WPS4176
Controls on Capital Inflows and External
Shocks
Antonio C. David1
adavid2@worldbank.org
Abstract
In this paper we attempt to analyze whether price-based controls on capital inflows are
successful in insulating economies against external shocks. We present results from VAR
models that indicate that Chile and Colombia, countries that adopted controls on capital
inflows, seem to have been relatively well insulated against external disturbances.
Subsequently, we use the ARDL approach to co-integration in order to isolate the effects of
the capital controls on the pass-through of external disturbances to domestic interest rates in
those economies. We conclude that there is evidence that the capital controls allowed for
greater policy autonomy.
JEL Classification: E60, F33, F34, O16, O54.
Keywords: Capital Flows, Capital Controls, Developing Countries.
World Bank Policy Research Working Paper 4176, March 2007
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage
the exchange of ideas about development issues. An objective of the series is to get the findings out
quickly, even if the presentations are less than fully polished. The papers carry the names of the
authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in
this paper are entirely those of the authors. They do not necessarily represent the view of the World
Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are
available online at http://econ.worldbank.org.
1I would like to thank Edouard Challe, Carmen Li, Chris Meissner, Gabriel Palma, John Eatwell and
participants at the Annual MMF Research Group conference in September 2005 for comments. I also
would like to thank Francisco Gallego, Hernan Rincon and Andre Loureiro for help with data.
1
Introduction
It is usually recognized by analysts that Chile and Colombia were spared from
the financial turmoil that originated in Mexico by the end of 1994; nevertheless, it is
also argued that those two countries suffered difficulties following the Asian and
Russian debacle at the end of that decade. Contrary to a large number of countries in
the region, Chile and Colombia adopted price-based capital account regulations in
order to manage their integration to international financial markets and avoid some of
the deleterious side effects of liberalization such as excessive real exchange rate
appreciations. The effectiveness of these regulations has been subject to heated debate
(see David, 2005 for a recent survey). In this paper, we will attempt to analyze
whether the controls on capital inflows adopted by the two countries were successful
in reducing the vulnerability of those economies to external shocks.
This paper is divided into three sections. Firstly, we will review the channels
through which the reserve requirements on capital inflows could possibly affect the
transmission of international shocks to the domestic economy. Subsequently, we will
present results from vector autoregressive models in order to assess the transmission
of global financial shocks to the Chilean and Colombian economies. Finally, in the
third section, we will use the ARDL approach to co-integration in order to attempt to
isolate the effects of the capital controls on the pass-through of external disturbances
to domestic interest rates.
2
1. The Unremunerated Reserve Requirement and International Shocks
Asymmetrici taxes, such as the unremunerated reserve requirements adopted
by Chile and Colombia, could affect the transmission of international shocks to the
domestic economy through several channels, including correcting market failures due
to moral hazard (see for instance Dooley & Walsh, 2000). They could act as "speed
bumps" in periods of excessive liquidity and expansionary pressures, which are a
characteristic of the upswing of financial cycles (Palma, 2000 and 2003). During this
credit expansion phase, there is an increase in credit risk because the quality of
financial intermediaries' portfolios decreases, leading to an increase in non-
performing loans.
This is aggravated by the fact that financial market agents frequently have
inappropriate responses to changes in risk over the economic cycle, such that risks are
underestimated during booms and overestimated during recessions. In the presence of
these dynamics, the adoption of restrictions on capital inflows in a countercyclical
manner in order to mitigate the expansionary effects associated with surges in capital
inflows could be an appropriate policy response in capital importing countries with
the objective of reducing economic volatility (Palma, 2003).
In addition, the currency mismatch strand of the financial crises literature
argues that if a substantial amount of debt is denominated in foreign currency due to
"Original Sin"ii, a country could become vulnerable to self-fulfilling crises (Aghion et
al., 2000). By taxing capital inflows it is possible to restrict the level of external
indebtedness and limit the negative real effects of devaluations due to the exposure of
domestic firms' balance sheets. In fact, benefits arise because the decision by an
individual firm to borrow in foreign currency imposes costs on the rest of the
economy that are internalized by the tax on inflows.
Furthermore, the maturity structure of external debt seems to matter for the
occurrence and severity of currency collapses (Carlson and Hernandez, 2002). This
fact has been addressed theoretically by the maturity mismatch literature (Rodrik &
Velasco, 1999). This type of model emphasizes that in the presence of market failures,
borrowers would not consider the effects of the stock of short-term debt on
contractual interest rates and would choose the privately less costly option of taking
short-term rather than long-term debt. Hence, reserve requirements on short-term
capital inflows could play a role in reducing the likelihood of liquidity problems by
mitigating the externality associated with short-term debt.
3
Moreover, the unremunerated reserve requirements (URR) might be able to
reduce the vulnerability of a country to external shocks through prevention of real
exchange rate overvaluation. Dornbusch, Goldfajn and Valdes (1995) show that an
overvalued real exchange rate is a fairly good predictor of future currency crises;
therefore the impact of the reserve requirements on the real exchange rate seems to be
of considerable importance.
There is a vast literature that focuses on specific transmission channels of
crises from one country/region to another and emphasizes mostly trade and financial
links as the main carriers of spill-over effects associated with financial stress (see
Dungey et al., 2003 for a survey). The spread of crises through trade links can take
place via competitive devaluations or simply by the reduction in demand for exports
due to a weaker economic situation in the "ground zero" and other affected countries.
The role of financial links is slightly more complex. While some authors
concentrate on common lender effects (a lender is affected by crisis in one country,
faces liquidity problems such as margin calls and has to liquidate its assets
elsewhere); others focus on "rebalancing effects" in the context of standard portfolio
models, where investors pull away from risky assets (emerging markets) when faced
with a large negative shock, i.e. an increase in volatility (Schinasi and Smith, 1999).
The latter effect is akin to a simple increase in risk aversion by international players.
Calvo and Mendoza (2000) emphasize the importance of information asymmetries
and costs in acquiring information about specific markets, which encourage herd
behavior by international investors as a major source of "contagion". We could also
consider a more general story, where in the presence of multiple equilibria, a crisis in
one country can act as a sunspot and coordinate investors' expectations. In this case,
the transmission of crises occurs purely because of changes in investor's beliefs.
What role could the URR play in addressing those transmission channels of
crises? This type of controls on capital inflows is capable of mitigating problems
related to liquidity risk by increasing the maturity structure of external debt. Hence it
may reduce the exposure to sudden capital outflows associated with the "common
lender", "risk aversion", "Calvo and Mendoza-style herd behavior" and other multiple
equilibria stories. Nonetheless, the reserve requirements cannot affect trade links in a
direct clear-cut way and therefore cannot prevent distress when trade is a major
transmission channel of shocks.
4
We should conclude this section by noting that controls on capital inflows may
be ineffective against runs by domestic residents, which were particularly important in
the Mexican and Brazilian crisis episodes (Frenkel & Schmukler, 1996 and Goldfajn,
2000). Nevertheless, one has to note that by changing the maturity structure of
external debt, the controls make runs more expensive and hence could play an indirect
role in preventing them.
2 Evidence from Vector Autoregressive Models
Our objective in this section is to obtain some stylized facts about the response
of domestic macroeconomic variables to external shocks, measured by the spread paid
by emerging markets over U.S. instruments of the same maturity. For this purpose, we
will use the global Emerging Markets Bond Index (EMBI) spreads constructed by J.P.
Morgan. This index tracks the traded market for U.S. dollar denominated Brady and
other similar bonds and is widely regarded as an appropriate measure of foreign
investors' risk perceptions regarding emerging markets. We chose to use the EMBI
rather than the commonly used EMBI+ spreads (which is a less restrictive index,
including Eurobonds and other sovereign debt instruments) because of data
availability as the latter series only starts in 1998iii.
We will implement reduced-form models of the transmission of international
shocks to macroeconomic variables using the vector autoregressive (VAR)
methodology. Other authors such as Edwards (2000) and Edison and Reinhart (2001)
have previously used this framework for similar purposes.
When specifying a reduced-form VAR model it is crucial to assess whether
the variables included in the system are stationary or not for inference to be valid,
since it is well known that testing hypotheses on coefficients of integrated variables
requires non-standard asymptotic theoryiv (Canova, 1995). In our first model, we will
use a foreign exchange market pressure index for Argentina, Chile, Colombia and
Mexico as endogenous variables in the system.
We construct the pressure index following the empirical literature on currency
crises in order to account for the fact that in periods of stress due to external shocks
the burden of adjustment does not fall exclusively on interest rates, but it is also
reflected in changes in international reserves and changes in the nominal exchange
rate. Our index is a weighted average of changes in domestic interest rates; nominal
5
exchange rates and the log of foreign reserves (see Eichengreen, Rose and Wyploz,
1996 for a similar index). Hence, the pressure index on foreign exchange markets for
country j at time t is given by:
empjt = it + st -
1 1
i s R1 Rt
We estimated the model from May 1991 to June 2001 and included the global
EMBI spread and the pressure indexes, as constructed above for the different
countries, as endogenous variables in our system. In addition, the junk bond spread
(this variable has been used in the literature as a measure of international investor's
risk appetite, see Mody & Taylor, 2002) and the log of world oil price index (included
to capture real shocks) were used as exogenous variables in the system. The data
definitions and sources are described in Appendix B.
We chose a lag structure of 4 for this model, as there was a conflict between
the different information criteria; the LM test does not detect serial correlation of the
residuals for this specification. The unit root tests performed and reported in
Appendix A showed that all the pressure indexes are stationary. Note also that both
tests reject the null of a unit root for the EMBI spread, the junk bond spread and the
world oil price series at the 5% level.
It is important to note that we decided not to include Brazil in our models due
to the fact that the 1994 Real stabilization plan represents a significant structural break
that would have reduced the sample size of the specifications (prior to 1994 this
country was experiencing a turbulent period of high inflation). In addition, Brazil
adopted controls on inflows of a quite different form (see David, 2007) and, therefore,
neither constitutes an adequate "observation" of a country that imposed reserve
requirements nor an appropriate "counterfactual" for our analysis.
The generalized impulse responses presented in Figure 1 demonstrate that the
pressure indexes for Chile and Colombia do not respond significantlyv to shocks to the
EMBI spread nor do they respond significantly to shocks to the other countries'
(Argentina and Mexico) pressure indexes. Nevertheless, the Argentinean and Mexican
indexes do respond to EMBI shocks and Mexico is the country that presents the
strongest response. It is also of interest to note that the Argentinean pressure index
presents a statistically significant response to shocks in the Mexican pressure index
indicating that financial "contagion" occurred between these two countries.
6
The advantage of using generalised impulse responses when compared to
standard Cholesky ones is that they do not depend on the ordering of the variables in
the system. Generalized impulse responses compare the conditional expectation of a
variable in the model, given a shock and the history of the model, to the conditional
expectation of that variable given the historically observed information of the model
(Koop, Pesaran & Potter, 1996 for details).
Table 1 shows the variance decomposition for the VAR model estimated
above. One should note that only a small percentage of the forecast errors in the
Chilean and Colombian pressure indexes can be attributed to EMBI shocks. In fact,
the figures are 0.07% and 1.17% for the first month for each country respectively.
Nevertheless, when we look at the Argentinean and Mexican indexes for the same
horizon, those figures become much larger (15.69% and 40.26% respectively).
These results indicate that the unremunerated reserve requirements (combined
with other capital account polices) might have helped to insulate the Chilean and
Colombian economies from certain types of global external financial shocks, namely
the ones captured by the EMBI spread. Evidently, at this stage, we cannot distinguish
whether this difference is due to capital account policies, other macroeconomic
policies, or simply the type of exchange rate regime adopted by the different
countries. One also has to note that the precise role played by the capital controls in
insulating those economies was not clarified in our empirical analysis so far.
In order to confirm the validity of our results we decided to estimate models
for Chile and Colombia individually, including a wider selection of macroeconomic
variables. The theoretical foundations of those VAR models are standard New
Keynesian sticky-price models of the monetary transmission mechanism applied to
small open economies, assuming that the domestic central bank responds (possibly
with a lag due to information frictions and measurement limitations) to deviations of
domestic inflation from the inflation target, to the output gap and to external
conditions captured by the terms of trade, the real exchange rate gap, the sovereign
risk premium and the EMBI spread.
First, we estimated a model of the Chilean economy using monthly data from
January 1991 to December 2000, including as endogenous variables the EMBI spread,
the output gap, the difference between inflation in the past 12 months (change in the
consumer price index) and the inflation target, the exchange rate indexed deposit rate
(this is the policy instrument), the Chilean sovereign risk premium (this variable was
7
included to capture Chile's default risk as perceived by foreign investors) and the
cyclical component of the real exchange rate. We also included as exogenous
variables the logarithm of the Chilean terms of trade in order to capture real external
shocks and the Junk Bond spread (for reasons already outlined). Since we assume that
Chile is a small open economy, it is intuitive to think that no biases arise by
considering those two variables as exogenous. The Akaike, Schwartz and H-Q
information criteria suggested a model with two lags and there is no evidence of serial
correlation for this specification, according to the LM test.
Once again variable sources and definitions are presented in the Appendix B.
The unit root tests performed (Appendix A) show that the null of non-stationarity is
rejected at the 5% level for the inflation, interest rate and country risk series. Non-
stationarity is also rejected for the terms of trade series at the 10% level and for the
output gap and real exchange rate series at the 1% level.
The generalized impulse responses are presented in Figure 2. One should note
that the domestic exchange-rate-indexed deposit rate does not present a statistically
significant response to shocks to the EMBI spread, whereas the real exchange only
presents a marginally significant response between the second and fourth months.
Therefore, the impact of EMBI shocks on the real exchange rate seems to be relatively
short-lived and small. In addition, the domestic deposit rate also seems to be resilient
to shocks to the Chilean sovereign risk premium and so does the real exchange rate.
These conclusions are confirmed by the variance decomposition analysis (see
Table 2). Shocks to the EMBI spread are only responsible for a small part of the
forecast errors in the real exchange rate and the domestic deposit rate. In the first
period, the EMBI accounts for 1.13% of forecast errors in the interest rate and 1.17%
in the real exchange rate, whereas in period 10 the figures are 4.79% and 10.43%
respectively. Hence, it seems that interest rates were insulated against external shocks
in Chile, whereas the real exchange rate is slightly more vulnerable.
To sum up, the URR and other capital account policies seem to have been
capable of reducing the pass-through of external shocks to the Chilean economy,
especially as far as deposit interest rates are concerned. Nonetheless, the capital
account policies did not completely insulate Chile, as the real exchange rate was more
vulnerable to shocks.
In addition, we estimated a similar model for the Colombian economy for the
period from January 1993 to June 2002. We included as endogenous variables the
8
EMBI spread, the output gap, the difference between the inflation rate and the annual
inflation target set by the Central Bank, the 90-day real deposit rate, the Colombian
sovereign risk premium and the cyclical component of the real exchange rate. We also
included as exogenous variables the log of the world oil price index (this variable was
included to capture real shocks, since we do not have a reliable series for the
Colombian terms of trade that spans the whole period), the junk bond spread and a
dummy variable that takes the value of zero before September 1999 and one
thereafter, to account for the change in the exchange rate band.
The Akaike information criterion suggests a model with two lags and the LM
test does not detect serial correlation for that specification. Unit root tests show that
all the variables included are stationary, except for the country risk premium
(Appendix A). We decided to estimate a model including this variable in levels to
make the interpretation of the results more clear. McCallum (1993) argues that if the
residuals of each equation in the system are stationary and there is no evidence of
serial correlation, the model in levels can be correctly estimated. The unit root tests
presented in Appendix A indeed show that the residuals of each equation in the
system are stationary.
The impulse response functions obtained are reported in Figure 3. One can
observe that the domestic interest rate presents a marginally significant response to
EMBI shocks between periods 2 and 5. The real exchange rate also presents a
marginally significant response to EMBI shocks from period two to four. Hence, it
seems that both domestic interest rates and the real exchange rate were relatively
insulated against external financial disturbances, as the effects of EMBI shocks were
small (barely different from zero) and short-lived.
When shocks to the country risk premium are considered, it is possible to
observe that according to this model, the domestic deposit rate only presented a
marginally significant response to country risk shocks. Nonetheless, the country risk
variable seems to have larger effects on the real exchange rate (the response of the
real exchange rate to country risk shocks is significant until the forth month and the
shock dies out after 6 months).
The variance decomposition analysis (see Table 3) shows that shocks to the
EMBI spread explain 23.24% of the forecast errors in the Colombian sovereign risk
after 2 months, but only 3.32% of errors in interest rates and 2.66% of errors in the
real exchange rate. These results clarify the conclusions obtained from the impulse
9
response functions. The Colombian country risk seems to co-move with global risk
premia, whereas domestic interest rates and the real exchange rate are relatively
insulated from those shocks. We may conclude that the capital account management
policies were relatively successful in insulating the Colombian economy against
global shocks (as measured by the EMBI spreads).
4. Disentangling the Effects of the URR
In this section, we will attempt to isolate the possible role of the reserve
requirements on capital inflows as far as the pass-through of external shocks to
domestic interest rates is concerned and confirm whether those capital account
restrictions contributed towards a greater autonomy for domestic monetary policy.
Indeed, following the results from previous sections, our main conjecture is that the
URR did reduce the effect of foreign shocks on domestic rates even in the context of a
crawling band exchange rate regime.
It is well known in the econometrics literature that interest rates frequently
behave in ways close to unit roots in finite samples. Nonetheless, it seems
counterintuitive that they would present unit roots as this would mean that some series
could go to infinity and others reach negative values. Series that are quite persistent in
this way can present the same estimation problems in standard models as regressions
involving non-stationary variables. In addition, the application of standard co-
integration analysis tools such as the Johansen procedure requires the classification of
the variables included in the model into I(1) or I(0), which is difficult in this case.
We will estimate error correction models that circumvent that issue for the
Chilean indexed policy interest rate and include the 90-day Libor interest rate (in real
terms) and the tax equivalent of the reserve requirement as explanatory variables.
Hence, we implemented a model describing the dynamics of the Chilean interest rate,
which is given by:
n m k
it = C + iit + it
-i *
j - j+ wURRt
-w +(it -1it -2URRt )
*
-1 -1 -1
i=1 j=1 w=1
This expression determines the first differences in the domestic interest rate as
a function of past changes of the rate itself, past changes in the LIBOR rate (i* ),
changes in the URR and an error-correction term.
10
As mentioned before, it is not possible to determine unambiguously whether
the variables that would form the long run relationship are stationary or not.
Therefore, we decided to adopt the autoregressive-distributed lag (ARDL) approach to
co-integration proposed by Pesaran et al. (2001) that does not require the
classification of variables into stationary and non-stationary. This approach consists of
two stages.
First, a test for the existence of a long-run relation between the variables is
implemented. It amounts to a simple F-test for the joint significance of the lagged
levels of the variables in the error-correction model outlined above. This F-statistic
has a non-standard distribution irrespective of whether the variables are I(0) or I(1)
and the appropriate critical value bounds have been tabulated by Pesaran et al. in
order to test for the null of no co-integration (i.e. coefficients are not significantly
different for zero).
In fact, there are two sets of critical values: one assuming that all the variables
in the model are I(0) and the other assuming that all the variables are I(1). This
provides a band covering all possible classification of variables. If the computed F-
statistic falls outside the band, it is possible to reach a conclusion regarding the
existence of a co-integration relationship. On the other hand, if the statistic falls inside
the band the result of the inference is inconclusive and depends whether the variables
are I(1) or I(0). The second stage of the analysis refers to estimating the coefficients
for the long-run relationships and making inferences about their values.
We estimated models for the Chilean rate on lagged values of itself, lagged
values of the LIBOR interest rate and lagged values of the tax equivalent of the
reserve requirementsvi for the period from January 1991 to December 2000 (six lags
were used in all specifications)vii. We experimented two basic different models: one
with a dummy for the turbulent period following the Russian default in 1998, which
represents a transition away from the use of controls on inflows and another model
including a dummy for the change in the exchange rate regime in September 1999.
As argued by Pesaran et al. (2001), ARDL estimation is applicable even when
the explanatory variables are endogenous, provided that the order of the model is
large enough to account for contemporaneous correlations between errors in the data
generating process. In our case, the foreign interest is evidently long-run forcing and
our tests for a model including the URR as the endogenous variable fail to reject the
11
null that the level variables do not enter significantly in the equation, which indicates
that the URR can be treated as an exogenous variableviii.
The F-statistic for those models of Chilean interest rates are 11.834 and 4.947
respectively, whereas the critical bounds at the 5% level, calculated by Pesaran et al.
are 3.10 and 3.87 (the critical bounds at the 1% level are 4.13 and 5.00). Hence, we
can conclude that there exists a long run relationship between those variables for the
period analyzed, irrespectively of their order of integration, as the test statistic
exceeds the critical bounds.
We now proceed to the estimation of the ARDL model. The Akaike
information criterion suggests a ARDL(2,0,4) model for the specification including
the 1998 dummy. If we consider the long-run relationship between the variables, the
estimated coefficients are (with p-values in brackets):
it = 4.51+ 0.27it + 0.20URRt + 2.96 Dummy1998
*
[0.00] [0.00] [0.00] [0.00]
One should note that all the coefficients are highly significant (even at the 1%
level). The coefficient of the URR is positive indicating that the URR allowed for
higher interest rates in the long-run than would have been the case in the absence of
capital controls. An increase in the URR by one unit would lead to an increase in the
interest rate of 0.2 percentage points.
Hence, the adoption of the URR mitigated part of the expansionary pressures
due to high capital inflows and allowed for greater monetary policy autonomy. One
also has to note that the coefficient for the foreign interest rate is only 0.27, therefore
indicating that the adjustment of local rates to changes in foreign rates is less than
proportional, thus confirming the evidence of relative monetary independence.
The estimation results for the error-correction model outlined above are
presented in Table 4. All regressors are statistically significant and the diagnostic
statistics do not show any evidence of serial correlation of the residuals and the high
R-squared (0.56) indicates a good fit.
One interesting feature of the results presented above concerns the coefficient
for the error correction term, which is relatively small (-0.306). This implies that the
half-life, calculated as ln(0.5)/ln(1+error-correction coefficient) is of approximately 2
months. This suggests that the speed of adjustment of domestic interest rates to
foreign ones is only moderate, hence indicating greater monetary policy independence
in the Chilean case. Therefore, we can conclude that the evidence so far corroborates
12
our previous inferences concerning the insulating properties of controls on capital
inflows.
Subsequently, we also applied the ARDL approach to analyse the effects of
the reserve requirements on Colombian 90-day deposit interest ratesix. We estimated a
very similar model to the one outlined for Chile. The sample period goes from
January 1991 to December 2000 and a dummy variable taking the value of 1 after
September 1999, when the exchange rate band was abandoned was included in the
regressions. The F-statistic for this model is 4.242, which lies outside the bounds at
the 5% level, but lies inside the bounds at the 1% level, therefore yielding
inconclusive results. Nonetheless, given the small sample size, significance at the 5%
level is more than adequate.
Hence, we can be reasonably confident that a long run relationship between
those variables exists, irrespective of the order of integration of the variables. In
addition, the tests indicate that the URR can be treated, once again, as a long-run
forcing variable (the F-statistic for the model with the URR as the dependent variable
is 1.84, which is below the bounds at the 5% level).
The Akaike information criterion suggests a ARDL (4,1,6) model. The
estimated coefficients for the long run relationship are (with p-values in brackets):
it = 3.96+1.74it + 0.11URRt - 5.08Dummy1999
*
[0.07] [0.02] [0.08] [0.06]
One should note that all the coefficients are significant at the 10% level and
have the expected signs. The coefficient of the URR is positive permitting us to infer
that the URR allowed for higher interest rates in the long-run. Nonetheless, the
coefficient is smaller than in the Chilean case. An increase in the URR by one unit
would lead to an increase in the domestic interest rate of 0.11 percentage points.
The coefficient for the foreign interest rate exceeds 1, implying more than
proportionate changes in domestic rate when the foreign rate changes. This is not very
surprising for an emerging market and it's in line with the evidence towards over-
adjustment of interest rates in developing countries to shocks in foreign interest rates
provided by Frankel et al. (2004).
The estimation results for the error-correction model are presented in Table 5.
Most regressors are statistically significant at conventional levels and the diagnostic
statistics do not show any evidence of serial correlation of the residuals. Once again,
13
the relatively high R-squared indicates a good fit of the model (although the normality
of the residuals is rejected at the 5% level, but not at the 1% level).
The coefficient for the error correction term is small (-0.145), which indicates
that adjustment towards equilibrium occurs slowly. In fact the half-life for Colombian
rates is approximately 4 months (twice the one of Chile). This suggests, once again, a
degree of monetary autonomy. Therefore, the evidence seems to indicate that the
controls on capital inflows adopted were capable of mitigating the impact of external
shocks on the domestic economy for both the Chilean and Colombian cases.
Conclusion
Unremunerated reserve requirements have the potential to be an effective
flexible instrument to balance macroeconomic disequilibria arising from surges in
capital inflows, such as large current account deficits linked to an excessive real
exchange rate appreciation. Furthermore, this type of capital controls could be
instrumental for the application of countercyclical macroeconomic policies; since they
allow for greater policy autonomy in the face of pro-cyclical capital flows, as
confirmed by the evidence presented in this paper.
We conclude that, when a very specific type of financial shock is considered,
namely shocks to the EMBI index, the Chilean and Colombian economies seem to
have been relatively well insulated against external disturbances, when compared to
other large Latin American countries. In addition, when we look at the two countries
individually, we found that, in the case of Chile, global financial shocks did not affect
domestic interest rates and only marginally affected the real exchange rate. The
Colombian economy also seems to have been relatively well insulated against external
disturbances, as EMBI shocks only marginally affected both interest rates and the real
exchange rate. One has to bear in mind that the reserve requirements are not the only
policy measures that could possibly have reduced the external vulnerability in those
countries.
Furthermore, by disentangling the specific effects of the capital controls on
domestic interest rates, we uncovered evidence that the reserve requirements allowed
for greater policy autonomy and did reduce the pass-through of external disturbances
to the economies considered. In fact, the effects of foreign interest rate shocks on
domestic rates were reduced even in the context of a crawling band exchange rate
14
regime. Despite the fact that controls on capital inflows have the potential to be
beneficial in terms of reducing financial vulnerability, this does not mean that these
policies were applied "optimally" by the two countries at all moments in time (see
Ffrench-Davis & Tapia, 2001).
15
REFERENCES
Aghion, P., Bachetta, P., Banerjee, A. (2000) "A simple model of monetary policy
and currency crises" European Economic Review, 44 728-738.
Bennett, H. & Valdes, R. (2001) "Series de terminos de intercambio de frecuencia
mensual para la economia chilena: 1965-1999." Working Paper 96, May. Santiago:
Central Bank of Chile.
Calvo, G.A. & Mendoza, G. (2000) "Rational Contagion and the Globalization of
Securities Markets" Journal-of-International-Economics; 51(1), June p. 79-113.
Canova, F. (1995) "Vector Autoregressive Models: Specification, Estimation,
Inference, and Forecasting" in Pesaran,-M.-Hashem; Wickens,-Michael-R., eds.
Handbook of applied econometrics. Volume 1. Macroeconomics. Handbooks in
Economics. Oxford and Malden, Mass.: Blackwell, 73-138.
Carlson, M. & Hernandez, L. (2002) "Determinants and Repercussions of the
composition of capital inflows." Board of Governors of the Federal Reserve System
International Finance Discussion Papers 717, January. Washington:DC.
David, A.C. (2005) "Are Price-Based Capital Account Regulations Effective?
Evidence from the Experiences of Brazil, Chile and Colombia in the 1990s." Faculty
of Economics, University of Cambridge, Ph.D., Dissertation.
David, A.C. (2007) "Revisiting Price-Based Capital Account Regulations in a
Sophisticated Emerging Market" World Development, Forthcoming.
Dooley, M.P. & Walsh, C. (2000) "Capital Controls and Crises: Theory and
Experience" University of California Santa Cruz. mimeo.
Dornbusch, R., Goldfajn, I. & Valdes, R. (1995) "Currency Crises and Collapses"
Brookings-Papers-on-Economic-Activity; 0(2): 219-70
Dungey, M. et al. (2003) "Empirical Modelling of Contagion: A Review of
Methodologies" CERF Working Paper Series n 8 University of Cambridge.
Edison, H. & Reinhart, C.M. (2001) "Stopping hot money" Journal of Development
Economics, vol.66, 533-553.
Edwards, S. (2000) "Contagion" World-Economy. July; 23(7): 873-900.
Eichengreen, B., Rose, A. & Wyplosz, C. (1996) "Contagious Currency Crises" The
Scandinavian Journal of Economics, vol. 98 (4).
Eichengreen, B. & Hausmann, R, (1999) "Exchange Rates and Financial Fragility" in
New Challenges for Monetary Policy, Federal Reserve Bank of Kansas City, Jackson
Hole Conference proceedings, forthcoming.
16
Ffrench-Davis, R. & Tapia, H. (2001) "Three Varieties of Capital Surge Management
in Chile" in Ffrench-Davis, R. (ed.) Financial Crises in Successful Emerging
Economies Brookings Institution Press Washington, D.C.
Frankel, J. et al. (2004) "Global transmission of interest rates, monetary independence
and currency regime" Journal of International Money and Finance 23, p. 701-733.
Frenkel, J. & Schmukler, S. (1996) "Country Funds Discounts and the Mexican Crises
of December 1994: Did local residents turn pessimistic before international
investors?" Federal Reserve Board working paper, IFDP 1996-563, September.
Gallego, F. Et al. (2002) "Capital Controls in Chile: Were they Effective?" in
Hernandez, L. & Schmidt-Hebbel eds. Banking, Financial Integration and
International Crises. Santiago de Chile: Central Bank of Chile.
Goldfajn, I. (2000) "The Swings in Capital Flows and the Brazilian Crisis" Pontificia
Universidade Catolica do Rio de Janeiro, mimeo.
Koop, G., Pesaran, M.H., Potter,S (1996) "Impulse Response Analysis in Nonlinear
Multivariate Models" Journal-of-Econometrics. September 74(1): 119-47.
Mody, A. & Taylor, M.P. (2002) "International Capital Crunches: The time-varying
Role of Informational Asymmetries" IMF Working Paper 43 February, Washington:
DC.
Palma, J.G. (2000) "The Three Routes to Financial Crises: the need for Capital
Controls" University of Cambridge, mimeo.
Palma, J.G. (2003) "Notes on Capital Flows to Developing Countries" University of
Cambridge, mimeo.
Pesaran, H. et al. (2001) "Bounds Testing Approach to the Analysis of Level
Relationships" Journal of Applied Econometrics 16: 289-326.
Rincon, H. & Villar, L. (2000) "The Colombian Economy in the1990s: Capital Flows
and Foreign Exchange Regimes" Borradores de Economia no 149 Banco de la
Republica. Bogota: Colombia.
Rodrik, D. & Velasco, A. (1999) "Short-Term Capital Flows" NBER Working Paper
7364 Cambridge:MA.
Schinasi, G.J. & Smith, R.T. (1999) "Portfolio Diversification, Leverage and
Financial Contagion". IMF Working Paper 99/136.
Sims, Stock & Watson (1990) "Inference in Linear Time Series Models with Some
Unit Roots" Econometrica-. January; 58(1): 113-44.
17
Figure 1
Response to Generalized One S.D.Innovations ± 2 S.E.
Response of Chile to EMBI Response of Chile to Argentina Response of Chile to Mexico Response of Colombia to EMBI
3 3 3 3
2 2 2 2
1 1 1 1
0 0 0 0
-1 -1 -1 -1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of Colombia to Argentina Response of Colombia to Mexico Response of Argentina to EMBI Response of Argentina to Argentina
3 3 3 3
2 2 2 2
1 1 1 1
0 0 0 0
-1 -1 -1 -1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of Argentina to Mexico Response of Mexico to EMBI Response of Mexico to Argentina Response of Mexico to Mexico
3 3 3 3
2 2 2 2
1 1 1 1
0 0 0 0
-1 -1 -1 -1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
18
Figure 2
Response to Generalized One S.D. Innovations ± 2 S.E.
Response of RISK to EMBI Response of RISK to RISK
.6 .6
.4 .4
.2 .2
.0 .0
-.2 -.2
-.4 -.4
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of Interest Rate to EMBI Response of Interest Rate to RISK
.6 .6
.4 .4
.2 .2
.0 .0
-.2 -.2
-.4 -.4
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of Real Exchange Rate to EMBI Response of Real Exchange Rate to RISK
.6 .6
.4 .4
.2 .2
.0 .0
-.2 -.2
-.4 -.4
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
19
Figure 3
Response to Generalized One S.D. Innovations ± 2 S.E.
Response of RISK to EMBI Response of RISK to RISK
1.0 1.0
0.5 0.5
0.0 0.0
-0.5 -0.5
-1.0 -1.0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of Interest Rate to EMBI Response of Interest Rate to RISK
1.0 1.0
0.5 0.5
0.0 0.0
-0.5 -0.5
-1.0 -1.0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of Real Exchange Rate to EMBI Response of Real Exchange Rate to RISK
1.0 1.0
0.5 0.5
0.0 0.0
-0.5 -0.5
-1.0 -1.0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
20
Table 1
Variance Decompositions
Percentage of Variance Associated with EMBI Shocks
Horizon Chilean Pressure Colombian Argentinean Mexican
Index Index Index Index
1 0.07 1.17 15.69 40.26
2 1.23 3.25 15.54 39.50
5 4.49 5.11 20.12 35.98
10 5.29 6.18 20.25 34.87
Table 2
Variance Decompositions
Percentage of Variance Associated with EMBI Shocks
Horizon Chilean Interest Chilean Country Chilean Real
Rate Risk Exchange Rate
1 1.13 17.28 1.17
2 1.37 29.54 3.67
5 3.01 33.69 8.87
10 4.79 28.83 10.43
Table 3
Variance Decompositions
Percentage of Variance Associated with EMBI Shocks
Horizon Colombia Colombia Colombia Real
Interest Rate Country Risk Exchange Rate
1 0.00 14.31 0.70
2 3.32 23.24 2.66
5 7.19 20.43 7.83
10 4.80 20.12 5.81
21
Table 4
Error-Correction Model for the Chilean Interest Rate (1991-2000)
Variable Coefficient R-squared Q-Statistic at lag 3 Normality
[P-values]
C 1.379 0.560 2.507 71.18
[0.00] [0.43] [0.00]
it 0.439
-1 [0.00]
it * 0.083
[0.00]
URRt -0.075
-1 [0.01]
URRt 0.163
-2 [0.00]
URRt -0.053
-3 [0.10]
Error-Correction-Term -0.306
[0.00]
Q-statistic refers to the test for serial correlation of the residuals of the model. Normality
is the Jarque-Bera test for the normality of the residuals in the regression. P-values for all the test
statistics are presented in brackets.
Table 5
Error-Correction Model for the Colombian Interest Rate (1991-2000)
Variable Coefficient R-squared Q-Statistic at lag 3 Normality
[P-values]
C 0.576 0.331 0.877 8.67
[0.13] [0.83] [0.01]
it 0.352
-1 [0.00]
it -0.010
-2 [0.90]
it 0.231
-3 [0.01]
it * -0.050
[0.85]
URRt -0.032
-1 [0.31]
URRt -0.000
-2 [0.98]
URRt 0.053
-3 [0.09]
URRt -0.001
-4 [0.94]
URRt -0.073
-5 [0.03]
Error-Correction-Term -0.145
[0.00]
Q-statistic refers to the test for serial correlation of the residuals of the model. Normality
is the Jarque-Bera test for the normality of the residuals in the regression. P-values for all the test
statistics are presented in brackets.
22
APPENDIX A
Unit Root Tests for Selected Variables
ADF-GLS test statistic Ng-Perron test statistic
EMBI Spread -2.284 (**) -2.223 (**)
Argentina Pressure Index -4.628 (***) -4.056(***)
Chile Pressure Index -4.809(***) -4.037(***)
Colombia Pressure Index -7.817(***) -5.185(***)
Mexico Pressure Index -5.353(***) -8.804(***)
Junk Bond Spread -2.198 (**) -2.205 (**)
Chile Output Gap -10.477 (***) -6.534 (***)
Chile Inflation Gap -2.204(**) -2.136 (**)
Chile Indexed Interest Rate -2.711(***) -2.598 (***)
Chilean Country Risk -2.659 (***) -2.496 (**)
Chile Terms of Trade -1.879 (*) -1.720 (*)
Chile Real Exchange Rate Gap -5.487 (***) -5.419 (***)
Colombia Output Gap -4.170 (***) -5.117 (***)
Colombia Inflation -2.413 (**) -2.546 (**)
Colombian Country Risk -1.241 -1.304
Colombia Real Interest Rate -1.620 (*) -1.664 (*)
Colombia Real Exchange Rate -1.743 (*) -9.615 (***)
Gap
World Oil Prices -2.755(***) -2.847(***)
(**) Denotes significance at the 5% level and (*) significance at the 10% level. Lag selection
based on Schwartz information criterion.
Unit Root Tests for Residuals in Colombian VAR
ADF-GLS Ng-Perron
EMBI Equation -8.134 (***) -5.090 (***)
Country Risk Equation -2.888 (***) -2.174 (**)
Output Gap Equation -2.549 (**) -1.958 (*)
Inflation Gap -5.325 (***) -4.066 (***)
Real Interest Rate -10.496 (***) -5.267 (***)
Real Exchange Rate -3.339 (***) -2.699 (***)
(***) Denotes significance at the 1% level (**) Denotes significance at 5% level and (*)
significance at 10% level. Lag selection based on Schwartz information criterion.
23
Appendix B
Overview of Data & Sources
Series Description/Notes Source
Domestic Colombia: 90-day real deposit rate. Chile: exchange Chilean Central Bank Website
Interest Rate rate indexed policy rate. (www.bcentral.cl) and Colombian
Central Bank Website
(www.banrep.gov.co).
EMBI Spread Global EMBI Spread. J.P. Morgan, obtained from
Thomson's Datastream.
Nominal IMF's International Financial
Exchange Statistics Database.
Rates
International IMF's International Financial
Reserves Statistics Database.
Output Gap Chile: difference between the log of the economic Author's calculations based on raw
activity index (IMACEC) and its trend and irregular data from respective Central Banks.
components obtained using a structural time series
model (estimated by Kalman filtering techniques).
Colombia: difference between the log of the industrial
production index and its trend, seasonal and irregular
components using a structural time series model.
Inflation Gap Deviation of actual Inflation from Inflation target. Chile: The monthly inflation target
series provided by Gallego at al.
(2002).
Colombia: Central Bank Website
(www.banrep.gov.co).
Real Exchange Chilean Central Bank Website
Rate (www.bcentral.cl) and Colombian
Central Bank Website
(www.banrep.gov.co).
Real Exchange Cyclical Component of Real Exchange Rate, obtained Author's Calculations.
Rate Gap by fitting a structural time series model (Kalman Filter)
to real exchange rate series.
Country Risk Chile: premium on international bonds issued by Central Bank of Chile, Rincon &
Premium Chilean corporations. Colombia: Series constructed Villar (2000) & J.P. Morgan,
using spread of Colombian bonds in international obtained from Thomson's
markets over US T-bills. Updated using EMBI+ until Datastream.
2002
Terms of Bennett & Valdes (2001) &
Trade Colombian Central Bank Website
(www.banrep.gov.co).
Junk Bond Difference in Yields between US High-Yield Bonds Merrill Lynch, obtained from
Spread and 10-Year US Treasury Bonds. A number of Thomson's Datastream.
different High-Yield indexes where used such as
Master II (H0A0) and High Yield 175 (X0A0)
World Oil IMF's International Financial
Price Index Statistics Database.
International 90-day Libor interest rate (in real terms) IMF's International Financial
Interest Rate Statistics Database.
24
Appendix C
Calculation of the Tax Equivalent
The vast majority of the literature adopts a "naïve" measure of this tax, which
is constructed by combining an arbitrage condition and the uncovered interest parity
equation. Consider the case where the maturity of the loan (capital inflow) k is larger
than the holding period of the deposit requirement (h ). Assuming that the exchange
rate is fixed, the uncovered interest parity would take the form:
(1+ i) = (1+ i*) +
Where i is the domestic interest rate, i* is the international interest rate and is the
tax equivalent of the reserve requirement. The following arbitrage equation is also
used:
(1- u)(1+ i)k + u(1+ E[max(i,i*)])k -h = (1+i*)k
where u is the unremunerated reserve requirement. The above expression
simply states that the return from investing in Chile (considering the reserve
requirements) has to be equal to the return obtained from investing abroad, i.e. the
international interest rate (opportunity cost). Using the UIP condition to substitute for
domestic interest rates and using the approximation that (1+ j)x (1+ xj) for a small
j , we can derive the usual naïve equation for the tax equivalent:
= u h i*
(1- u) k
Note that the tax-equivalent is inversely related to the maturity of the capital
inflow (k ), i.e. it falls more heavily on short-term capital inflows. This widely used
measure of the tax equivalent of the reserve requirements has significant limitations. It
does not take the country risk premium into account and does not consider variations
in the exchange rate. In our analysis, we will adopt a measure of the tax-equivalent
cost that incorporates those factors. Firstly, we will consider the case where the
holding period of the deposit (h ) is equal to the maturity of the loan (k ). The UIP
condition at present is given by:
(1+ i) = (1+ i*)(1+ )(1+ Se)(1+ )
where is the country risk premium and Se is the expected change in the exchange
rate. The arbitrage condition is given by:
[(
1+ i*)(1+ )(1+ Se) = (1- u)(1+ i)k + u(1+ Se)h
]
k
where the term of the left-hand side is the opportunity cost of the investment decision
(foreign interest rate plus country and currency risk), the first term on the right-hand
side represents the returns on the amount actually invested (i.e. minus the reserve
requirements) and the second term represents possible capital gains (or losses) from
changes in the exchange rate. By manipulating those two equations we can obtain the
following result:
1
(1+ ) = 1 [[(
1+ i*)(1+ )(1+ S e) - u(1+ S e)h
]
k ] k
(1+ i*)(1+ )(1+ S e) 1- u
1
Note that the same formula is valid for the case when the holding period of the
deposit requirement is larger than the maturity of the loan, i.e. the case where h > k .
This occurs because the arbitrage condition is very similar to the one outlined above,
more specifically:
25
[( 1+ i*)(1+ )(1+ Se) = (1- u)(1+ i)k + u(1+ Se)h
]
h
By manipulating this expression we obtain:
1
(1+ ) = 1 [[(
1+ i*)(1+ )(1+ Se) - u(1+ Se)h
]h ] k
(1+ i*)(1+ )(1+ Se) 1- u 1
Now we will proceed to calculating the tax equivalent when h < k i.e. when
the maturity of the investment is larger than the holding period of the reserve
requirement. If we assume that the proceeds of the deposit requirement are reinvested
abroad, the arbitrage condition becomes:
[( 1+ i*)(1+ )(1+ S e) = (1- u)(1+ i)k + u[(1+ Se)h (1+ i*)k ]
]
k -h
By manipulating this we obtain:
1
(1+ ) = 1 [[( ]k ] k
(1+ i*)(1+ )(1+ Se) 1- u 1 1+ i*)(1+ )(1+ Se) - u[(1+ Se)h (1+ i*)k-h ]
Evidently, those measures only constitute a proxy for the costs of the reserve
requirements and depend on a certain number of restrictive assumptions. In the
empirical implementations, we decided to use an average of the tax equivalent
variable for different maturities and we experimented with different measures for the
expected rate of exchange rate depreciation (or appreciation).
i Those taxes are asymmetric in the sense that they fall more heavily on short-term capital inflows and
did not apply to all capital flows (trade credits were usually exempt).
ii Original Sin is a term coined by Eichengreen and Hausman (1999) that essentially refers to the
inability of developing countries to issue debt abroad in their own currency.
iiiChilean securities are not included in the EMBI index and Colombian ones were only included after
1998. In any case, the VAR methodology would be able to tackle the possible endogeneity problems
that might arise.
iv Although Sims, Stock & Watson (1990) prove that coefficients are consistently estimated
independently of the order of integration.
v Throughout this paper a "significant" impulse response for each month following a shock means that
the interval defined by the error bands does not contain the value zero. Note that error bands were
calculated using Monte Carlo simulations (1000 repetitions).
viIn our specifications, we used an average of the tax equivalent for different maturities of capital
inflows.
vii One of the reasons why we did not extend our sample beyond December 2000 is that the inclusion of
the 1999 dummy variable would change the asymptotic results of the tests, if the fraction of the sample
during which the dummy is different from zero does not go to zero as the sample size increases (see
Pesaran et al, 2001, p.307).
viiiThe test statistic is 1.83, which is below the critical bounds both at the 5% level and at the 10%
level.
ix We used real (ex post) interest rates in this case to facilitate the comparison with the Chilean
experience and eliminate the noise created by volatile inflation.
26