Chapter

4 Financial Dollarization in Latin America

Author(s):
Adrián Armas, Eduardo Levy Yeyati, and Alain Ize
Published Date:
July 2006
Share
  • ShareShare
Show Summary Details
Author(s)
Robert Rennhack and Masahiro Nozaki1 

4.1 Introduction

In the past fifteen to twenty years, many developing countries have experienced a process known as financial dollarization (FD), in which residents hold deposits denominated in foreign currency – the US dollar in many cases.2 In several countries, this has been accompanied by dollarization of the real sector, with a large share of purchases of goods and services and payment of wages taking place in foreign currency, or by currency substitution, where foreign currency also serves as a means of payment. The process of FD has usually occurred in the aftermath of a severe economic crisis involving high inflation that undermines confidence in the local currency. Moreover, in many of these countries, dollarization remains very high, even though economic performance has improved and inflation has subsided.

Over the past decade, concerns about the effects of FD have increased. FD can help an economy by discouraging capital flight and encouraging residents to keep their savings in the domestic financial system. Yet it also carries potentially significant drawbacks, especially by narrowing the scope for policy manoeuvre during a crisis.3 If residents maintain significant cash balances in foreign currency, monetary policy may be less effective in managing domestic liquidity to control inflation or to dampen the effects of banking difficulties through lender of last resort financing. More importantly, banks in highly dollarized countries tend to lend in foreign currency to borrowers with little or no foreign exchange earnings. This could weaken balance sheets by creating a significant currency mismatch. Banks could suffer severe losses in the event of a sharp real depreciation, which would drive up the costs of servicing foreign currency debt without necessarily raising the borrowers’ income. Governments in highly dollarized countries also face this risk, as they tend to collect revenues in local currency while servicing debts in foreign currency. In this situation, high FD can deepen an economic crisis, such as in the case of Argentina in 2001 and Uruguay in 2002.

For this reason, the policy debate has focused on the causes of FD and the best policies to promote a recovery in the use of local currency for financial transactions and savings. This chapter seeks to test several explanations for FD, with an emphasis on Latin America – a region that includes countries which have avoided FD as well as those with persistently high FD. And in the past few years several countries in the region – most notably Paraguay and Peru – have been able to reduce the extent of FD. Section 4.2 reviews the empirical trends in FD. Section 4.3 assesses whether FD has been a rational response to inflation uncertainty. Section 4.4 looks at the role of exchange rate policy and currency mismatches in encouraging and perpetuating FD. Section 4.5 reviews the policy implications of the results.

4.2 Trends in financial dollarization

FD increased in most developing country regions between the mid-1990s and early this decade (Table 4.1). The use of foreign currency rose most rapidly in the transition economies, with almost half of all bank deposits denominated in foreign currency by 2001. FD rose in Latin America and Africa, while holding steady in Asia during this period. This trend occurred despite a significant decline in inflation after 1995 in most regions (Table 4.2).

In the early 1990s, Latin America, Africa and the transition economies experienced high inflation on average. Asia experienced a moderate rise in inflation around the time of the Asian crisis in 1997-8. However, by the late 1990s, all of these regions had rates of inflation close to industrialized country levels.

Table 4.1Dollarization by region, 1995 and 2001
19952001
Transition economies34.447.8
Of which:Bosnia and Herzegovina62.5
Bulgaria29.557.2
Hungary30.520.5
Poland27.618.9
Russia28.534.3
Slovenia42.136.1
Ukraine36.832.4
Asia31.030.3
Of which:Indonesia19.720.1
Korea0.53.5
Lao People’s Democratic Republic57.382.7
Philippines24.730.7
Thailand0.31.3
Vietnam34.643.4
Africa23.231.9
Of which:Angola25.481.0
Ghana25.6
Nigeria4.15.0
South Africa0.76.2
Zambia20.142.7
Latin America39.844.3
Table 4.2Average inflation by region (in per cent per year)
1990–41995–92000–3
Africa469.8127.037.6
Asia7.311.24.4
Industrialized4.62.12.2
Latin America365.614.89.3
Transition873.044.110.4
Source: International Financial Statistics (IFS).

Looking more closely at Latin America, FD picked up sharply between 1990 and 2001 (Table 4.3). Foreign currency deposits (FCDs) as a share of total deposits rose significantly in countries that were already highly dollarized, such as Bolivia and Uruguay. Dollarization also increased in countries with lower levels of dollarization in 1990, such as Costa Rica, the Dominican Republic and Honduras, Nicaragua and Paraguay. Early this decade, Ecuador and El Salvador opted for full, official dollarization, each under very different circumstances. Five countries in Latin America – Brazil, Chile, Colombia, Mexico and Venezuela – have avoided significant dollarization, even though they also have experienced severe macroeconomic problems since 1980. These countries preserved the demand for their currencies through a combination of sound economic policies, indexed financial instruments and legal restrictions on dollarized transactions. Except for Venezuela, residents of these countries placed their foreign currency assets abroad, but even in these cases, total foreign exchange deposits (including offshore deposits) were less than in the highly dollarized countries.4 Moreover, by shifting the FCDs abroad, these countries insulated their domestic banking systems from the risks associated with FD.

Table 4.3Selected Latin American countries: deposit and loan dollarization
Foreign currency-denominated deposits (in per cent of total deposits)Foreign currency-denominated loans (in per cent of total loans)
199020012002200320042001200220032004
Argentina47.271.54.26.710.780.07.27.114.1
Bolivia80.791.590.890.085.397.097.397.797.7
Brazil0.06.16.56.518.019.412.0
Chile16.314.012.813.211.913.813.010.310.3
Colombia0.30.50.40.02.011.011.68.86.1
Costa Rica26.849.150.050.256.667.253.055.553.3
Dominican Republic12.223.926.127.525.127.630.937.027.3
Ecuador13.3100.0100.0100.0100.0100.0100.0100.0100.0
El Salvador4.1100.0100.0100.0100.0100.0100.0100.0100.0
Guatemala0.05.18.812.414.915.316.717.7
Honduras1.833.434.235.135.722.222.826.430.9
Mexico10.18.14.64.55.420.512.912.39.8
Nicaragua40.370.672.169.668.783.683.184.385.0
Paraguay133.966.668.763.047.052.858.255.751.7
Peru62.574.373.270.664.180.579.777.975.9
Uruguay288.683.090.093.083.066.081.076.070.0
Venezuela0.20.20.20.10.70.80.70.6
Sources: Central banks; and IMF staff estimates.

Since 2001, FD has declined in some Latin American countries. Argentina forced its residents to convert foreign currency into pesos, thereby reducing its dollarization sharply. Bolivia, Peru and Uruguay have experienced moderate declines in FCDs as a share of total deposits, while FD fell sharply in Paraguay in 2004. Nonetheless, the extent of FD still remains high in many of these countries.

This persistence of FD seems puzzling because most of Latin America made significant gains in macroeconomic stability in this period. Both the rate and volatility of inflation declined significantly since the mid-1990s. Also the real exchange rate became more volatile, compared with the previous fifteen years, which would tend to discourage FD (Table 4.4). The rise in the volatility of the real exchange rate probably results from the adoption of flexible exchange rate regimes in the late 1990s by many Latin American countries. The central government deficit declined as well, while financial systems appear to have deepened. Real economic growth has remained steady at 2.5 per cent a year on average, while real lending interest rates have become positive in real terms.

Table 4.4Indicators of macroeconomic stability, 1980-2003 (in annual per cent change, unless specified otherwise)
1980-951996-2003
Latin America1Outside

Latin America
Latin AmericaOutside

Latin America
MeanVolatility2MeanVolatility2MeanVolatility2MeanVolatility2
Nominal stability
M2216.34.075.54.415.41.030.25.0
CPI244.84.489.69.411.21.321.08.1
Nominal exchange rate234.74.884.217.214.21.943.110.9
Lending interest rate81.14.727.43.129.20.620.41.1
Government deficit
(in per cent of GDP)-3.91.2-4.71.9-3.00.9-3.31.7
M2 (in per cent of GDP)29.60.552.80.933.90.451.20.9
Real stability
Real GDP2.51.82.13.52.51.43.81.2
Real exchange rate5.310.4-0.285.0-0.336.31.111.5
Real lending interest rate-147.08.2-4.838.119.71.18.24.3
Exports (in per cent of GDP)22.20.443.21.327.50.443.40.6
Sources: IFS and IMF, World Economic Outlook (WEO).

The persistence of dollarization through 2001 could reflect a historical legacy. Inflation in many Latin American countries during the period 1980-95 was extremely high by historical standards and compared with other developing country regions. In the period 1980-2003, there were a total of 56 so-called free-fall events – defined as years when broad money or consumer prices rose or the currency depreciated by over 1,000 per cent or when deposit or lending interest rates exceeded 100 per cent (Table 4.5). Three quarters of these events occurred in six Latin American countries (Argentina, Bolivia, Brazil, Nicaragua, Peru and Uruguay).

Table 4.5Summary of free-fall events, 1980-20031
CountryYearM2

(per cent

change)
CPI

(per cent

change)
Exchange

rate2

(per cent

change)
Deposit

interest

rate

(per cent)
Loan

interest

rate

(per cent)
Angola1993657.21379.4958.1
Angola19943304.9948.82137.3
Angola1995475.92671.84521.1125.9206.3
Angola19963804.64145.14555.2147.1217.9
Angola2000303.7325.0372.739.6103.2
Argentina1981118.3104.5139.6157.1
Argentina1982131.5164.8488.8126.2
Argentina1983403.0343.8306.2281.3
Argentina1984603.7626.7542.4396.8
Argentina1985435.0672.2789.6630.0
Argentina1987163.7131.3127.4175.9
Argentina1988441.5343.0308.2371.8
Argentina19892283.23079.84736.717235.8
Argentina19901059.42314.01051.81517.9
Bolivia19841421.11281.41253.8108.3120.7
Bolivia19857035.311749.613943.268.8172.2
Brazil1980115.0
Brazil198188.1101.776.7108.0
Brazil198284.0100.592.8156.1
Brazil1983135.8135.0221.4154.6
Brazil1984270.1192.1220.3267.6
Brazil1985322.5226.0235.5295.4
Brazil1986289.2147.1120.2109.5
Brazil1987213.7228.3187.3401.0
Brazil19881511.9629.1568.9859.4
Brazil19891461.91430.7980.55845.0
Brazil19901147.52947.72310.19394.3
Brazil1991705.3432.8495.3913.5
Brazil19921651.7951.61009.91560.2
Brazil19932979.81928.01859.93293.5
Brazil19941035.72075.91887.75175.2
Bulgaria1996124.5164.874.7123.5
Israel1980131.0176.9
Israel1981829.3116.8123.1170.6
Israel1982141.8120.4112.3140.2
Israel1983206.9145.6131.6132.9186.2
Israel1984510.2373.8421.6438.4823.0
Israel1985168.5304.7302.1178.8503.4
Nicaragua198812360.010205.0262676.7107379.1121906.0
Nicaragua19892746.84770.25703.71585.9558.0
Nicaragua19908603.87485.54401.09.522.0
Nicaragua19911428.42945.12930.611.617.9
Peru1988624.7667.0665.2161.8174.3
Peru19892015.03398.71969.51135.61515.9
Peru19906311.57481.76947.02439.64774.5
Peru1991236.1409.5311.2170.5751.5
Peru199255.573.561.359.7173.8
Poland1989236.0244.6234.3100.064.0
Poland1990121.9555.4560.141.7504.2
Uruguay198887.262.259.067.8101.5
Uruguay1989118.780.473.384.7127.6
Uruguay1990123.0112.588.397.8174.5
Uruguay199178.8102.072.575.2152.9
Uruguay199245.468.549.954.5117.8
Uruguay200215.814.059.6126.1
Zambia1993101.5183.3162.9113.3
Sources: WEO and IFS.

4.3 Financial dollarization as a rational response to inflation uncertainty

Theoretical overview5

Even though inflation may have declined in countries with high FD, doubts may linger about the credibility of monetary policy, and residents resort to FCDs to protect their purchasing power measured in local currency from the risk of a surge in inflation. The yield curves in the six highly dollarized countries at end-2004 suggest that markets still wonder about the future stance of monetary policy (Figure 4.1).

In Bolivia, Paraguay, Peru and Uruguay, the gap between the yield curve for domestic currency deposits and for FCDs widens over time to well in excess of the inflation differential in most of these countries. In Costa Rica, the difference between the yield curves is closer to the inflation differential, yet the differential still widens gradually at longer maturities, suggesting concerns about the future stance of monetary policy.

Figure 4.1Yield curve for deposit interest rates, 2004

Source: Country authorities.

Ize and Levy Yeyati (2003) emphasize the importance of the relative volatility of inflation for determining the degree of FD. They argue that residents will prefer to hold FCDs if the risk of unexpected inflation is high. Specifically, residents look at the volatility of inflation relative to that of the change in the real exchange rate and hold a larger share of their portfolio in foreign currency assets as inflation becomes relatively more volatile and as the real exchange rate becomes more stable. In this situation, the real value of foreign currency assets – measured in terms of domestic purchasing power – is more stable. Similarly, stable domestic inflation and a volatile real exchange rate will make domestic currency assets a better store of value. This approach assumes that arbitrage tends to equalize the rates of return on domestic and foreign currency assets, which implies that shifts in inflation or interest rates will not affect the decision to hold foreign currency assets. Ize and Levy Yeyati develop a variable that measures the portfolio share allocated to foreign currency assets that minimizes the variance of a portfolio with local currency and foreign currency interest-bearing assets. They show that this share of the MVP raises as domestic inflation becomes more variable relative to the real exchange rate. They present empirical support for their view that a larger MVP contributes to more FD.

Weak institutions undermine the credibility of policies, as residents may fear that governments will erode the value of financial assets by generating unexpected inflation. De la Torre and Schmukler (2004) add that weak institutions can also raise doubts about the enforceability of contracts and encourage residents to shorten the duration of contracts or undertake transactions offshore in countries with more secure legal frameworks. Indicators of the quality of institutions have been developed by the World Bank, with a database starting in 1996 that includes measures of political stability, government effectiveness, regulatory quality, rule of law and control of corruption, and voice and accountability. Other agencies have compiled longer time series on institutional variables, such as political stability, bureaucratic effectiveness and ethnic tensions.

The current macroeconomic situation can also influence the degree of dollarization. Guidotti and Rodríguez (1992) and Uribe (1997) develop models of currency substitution to explain how high inflation lowers demand for domestic currency as a means of payment and unit of account, contributing to dollarization. Their models also show that currency substitution can remain high even when inflation declines. Guidotti and Rodríguez point to costs associated with re-denominating transactions back into domestic currency, while Uribe attributes persistent dollarization to network effects – the cost of using foreign currency declines as more residents rely on this means of payment. In both models the demand for domestic currency will recover if inflation falls by enough to justify incurring the costs of the transition. While these models were developed to explain currency substitution, the results can also apply to FD, especially in economies where financial innovations allow broader forms of money to also serve as a means of payment. The fiscal deficit can also affect the degree of dollarization. In many dollarized countries, the surge in inflation that cut confidence in the domestic currency arose from a wide fiscal deficit that had to be financed with money creation. For this reason, fiscal discipline might help reduce dollarization by strengthening confidence.

Empirical results

We estimated equations that sought to explain FD in terms of the MVP, inflation, the central government deficit, indices of institutional quality and political stability and legal restrictions on dollarization. The dependent variable is the ratio of FCDs to total deposits, which – while imperfect – provides the most widely available measure of dollarization. We first estimate a cross-section model for a sample of over 62 countries with broad regional coverage that includes OECD countries, transition economies, Asia and Africa, as well as Latin America for the period 1990-2001.

The results of the cross-section equations confirm the results of de Nicoló, Honohan and Ize (2005) and Ize and Levy Yeyati (2003) (Table 4.6). Equation 1 shows that the minimum variance portfolio explains an important part of dollarization, with a 10 per cent increase in the MVP raising deposit dollarization by 5 per cent. Inflation plays an important role as well. In addition, legal restrictions on FCDs appear to be effective in reducing deposit dollarization. The coefficient on the central government deficit – both as a share of GDP and of broad money – is not statistically significant. While surprising, this result could reflect a measurement problem – the central government deficit is the most widely available measure but may not be sufficiently comprehensive. Possibly, the current fiscal position may not reflect lingering uncertainty about future fiscal policy, or institutional changes – such as eliminating central bank financing to the government – may have eased concerns about the risk of monetizing large fiscal imbalances.

Table 4.6Deposit dollarization: results of cross-country regressions1
(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)
EquationIncluding OECD countriesExcluding OECD countriesFree-fall
MVP0.50.460.460.450.440.420.420.410.420.48
(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)
Inflation0.0340.0370.0360.0370.0370.0380.0370.040.0380.035
(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.01)(0.00)(0.00)(0.00)
Restriction-5.2-5.9-5.3-5.4-5.8-5.8-5.7-5.8-5.7-4.4
(0.06)(0.03)(0.07)(0.04)(0.03)(0.07)(0.07)(0.07)(0.07)(0.09)
Government balance-0.87
(0.11)
Voice and accountability-4.6-0.19
(0.04)(0.96)
Regulatory quality-5.6-1.58
(0.04)(0.72)
Rule of law-3.71.67
(0.07)(0.64)
Control of corruption-4.10.10
(0.03)(0.98)
Free-fall (1980s)0.1
(0.13)
Constant9.114.315.214.014.217.017.217.317.010.6
(0.00)(0.00)(0.000)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)
No. of observations62636363634444444463
R-squared0.660.670.670.620.670.600.600.610.600.66
Sources: Data sources, variable definitions and estimation methodology are presented in Appendices, 4.1 and 4.2.

In Equations 2 through 5, the coefficients on the indicators of institutional quality are statistically significant and have the correct sign in the full sample, which includes OECD countries.6 Looking at indicators of institutional quality developed by the World Bank, there is no significant difference between the quality of institutions in Latin America and Asia, Africa, or the transition economies. However, OECD countries clearly have much stronger institutions than developing countries. When OECD countries are excluded from the sample, the coefficients on the institutional variables are no longer statistically significant (equations 6-9). This result could suggest that significant gains in institutional quality are required to bolster confidence and discourage dollarization.

We tested whether FD was higher in countries that experienced so-called free-fall events during the 1980s. Equation 10 includes a dummy variable for those countries, and the results suggest that this factor does not explain differences in FD across countries.

We tried to assess how quickly the MVP, inflation, institutional quality and other factors affect the level of dollarization by estimating these equations using a panel data set with a lagged dependent variable (Table 4.7). These equations were estimated using the two-step system generalized method of moments (GMM) method developed by Blundell and Bond (1998).7 Equation 11 suggests a high degree of persistence to dollarization, as the coefficient on the lagged dollarization ratio is quite high at 0.95. The MVP has a statistically significant effect on dollarization, although relatively small in the near term, as a 10 per cent decline in the MVP would lead to just a 0.3 per cent decline in FD after one year. The longer-term effect is much larger – and similar to the elasticity estimated in the cross-country regressions – with a 10 per cent decline in the MVP leading to a 5 per cent decline in deposit dollarization. Inflation and the nominal rate of depreciation have a statistically significant but small impact on dollarization, while the central government deficit has no significant impact on dollarization (Equations 11-13).8 Equations 14 through 18 suggest that none of the measures of institutional quality or political stability – apart from the quality of the bureaucracy – have a significant effect on deposit dollarization, possibly reflecting the fact that there is insufficient variation in these variables over time.

Table 4.7Deposit dollarization: results of panel data regressions1
(11)(12)(13)(14)(15)(16)(17)(18)(19)
EquationMacroInstitutionalLatin

America
Highly

dollarized
Dt-10.950.940.960.940.940.940.950.720.77
(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)(0.00)
MVPt0.0330.0350.0260.0360.0270.0350.0280.0780.036
(0.07)(0.06)(.105)(0.1)(0.06)(0.06)(0.07)(0.01)(0.02)
Inflationt0.001
(0.08)
Depreciationt0.001
(0.10)
Government balancema(3)0.063
(0.56)
Democratic process0.388
(0.3)
Bureaucracy-0.643
(0.09)
Control of corruption-0.02
(0.95)
Internal conflict-0.09
(0.62)
Law and order
Dt-1 Latin America0.25
(0.00)
MVPt Latin Amercia-0.09
(0.01)
Dt-1 highly dollarized0.21
(0.03)
MVPt highly dollarized-0.04
(0.03)
Constant1.471.51.36-0.123.091.592.193.373.3
(0.01)(0.01)(.016)(0.932)(0.03)(0.07)(0.23)(0.00)(0.00)
No. of countries474747474747474747
No. of observations338338338338338338338338338
F-statistic604.6807.61282.9836.41161.1781.91025.41300.23224.1
Sources: Data sources, variable definitions and estimation methodology are presented in Appendices 4.1 and 4.2.

There is some evidence that the persistence of FD is higher in Latin America and in highly dollarized countries (with dollarization ratios above 40 per cent). Equation 19 includes an interactive dummy variable for Latin America on the coefficient for the lagged dependent variable and for the MVP. The results indicate that persistence is much lower outside Latin America, as the coefficient on the lagged dependent variable declines to 0.72 for these countries. The coefficient on the MVP is considerably higher for countries outside Latin America. Equation 20 includes a similar interactive dummy variable for the lagged dependent variable and the MVP but this time for highly dollarized countries, and the results show that persistence is higher, and the effect of the MVP is lower, in these countries.

4.4 The role of credit risk

Ize and Powell (2004) and Ize (Chapter 2 of this volume) broaden the explanation of FD to include credit risk arising from a shift in interest rates or exchange rates. They emphasize the role of expected bankruptcy costs – which are often high in many developing countries because of non-transparent accounting and lengthy and at times unreliable judicial proceedings. From the creditors’ perspective, the value of claims and collateral in local currency can also be diluted by surprise inflation. These authors show that economies settle into equilibria using the currency or mix of currencies that limit expected bankruptcy costs. Ize (Chapter 2 of this volume) shows that equilibria with high FD are possible with an inflexible and asymmetric exchange rate policy, prudential regulations that encourage moral hazard and strong concerns about financial stress arising from a currency mismatch.9

The exchange rate policy of central banks in many developing countries may encourage dollarization by limiting exchange rate flexibility, which reduces the risk of holding foreign currency assets and of lending in foreign currency to all sectors, including non-tradable. For this group of countries and time period, we calculated the Calvo-Reinhart index of fear of floating, which measures the variability of the rate of depreciation in the nominal exchange rate relative to the sum of the variability of net international reserves and the variability of short-term interest rates (Table 4.8). This index ranges from zero in the case of an exchange rate peg to infinity in the case of full exchange rate flexibility. According to this index, the dollarized Latin American countries tended to have significantly less exchange rate flexibility in the period 1990-2004 than the countries in the region that have avoided significant dollarization. However, there is considerable variation among these countries, with Bolivia and Honduras showing similar degrees of flexibility as Guatemala and Mexico over this period.

Table 4.8Latin America: indicators of exchange rate policy
Asymmetry2
De facto flexibility1BiasSkewnessDollarization3
Highly dollarized
Bolivia0.110.920.0985.3
Costa Rica0.010.320.1456.6
Dominican Republic0.650.920.6525.0
Honduras0.230.810.9435.7
Nicaragua0.001.00-0.1868.7
Paraguay0.140.541.4047.1
Peru0.110.380.6064.1
Uruguay0.050.77-0.3283.0
Average0.160.710.4258.2
Low dollarization
Brazil0.340.540.570.0
Chile0.930.26-0.0511.9
Colombia0.790.410.452.0
Guatemala0.120.090.3314.9
Mexico0.250.190.325.4
Venezuela1.300.671.080.0
Average0.620.360.455.7
Source: Authors’ estimates.

An asymmetric exchange rate policy – one that allows for some nominal currency depreciation but always resists nominal currency appreciation – can provide a oneway bet for holding FCDs and encourage dollarization, especially if combined with limited exchange rate flexibility. With this type of exchange rate policy, residents would preserve their purchasing power in local currency by holding foreign currency assets, which would benefit from higher average returns as well as lower risk.

We looked at several measures of the asymmetry of exchange rate policy. First, we constructed an index of bias in exchange rate policy for the period 1990-2004 by assigning a value of -1 in months of currency appreciation and 1 in months of currency depreciation, and then finding the average for the year.10 Values of this index close to 1 indicate a bias towards currency depreciation, while a value close to -1 suggests a bias in the other direction. The results indicate that the highly dollarized Latin American countries, especially Bolivia, Nicaragua and Costa Rica, have had a stronger bias towards nominal currency depreciations than the other countries in the region.

Asymmetry of exchange rate policy could also refer to infrequent but sizable currency depreciations. For this reason, we also estimated the skewness of the distribution of monthly currency depreciations for two periods – 1990-2004 and 1980-9. According to this measure, the more positive the degree of skewness, the greater the bias towards currency depreciations; more negative degrees of skewness indicate a bias towards currency appreciation. For the period 1990-2001, we found that there were no noticeable differences in this measure of asymmetry between highly and less dollarized countries, as both groups of countries had the same average degree of skewness over this time period. However, these averages mask considerable variation over time and across countries.

Prudential guidelines, such as capital adequacy requirements or deposit insurance, can encourage banks to engage in excessive foreign currency lending. These guidelines may not force creditors and borrowers to internalize the true costs of loans in foreign currency, which should include a premium for currency risk. The highly dollarized countries in Latin America tend to have prudential requirements that are largely neutral with respect to currency denomination (Table 4.9). Honduras is the only country that limits lending in foreign currency – both overall and to non-exporting clients. Bolivia, Honduras, Peru and Paraguay apply higher reserve or liquid asset requirements on FCDs. All the highly dollarized countries in the region apply the same capital adequacy requirement to foreign and local currency assets and extend the same deposit insurance coverage to all deposits, regardless of currency denomination. These countries limit banks’ net position in foreign exchange, and restrictions on the net long positions in foreign currency might create an incentive to onlend FCDs.

Table 4.9Risk management arrangements in selected highly dollarized economies, 2004
BoliviaCosta RicaDominican RepublicHondurasNicaraguaParaguayPeruUruguay
Credit riskNo specific limitsNo specific limitsNo specific limitsYesaNo specific limitsNo specific limitsNo specific limitsbNo specific limits
Liquidity risk
Differential liquidity/reserve requirementsYesNoNoYesNoYesYesNo
If yes, the requirements (in per cent of eligible deposits)14 per cent + marginal on foreign currency12 per cent reserve requirement and 2 per cent forced investment for all currencies 38 per cent liquidity only for foreign currency26.5 per cent on foreign currency 15 per cent on local currency30 per cent on foreign currency 6 per cent on local currency
12 per cent – marginal on local currency
Capital adequacy requirements
Different for foreign currency?NoNoNobNo
Deposit insuranceNoYesYesYesYes
Limited coverageUS$17,000US$10,00075 times minimum monthly wageS/72,794
What is the limit?
Does the limit differ by currency?NoNoNo
Lender of last resort
Operations in foreign currencyYesNoNoNoNoNoYesYes
Limits on banks’ FX position
(in per cent of capital)
Long position80 per cent100 per centNo4 per cent1100 per cent150 per cent
Short position20 per centNone15 per cent4 per cent1100 per cent150 per cent
Indexed domestic currency instrumentsUFV introduced in 2002NoLimited useNoNoNoLimited use
Sources: Singh et al. (2005); and Fund staff.

Countries with high FD face the potential for financial stress arising from sizable currency mismatches – liabilities in foreign currency that are not fully backed by assets or income streams also in foreign currency. This mismatch can make unwinding dollarization more risky and costly, especially if this happens in the context of a real exchange rate depreciation that could impose large losses on banks. Banks in highly dollarized countries often lend in foreign currency to many different sectors, including construction, wholesale and retail, trade and mortgages (Figure 4.2). In Costa Rica – where foreign currency loans accounted for about two-thirds of total loans in 2004 – loans to these sectors account for well over half of total loans in foreign currency. This most likely reflects confidence in the stability of Costa Rica’s real exchange rate. Moreover, lending in foreign currency for mortgages adds a political economy dimension to exchange rate policy, as governments would come under strong pressure for a bail-out if homeowners ran into difficulties in paying their mortgages after a sharp real depreciation. A similar pattern of lending occurs in Honduras and Peru (Figure 4.2).

Figure 4.2Sectoral composition of commercial bank loans, various countries (in per cent)

Source: Country authorities.

The balance sheets of non-financial corporations also show a similar pattern. In 2001, a large share of the total liabilities of the corporate sector in Argentina, Bolivia, Costa Rica, Peru and Uruguay was in foreign currency, ranging from 53 per cent in Bolivia to 78 per cent in Uruguay (Table 4.10). These corporations appear to have been reacting in part to the high variability of domestic inflation. In addition, the Latin American countries that impose legal restrictions on dollarization, such as Colombia and Brazil, as well has having relatively stable domestic inflation, had relatively low levels of dollarization of corporate liabilities. The non-exporting sector in the highly dollarized countries also tended to have a relatively high share of dollarized liabilities to total liabilities. In Uruguay, for example, non-exporting firms had, on average, 78 per cent of their loans denominated in foreign currency. Moreover, the corporations with higher liability dollarization tended to have large net short positions in foreign currency.

Table 4.10Corporate sector dollar-denominated liabilities, 2001 (in per cent of total liabilities)
All firmsNon-exporting
Argentina60.153.8
Bolivia52.947.9
Brazil20.421.5
Chile20.513.8
Colombia6.45.1
Costa Rica64.3n.a.
Mexico33.314.5
Peru63.561.3
Uruguay77.677.5
Venezuela34.3n.a.
Source: Inter-American Development Bank.

As a rough measure of the extent of currency mismatches, we looked at the share of a country’s FCDs in relation to its exports.11 This measure tries to capture the extent of the banking system’s vulnerability to losses from foreign exchange risk through its loan portfolio. In many dollarized countries, the level of FCDs in banks is similar to the level of foreign currency loans made by banks, because most countries impose limits on banks’ net foreign exchange positions. At the same time, foreign exchange earnings of bank clients should ultimately come from the country’s exports of goods and services. This measure only moderately correlates with the FCD ratio, with a correlation coefficient of 0.48. The countries with relatively high FD are fairly open to international trade, with exports of goods and services being about a third of GDP in the period 2000-4 (Table 4.11). FCDs in these countries averaged about 60 per cent of total deposits and about 90% of export earnings. The less dollarized countries in the region had lowered a lower share of exports (about a quarter of GDP) but also had much less FD, resulting in FCDs equivalent to about 10 per cent of exports.

Table 4.11Latin America: indicators of currency mismatch, 2000-4 (in per cent)
Exports1FCD2FCD/Exports3
Highly dollarized
Bolivia20.790.0149.0
Costa Rica44.948.337.0
Dominican Republic45.425.719.0
Honduras39.133.435.0
Nicaragua23.470.3145.0
Paraguay39.761.534.0
Peru17.470.1102.0
Uruguay23.686.1202.0
Average31.860.790.4
Less dollarized
Brazil14.86.410.0
Chile34.911.513.0
Colombia18.80.71.0
Guatemala17.88.311.0
Mexico18.97.18.0
Venezuela30.60.20.0
Average22.75.77.2
Sources: IFS, WEO and de Nicoló, Honohan and Ize (2005).

Empirical results

We estimated cross-section and panel data regressions that include the variables for the degree of exchange rate flexibility, bias of exchange rate policy and for the extent of currency mismatch, as well as the MVP and the rate of inflation (Table 4.12).12 Equation 20 suggests that these variables explain a significant share of the differences in FD across countries. The MVP, inflation and restrictions on FD remain statistically significant, plus the bias towards currency depreciation and larger currency mismatches also explain FD. The degree of exchange rate flexibility (the variable ‘Float’) is not statistically significant, possibly because its effects are captured by the MVP. We must caution that the cross-section equation could be picking up the endogeneity of the bias towards currency depreciation and the currency mismatch, which could be affected by the extent of FD. Equation 21 looks at the evolution of FD over time, and uses the lagged values of these variables to minimize the endogeneity problem. This equation shows that FD is explained by the bias towards currency depreciation, the currency mismatch as well as inflation and the MVP, but the degree of exchange rate flexibility is not statistically significant. Interestingly, the central government balance now becomes statistically significant, with larger surpluses contributing to lower FD. The measures of the degree of skewness of the distribution of currency depreciations for 1990-2001 and 1980-9 were not statistically significant in either equation (not shown in the table), probably because of the significant variation in the skewness measure across countries and over time.

Table 4.12Deposit dollarization: effect of exchange rate policy1
(20)(21)
EquationCross-countryPanel data
MVPt0.350.04
(0.00)(0.03)
Inflationt0.040.002
(0.00)(0.15)
Floatt-1-0.910.00
(0.22)(0.85)
Restriction-2.7
(0.19)
Government balancema(3)-0.05-.40
(0.91)(0.01)
Asymmetryt-113.41.83
(0.02)(0.07)
Currency mismatcht-12214.33
(0.00)(0.00)
Constant5.817.9
(0.03)(0.00)
No. of observations61331
No. of countries6146
R-squared0.80
Sources: The data sources, variable definitions and estimation methodology are presented in Appendices 4.1 and 4.2.

There appear to be two equations that provide good explanations of FD – equation 11 with a lagged dependent variable, the MVP and inflation and equation 21 with the MVP, inflation, exchange rate flexibility, central government balance, asymmetry and the currency mismatch. The out-of-sample forecasts for the period 2002-4 suggest that both models capture the main trends in FD in some of the highly dollarized countries in the region (Figure 4.3). Both models correctly point to declines in FD in Bolivia, Nicaragua, Peru and Uruguay and to broadly stable FD in the Dominican Republic and Honduras. However, the models miss the rise in FD in Costa Rica and the sharp drop in FD in Paraguay in 2004.

Figure 4.3Deposit dollarization: out-of-sample forecast

Source: Authors’ calculation.

4.5 Concluding remarks

These results provide evidence that FD is a rational response to uncertainty about inflation. FD tends to remain high in countries with unstable and high domestic inflation and with institutions that undermine confidence in the outlook for inflation. The evidence on the role of the central government balance is mixed, although equation 21 supports the view that larger fiscal surpluses do help reduce FD. Legal restrictions may have been effective in preventing FD, most likely in countries with low inflation or effective indexation mechanisms to preserve purchasing power in local currency. In countries that already have high FD, imposing such restrictions could create strong incentives to place financial savings offshore, leading to a costly economic adjustment. The study also finds that an exchange rate policy that is biased towards depreciation tends to contribute to high FD, although the skewness measure of asymmetry is not statistically significant. The degree of exchange rate flexibility probably also matters, but these effects appear to be captured by the MVP. This exchange rate policy cuts the risk of lending and saving in foreign currency and tends to enhance the rate of return on foreign currency assets. Countries with high FD also have significant currency mismatches, which are encouraged by exchange rate policy as well as by prudential regulations that are largely currency neutral.

The results also point to strong persistence in FD, with a high and statistically significant coefficient on the lagged dependent variable. However, this persistence does not appear to reflect the legacy of high inflation in the 1980s, as the free-fall indicator and the skewness of exchange rate policy during the 1980s do not appear to have a significant effect on FD. The persistence could reflect the effect of currency mismatches and policies – such as exchange rate policy and prudential regulations – that create incentives for residents to continue to hold FCDs. At the same time, the extent of FD probably also explains currency mismatches and imposes limits on exchange rate policy, and the causality implied by the econometric results needs to be interpreted with caution. But this is precisely the point of the explanation of dollarization in Ize (Chapter 2 of this volume) – economies with highly variable inflation and financial market imperfections can find themselves locked into an equilibrium with high FD because of the very high economic costs of moving to a low dollarization equilibrium.

These results suggest that countries with significant FD should strive to encourage the use of domestic currency by maintaining macroeconomic stability, with low and stable inflation; allowing for more exchange rate flexibility and less bias towards depreciation; and strengthening institutions to improve confidence in the sustain-ability of economic policies. Highly dollarized countries should adapt their prudential regulations to ensure that creditors and debtors internalize the costs associated with FD. At the same time, restoring confidence in the domestic currency may take many years of sound policies and may require a careful approach to limit the transition costs of returning to a low dollarization equilibrium.

Appendix 4.1: Data description

Variable description

Variable nameVariable descriptionSource
Dependent variable
Deposit dollarization ratioFCDs in per cent of total bank deposits.DNHI
Independent variables: macro indicators
Dt-1Lagged deposit dollarization ratio.DNHI
Dt-1 Latin AmericaLagged deposit dollarization ratio interacted with a dummy variable for Latin America (1 for Latin American countries and 0 otherwise).
MVPMinimum variance portfolio as constructed by Ize and Levy Yeyati (2003). The unit is in percentage, not in decimal fraction. See below for details of computation.IFS
MVP Latin AmericaMVP interacted with a dummy variable for Latin America.
RestrictionIndex for restriction on FCDs. ‘0’ represents no restriction and higher scores represent heavier restriction.DNHI
Inflation3-year backward-looking average inflation rate calculated as a percentage change in CPI.IFS
Government balance3-year backward-looking average of the government balance in per cent of GDP. Negative figures imply deficits.WEO
Free-fall (80s)The number of months during the 1980s in which 12-month inflation is above 40 per cent.RR
Depreciation3-year backward-looking average of nominal depreciation, i.e., percentage change of the exchange rate measured by national currency unit per US dollar.IFS
FloatCalvo and Reinhart index of exchange rate flexibilityIFS
AsymmetryIndex of asymmetry of exchange rate movements. Constructed by assigning a value of -1 in months of currency appreciation and 1 in months of currency depreciation, and then averaging for the year.IFS
FCD/ExportFCDs divided by exports. The unit is not in per cent, but in decimal fraction.DNHI, WEO, IFS
Governance variables1
Voice and accountabilityIndex of voice and accountability by Kaufmann, Kraay and Mastruzzi. Average for the years 1996, 1998, 2000 and 2002.WBGOV
Regulatory qualityIndex of regulatory quality by Kaufmann, Kraay and Mastruzzi. Average for the years 1996, 1998, 2000 and 2002.WBGOV
Rule of lawIndex of rule of law by Kaufmann, Kraay and Mastruzzi. Average for the years 1996, 1998, 2000 and 2002.WBGOV
Control of corruption (WB) used in Table 4.6Index of control of corruption by Kaufmann, Kraay and Mastruzzi. Average for the years 1996, 1998, 2000 and 2002.WBGOV
Democratic processPolitical risk rating on democratic accountability.PRS
BureaucracyPolitical risk rating on bureaucracy quality.PRS
Control of corruption used in Table 4.7Political risk rating on corruption.PRS
Internal conflictPolitical risk rating on internal conflict.PRS
Law and orderPolitical risk rating on law and order.PRS

Computation of MVP

MVP for year t was computed using a formula

where πt and nt represent inflation and depreciation of the nominal exchange rate respectively.13 The correlation coefficient and standard deviations were estimated using quarterly data over a ten-year horizon; that is, to estimate MVP for year t, we used quarterly data of inflation and depreciation from year t-9 to year t.

MVP represents domestic agents’ optimal portfolio of FCDs over total deposits, and the agents cannot usually have a short position. To incorporate this, we assigned MVP a value of 0 if the estimate of Corr (nt, pt) was negative and a value of 100 if the estimate of MVP exceeded 100.

Data sources

DNHI: de Nicoló, G., P. Honohan and A. Ize (2005) ‘Dollarization of Bank Deposits: Causes and Consequences’, Journal of Banking and Finance, Vol. 29, No. 7, pp. 1697-727.

International Financial Statistics (IFS), International Monetary Fund.

Political Risk Rating (PRS): political risk rating by the Political Risk Services Group (www.prsgroup.com).

RR: Reinhart, C. and Rogoff, K. (2002) ‘The Modern History of Exchange Rate Arrangements: A Reinterpretation’, NBER Working Paper No. 8963 (Cambridge, MA: National Bureau of Economic Research).

WBGOV: Kaufmann, D., A. Kraay and M. Mastruzzi (2003) ‘Governance Matters III: Governance Indicators for 1996-2002’, World Bank Policy Research Working Paper No. 3106, 30 June.

World Economic Outlook (WEO) (2004), International Monetary Fund.

Appendix 4.2: Estimation methodology

1. Cross-country regressions (Tables 4.6 and 4.12)

For cross country regressions, we converted our panel dataset into a cross-country dataset by averaging variables over years for each country. Then, we ran ordinary least squares (OLS) regressions of deposit dollarization ratio on the independent variables specified in the tables.

2. Panel data regressions (Tables 4.7 and 4.12)

We used an unbalanced panel dataset with 47 countries and the period 1990-2001 and estimated two models:

  • Model 1 (without the lagged dollarization ratio): Dit = c + β*MVPit + γXit + ut + νit

  • Model 2 (with the lagged dollarization ratio): Dit = c + αDit-1 + β*MVPit + γXit + ui + νit

where Dit represents the deposit dollarization ratio, MVPit the minimum variance portfolio, Xit other independent variables and ui country-specific effects. We make standard assumptions for the disturbance term νit:

Model 1 was estimated with standard fixed and random effect models without instrumental variables.

Model 2 was estimated with the two-step system GMM method developed by Blundell and Bond (1998).14 We chose this estimator over a standard fixed or random effect estimator because the latter generates biased coefficient estimates under the presence of the lagged dependent variable in the right-hand side. We chose the system GMM method over the so-called difference GMM method because the difference GMM method is known to suffer from weak instrumental variables problems when the coefficient on the lagged dependent variable is close to one, and the system GMM method can circumvent this problem.15 To address small-sample downward biases on standard errors in two-step estimations, we used a corrective method invented by Windmeijer (2005).

Our result for the baseline regression (equation 11 in Table 4.7) passes standard diagnostic tests. The Sagan test does not reject the hypothesis of no over-identifying restrictions. The second-order serial correlation in the first differences of residuals is not detected, thereby validating the foundation of the GMM moment conditions. In addition, the coefficient estimates do not change significantly when a smaller number of moment conditions are used or when the constant term is excluded from the right-hand side.

Notes

The authors would like to thank Roberto Garcia-Saltos, Kevin Cown, Alain Ize, Eduardo Levy Yeyati, Chris To we and the participants in a seminar at the IMF for their comments. They would also like to thank Genevieve Mendiola for superb research assistance.

This differs from official, full dollarization, which entails the legal adoption of a foreign currency as the sole monetary unit of a country. Currently, Ecuador, El Salvador and Panama are the only three Latin American countries with this monetary regime.

The costs and benefits of FD are discussed fully in Baliño, Bennett and Borensztein (1999). Reinhart, Rogoff and Savastano (2003) challenge the notion that dollarization limits the scope for an independent monetary policy.

The companion papers in this volume by Ize (Chapter 2) and Ize and Levy Yeyati (Chapter 3) more thoroughly review the theoretical explanations of FD.

When all of the indicators of institutional quality are included in one equation, none of the coefficients is statistically significant, suggesting the presence of multi-colinearity.

The estimation methodology is described in more detail in Appendix 4.2.

When both inflation and nominal exchange rate depreciation are included in the same equation, both coefficients are statistically insignificant.

The model makes the crucial assumption that projects returns rise with a real exchange rate depreciation. This means that borrowers in foreign currency would perceive that they would benefit as well from a real depreciation.

Periods of no change in the exchange rate were assigned a value of 0.

Goldstein and Turner (2004) propose a broader, more aggregated measure of the currency mismatch for an economy.

Including a lagged dependent variable in equation 21 led to counterintuitive results.

Under an assumption that agents regard the US inflation as fixed, our formula is equivalent to the original definition from Ize and Levy Yeyati (2003).

This method requires additional assumptions that E(Di1νit) = 0∀t and E((ui + vi3Di2) = 0 for each i.

In fact, the difference GMM method produces a coefficient estimate on the lagged dependent variable of around 0.36, much smaller than the estimate by the system GMM method.

References

    Baliño, T., A.Bennett and E.Borensztein (1999) ‘Monetary Policy in Dollarized Economies’, IMF Occasional Paper No. 171 (Washington, DC: International Monetary Fund).

    Blundell, R. and S.Bond (1998) ‘Initial Conditions and Moment Restrictions in Dynamic Panel Data Models’, Journal of Econometrics, Vol. 87, pp. 11543.

    Calvo, G. and C.Reinhart (2002) ‘Fear of Floating’, Quarterly Journal of Economics, Vol. 117, No. 2, pp. 379408.

    de la Torre, A. and S.Schmukler (2004) ‘Coping with Risk Through Mismatches: Domestic and International Financial Contracts for Emerging Economies’, Working Paper No. 3212 (Washington, DC: World Bank).

    de Nicoló, G., P.Honohan and A.Ize (2005) ‘Dollarization of Bank Deposits: Causes and Consequences’, Journal of Banking and Finance, Vol. 29, No. 7, pp. 1697727.

    Goldstein, M. and P.Turner (2004) Controlling Currency Mismatches in Emerging Economies (Washington, DC: Institute for International Economics).

    Guidotti, P. and C. A.Rodríguez (1992) ‘Dollarization in Latin America: Gresham’s Law in Reverse?’, Staff Papers, International Monetary Fund, Vol. 39 (September), pp. 51844.

    Ize, A. and E.Levy Yeyati (2003) ‘Financial Dollarization’, Journal of International Economics, Vol. 59 (March), pp. 32347.

    Ize, A. and A.Powell (2004) ‘Prudential Responses to De Facto Dollarization’, IMF Working Paper 04/66 (Washington, DC: International Monetary Fund). Revised version published in the Journal of Policy Reform, Vol. 8, No. 4 (2005) pp. 24162.

    Reinhart, C., K.Rogoff and M.Savastano (2003) ‘Addicted to Dollars’, NBER Working Paper No. 10015 (Cambridge, MA: National Bureau of Economic Research).

    Singh, A., A.Belaisch, C.Collyns, P.De Masi, R.Krieger, G.Meredith and R.Rennhack (2005) ‘Stabilization and Reform in Latin America: A Macroeconomic Perspective on the Experience since the Early 1990s,’ IMF Occasional Paper No. 238 (Washington, DC: International Monetary Fund).

    Uribe, M. (1997) ‘Hysteresis in a Simple Model of Currency Substitution’, Journal of Monetary Economics, Vol. 40 (September), pp. 185202.

    Windmeijer, F. (2005) ‘A Finite Sample Correction for the Variance of Linear Efficient Two-step GMM Estimators’, Journal of Econometrics, Vol. 126, pp. 2551.

    Other Resources Citing This Publication