and currency runs. Note that
an appreciation can cause the central bank to build reserves if the latter are held for precautionary motives, as a
stronger currency “deteriorates” the reserve-to-money coverage ratio because of valuation changes. The change in the
reserve-to-M2 ratio used as a proxy for intervention below mitigates this potential bias.
10

Page 10 of 40
excluding gold). This measure ensures that precautionary reserve purchases or sales to make up
for changes in broad money a la Obstfeld et al (2010) are not spuriously reflected in the
intervention variable.

In turn, a natural approach would be to model the variation of the RER, controlled for deviations
from its long-run fundamental value and for short-run drivers as we did in Table 1, augmented
with a proxy of central bank intervention as defined above.14 We use the ERER based on trend
fundamentals estimated based on Table 1 to compute the deviation of the RER from its
equilibrium level. In turn, we use this deviation, along with changes in fundamentals, in an error
correction equation augmented with the intervention variable. Finally, to correct for the
endogeneity bias (the fact that intervention may respond to changes in the exchange rate as much
as the other way around); we instrument intervention with the change in the M2-to-GDP ratio.

More formally, our econometric specification is given by:

, (1)

where the first two terms on the right-hand side are the lagged dependent variable and the error
correction terms, respectively. The vector of the first differences of the fundamental drivers of the
ERER is given by:

to which we add a set of financial drivers that might drive exchange rate movements in the short-
term:

14
As noted, the specification is leaving aside other policy measures that may affect the exchange rate: while some of
them (changes in reserve requirements, capital controls) have a low frequency that is unlikely to alter significantly the
results of the test, the omission of intervention in derivative markets or through interest rates may bias somewhat the
results.
11

Page 11 of 40
where the VIX, an index of implied volatility of options on the S&P500, proxies for global risk
appetite, and the interest rate differential with respect to the USD interest rate (carry) controls for
short-term speculative demand for the currency.

Our variable of interest is the intervention variable is .

Finally, the error term is assumed to comprise a country-specific fixed effect ( ) and an
idiosyncratic error term

3. Empirical results

This section presents the main results from an error-correction panel model of the dynamics of the
real exchange rate using the sample described in the previous section. Furthermore, we analyse
the robustness of our results to different definitions of the equilibrium RER, estimation methods,
alternative samples, and regional differences in the effect of FX intervention. In addition, we
explore a series of non-linearities and interactions with global risk and USD shocks.

a. Baseline results

Table 4 reports the results from the estimation of equation (1).15 Focusing on the first four
columns, a series of interesting results emerge. First, FX interventions have a significant impact
on of the real exchange rate, according to column (1) in Table 4. In terms of magnitude, the
estimated effects are non-negligible but rather small: the coefficient shows that a one standard
deviation increase in INT is associated with a 0.4% depreciation of the real exchange rate.16
Second, the error correction term is negative, as expected, and significant at 1%. Therefore, when
the real exchange rate equilibrium appreciates relative to its equilibrium value, between 2.7% and

15
The estimates use the ERER based on the whole sample (emerging and developed economies) from column 6 in
Table 1. Results are robust to this choice. They are omitted here due to space considerations, but available upon
request
16
The standard deviation of INT in our sample is 2.1%.
12

Page 12 of 40
3.2% of the misalignment –depending on the specification– is reverted within the following
month. This magnitude is in line with the literature, as the implied half-life in this case would be
between 24 and 27 months, close to the 2.5 years often found in earlier studies.17 This implies that
the adjustment to the equilibrium exchange rate after an intervention would also take place at a
similar pace. For the estimates in column 1, it would take 24 months to halve the effect, while
considering the estimates in column 4 it would take around 30 months. Thus, although we are
primarily interested with the contemporaneous “leaning against the wind” effect, our estimates
show that the effects could be fairly persistent. Finally, the results also show a negative and
significant impact of changes in global risk aversion – proxied by changes in the VIX – on the real
exchange rate in emerging markets (EMs). This is in line with the anecdotal evidence that global
risk-off events are typically associated with emerging currency sell offs.18

Clearly, our intervention variable could be endogenous, because of valuation effects or because
the central bank might intervene by buying reserves during appreciation episodes and selling
during depreciations. It is important to point out that this would be an attenuation bias, reducing
the size of the coefficient towards zero compared to the true coefficient. Therefore, the estimate in
column 1 would be a lower bound estimate of the true effect. In order to address this potential
endogeneity, we estimate a country-specific reaction function that includes all additional controls
from equation (1) and the lagged values of the reserves to GDP ratio, the volatility of the nominal
exchange rate against the USD and the first log difference of M2 as instruments. Reserves to GDP
and changes in M2 are included to capture interventions due to precautionary savings motives.19
Similarly, there is a large literature that suggests that central banks intervene in periods of high
volatility. F. In column (2) we present the results using the instrumented intervention variable.
Consistent with the attenuation bias of the reverse causality described above, once instrumented
the point estimate of the FX intervention effect increases almost two times compared to column
(1). Now a one-standard deviation intervention leads to a change of 0.8% in the real exchange
rate.

17
See e.g. Kubota (2013) and references therein.
18
Alternatively, using the yield of low-grade US corporate bonds or a combination of the latter and the VIX renders
the same result.
19
Note that, when estimating the pooled first stage regressions, both Reserves to GDP and M2 growth have the
expected signs and are significant, while the volatility of the exchange rate has a positive but insignificant effect..
13

Page 13 of 40
 [Insert Table 4 here]

Next, in columns (3) and (4) we explore whether there are significant differences between the
effects of buying or selling reserves. Results are mixed; while for the original specification there
seems to be no significant asymmetry, in the instrumented version the effect of foreign currency
sales is significantly smaller (around half the size) than that of purchases.

Our baseline results (columns 1 to 4) are estimated using the least-squared dummy variable
(LSDV) estimate, which generally is biased due to the endogeneity of the lagged dependent
variable although the bias approaches zero for a large number of observations in the time
dimension (Nerlove, 1971; Nickell, 1981). Therefore, in columns (5) to (8) we replicate the
estimation using a bias-corrected LSDV estimate based on Kiviet (1995, 1999) and extended to
unbalanced panels by Bruno (2005a; 2005b).20 Despite somewhat overall less significant
coefficients (e.g. for the error correction term and several of the controls), the main results
regarding central bank intervention are remarkably stable. The economic significance of the
effects of interventions barely changes from the original or the instrumented version. The same is
holds for the difference between sales and purchases.

b. Robustness and extensions

The results discussed in the previous section bring to mind a number a basic robustness checks,
covering issues of causality21, initial conditions, definition of the equilibrium RER, sample and
country characteristics, as well as the intervention proxy used. This section summarizes the main
findings of the many tests we run to examine the robustness of our results.

20
The literature shows that the LSDVC estimator performs well in small samples. For example, as shown by Judson
and Owen (1999), using Monte Carlo simulations, for panels of all sizes the bias corrected estimator consistently has
the lowest root mean squared error in comparison to OLS, Anderson-Hsiao and GMM estimators. In this paper, we
use the Bruno (2005b) implementation of the xtlsdvc routine in STATA, which is the software used in all
econometric estimations in this paper.

21
Similar results are obtained using GMM estimators to address the causality issue. However, the number of
instruments in difference (Arellano and Bond, 1991) and system GMM (Blundell and Bond, 1998) tends to explode
with large T. This weakens the Hansen test of overidentifying restrictions to the point where it generates implausibly
good p-values of 1 in all of the regressions using our dataset. Since GMM estimators with too many overidentifying
restrictions perform poorly in small samples, we do not include the resulting estimates for space considerations, but
they are available upon request.
14

Page 14 of 40
First, we test whether our results are sensitive to the specific methodology we use for computing
the equilibrium RER. One technical concern pertains to the number of leads and lags included in
the estimation of the equilibrium RER, which was set to one in our baseline specification. The
Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) do not agree on
the number of leads and lags to be included in the estimation.22 Given the remaining uncertainty
regarding the lag structure, in Table 5 we present the results for our baseline regression with the
equilibrium RER estimated using 0 (Column 1 and 2), 1 (Column 3 and 4) and 2 leads/lags
(Column 5 and 6) as well as those chosen for the different regions according to the AIC (Column
7 and 8) and the BIC (Column 9 and 10). We find that the particular lead and lag structure used
for computing the equilibrium RER has little bearing on our main results.

[Insert Table 5 here]

A second potential criticism concerns the specific methodology used for computing the
equilibrium RER. In a first step, we estimate the equilibrium RER for all emerging countries as
well as all countries in our dataset including advanced economies together instead of performing
separate estimations for ASIA, EMEA and LAC. Furthermore, we use two non-parametric
methodologies to estimate alternative equilibrium RER: the Hodrick-Prescott filtered series of the
RER23 and the centered 3-year and 5-year moving averages of the RER. Table 6 presents the
results using these alternative definitions and show that the coefficient of the intervention proxy
remains essentially unchanged: our main results are robust to alternative definitions of the
equilibrium RER.

[Insert Table 6 here]

Is the intervention-exchange rate link just reflecting a valuation effect on the reserve-to-M2 ratio?
Because there are several episodes in our sample in which INT increases (decreases) due to a
decline (increase) in the denominator (M2) rather than to an increase in reserves, it could be

22
The AIC selects 0 leads/lags for LAC, 1 lead/lag for ASIA and 2 leads/lags for EMEA, while the BIC prefers 0
leads/lags for LAC, 0 leads/lags for ASIA and 1 lead/lag for EMEA.
23
We use for the monthly RER in accordance with Ravn and Uhlig (2002).
15

Page 15 of 40
argued that the negative association between the intervention and the RER is at least in part driven
by valuation effects due to sudden exchange rate changes unrelated to intervention. For example,
a currency selloff due to a peak of global risk aversion may lead to an increase in the reserve to
M2 ratio regardless of (and generally before any) central bank intervention. To test whether our
findings are robust to the exclusion of these episodes, we use a “restricted sample” that excludes
observations for which the monthly change in reserves had the opposite sign as the change in
reserves as a ratio of M2 (cases for which changes in the intervention variable can be attributed to
changes in the exchange rate). Reassuringly, the results presented in columns (1) to (4) of Table 7
are comparable, albeit somewhat smaller, to those in Table 4. Furthermore, there is no evidence of
sales having a different impact from purchases.

Is intervention in emerging markets equally effective across emerging market groups? Differences
across regions could arise, for example, from structural differences in their financial markets that
might affect the effectiveness of intervention (as discussed in more detail below). In columns (5)
and (6) of Table 7, we report estimates with region-specific interactions for LatAm, EMEA and
Emerging Asia (the omitted interaction). The results show no significant differences between
Latin American and Emerging Asian economies, but a higher effectiveness in emerging Europe.24

Is the effectiveness of intervention related to the initial misalignment? One could argue that wide
deviations from the ERER create expectations of a reversal to equilibrium, amplifying the effect
of any intervention by the central bank to that end. By contrast, intervention when the currency is
near its equilibrium value is more likely to be offset by the market, anticipating a reversal to
normal values. We test this hypothesis by focusing on the size of the gap with respect to the
equilibrium exchange rate, interacting the intervention variable with a “large deviation” dummy
(where “large deviation” is defined as one exceeding 10%, the upper quartile of the absolute size
deviations in our sample). Columns (7) and (8) report the results, supporting the previous
hypothesis: the interaction has the expected negative sign, indicating that interventions are more
effective the larger the deviation, and is statistically significant for the “restricted sample”.25

24
Similar results are obtained when estimating separate regressions for each region that all for differences in all
coefficients, not included for space considerations but available upon request.
25
Including the large deviation dummies as an additional control does not alter these results. They are not shown here
for the sake of brevity but are available upon request.
16

Page 16 of 40
Finally, we test whether the interaction displays asymmetric effects according to whether the
central bank is buying or selling reserves. Again, we find that, while both for the full and the
restricted sample intervention is more effective in the face of large deviations, the pattern seems to
be stronger when intervention takes the form of reserve purchases to curb appreciations.

[Insert Table 7 here]

Is the effectiveness of intervention conditioned by the characteristics of local financial markets?

The economic literature has identified two main channels through which sterilized interventions
can exert this influence: the portfolio and the signaling channel (Sarno and Taylor, 2000). The
latter assumes that interventions could be seen as signals of future monetary and exchange rate
policy (from monetary authorities with private information about their inflation and exchange rate
objectives) and could therefore affect the spot exchange rate by altering exchange rate
expectations. The portfolio channel, in turn, works through the imperfect substitutability of local
and foreign currency assets that, in the event of a sterilized intervention that modifies the relative
supply of local currency assets, should induce a compensating change in cross-asset returns
(namely, a depreciation or appreciation consistent with the change in the currency composition of
the market portfolio). The nature of domestic financial markets can, a priori, influence the
effectiveness of these two channels. For example, emerging market currencies, which are
generally regarded as riskier than reserve currencies, face limited levels of cross-currency
substitutability, which should strengthen the portfolio channel (Disyatat and Galati, 2005). By
contrast, the incidence of the nominal uncertainty and poor track record characteristic of many
emerging economies on intervention effectiveness through the signalling channel is ambiguous: it
may detract from the impact of intervention on expectations (if monetary policy is seen as erratic
and noisy) or enhance it (it policy is perceived as opaque but consistent).

In order to explore these issues we proceed in two steps. First, we estimate our baseline model
allowing for a country-specific intervention coefficient (by interacting intervention with a set of
country dummies). Next, we analyse the correlation between this country-specific measure of

17

Page 17 of 40
intervention effectiveness and the average value of financial and macroeconomic variables that
capture relevant aspects of the signalling or portfolio channels. More precisely, we proxy currency
substitutability with both a measure of “original sin” (OSIN3) –the degree to which countries can
borrow abroad in their own currency– computed by Eichengreen, Hausmann and Panizza (2005)

as: , and a proxy for financial dollarization (the deposit

dollarization ratio compiled in Levy Yeyati, 2006). In turn, to test the signalling channel, we
include the average CPI inflation rate as an indicator of the Central Bank’s monetary policy track
record. Additionally, we test domestic financial development (measured as credit to the private
sector over GDP) and two measures of financial openness: a de jure measure of capital account
openness from the updated version of Chinn and Ito (2008), and a de facto measure using the sum
of external assets and liabilities as a share of GDP based on Lane and Milesi-Ferretti (2007).

The results (shown in Figure 3) indicate that:

i) Intervention effectiveness declines with inflation and financial dollarization. This is a
priori consistent with the premise that higher (and more volatile) levels of inflation
detract from the central bank´s credibility, as a result, from the signalling role of
interventions; and with the fact that, in financially dollarized economies, greater
substitutability between local and foreign currency assets weakens the portfolio
channel. Note that inflation and financial dollarization in emerging economies are
themselves connected, since the pass-through from the exchange rate to CPI inflation
tends to increase with inflation (see e.g. Mihaljek and Klau, 2008).
ii) Interventions are more effective in countries with more developed financial markets, a
result that may be capturing the fact that local market development is generally
associated with a greater demand for local currency and lower currency substitution.26
iii) Intervention effectiveness does not appear to be correlated with de jure and de facto
measures of financial openness.

[Insert Figure 3 here]

26
We also estimated one-step models that include these variables as interaction with the intervention in the original
panel regressions, with similar results.
18

Page 18 of 40
Is intervention more effective when it is leaning against the global wind? We analyse whether the
impact of central bank interventions increases as the global financial cycle intensifies;
particularly, we examine whether is more effective to contain depreciation during peaks of global
risk aversion (Table 8) or exchange rate pressures (in either direction) fuelled by large swings in
the global value of the USD (Table 9), as opposed to foreign exchange sales and purchases in
times of “normal” dollar behaviour.

Columns (1) and (2) in Table 8 report estimates for periods where global risk aversion (as proxied
by the VIX) is above or below its median. The comparison shows that, overall, interventions are
slightly more effective during periods of high VIX. In Column (3) we pool observations and
interact the intervention coefficient with a high-risk-aversion dummy and find similar results: the
point estimate for intervention in risk-off periods is 20 percentage points larger. When we look at
the effectiveness of FX sales during risk aversion episodes (column 4), the difference relative to
other periods is small but significant at 10%. Defining risk-off periods differently, as months
when global risk aversion increases by more than 10%, we get comparable results (columns 5 to
8). In sum, intervention is also effective during spells of high global risk aversion, although not
particularly more effective.

[Insert Table 8 here]

In Table 9, we replicate the exercise for the case of globally-driven exchange rate pressure,
captured by large variations in the global value of the USD, where large is defined as log changes
of the DXY –the nominal effective exchange rate of the USD against other major developed
economies’ currencies– is either at the lower decile (a depreciation of at least 2.6%) or the upper
decile (an appreciation of at least 1.8%) of its sample distribution. This time, the results are more
convincing: the estimates reported in columns (1) and (2) of Table 9 show that interventions
during shocks to the global value of the UD dollar are indeed significantly more effective than
otherwise. The effect appears to be symmetric: purchases during USD appreciations and sales
during USD depreciations show up with comparable coefficients (columns 3 to 6). Reassuringly,

19

Page 19 of 40
when looking at sales and purchases (columns 4-7) are not significantly different for or restricting
the sample to just USD appreciations and depreciations episodes (column 7 and 8), intervention
coefficients are somewhat larger than in the whole sample. In sum, there is evidence that leaning
against the wind seems to be particularly effective during periods of global dollar depreciation and
appreciation.
[Insert Table 9 here]

Finally, we present some robustness checks using an alternative intervention variable. In

particular, instead of using the change in the ratio of reserves to M2, we use the change in reserves
normalized by GDP, trend GDP (HP filtered) and M2. These proxies have also been recently in
the literature by Blanchard et al (2015) and Adler et al (2015). In Table 10, we present different
estimates. For all cases, we find a positive and significant impact of this alternative intervention
variable. However, once we include the change in M2 as an additional control, the point estimates
turn negative and significant for the GDP and trend GDP normalization, as in our baseline
estimates, and insignificant for the normalization by M2. This shows the importance of taking into
account changes in the money stock to assess the effects of intervention, as we argued above.
Next, we instrumented the intervention variables using the country-specific GARCH estimate of
the volatility of the nominal exchange rate and the lagged ratio of reserves to GDP as before. In
all cases now the estimates coefficient is negative and statistically significant.

4. Conclusions

In this paper, by defining intervention as changes in the reserve-to-M2 ratio, we showed that
exchange rate interventions in EM are generally effective, in the sense that they move the real
exchange rate in the desired direction of the intervention. Our findings are robust to endogeneity
corrections (indeed, since the reverse causality works against the empirical measure of
effectiveness, the use of instrumental variables strengthen the results), as well as to the use of
different samples and specifications. We find little evidence of asymmetries related to the
direction of intervention (sales vs. purchases) or to the direction of the global shock (global
20

Page 20 of 40
appreciations versus depreciations of the US dollar), and no significant difference in the
effectiveness of intervention according to the nature of the trigger (for example, depreciation
pressures due to spikes in global risk aversion). We document that interventions tend to be more
effective when the real exchange rate exhibits a large deviations from its long run equilibrium
level. Regarding the incidence of country characteristics, we find that interventions are less
effective in economies with high levels of financial or external liability dollarization (in line with
the existence of a portfolio channel) and under high inflation (in line with the signalling channel).
The evidence presented here supports the view that central bank intervention can indeed influence
the exchange rate in the short run, a premise that, although accepted by most practitioners, has had
so far received only partial confirmation in the empirical academic literature. In addition, we find
preliminary evidence consistent with the portfolio and signalling channels highlighted by the
theoretical literature, a starting point for further research on the preconditions and the transmission
channels of foreign exchange interventions in emerging markets.

21

Page 21 of 40
REFERENCES
ADLER, G., N. LISACK, and R. MANO (2015), “ Unveiling the Effects of Foreign Exchange
Intervention: A Panel Approach,” IMF Working Paper, WP/15/130.
ADLER, G. and C. E. TOVAR (2011), “Foreign Exchange Intervention: A Shield Against
Appreciation Winds?” IMF Working Paper WP/11/165, International Monetary Fund,
Washington DC.
AIZENMAN, J. and J. LEE (2007), “International Reserves: Precautionary Versus Mercantilist
Views, Theory and Evidence,” Open Economies Review 18(2), pp. 191-214.
ARELLANO, M. and S. BOND (1991), “Some tests of specification for panel data: Monte Carlo
evidence and an application to employment equations,” Review of Economic Studies 58,
pp. 277-297.
BELLO, O., R. HERESI, and R. PINEDA (2010), “El Tipo de Cambio Real de Equilibrio: Un
Estudio para 17 países de América Latina,” Serie Macroeconomía para el Desarrollo, No.
82, UN-ECLAC.
BIS (Bank for International Settlements) (2005), “Foreign Exchange Market Intervention in
Emerging Markets: Motives, Techniques, and Implications,” BIS Papers, No 24.
BIS (2013), “Market volatility and foreign exchange intervention in EMEs: what has changed?”
BIS Papers No. 73.
BLANCHARD, O., G. ADLER and I. DE CARVALHO FILHO (2015), “Can Foreign Exchange
Intervention Stem Exchange Rate Pressures from Global Capital Flow Shocks?” IMF
Working Paper WP/15/159.
BLUNDELL, R. and S. BOND (1998), “Initial conditions and moment restrictions in dynamic
panel data models,” Journal of Econometrics 87, pp. 115-143.
BRUNNERMEIER, M. K., S. NAGEL, and L.H. PEDERSEN (2008), “Carry Trades and
Currency Crashes”. K. Rogoff, M. Woodford, and D. Acemoglu (eds., NBER
Macroeconomics Annual 2008, 23, pp. 313-47.
BRUNO, G.S.F. (2005a), “Approximating the bias of the LSDV estimator for dynamic
unbalanced panel data models,” Economics Letters 87, pp. 361-366.
BRUNO, G.S.F. (2005b), “Estimation and inference in dynamic unbalanced panel data models
with a small number of individuals,” Stata Journal 5(4), pp. 473-500.

22

Page 22 of 40
CALVO, G.A. and C.M. REINHART (2002), “Fear of Floating,” Quarterly Journal of Economics
117(2), pp. 379-408.
CARDENAS, M., K. FODA, C. HENAO and E. LEVY YEYATI (2011), Latin America
Economic Perspectives: Shifting Gears in an Age of Heightened Expectations,
Brookings, Washington DC.
CHINN, M. D. and H. ITO (2008). “A New Measure of Financial Openness,” Journal of
Comparative Policy Analysis 10 (3), pp. 309-322.
DE LA TORRE, A., E. LEVY YEYATI and S. PIENKNAGURA (2013), “Latin America’s
Deceleration and the Exchange Rate Buffer.” LAC Semiannual Report (October), World
Bank, Washington, DC.
DISYATAT, P. and G. GALATI (2005), “The effectiveness of foreign exchange intervention in
emerging market countries,” BIS Papers No. 24, pp.97-113.
EICHENGREEN, B., R. HAUSMANN and U. PANIZZA (2005), “The Pain of Original Sin,” in
Other People’s Money, B. Eichengreen and R. Hausmann (eds.), Chicago University
Press, Chicago.
GLÜZMANN, P.A., E. LEVY YEYATI and F. STURZENEGGER (2012), “Exchange rate
undervaluation and economic growth: Diaz Alejandro (1965) revisited,” Economics
Letters 117, pp. 666-672.
HAUSMANN, R., U. PANIZZA and E. STEIN (2001), “Why do countries float the way they
float?” Journal of Development Economics 66(2), pp. 387-414.
HAUSMANN, R., L. PRITCHETT and D. RODRIK (2005), “Growth Accelerations,” Journal of
Economic Growth 10, pp. 303-329.
IMBS, J., H. MUMTAZ, M. RAVN and H. REY (2003), “Non-Linearities and Real Exchange
Rate Dynamics,” Journal of the European Economic Association 1 (2-3), pp.639-649.
IMF (2006), “Methodology for CGER Exchange Rate Assessments,” International Monetary
Fund, November, Washington DC.
IMF (2013), The External Balance Assessment (EBA) Methodology, IMF Working Paper 13/272.
JOHNSON, S., J.D. OSTRY and A. SUBRAMANIAN (2010), “Prospects for Sustained Growth
in Africa: Benchmarking the Constraints,” IMF Staff Papers vol. 57(1), pp. 119-171.
JUDSON, R.A. and A.L. OWEN (1999), “Estimating Dynamic Panel Data Models: A Guide for
Macroeconomists”, Economic Letters, 65(1), pp. 9-15.

23

Page 23 of 40
KIVIET, J.F. (1995), “On Bias, Inconsistency, and Efficiency of various Estimators in Dynamic
Panel Data Models,” Journal of Econometrics, 68, pp. 53-78.
KIVIET, J.F. (1999), “Expectation of Expansions for Estimators in a Dynamic Panel Data Model;
some Results for Weakly Exogenous Regressors”, in: C. Hsiao, K. Lahiri, L.-F. Lee and
M.H. Pesaran (Eds.),,Analysis of Panel Data and Limited Dependent Variables,
Cambridge University Press, Cambridge.
KUBOTA, M. (2013), “Estimating the Half-life of Theoretically Founded Real Exchange Rate
Misalignments,” Policy Research Working Paper 6411, World Bank, Washington DC.
LANE, P. R. and G.M. MILESI-FERRETTI (2007), “The external wealth of nations mark II:
Revised and extended estimates of foreign assets and liabilities, 1970–2004”, Journal of
International Economics 73, November, 223-250.
LEVY YEYATI, E. (2006), “Financial Dollarization: Evaluating the Consequences”, Economic
Policy, January 2006, pp. 61-118.
LEVY YEYATI, E. (2010), “What drives reserve accumulation (and at what cost)?” Article
published in voxeu.org, September 30. http://www.voxeu.org/article/what-drives-reserve-
accumulation-and-what-cost
LEVY YEYATI, E. and F. STURZENEGGER (2010), “Monetary and exchange rate policies,”
Chapter 64 in Handbook of Development Economics (D. Rodrik and M. Rosenzweig,
eds.), Vol. 5, pp. 4215–4281.
LEVY YEYATI, E., F. STURZENEGGER and P.A. GLÜZMANN (2013), “Fear of
appreciation,” Journal of Development Economics 101, pp. 233-247.
MARK, N. C. and D. SUL (2003), “Cointegration Vector Estimation by Panel DOLS and Long-
run Money Demand”, Oxford Bulletin of Economics and Statistics, 65(5), pp. 655-680.
MIHALJEK, D. and M. KLAU (2008), “Exchange rate pass-through in emerging market
economies: what has changed and why?” BIS Papers No. 35, pp.103-130.
NERLOVE, M. (1971), “Further Evidence on the Estimation of Dynamic Economic Relations
from a Time Series of Cross-Sections”, Econometrica: Journal of the Econometric
Society, pp. 359-382.
NICKELL, S. (1981), “Biases in Dynamic Models with Fixed Effects”, Econometrica: Journal of
the Econometric Society, pp. 1417-1426.

24

Page 24 of 40
OBSTFELD, M., J. C. SHAMBAUGH, and A. M. TAYLOR (2010), “Financial Stability, the
Trilemma, and International Reserves,” American Economic Journal Macroeconomics 2,
pp. 57-94.
RAJAN, R. and A. SUBRAMANIAN (2011), “Aid, Dutch Disease and Manufacturing Growth,”
Journal of Development Economics 94(1), pp. 106-118.
RAVN, M. O. and H. UHLIG (2002), “On Adjusting the Hodrick-Prescott Filter for the
Frequency of Observations”, Review of Economics and Statistics 84:2, pp. 371-376.
RODRIK, D. (2008), “The real exchange rate and economic growth: theory and evidence,”
Brookings Papers on Economic Activity, Fall 2008. pp. 365–412.
SARNO, L. AND M. P. TAYLOR (2001), “Official Intervention in the Foreign Exchange
Market: Is It Effective and, If So, How Does It Work?” Journal of Economic Literature,
39(3), pp. 839-868.
STOCK, J. H. and M. W. WATSON (1993), “A Simple Estimator of Cointegrating Vectors in
Higher Order Integrated Systems”, Econometrica, 61(4), pp. 783-820.

25

Page 25 of 40
Figure 1: Effective Real Exchange Rates and equilibrium estimates by country

26

Page 26 of 40
Figure 1: Effective Real Exchange Rates and equilibrium estimates by country (continued)

27

Page 27 of 40
Figure 2: Average difference between equilibrium real exchange rate estimates at current versus trend values by regions

28

Page 28 of 40
 Figure 3: Correlations between intervention effectiveness and financial market characteristics

ARG ARG ARG

2

2
PER PHL
TUR TUR
PHL
PER PHL
TUR BRA PER
BRA BRA
1

1
RUS RUS RUS
Intervention

Intervention

Intervention
KOR KOR KOR
POL CHL POL CHL POL CHL
0

0
COL HUN COL HUN COL HUN
ROM ROM
CZE CZE CZE
IDN IDN IDN
-1

-1

-1
MYSZAF ZAF MYS ZAF MYS
-2

-2

-2
-3

-3

-3
MEX MEX MEX

-1 -.5 0 .5 1 1.5 -.5 0 .5 1 1.5 2 -2 -1 0 1 2

Domestic Credit/GDP in logs sample avg(Foreign Asset + Liabilities)/GDP sample avgDe Jure Financial Openness sample avg
coef = -.75183486, se = .50619838, t = -1.49 coef = -.13070909, se = .58650642, t = -.22 coef = -.12902752, se = .29719675, t = -.43

ARG ARG ARG

2
2

2
PERPHL TUR PHL TUR
PHL
PER
BRA PER BRA
TUR
1

1
RUS RUS
1
Intervention

Intervention

Intervention
KOR RUS KOR
CHL
POL POL CHL
0

0
COL
HUN COL
HUN
CHL ROM
0

CZE POL CZE

IDN COL HUN IDN
-1

-1
ROM
CZE
MYS ZAF MYS
-1

IDN ZAF
-2

-2
MYS
ZAF
-3

-3
MEX MEX
-2

-.05 0 .05 .1 .15 -.2 0 .2 .4 .6 -.6 -.4 -.2 0 .2 .4

Inflation sample avg Deposit dollarization in 2003 Original Sin sample avg
coef = 10.13461, se = 6.7879403, t = 1.49 coef = 2.7491383, se = 1.5243176, t = 1.8 coef = 2.2089913, se = 1.2338429, t = 1.79

29

Page 29 of 40
Table 1 – Motives, objectives and tools for FX market interventions reported by Central Banks

Motives for Intervention Importance in 2005 Importance in 2012-2013

High Moderate Low High Moderate Low
Low
1. Curb excessive speculation 42% 21% 0% 60% 20% 0%
2. Maintain monetary stability 36% 11% 11% 55% 10% 10%
3. Discourage sharp capital flows 21% 16% 5% 30% 25% 5%
4. Build Reserves 36% 0% 11% 30% 10% 10%
5. Smooth the impact of commodity price movements 16% 5% 16% 20% 5% 15%
6.Maintain or enhance competitiveness 11% 11% 16% 25% 5% 15%

Immediate Objectiles 2005 2012 Intervention Tools Regularly Occasionally

1. Influence the level of the exchange rate 21% 20% Market based
2. Smooth the path of the exchange rate 36% 35% 1. Spot market intervention 91% 9%
3. Limit Volatility 57% 60% 2. Forward market 9% 36%
intervention
4. Limit pressures arising from capital flows 42% 45% 3. Use of derivatives 14% 32%
5. Provide liquidity to thin exchange rate market 10% 20% Other interventions
4. Capital Controls 12% 18%

Note: The results are based on a 2013 BIS survey and a WB survey conducted in 2013. Respondents in these three surveys include: Argentina, Brazil, Chile,
Colombia, Czech Republic, Hungary, Hong Kong SAR, Rep. of Korea, India, Indonesia, Malaysia, Mexico, New Zealand, Peru, Philippines, Poland, Singapore, South
Africa, Thailand, Turkey, Uruguay, and Venezuela.

30

Page 30 of 40
Table 2 – Summary Statistics

Mean Median SD Minimum Maximum

log(RER)(t)-log(RER)(t-1) 0.001 0.002 0.023 -0.145 0.093

Error correction equilibrium RER(t-1) 0.009 0.015 0.104 -0.623 0.328
Net Foreign Assets over GDP -0.306 -0.290 0.256 -1.290 0.168
Trade Openness 1.118 0.987 0.452 0.369 2.284
Government Consumption 0.803 0.782 0.213 0.423 1.243
Productivity 0.421 0.391 0.188 0.087 0.836
Commodity Terms of Trade 100.010 99.961 1.451 90.802 109.154
Delta VIX 0.032 -0.320 5.735 -28.510 19.160
log(DXY)(t)-log(DXY)(t-1) -0.002 -0.004 0.019 -0.048 0.065
Carry USD 1.306 0.824 2.304 -1.358 41.384
(R/M2)(t)-(R/M2)(t-1) -0.000 -0.001 0.022 -0.249 0.163
Instrumented Intervention -0.000 -0.001 0.012 -0.102 0.123
Note: Summary statistics are for the sample used in the baseline specification in Table 4.

31

Page 31 of 40
Table 3 – Equilibrium RER as a function of fundamentals

(1) (2) (3) (4) (5) (6)

Variables LAC ASIA EMEA Emerging Advanced All
NFA 0.406*** -0.113 -0.765*** -0.120 -0.153*** -0.146**
(3.45) (-1.27) (-3.83) (-1.53) (-2.68) (-2.54)
Log(Open) -0.784*** -0.452*** -0.688*** -0.512*** -0.684*** -0.540***
(-2.83) (-6.31) (-4.42) (-6.53) (-10.05) (-6.70)
Log(Commodity 2.286** -0.360 0.276 -0.190 0.154 0.243
TOT)
(2.15) (-0.60) (0.58) (-0.43) (0.54) (0.79)
Log(G/ GDP) 0.873*** 0.455*** 0.389'''' 0.379*** -0.0313 0.174
(6.51) (3.31) (2.43) (3.21) (-0.21) (1.37)
Log(Productivity) 0.769** 0.764*** 0.0587 0.735*** 0.313 0.691***
(2.56) (7.65) (0.31) (6.29) (1.09) (6.40)
N 325 390 455 1170 715 1885
Countries 5 6 7 18 11 29
R-sq Within 0.6915 0.7321 0.7536 0.6226 0.6445 0.5889

Note: LAC includes Brazil, Chile, Colombia, Mexico, and Peru; ASIA includes India, Indonesia, Rep. of Korea, Malaysia, Philippines, and
Thailand; EMEA includes Czech Republic, Hungary, Poland, Rumania, Russia, South Africa and Turkey; Advanced includes Australia,
Canada, Israel, Japan, New Zealand, Norway, Singapore, Sweden, Switzerland, and UK. Estimates based on a dynamic OLS specification. t
statistics in parentheses using clustered standard errors at the country level The asterisks indicate statistical significance,* p<0.10, **p<0.05,
***p<0.01.

32

Page 32 of 40
Table 4 – Baseline estimates and bias-corrected estimates
(1) (2) (3) (4) (5) (6) (7) (8)
Estimation Method LSDV LSDV LSDV LSDV LSDVC LSDVC LSDVC LSDVC
Lagged Dep. variable 0.128 0.134 0.126 0.128 0.144 0.150 0.143 0.145
(0.036)*** (0.035)*** (0.036)*** (0.035)*** (0.023)*** (0.023)*** (0.023)*** (0.023)***
Error correction equilibrium RER (t-1) -0.032 -0.028 -0.032 -0.027 -0.017 -0.013 -0.017 -0.012
(0.006)*** (0.006)*** (0.007)*** (0.006)*** (0.018) (0.018) (0.018) (0.018)
ΔNFA 0.143 0.145 0.137 0.132 0.110 0.113 0.104 0.100
(0.036)*** (0.039)*** (0.034)*** (0.043)*** (0.186) (0.182) (0.186) (0.182)
Δ Trade Openness (logs) -0.522 -0.482 -0.518 -0.482 -0.511 -0.472 -0.507 -0.471
(0.056)*** (0.048)*** (0.056)*** (0.046)*** (0.121)*** (0.119)*** (0.120)*** (0.118)***
Δ Government consumption (logs) 0.031 0.024 0.031 0.023 0.034 0.027 0.034 0.026
(0.019) (0.019) (0.019) (0.017) (0.077) (0.076) (0.077) (0.076)
Δ Productivity (logs) -0.076 -0.077 -0.075 -0.075 -0.065 -0.066 -0.064 -0.064
(0.012)*** (0.013)*** (0.013)*** (0.014)*** (0.182) (0.177) (0.182) (0.177)
Δ VIX -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)** (0.000)** (0.000)** (0.000)**
Carry USD -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
(R/M2)(t) - (R/M2)(t-1) -0.177 -0.234 -0.179 -0.234
(0.057)*** (0.089)** (0.075)** (0.135)*
Instrumented Intervention -0.382 -0.503 -0.383 -0.502
(0.093)*** (0.134)*** (0.107)*** (0.161)***
Sales of reserves 0.108 0.107
(0.079) (0.219)
Instrumented sales of reserves 0.244 0.241
(0.100)** (0.255)
2
R 0.26 0.29 0.26 0.32
N 1,834 1,834 1,834 1,834 1,824 1,824 1,824 1,824
* p<0.1; ** p<0.05; *** p<0.01. Dependent variable: first log difference of the real exchange rate. All regressions include country dummies. Robust standard errors in
parenthesis for (1) – (4). Bootstrapped standard errors (500 reps) for (5) – (8).

33

Page 33 of 40
Table 5 – Robustness: Leads and lags in equilibrium RER estimation
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Lead/Lag Lead/Lag Lead/Lag Lead/Lag
Lead/Lag=0 Lead/Lag=0 Lead/Lag=1 Lead/Lag=1 Lead/Lag=2 Lead/Lag=2 (AIC) (AIC) (BIC) (BIC)
Lagged Dep. variable 0.129 0.134 0.128 0.134 0.127 0.133 0.128 0.133 0.128 0.133
(0.034)*** (0.033)*** (0.036)*** (0.035)*** (0.034)*** (0.033)*** (0.034)*** (0.033)*** (0.034)*** (0.033)***
Error correction
equilibrium RER (t-1) -0.035 -0.030 -0.032 -0.028 -0.033 -0.029 -0.034 -0.029 -0.034 -0.029
(0.008)*** (0.009)*** (0.006)*** (0.006)*** (0.007)*** (0.007)*** (0.007)*** (0.007)*** (0.007)*** (0.007)***
ΔNFA 0.056 0.070 0.143 0.145 0.131 0.135 0.125 0.129 0.120 0.124
(0.045) (0.049) (0.036)*** (0.039)*** (0.035)*** (0.038)*** (0.035)*** (0.038)*** (0.036)*** (0.038)***
Δ Trade Openness (logs) -0.523 -0.485 -0.522 -0.482 -0.525 -0.486 -0.525 -0.486 -0.526 -0.487
(0.056)*** (0.049)*** (0.056)*** (0.048)*** (0.057)*** (0.049)*** (0.057)*** (0.049)*** (0.056)*** (0.049)***
Δ Government
consumption (logs) 0.032 0.025 0.031 0.024 0.034 0.027 0.033 0.026 0.033 0.025
(0.018)* (0.018) (0.019) (0.019) (0.019)* (0.018) (0.018)* (0.018) (0.018)* (0.018)
Δ Productivity (logs) -0.048 -0.053 -0.076 -0.077 -0.042 -0.047 -0.041 -0.046 -0.042 -0.047
(0.013)*** (0.013)*** (0.012)*** (0.013)*** (0.012)*** (0.013)*** (0.012)*** (0.013)*** (0.012)*** (0.013)***
Δ VIX -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)***
Carry USD -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
(R/M2)(t) - (R/M2)(t-1) -0.176 -0.177 -0.177 -0.177 -0.177
(0.057)*** (0.057)*** (0.057)*** (0.057)*** (0.057)***
Instrumented
Intervention -0.377 -0.382 -0.380 -0.379 -0.379
(0.092)*** (0.093)*** (0.092)*** (0.092)*** (0.092)***
N 1834 1834 1834 1834 1834 1834 1834 1834 1834 1834
R-squared 0.258 0.286 0.257 0.286 0.258 0.287 0.258 0.287 0.258 0.287
Notes: * p<0.1; ** p<0.05; *** p<0.01. Robust standard errors in parentheses. The estimates in Column 3 and 4 correspond to our baseline specification in Table 4 (Column 1 and 2). Lead/Lag (AIC)
includes estimates of the equilibrium RER using the Akaike Information Criterion for lead and lag selection. Lead/Lag (BIC) includes estimates of the equilibrium RER using the Bayesian Information
Criterion for lead and lag selection.

34

Page 34 of 40
Table 6 – Robustness: Definition of equilibrium RER
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Region Region Emerging Emerging All All HP filter HP filter 3yr MA 3yr MA 5yr MA 5yr MA
Lagged Dep. variable 0.128 0.134 0.127 0.133 0.128 0.134 0.167 0.170 0.194 0.197 0.172 0.177
(0.036)*** (0.035)*** (0.036)*** (0.034)*** (0.036)*** (0.035)*** (0.036)*** (0.035)*** (0.034)*** (0.033)*** (0.037)*** (0.036)***
ECM equilibrium RER (t-
1) -0.032 -0.028 -0.031 -0.026 -0.033 -0.029 -0.078 -0.073 -0.125 -0.116 -0.080 -0.073
(0.006)*** (0.006)*** (0.007)*** (0.007)*** (0.008)*** (0.008)*** (0.013)*** (0.013)*** (0.017)*** (0.016)*** (0.017)*** (0.016)***
ΔNFA 0.143 0.145 0.139 0.141 0.145 0.147 0.146 0.152 0.280 0.277 0.273 0.262
(0.036)*** (0.039)*** (0.036)*** (0.038)*** (0.036)*** (0.039)*** (0.048)*** (0.048)*** (0.067)*** (0.065)*** (0.066)*** (0.066)***
Δ Trade Openness (logs) -0.522 -0.482 -0.518 -0.479 -0.516 -0.478 -0.491 -0.454 -0.469 -0.440 -0.517 -0.481
(0.056)*** (0.048)*** (0.056)*** (0.048)*** (0.056)*** (0.048)*** (0.054)*** (0.047)*** (0.050)*** (0.044)*** (0.056)*** (0.049)***
Δ Government
consumption (logs) 0.031 0.024 0.029 0.023 0.027 0.020 0.028 0.021 0.026 0.021 0.028 0.0240
(0.019) (0.019) (0.019) (0.018) (0.019) (0.018) (0.017) (0.017) (0.018) (0.018) (0.020) (0.021)
Δ Productivity (logs) -0.076 -0.077 -0.067 -0.069 -0.073 -0.073 -0.146 -0.142 -0.176 -0.168 -0.228 -0.222
(0.012)*** (0.013)*** (0.012)*** (0.013)*** (0.012)*** (0.013)*** (0.020)*** (0.020)*** (0.022)*** (0.022)*** (0.047)*** (0.046)***
Δ VIX -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)** (0.000)** (0.000)** (0.000)**
Carry USD -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 0.000 0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
(R/M2)(t) - (R/M2)(t-1) -0.177 -0.178 -0.177 -0.172 -0.159 -0.160
(0.057)*** (0.058)*** (0.057)*** (0.054)*** (0.053)*** (0.055)***
Instrumented
Intervention -0.382 -0.382 -0.380 -0.373 -0.372 -0.395
(0.093)*** (0.094)*** (0.094)*** (0.090)*** (0.088)*** (0.095)***
N 1834 1834 1834 1834 1834 1834 1834 1834 1511 1511 1283 1283
R-squared 0.257 0.286 0.256 0.285 0.258 0.286 0.273 0.302 0.292 0.320 0.290 0.324

Notes: * p<0.1; ** p<0.05; *** p<0.01. Robust standard errors in parentheses. The estimates in Column 1 and 2 correspond to our baseline specification in Table 4 (Column 1 and 2). “Region” refers to
estimates of the equilibrium RER performed separately for ASIA, EMEA and LAC. “Emerging” refers to estimates of the equilibrium RER using the full set of emerging economies in our dataset. “All”
refers to estimates of the equilibrium RER using the full set of countries in our dataset including advanced economies. “HP filtered” includes Hodrick-Prescott filtered series of the RER as estimates for
the equilibrium RER. “3yr MA” and “5yr MA” use centered 3-year and 5-year moving averages as estimates for the equilibrium RER.

35

Page 35 of 40
Table 7 – Robustness: Sample, regions, large deviations from equilibrium RER and sales versus purchases
1 1 1
Restricted Restricted Restricted Restricted Regional Regional Deviation Deviation Sales vs. Sales vs.
Sample Sample Sample Sample estimates estimates from ERER from ERER purchases purchases
Lagged Dep. variable 0.143 0.151 0.143 0.153 0.132 0.153 0.133 0.150 0.131 0.150
(0.036)*** (0.037)*** (0.036)*** (0.037)*** (0.035)*** (0.039)*** (0.034)*** (0.036)*** (0.035)*** (0.036)***
ECM equilibrium RER (t-1) -0.030 -0.026 -0.030 -0.026 -0.029 -0.028 -0.027 -0.023 -0.026 -0.023
(0.007)*** (0.008)*** (0.007)*** (0.008)*** (0.006)*** (0.007)*** (0.007)*** (0.008)*** (0.006)*** (0.008)***
Δ NFA -0.003 0.016 -0.001 0.019 0.138 0.029 0.145 0.026 0.133 0.025
(0.087) (0.086) (0.085) (0.085) (0.039)*** (0.086) (0.041)*** (0.086) (0.047)** (0.086)
Δ Openness (logs) -0.483 -0.457 -0.484 -0.457 -0.481 -0.454 -0.481 -0.457 -0.482 -0.457
(0.080)*** (0.073)*** (0.081)*** (0.073)*** (0.049)*** (0.074)*** (0.049)*** (0.072)*** (0.047)*** (0.072)***
Δ Govt. consumption (logs) 0.033 0.027 0.033 0.027 0.027 0.030 0.024 0.025 0.021 0.025
(0.015)** (0.016) (0.015)** (0.017) (0.019) (0.016)* (0.018) (0.016) (0.016) (0.016)
Δ Productivity (logs) -0.194 -0.196 -0.194 -0.196 -0.074 -0.188 -0.076 -0.203 -0.071 -0.203
(0.022)*** (0.021)*** (0.021)*** (0.021)*** (0.014)*** (0.023)*** (0.013)*** (0.022)*** (0.016)*** (0.023)***
Δ VIX -0.000 -0.000 -0.000 -0.000 -0.001 -0.000 -0.001 -0.000 -0.001 -0.000
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)***
Carry USD 0.000 -0.000 0.000 -0.000 -0.000 -0.000 -0.000 0.000 -0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
(R/M2)(t) - (R/M2)(t-1) -0.069 -0.053
(0.023)*** (0.041)
Instrumented Intervention -0.249 -0.206 -0.301 -0.198 -0.339 -0.200 -0.336 -0.200
(0.069)*** (0.063)*** (0.091)*** (0.040)*** (0.099)*** (0.061)*** (0.098)*** (0.061)***
Sales of reserves -0.031
(0.072)
Instrumented sales of res. -0.114
(0.090)
LAC dummy x Instr. Int. 0.071 0.043
(0.163) (0.080)
EMEA dummy x Instr. Int. -0.294 -0.316
(0.141)* (0.181)*
Instr. Int. x large dev. ERER -0.134 -0.180
(0.120) (0.088)*
Instr. Sales x large dev. ERER 0.142 -0.165
(0.130) (0.145)
Instr. Purch. x large dev ERER -0.360 -0.186
(0.165)** (0.097)*
2
R 0.21 0.23 0.21 0.23 0.30 0.24 0.29 0.23 0.30 0.23
N 1,137 1,137 1,137 1,137 1,834 1,137 1,834 1,137 1,834 1,137
See Table 1 for details. The restricted sample is comprised of observations for which the change in reserves has the same sign as the intervention variable. 1 restricted

36

Page 36 of 40
sample only.

Table 8 – The effectiveness of FX intervention and global risk aversion shocks

(1) (2) (3) (4) (5) (6) (7) (8)

High VIX Low VIX Full sample Full sample ΔVIX > 10% ΔVIX <= 10% Full sample Full sample
sample sample
Lagged Dep. Variable 0.120 0.136 0.131 0.128 0.093 0.112 0.134 0.128
(0.046)** (0.043)*** (0.035)*** (0.035)*** (0.067) (0.032)*** (0.035)*** (0.035)***
ECM equilibrium RER (t-1) -0.044 -0.030 -0.028 -0.027 -0.033 -0.029 -0.028 -0.027
(0.014)*** (0.010)*** (0.006)*** (0.006)*** (0.010)*** (0.008)*** (0.006)*** (0.006)***
Δ NFA 0.143 0.227 0.146 0.133 0.021 0.155 0.143 0.132
(0.070)* (0.058)*** (0.040)*** (0.042)*** (0.116) (0.062)** (0.040)*** (0.043)***
Δ Openness (logs) -0.474 -0.453 -0.480 -0.482 -0.493 -0.479 -0.484 -0.482
(0.066)*** (0.061)*** (0.047)*** (0.046)*** (0.075)*** (0.060)*** (0.048)*** (0.046)***
Δ Govt. consumption (logs) -0.035 0.059 0.023 0.023 0.045 0.022 0.021 0.023
(0.034) (0.020)*** (0.018) (0.017) (0.058) (0.015) (0.018) (0.017)
Δ Productivity (logs) -0.096 -0.086 -0.079 -0.075 -0.204 -0.021 -0.077 -0.075
(0.024)*** (0.042)* (0.014)*** (0.014)*** (0.039)*** (0.013) (0.013)*** (0.014)***
Δ VIX -0.001 -0.001 -0.001 -0.001 -0.001 -0.000 -0.001 -0.001
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000) (0.000)*** (0.000)***
Carry USD -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)* (0.000) (0.000) (0.000)
Instrumented Intervention -0.460 -0.270 -0.256 -0.503 -0.449 -0.336 -0.340 -0.503
(0.116)*** (0.067)*** (0.067)*** (0.134)*** (0.141)*** (0.067)*** (0.077)*** (0.134)***
Instr. Intervention x High VIX -0.201
(0.068)***
Instrumented sales of reserves 0.255 0.246
(0.139)* (0.110)**
Instrumented sales x High VIX -0.018
(0.090)
Large ΔVIX x Instr. Intervention -0.134
(0.087)
Large ΔVIX x Instr. Sales of reserves -0.005
(0.078)
2
R 0.33 0.20 0.29 0.29 0.31 0.24 0.29 0.29
N 925 909 1,834 1,834 454 1,380 1,834 1,834

37

Page 37 of 40
Table 9 – The effectiveness of FX intervention and USD shocks
(1) (2) (3) (4) (5) (6) (7) (8)
Large USD Large USD
Appreciations Depreciations
Lagged Dep. Variable 0.125 0.130 0.128 0.124 0.134 0.132 0.006 0.217
(0.037)*** (0.036)*** (0.037)*** (0.038)*** (0.035)*** (0.036)*** (0.097) (0.102)**
ECM equilibrium RER (t-1) -0.031 -0.027 -0.032 -0.031 -0.028 -0.028 -0.044 -0.046
(0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.006)*** (0.017)** (0.015)***
Δ NFA 0.130 0.133 0.140 0.113 0.142 0.126 0.271 0.160
(0.037)*** (0.043)*** (0.036)*** (0.038)*** (0.040)*** (0.041)*** (0.135)* (0.200)
Δ Openness (logs) -0.517 -0.483 -0.521 -0.514 -0.481 -0.479 -0.574 -0.377
(0.054)*** (0.047)*** (0.056)*** (0.054)*** (0.047)*** (0.047)*** (0.104)*** (0.119)***
Δ Govt. consumption (logs) 0.027 0.023 0.031 0.033 0.025 0.025 -0.005 0.104
(0.019) (0.017) (0.019) (0.019) (0.019) (0.019) (0.087) (0.051)*
Δ Productivity (logs) -0.076 -0.077 -0.077 -0.074 -0.077 -0.076 -0.107 -0.300
(0.013)*** (0.014)*** (0.012)*** (0.013)*** (0.013)*** (0.013)*** (0.049)** (0.106)**
Δ VIX -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 -0.000
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.001)** (0.000)
Carry USD -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)** (0.000)**
Intervention = (R/M2)(t) - (R/M2)(t-1) -0.137 -0.173 -0.159
(0.049)** (0.060)** (0.055)***
USD shock x Intervention -0.198
(0.065)***
Instrumented Intervention -0.315 -0.381 -0.370 -0.455 -0.382
(0.077)*** (0.094)*** (0.085)*** (0.166)** (0.131)***
USD shock x Instrumented Intervention -0.235
(0.096)**
USD shock appreciation x Sales of FX 0.028 -0.174
(0.137) (0.158)
USD shock depreciation x Purchases of FX -0.118 -0.120
(0.104) (0.086)
USD appreciation dummy -0.006 -0.004
(0.003)** (0.002)
USD depreciation dummy -0.001 -0.002
(0.002) (0.002)
USD shock appreciation x instr. Sales of FX 0.207 0.130
(0.211) (0.202)
38

Page 38 of 40
USD shock depreciation x instr. Purchases of FX -0.113 -0.089
(0.152) (0.132)
2
R 0.26 0.29 0.26 0.26 0.29 0.29 0.39 0.38
N 1,834 1,834 1,834 1,834 1,834 1,834 216 168

Table 10 – Estimates using alternative intervention proxy

(1) (2) (3) (4) (5) (6) (7) (7) (7)
Change in Change in Change in Change in Change in Change in Change in Change in Change in
Reserves/GDP Reserves/GDP Reserves/GDP Reserves / Reserves / Reserves / Reserves/M2 Reserves/M2 Reserves/M2
Trend GDP Trend GDP Trend GDP
Lagged Dep. Variable 0.131 0.155 0.158 0.132 0.154 0.155 0.131 0.151 0.155
(0.026)*** (0.024)*** (0.024)*** (0.026)*** (0.024)*** (0.024)*** (0.026)*** (0.024)*** (0.024)***
ECM equilibrium RER (t-1) -0.032 -0.023 -0.022 -0.033 -0.023 -0.021 -0.033 -0.024 -0.022
(0.007)*** (0.006)*** (0.006)*** (0.007)*** (0.006)*** (0.006)*** (0.007)*** (0.006)*** (0.006)***
Δ NFA -0.013 0.011 0.011 -0.007 0.006 0.001 -0.011 0.002 0.004
(0.066) (0.061) (0.061) (0.066) (0.061) (0.061) (0.066) (0.061) (0.061)
Δ Openness (logs) -0.498 -0.377 -0.376 -0.497 -0.377 -0.377 -0.502 -0.384 -0.377
(0.043)*** (0.040)*** (0.040)*** (0.043)*** (0.040)*** (0.040)*** (0.043)*** (0.040)*** (0.040)***
Δ Govt. consumption 0.038 0.022 0.019 0.038 0.022 0.020 0.035 0.025 0.024
(logs)
(0.026) (0.024) (0.024) (0.026) (0.024) (0.024) (0.026) (0.024) (0.024)
Δ Productivity (logs) -0.193 -0.179 -0.179 -0.193 -0.180 -0.180 -0.196 -0.181 -0.178
(0.083)** (0.077)** (0.077)** (0.083)** (0.077)** (0.077)** (0.083)** (0.077)** (0.077)**
Δ VIX -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)***
Carry_USD 0.000 0.000 -0.000 0.000 0.000 -0.000 0.000 0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Δ M2 (logs) 0.282 0.291 0.284 0.293 0.271 0.283
(0.020)*** (0.020)*** (0.020)*** (0.021)*** (0.019)*** (0.020)***
Intervention 0.238 -0.212 1.425 -1.255 0.071 0.003
(0.118)** (0.113)* (0.633)** (0.611)** (0.022)*** (0.021)
Intervention -0.464 -2.696 -0.080
instrumented
(0.172)*** (1.095)** (0.045)*
R2 0.21 0.33 0.34 0.21 0.33 0.34 0.21 0.33 0.33
N 1,144 1,144 1,137 1,144 1,144 1,137 1,144 1,144 1,137

39

Page 39 of 40
40

Page 40 of 40