Exchange Market Pressure, Currency Crises, and Monetary Policy
Additional Evidence From Emerging Markets

Contributor Notes

Author’s E-Mail Address: etanner@imf.org

This paper extends my previous work by examining the relationship between monetary policy and exchange market pressure (EMP) in 32 emerging market countries. EMP is a gauge of the severity of crises, and part of this paper specifically analyzes crisis periods. Two variables gauge the stance of monetary policy: the growth of central bank domestic credit and the interest differential (domestic versus U.S. dollar). Evidence suggests that monetary policy plays an important role in currency crises. And, in most countries the shocks to monetary policy affect EMP in the direction predicted by traditional approaches: tighter money reduces EMP.

Abstract

This paper extends my previous work by examining the relationship between monetary policy and exchange market pressure (EMP) in 32 emerging market countries. EMP is a gauge of the severity of crises, and part of this paper specifically analyzes crisis periods. Two variables gauge the stance of monetary policy: the growth of central bank domestic credit and the interest differential (domestic versus U.S. dollar). Evidence suggests that monetary policy plays an important role in currency crises. And, in most countries the shocks to monetary policy affect EMP in the direction predicted by traditional approaches: tighter money reduces EMP.

I. Introduction

In emerging markets, currency crises have occurred more frequently in recent years and continue to concern policy makers.2 To explain currency crises, much early work emphasized the role of monetary policy at or around the time of the crisis. In a familiar example, if the increase of the domestic credit portion of the monetary base (usually the counterpart of an unsustainable fiscal deficit) outstrips that of base money demand, foreign exchange reserves will be depleted and the exchange rate regime will collapse.3 Extending this logic, Krugman (1979) and many others thereafter explored the implications of market speculation against an exchange rate regime, including the precise timing of a regime’s collapse.4

Recent empirical work on currency crises has used a broader list of explanatory variables, often with the goal of forecasting currency crises over a medium-term horizon. For example, early warning systems developed by several authors have suggested that certain fundamental indicators (including real exchange rate overvaluation and growth rates of GDP, exports, and banking sector credit) help forecast exchange rate crises in advance (as much as two years).5

In recent years, as Kaminsky and Reinhart (1999) argue, currency crises have been increasingly associated with financial fragilities, like the Mexican and Asian crises. Accordingly, financial sector variables may also help to predict exchange rate crises. And, several researchers have tested for external common shocks or market contagion.6

While fiscal and financial fundamentals may help predict exchange rate crises over the medium term, and contagion may be important, monetary policy may nonetheless play an important role in exchange rate crises, especially at or around the time of the crisis. For example, a country’s central bank might accommodate pressures on the banking system and/or adverse external shocks, expanding the domestic portion of the money supply and sterilizing outflows as it provides liquidity to the banking system.7

In this light, the question of whether monetary policy is effective in preventing or forestalling crises has recently received renewed theoretical attention (Bensaid and Jeanne (1997), Drazen (1999), Furman and Stiglitz (1998), Flood and Jeanne (2000), and Lahiri and Végh (2000). For example, some have suggested that higher interest rates might place undue pressure on the banking system, thus causing a recession and exacerbating problems in exchange markets. Others have noted that without a fiscal adjustment, tighter monetary policy increases the burden on the intertemporal budget and may hence be counterproductive.

To empirically address questions like these, many authors have used measures of exchange market pressure (EMP) that include exchange rate depreciation and movements of international reserves (and, in some cases, the interest rate differential). Such a variable can be used for different exchange rate regimes, providing a more complete picture than either exchange depreciation or reserve movements in isolation.8

This paper extends a previous one also written by myself (Tanner (2001)). Like that paper, this one addresses several questions regarding EMP and monetary policy. For example, are substantial movements in EMP, including exchange rate crises and other periods of turbulence, due mainly to changes in money demand or money supply? Do the relationships between EMP, money supply, and money demand change during periods of common shocks or turbulence in global capital markets? If so, do these relationships change in a consistent way? What can we say about the stance of monetary policy at or around the time of exchange rate crises?

As in the previous paper, EMP is defined herein as the sum of exchange depreciation (in percent) and reserve losses (in percent of the monetary base).9 However, this paper examines 32 countries, as compared to 6 in the previous one. This measure of EMP has three components: a real money demand component, a money supply component and a real exchange rate component.10 Accordingly, this decomposition is well suited to address questions like those posed above.

Next, the relationship between EMP and monetary policy is examined in a vector autoregression (VAR) system that is estimated for most countries in the data sample.11 Such a framework is well suited to examine relationships between monetary policy and EMP in both directions. Does monetary policy affect EMP as expected direction? And, in the other direction, what, if any, is the feedback relationship from EMP to monetary policy? Is outflow sterilization, like that found in Mexico and certain Asian countries, evident elsewhere?

The paper is organized as follows. In Part II, measures of exchange market pressure (EMP) and the stance of monetary policy are discussed. Summary statistics regarding EMP and monetary policy are presented for 34 emerging market countries, including through selected common-shock windows, namely Mexican, Asian, Russian, and Brazilian Crises. In Part III, the VAR model is developed and estimated. Part IV presents a summary and some policy implications.

II. Exchange Market Pressure (EMP) and Its Components

Many countries limit exchange rate flexibility with purchases or sales of international reserves. In such regimes, when measuring exchange market pressures (EMP), the exchange rate and international reserves must be considered jointly. In this regard, Table 1 presents average values for the growth of the nominal exchange rate measured in domestic currency units per dollar (ε) and the change in international reserves (NIR) as a fraction of the monetary base (MB, r = ΔNIR/MB), for 34 emerging market countries in the Western Hemisphere (Latin America and Jamaica, hereafter referred to as WH), Asia, and Europe.

Table 1.

Summary Indicators of Exchange Rate Regime

article image
article image
INTIND = var (r)/[var(r)+(var(ɛ)], r = ΔR/MB, R = reserves, MB = monetary base, e = % ΔE, E = nominal exchange rate (domestic currency units per dollar).

For WH countries, calculations are limited to recent low inflation periods. For further information, see bottom of next page.

Notes: All variables are computed on a monthly basis. INTIND = var (r)/[var(r)+(var(e)], r = DR/MB, R = reserves, MB = monetary base, e = %DE, E = nominal exchange rate (domestic currency units per dollar).

Table 1 presents weighted regional averages (including calculations for WH countries that omit Argentina, Brazil, and Mexico for WH countries).12 For WH countries, calculations are presented only for the more recent period of lower inflation. Data are presented for both the entire sample period and certain subsamples that correspond to periods generally associated with common global shocks, namely “Tequila Spillover” (94:9-95:5, hereafter referred to as TS), Asian (97:4-98:4), and Russian/Brazilian (98:7-99:7) periods.13

Table 1 also presents calculations of an intervention index, INTIND, defined as the variance of (scaled) reserve movements r as a ratio of the sum of the variances of scaled reserves and exchange depreciation: INTIND = var(r)/[var(r)+var(ε)]). INTIND thus summarizes the degree of exchange market intervention. Under a perfectly fixed exchange rate regime, INTIND is unity.14

This table suggests most countries intervene to some degree in their foreign exchange markets. In this context, several authors have suggested using measures or indices of EMP that include both variables.15 One definition of EMP, due to Girton and Roper (1977) is the sum of exchange rate depreciation and reserve outflows (scaled by base money):

EMPt=εtrt(1)

where ε is the growth of the nominal exchange rate measured in domestic currency units per dollar and r is the change in international reserves as a fraction of the monetary base (ΔNIR/MB).16 To demonstrate the intuition behind this definition of EMP, suppose that the excess supply of money rises by 1 percent. Under a fixed exchange rate, r = -1 percent; under a flexible exchange rate regime, ε = 1 percent, and under a mixed regime, these two variables would move in some combination.17

In an appendix, EMP is shown to contain three elements, namely a nominal money supply element, a real money demand element, and a residual term that includes changes in both the bilateral real exchange rate and foreign (U.S.) inflation:

EMPtδtmt+λt(2)

where, δt is the change in net domestic credit scaled by the monetary base (ΔNDA/MB), mt is the growth of real base money, and λt=πt*+zt, where zt is the change in the bilateral real exchange rate and πt* is foreign inflation. Thus, λt may be thought of as a residual term that includes real exchange rate changes.

Equation (2) underscores the question of how to gauge whether monetary policy is “tight” or “loose.” Much recent work, applied to the U.S. and other industrialized countries, has emphasized interest rates (like the Federal Funds Rate in the U.S.) as an indicator of the stance of monetary policy (higher interest rates mean tighter money) since interest rates are the policy instrument, reflecting the tastes and preferences of policy makers.19

By contrast, older literature emphasizes monetary aggregates to gauge the stance of monetary policy.18 However, for open economies with fixed or managed exchange rates there are advantages to focusing on the domestically determined component of the monetary base (δ) much like the traditional monetary approach.19 A more formal modeling strategy that includes both interest rates and δ is presented in Part III. As an intuitive justification for using δ, note that even if the central bank targets an interest rate ex-ante, domestic credit adjusts ex-post in order that money markets to clear at that interest rate. So, while interest rates may reflect the intentions of the policy makers, domestic credit indicates ex-post whether monetary policy is tight or loose.20 Thus, according to equation (2), EMP increases when real money demand (m) decreases, when the domestic component of the money supply (δ) increases, when the real exchange rate depreciates and/or foreign inflation falls (λ).21

One might also ask about the relative importance of the elements in the right hand side of equation (2). Are movements in EMP primarily due to changes in money demand, money supply, or other factors contained in λ? To address this question, consider next this expression for the variance of EMP:

var(EMP)=var(δ)+var(m)+var(λ)+2[cov(δ,m)+cov(δ,λ)+cov(m,λ)](3)

Table 2 presents the means of EMP, δ, m, and λ, the variance of EMP, and the elements on the right hand side of equation (3) as a fraction of var(EMP), for the 32 emerging market countries. For WH, calculations are presented both for a longer data sample and recent, lower inflation periods. Unsurprisingly, EMP is higher and more variable in the WH and Europe than in Asia. However, much EMP in the WH reflects the high inflation periods in certain countries (notably Brazil). When high inflation periods for WH are excluded EMP falls and becomes less variable.

Table 2.

Exchange Market Pressure and its Components: Means and Variances

article image
article image
article image
article image
article image
article image
article image
article image
Notes: EMP= exchange market pressure, δt = change in domestic credit scaled by monetary base, mt= percent growth in real money base (deflated by CPI), λt= EMP — δt + mt= residual term (see text). All data are monthly. For time periods, including definitions of Tequila, Asia, and Russia/Brazil windows, and calculation of weighted averages, see Table 1.

An important message of Table 2 is that for most countries and time periods, money supply factors are an important element of EMP. Among WH and Asian countries, var(δ) as a fraction of var (EMP) generally exceeds either var(m) or var(λ). In these regions, in most cases, var(δ)/var(EMP) is close to (if not greater than) unity. In Europe, however, relative to other regions, var(δ)/var(EMP) is somewhat lower while var(m) and especially var(λ) are considerably greater.

The data suggest that periods of increased exchange market pressure are accompanied by expansions of domestic credit by the central bank, reductions in real money demand, and real depreciations. In Table 3, simple bivariate regression coefficients of EMP on δ, m, and λ, estimated country-by-country, are presented. These regressions are intended to summarize the data, not to imply causality. The estimated coefficients ∂EMP/∂δ are positive and almost always significant; those for ∂EMP/∂m are less than zero and in most cases significant; and those for ∂EMP/∂λ, are almost always positive and significant.

Table 3.

Bivariate Regressions of EMP with δ, m, λ

article image
Regression is: EMPt = a0 + a1 Zt + errort, Zt = (δt, mtt). EMP= exchange market pressure, δ = change in domestic credit scaled by monetary base, m = percent growth in real money base (deflated by CPI), λ = EMP − δ − m = residual term (see text). All data are monthly.

Focusing specifically on the common shock periods, casual inspection also suggests that currency crises tended to be more severe in those countries where monetary policy was looser, especially during the TS and Asian periods (but less so for the Russian and Brazilian episodes). Figures 1-4 show present average monthly EMP and δ – for those specific months when EMP peaked – for selected countries in each region.22 For the TS and Asian crisis periods the charts convey a positive cross-country relationship between EMP and δ.23

Figure 1.
Figure 1.

Tequila Crisis Window Exchange Market Pressure (EMP) and Domestic Credit Growth (δ)

Citation: IMF Working Papers 2002, 014; 10.5089/9781451843132.001.A001

Figure 2.
Figure 2.

Asia Crisis Window Exchange Market Pressure (EMP) and Domestic Credit Growth (δ)

Citation: IMF Working Papers 2002, 014; 10.5089/9781451843132.001.A001

Figure 3.
Figure 3.

Russia Crisis Window Exchange Market Pressure (EMP) and Domestic Credit Growth (δ)

Citation: IMF Working Papers 2002, 014; 10.5089/9781451843132.001.A001

Figure 4.
Figure 4.

Brazil Crisis Window Exchange Market Pressure (EMP) and Domestic Credit Growth (δ)

Citation: IMF Working Papers 2002, 014; 10.5089/9781451843132.001.A001

There are, however, some important cases where EMP appears somewhat higher or lower than the contemporaneous value of δ might otherwise indicate. For example, during the TS period, EMP was high in Romania despite a tight monetary policy. During the Asian window, in Chile, Venezuela, and Malaysia, EMP seems somewhat higher than (low) δ might otherwise suggest. To help explain Malaysia, controls on foreign capital and ceilings on domestic bank credit may play a role. 24 By contrast, EMP in Indonesia during this period is somewhat less than indicated by high δ during that period. And, three European countries, namely Poland, Romania, and Bulgaria substantially expanded domestic credit at this time while experiencing only modest EMP. Note also that, in Brazil, during the Russian window of late 1998 and the Brazilian window of early 1999, periods of loose fiscal policy, EMP is high despite low δ. Note however, that the central bank did not raise interest rates during this episode. Rather, money demand fell dramatically.

III. EMP and Monetary Policy: A Vector Autoregression (VAR) Analysis

The purpose of this section is to develop and test an empirical model that links EMP to monetary policy. Extending my previous work (Tanner (2001)), a vector autoregression (VAR) framework will be used.25 The main question this model seeks to answer is whether monetary policy affects EMP in the “right” direction. That is, does tighter monetary policy reduces EMP? This question has recently received renewed theoretical attention. For example, several authors, including Flood and Jeanne (2000) and Lahiri and Végh (2000) suggest that defensive monetary policy may be ineffective or even counterproductive, since it implies a higher burden on the budget through higher interest payments.

An issue closely related to currency crises is sterilized intervention. As Flood and Marion (1998) note, within the context of a simple “modified first-generation model (of collapsing exchange rate regimes)…no fixed exchange rate regime can survive, even for a moment, if the monetary authority plans to sterilize an attack and those plans are understood by speculators.” Accordingly, the model provides estimate the reaction of monetary policy to EMP shocks, including sterilization of reserve outflows by the central bank.

A. Empirical Framework

To develop an empirical framework model, several preliminary issues must be addressed. One such issue is how to model monetary policy. Is the central bank’s instrument an interest rate or a monetary aggregate? For the econometrician, which variable better captures the stance of monetary policy?

A compromise strategy that permits a role for both a monetary aggregate (δ) and the interest differential (ϕ) is developed below. Consider the vector autoregression (VAR) system:

Xt=a0+a1Xt1+a2Xt2+.+vt(4)

where X = (δ, EMP, ϕ), is a vector of variables that includes scaled domestic credit growth (δ), exchange market pressure (EMP) and ϕ is the interest differential, ai is a vector of coefficients, and Vt = (V δ, V E, V ϕ) is a vector of error terms.26 A system like (4) permits testing for effects of past values of X on current values. Assumptions regarding the contemporaneous relationship between the variables in X, including the exogeneity of certain variables (like a policy variable), are easily incorporated into a system like (4). To do so, first note that each element of the error vector vt is in turn composed of “own” error terms wt = (wδ, we, wϕ) and contemporaneous correlations with “other” errors. That is:

vδ=wδ+βδEwE+βδϕwϕ(5a)
vE=βEδwδ+wE+βEϕwϕ(5b)
vϕ=βϕδwδ+βϕEwE+wϕ(5c)

Thus, as mentioned above, contemporaneous relationships among variables in X are reflected in both the expected signs of and a priori (exogeneity) restrictions on the β coefficients.27

As with any VAR system, exogeneity restrictions, an “ordering,” must be chosen to reflect an underlying structural model. Of course, a common criticism of VAR’s regards the choice of ordering. Orderings should not be chosen merely to obtain a desired result. Why is one ordering preferred to another? Such criticisms may confronted by clearly stating the assumptions behind the preferred ordering, but also providing results of alternative orderings.

Consider therefore the exogeneity restrictions for system (7a)-(7c) summarized by βδE = βδϕ = βδϕ = 0, a Choleski ordering of δ, BMP, ϕ. This ordering permits a role for both monetary aggregate (δ) and an interest rate variable (ϕ) as the central bank’s policy instrument, since the central banks’ preferences regarding δ and ϕ in the current period are assumed to be contained in the shock terms wδ and wϕ. This reflects a reality of policy in many central banks: both variables are instruments; even while a central bank sets an interest rate, it may also ration by quantity.

A feature of this ordering is that, while δ and ϕ are jointly determined, prices are assumed adjust to faster than quantities. Thus, current shocks to δ (wδ) are assumed to affect ϕ immediately, but shocks to ϕ (wϕ) only affect δ with a lag. Likewise, current shocks to δ (wδ) affect EMP immediately, but shocks to ϕ (wϕ) only affect EMP with a lag. Therefore, if monetary policy has a contemporaneous effect on EMP in the expected direction, β > 0.

We may think of shocks to EMP (wE) as exogenous changes in the willingness of international investors to keep their funds in the country. How might such a shock affect monetary policy and the interest differential? Consider first the demand for central bank lending by both the domestic banking system and the nonfinancial public sector, wδ(d):

wδ(d)t=(1k)wEt+a1(vϕtwϕt),a1<0(6)

Demand is negatively related to the interest rate. Also, since current innovations in the interest rate wϕ are assumed to affect the quantity of credit only with a lag, the current demand for credit wδ(d)t depends on the noninnovation portion of interest rate movements, namely vϕt - wϕt.

In this context, a policy of central bank accommodation, or sterilized intervention, is easily analyzed. An external shock wE causes a drain on banking system deposits. Under a fixed exchange rate regime, with no change in the interest differential ϕ, deposits likewise fall by wE. The banking sector thus demands to borrow more. Assuming a required reserve ratio is k (0 < k < 1), no excess reserves prior to the shock, and ϕ constant, the central bank thus accommodates these additional demands (sterilizing outflows) with a liquidity injection of (l-k)wE.28

Also, this framework permits an alternative interpretation of equation (5c): by inverting (6) and imposing equilibrium in credit markets (wδt = wδ(d)t), it is seen that current shocks to domestic credit (wδ) and EMP (wE) are assumed to affect the interest differential immediately.29

Interpreted thusly, the model has two additional implications. First, EMP shocks and the interest differential should be positively correlated, since βϕE = -(l-k)/a1 > 0. This should not be surprising: higher exchange depreciation and/or lower reserves signals more risk for investors. Second, if the central bank loosens up credit, ϕ should fall: βϕδ = 1/a1 < 0.30

We may now summarize policy issues in terms of the parameters. First, does monetary policy positively affect EMP (tighter monetary policy reduces EMP) as expected? Regarding the monetary aggregate δ as the policy variable, for contemporaneous effects, if monetary policy works in the expected direction, β should be positive and significant.

Note also that lagged effects are contained in corresponding IRF’s and F-tests (Granger causality) of δ on EMP in system (4). Regarding the interest differential ϕ, lagged effects are captured by corresponding IRF’s and F-tests (Granger causality) of ϕ on EMP in system (6). If monetary policy works in the expected direction, these effects should be negative and significant. If by contrast an interest rate defense exacerbates financial fragilities or the quasifiscal deficit (as mentioned by Flood and Jeanne (2000), Lahiri and Végh (2001), and others) the relationship between ϕ and EMP should be negative.31

Second, is monetary policy itself a function of EMP, as discussed above? Regarding the monetary aggregate δ, the reaction function is summarized by lagged IRF’s and F-tests (Granger causality) of EMP on δ. Regarding the interest differential ϕ, the reaction function is summarized by lagged IRF’s and F-tests (Granger causality) of EMP on ϕ.

B. Individual Country Estimates

VAR system (4) is estimated for 32 of the countries previously listed.32 In Tables 4 and 5, results of exclusion tests (F-statistics) and impulse response functions (IRF’s) are reported.33 As a preview of results discussed in detail below, there is substantial evidence that changes in monetary policy have an impact on EMP. In most cases, the sign of the impact is consistent with traditional models: expansions of domestic credit (δ) increase EMP, while in most cases an increase in the interest differential ϕ reduces EMP. However, as noted below, there are a few cases in which ϕ and EMP are positively (rather than negatively) related to EMP. In any event, evidence regarding the relationship between δ and EMP is somewhat stronger than that relating ϕ and EMP. Regarding feedback mechanisms in the other direction (see Table 5), EMP helps explain δ and/or ϕ in about half the sample.

Table 4.

Summary, Impacts of Monetary Policy on EMP, Individual Country Estimates

System (4): Xt = a0+ a 1Xt-1 + a2 Xt-2 + .... + vt,, X = (δ, EMP, Δϕ)

article image
Notes: F-test: *,* indicates significant F-statistic at 90, 95 percent levels, respectively; “a” indicates positive T-statistic of IRF in excess of |2.0|; Higher order: “+”/“-”: positive, negative IRFs with T-statistics in excess of 2.0, lag length in parentheses. Example: +(2,4) indicates positive shocks with T-statistic > 2.0 at lags 2 and 4. EMP = Exchange market pressure; 8 = change in domestic credit relative to money base; ϕ = interest differential.