Financial Dollarization of Households and Firms: Does It Differ?

Contributor Notes

Author’s E-Mail Address: jcorrales@fedesarrollo.org.co; pimam@imf.org

Using a newly complied and extended database from International Financial Statistics, and applying different panel-regression techniques, this paper documents the evolution of households’ and firms’ dollarization over the past decade. We assess the macroeconomic determinants of dollarization for households and firms and explore differences between high and low-income countries. We find that households’ and firms’ dollarization in loans and deposits are weakly explained by the currency substitution model, except in low income countries, where inflation plays a significant role. Instead, market development variables such as financial deepening, access to external debt and FX finance as well as other market considerations are key to explain the dynamics of deposits and loans dollarization, regardless of the level of income.These factors can account for a significant fraction of the dollarization, but using a variance decomposition model, there is evidence that a non-negligible portion has yet to be explained. This suggests that there are key determinants for household and firm dollarization that are not fully captured by traditional macroeconomic explanatory variables.

Abstract

Using a newly complied and extended database from International Financial Statistics, and applying different panel-regression techniques, this paper documents the evolution of households’ and firms’ dollarization over the past decade. We assess the macroeconomic determinants of dollarization for households and firms and explore differences between high and low-income countries. We find that households’ and firms’ dollarization in loans and deposits are weakly explained by the currency substitution model, except in low income countries, where inflation plays a significant role. Instead, market development variables such as financial deepening, access to external debt and FX finance as well as other market considerations are key to explain the dynamics of deposits and loans dollarization, regardless of the level of income.These factors can account for a significant fraction of the dollarization, but using a variance decomposition model, there is evidence that a non-negligible portion has yet to be explained. This suggests that there are key determinants for household and firm dollarization that are not fully captured by traditional macroeconomic explanatory variables.

I. Introduction

Dollarization—the use of foreign currency as a medium of exchange, store of value, or unit of account by residents—is a notable feature in low-income and emerging market countries of Latin America, Eastern Europe, and parts of Asia and sub-Saharan Africa (SSA) (see Corrales et al., 2015, Neanidis and Savva, 2009).

The economic literature has made great strides in helping us better understand the determinants of dollarization, which is often a reflection of past episodes of severe economic and political disruption, with both borrowers and lenders finding security in transacting in a foreign currency (see Levy-Yeyati, 2006)2. The chronic macroeconomic instability that has troubled countries in the past has, however, receded in most cases. Conflicts and social turmoil, while not eradicated, are now circumscribed. Institutions—notably the state—have been strengthened, governance is improving, and budgetary and monetary policies are becoming more sustainable in most low and middle-income countries. Similarly, while the financial system remains underdeveloped in many dollarized countries, there are signs of financial deepening emerging. In other words, the macroeconomic foundations are strengthening, and policymakers have gained in credibility, reducing the attractiveness of dollarization in many cases (see Corrales et al., 2015).

However, while the literature has focused on explaining the causes of overall dollarization, it has paid scant attention to the differences in the determinants between household and firm dollarization. Does household dollarization differ from firm dollarization? Do households and firms hold deposits and assets differently in foreign currencies, and if so, why? To gain new insights into the dollarization phenomenon, this study aims to delve more deeply into the impact of dollarization by circling in on dollarization at the household and firm level. This is an important question to answer, in the context of de-dollarization policies for instance (see Kokenyne et al., 2010). The literature has made it clear that dollarization is never easy to reverse, even if the underlying causes have been removed; having more information of the sectoral determinants of dollarization will therefore help better crystalize the policy implications. This is the first study to our knowledge to provide systematic empirical analysis regarding differences between macroeconomic determinants for households and firms’ dollarization across the world. This study therefore provides more targeted policies on how respectively household and firm dollarization can be addressed.

The main findings can be summarized as follows: Both household and firm deposit and loan dollarization tend to be largely driven by structural factors, rather than macroeconomic stability—though this could be because the period covered is one where across the world, inflation and exchange rate policies were largely kept in check in most countries. However, the impact of structural factors differs between households’ and firms’, and between deposit-and loan dollarization.

In the remainder of this paper, we briefly discuss the relevant literature on dollarization in section II, focusing on household and firms. Section III provides descriptive statistics on the development of dollarization in the last decade. The empirical approach for assessing the determinants of dollarization of households and firms and their contribution to the changes in dollarization levels in SSA over time are illustrated in section IV. Section V concludes.

II. Literature Review

The literature on dollarization has typically focused only on aggregate (household plus firm) deposit and/or credit dollarization (e.g. Nicolo, Honohan, and Ize, 2005, Levy-Yeyati, 2006, Stix, 2013). This is surprising, as households and firms have in principle different reaction functions to dollarization. This limited research interest by the profession so far is largely a reflection of the lack of cross-sectional data that allows disaggregated analysis. But even the theoretical literature has been rather mute on the differences between household and firm dollarization, with the notable exception of Ize (2005).

The initial focus of the literature was the currency substitution angle of the phenomenon, motivated by the history of high inflation in Latin America.3 While both households and corporations may want to save in FX in an unstable macroeconomic environment, to protect themselves from exchange rate depreciations or high inflation—and therefore the drivers of deposit dollarization may be similar—they may behave differently when it comes to borrowing—the drivers of loan dollarization may differ. Exporting firms for instance, which earn FX, have a natural hedge in FX borrowing and may therefore be more inclined to borrow in foreign currency. Households, on the other hand, typically earn in local currency and may therefore be less willing, and would take more risk, if they borrow in foreign currency. Financial institutions may also be reluctant to lend to households in foreign currency, given the risks involved, creating further impediments to borrow in FX for households (and firms that don’t export). Regulations may also impact the behavior of households and firms differently. This is because as a result of the differences in the composition of the balance sheet of households and corporations, regulators often impose different rules on borrowing, often prohibiting households to borrow in FX altogether, while allowing firms to borrow in FX as long as they export (see Cayazzo et al., 2006).4

The currency substitution view was, however, challenged by the persistence of dollarization in the 1990s, when inflation rates in dollarized economies declined significantly. Edwards and Magendzo (2003), analyzing the macroeconomic record of dollarized economies, report that inflation has been significantly lower in dollarized economies than non-dollarized ones. Although the currency substitution view attributes this persistence to long-lasting memories of past inflation that induce high inflation after years of price stability, the more recent literature that has emerged perceives dollarization more as an asset substitution phenomenon.

This emerging literature can be grouped into three main categories (Levy-Yeyati 2006):

  • (i) The portfolio view. This explains dollarization as the optimal portfolio choice for a given distribution of real returns in each currency. Thus, if the domestic deposit yields higher returns than a corresponding dollar deposit, deposit dollarization should be lower. According to this view, households and firms should act in an analogous manner when it comes to loans or deposits in FX. If the differential between domestic deposit rates and foreign deposit rates is high, they will tend to save in domestic currency, while they will prefer to borrow in foreign exchange, all else being equal. This assumes of course no restrictions such as regulations imposed on banks in their capacity to lend on foreign exchange.

  • (ii) The market development view. This looks at dollarization as the suboptimal response to a market imperfection. Restrictions on the use of foreign exchange reflected in limited current account openness, weak financial deepening (reflected in a low M2 to GDP ratio), with lack of investment opportunities in domestic currency—where economic agents prefer to hold currency in FX as the domestic financial sector is shallow and provides few saving options— are examples of market failures that can affect the dynamic of dollarization, with implications for the development of local currency markets. According to this view, households and firms should behave similarly—they will prefer to deposit savings in foreign currency for lack of domestic alternatives, and will be forced to borrow in foreign currency, given the limited options available in local currency. Even companies that have direct access to foreign capital markets—typically large firms—will act in the same manner. Only when domestic capital markets develop do households and firms start to borrow and save more in local currency.

  • (iii) The institutional view. The idea is that institutional failures can foster dollarization either by generating new distortions or by strengthening the channels highlighted in the first two categories. For example, the quality of institutions affects the credibility of monetary policy and commitment to an exchange rate regime. Political instability resulting from weak institutions, through its large fiscal costs and implications for inflation affects economic agents’ incentives to hold foreign currency denominated assets. Again, both households and firms will be impacted by the institutional view.

Not surprisingly, the empirical literature on dollarization of either households or firms has mostly focused on Latin America and more recently on Transition economies, given the availability of data. Most of the studies have also focused on the microeconomic determinants, rather than macroeconomic ones. This is because they often focus on a single country or a set of homogeneous countries for which data is available.

a. Households

While the phenomenon of corporate borrowing in FX could be explained by currency hedging of exporting firms, lending in FX currency to largely un-hedged households is harder to explain, as it poses a risk to financial stability. Only very few empirical studies have provided insights on the foreign currency lending behavior of households.

Pellényi and Bilek (2009) is one of the first studies that looked at the determinants of household loan dollarization, looking at household characteristics. By analyzing survey data of Hungarian households collected in 2008, they find that foreign currency borrowers are not statistically different from domestic currency borrowers with regards to income, age, and gender. In fact, FX borrowing is common in Hungary, and driven mostly by macroeconomic factors: high interest rate spreads, a relatively stable exchange rate and the competition by foreign owned banks. However, foreign currency borrowers tend to be more risk averse and more aware of currency risks. The study shows that being more sophisticated and aware of FX risks triggers risk mitigation tools such as insurance against the currency risk.

Beer, Ongena and Peter (2010), perform a similar analysis of the borrowing behavior of Austrian households. The authors, using multivariate multinomial logit models, on a survey of 2,556 Austrian households estimate the influence of household characteristics, which are split into subjective factors (e.g. risk perception, financial knowledge, and education) and objective factors (e.g. socio-demographics). According to their results, foreign currency borrowers are usually less risk averse, older, financially better educated, and wealthier. This confirms that individuals borrowing in FX are more sophisticated actors, knowing the risks and being able to better deal with them.

A more comprehensive study by Fidrmuc et al. (2013) looks at household survey data for 9 Central and Eastern European Countries (CEEC) over the 2007-2010 period. Using a two-stage Heckman selection approach allows them to relate the denomination of the currency loan to various socio-demographic and economic factors. Their results reveal that foreign currency loans are driven by households’ lack of trust in the stability of the local currency and in domestic financial institutions. Moreover, hedging factors (e.g. remittances and household income in FX), as well as expectations of euro adoption increase the probability of foreign currency loans. This finding can be interpreted as supporting the portfolio behavior view of households.

The few studies that looked at household dollarization—while insightful—are largely microeconomic in nature and mostly focused on a single country, looking at household characteristics, and tend to pay less attention to macroeconomic and institutional factors, something that our paper aims to correct.

b. Firms

The empirical research on firm-level data confirms a significant role for currency matching in the choice of the currency denomination by borrowers (Kedia and Mozumdar, 2003). Brown, Kirschenmann and Ongena (2010) consider several micro level determinants of firm borrowing in Bulgaria, by employing firm level loan data between 2003 and 2007. Their model incorporates both supply (bank characteristics) and demand determinants (firm characteristics) of foreign currency loans. Their results show that comparably larger and older firms as well as firms with lower distress costs in case of default demand more foreign currency loans. This confirms again that more sophisticated actors tend to borrow more in FX.

In a separate study, Brown et al. (2011) examine the firm- and country-level determinants of foreign currency borrowing by small firms, using information on loans extended to 3,101 firms in 25 transition countries between 2002 and 2005. Their results suggest that foreign currency borrowing is more strongly related to firm-level foreign currency revenues than it is to country-level interest rate differentials. This supports the conclusion that carry-trade behavior is not the key driver of foreign currency borrowing. Overall, the findings endorse the view that firms that take foreign currency loans are better equipped to bear the corresponding currency risks than is commonly thought.

Similarly, Mora et al. (2013), in one of the few studies outside of the CEEC region, investigate what induces small firms in an emerging market economy to borrow in dollars from domestic banks, by looking at Lebanon. The findings complement studies of large firms with foreign currency loans from foreign lenders. Exporters, naturally hedged against currency risk, are more likely to incur dollar debt. Firms also partly hedge themselves by passing currency risk to customers and suppliers. Firms reliant on formal financing (banks and supplier credit) are more likely to contract dollar debt than firms reliant on informal financing (family, friends and moneylenders). Bank relationships, however, do not increase the dollar debt likelihood. And finally, profitable firms are less likely to have dollar debt. Information frictions and limited collateral, therefore, constrain dollar credit even when it is intermediated domestically.

The only study that we are familiar with that looks at both household and firm dollarization is that of Basso et al. (2011). They use a newly compiled data set on 24 transition economies and, employing a standard panel as well as a panel-VAR methodology, find that increasing access to foreign funds leads to higher credit dollarization, while it decreases deposit dollarization. Their empirical results show that higher interest rate differentials on loans increases credit dollarization. On the other hand, deposit dollarization decreases when the interest rate differential on deposits increases. Hence interest rate differentials matter. They observe that household credit dollarization is lower compared to corporate dollarization, which might be comforting knowing that households usually have less hedging capabilities. An important distinction between households and firms is that a country’s openness to the international economy is contributing to corporate but not to household financial dollarization. Note that the explanatory power of their model is generally lower for household vis-a-vis total and corporate dollarization. Hence, this framework does not seem to capture all the main determinants of household dollarization.

Overall, the findings on the determinants of dollarization of households and firms currently in the literature are mostly aligned with expected behavior. They show that most individuals and entities engaging in dollarization for credit or deposits, are sophisticated, and aware of the risks. However, for our purposes, the above studies may also contain some biases, given their focus on Transition economies. Foreign currency borrowing behavior of individuals and firms is determined not only by economic and monetary policies, but also by more general political developments in CEEC countries, where accession to the European Union and/or to EMU is an important subject in the political agenda of governments. Additionally, once lenders get used to foreign currency deposits, it may take a rather long time to change their behavior again, which, in turn, indirectly impacts borrowing in foreign currency. Hence, both path dependence and expectations matter for the determination of the credit currency structure. We therefore aim to extend the scope of these studies, by building on the model of Corrales et al. (2015) and aim to offer a broad cross-country perspective on the drivers of dollarization by differentiating households from corporations.

III. Dynamics of Households and Firms’ Dollarization

Using the International Financial Statistics (IFS) database of the International Monetary Fund, we calculated deposit and loan dollarization, for households and firms, as the share of FX deposits/loans of total households/firms’ deposits/loans (e.g. firms’ loans dollarization equals firms’ loans in FX as percent of total firms’ loans). Therefore, dollarization measures are calculated separately for each group (appendix A2 covers the data source for each variable). The database includes countries from different regions and levels of income (see appendix A for more details) and it covers the period from 2001 to 2016. One caveat to our study is that many countries do not have refined data on household and firm credit or deposits, meaning that not all countries were included in the regressions. Another word of caution is that in many countries, particularly developing ones, individuals and firms may not be properly captured, potentially biasing some of the findings. Many big firms in developing and emerging countries belong to only one individual and depending on how their loans and deposits are registered (under the name of the firm of the owner or the individual directly) and treated for statistical purposes, this may impact the results.

Households and firms can be considered, in most of the countries, as the most important agents for financial activity. Between 2001 and 2016, households and firms combined represented 87.1 percent of total deposits and 92.1 percent of total loans in the financial sector, excluding the Central Banks. Moreover, in terms of dollarization, they represented 57.5 percent of deposits and 68.7 percent of deposits and loans respectively (Table 1). The level of dollarization for both deposits and loans appear greater for firms than for households. This is not a surprise, and could reflect that (exporting) firms are included in the data. As already discussed in the literature, these firms are more likely to both have access to FX, and to earn FX. Also, it is clear from the data that households have on net more deposits than loans in dollars. While we do not have the breakdown by households, it is likely that households are more conservative, and hold dollars as a form of saving. Firms on the other hand, are net borrowers in dollars, with significantly higher loans in foreign exchange.

Table 1.

Financial Sector: Loans and Deposits by Sector for all countries*. Average 2001-2016

article image
Source: International Financial Statistics – IMF, author’s calculations* Excludes observations for which dollarization is 0 across all five economic sectors

In terms of the evolution of dollarization and excluding countries with 0 and 100 percent level of dollarization5, it seems that households have reduced their median level of deposits and loans denominated in FX between the first and second half of the period from 2001 to 2016 (see figure 1). A similar result is found for firm’s deposit dollarization, but the opposite is true for firms’ FX loans (Panel 1). Since the sample is divided both before and after the world financial crisis and calculating the ratios of deposit and loan dollarization imply converting FX values into domestic currency, it is possible that the depreciation of domestic currencies, experienced after the crisis by many developing and transition economies, could explain the expansion on firm’s loan dollarization. Alternatively, the prevailing low interest rate environment following the global financial crisis—and the search for yield—may have encouraged local companies to borrow in foreign currency, at rates that are low by historic standards.

Figure 1.
Figure 1.

Private Sector Financial Dollarization

(Foreign currency deposits/loans as % of total. Excludes 0 or 100 dollarization)

Citation: IMF Working Papers 2019, 019; 10.5089/9781484393192.001.A001

Source: International Financial Statistics Database – IMF, authors calculations

Differences of dollarization by level of income

Disaggregating the data by level of income seems to provide more insights into the drivers of dollarization, since both deposit and loan dollarization differ significantly between higher and lower income6 countries (Table 2). Based on World Bank’s classification of countries by level of income, on average, low income countries show a greater level of deposit dollarization for households and firms. Several lower income countries have experienced episodes of hyperinflation, in part owed to political instability. Hence, these larger levels of deposit dollarization seem to reflect households and firms’ decision on hedging by using FX deposits as store of value. On the other hand, loan dollarization appears significantly higher in higher income countries. This may reflect the presence of large companies with better financial capabilities and revenues in FX currency that allow them to take FX loans.

Table 2.

Financial Dollarization by Level of Income*. 2001-2016

article image
*** p<0.01, ** p<0.05, * p<0.1Source: International Financial Statistics – IMF, author’s calculations* Excluding 0 or 100 percent levels of dollarization

Figure 2 shows loan dollarization on the x-axis and deposit dollarization in the y-axis. Most firms in both lower and higher income countries exhibit larger levels of loans dollarization relative to deposits dollarization. However, the vast majority is clustered around the 45-degree line, which suggests some precaution in terms of possible balance sheet effects. Nonetheless, there is a non-negligible share of countries where firms’ loans in foreign currency largely exceed their deposits in foreign currency, exposing them to currency risks. Households in both types of countries are more inclined to have deposits in foreign currency rather than loans. This could be due to bank systems allowing to take foreign currency deposits but not extending loans in a currency different than the domestic one.

Figure 2.
Figure 2.

Correlation between Loans and Deposits Dollarization. Average 2001-2016

(Foreign currency deposits/loans as % of total. Excludes 0 or 100 dollarization)

Citation: IMF Working Papers 2019, 019; 10.5089/9781484393192.001.A001

Source: International Financial Statistics – IMF, author’s calculations

A correlation analysis of key macroeconomic variables (Figures 3 and 4) suggests that firms’ dollarization of deposits and loans are mainly correlated with financial market development (proxied by M2 to GDP ratio) for both higher and lower income countries. Correlation between firms’ loan dollarization and inflation is stronger for higher income countries, but the correlation of firms’ deposits dollarization and inflation is stronger for lower income countries. In the case of household loan dollarization, lower income countries show larger correlations with macroeconomic variables than higher income countries. Household deposits dollarization appears strongly correlated with financial market development. The econometric section of this paper explores these results.

Figure 3.
Figure 3.

Correlations: Firms Dollarization and Macroeconomic Variables. Avg. 2001-16

Citation: IMF Working Papers 2019, 019; 10.5089/9781484393192.001.A001

Source: International Financial Statistics – IMF, author’s calculations
Figure 4.
Figure 4.

Correlations: Households Dollarization and Macroeconomic Variables. Avg. 2001-16

Citation: IMF Working Papers 2019, 019; 10.5089/9781484393192.001.A001

Source: International Financial Statistics – IMF, author’s calculations

IV. Determinants of Households’ and Firms’ Dollarization

a. Data and Model

The effective dataset of econometric analysis is an unbalanced panel consisting of 52 (39) countries for households’ deposit (loan) dollarization and 49 (63) countries for firms’ deposit (loan) dollarization. While dollarization data are available for more countries, the effective sample is reduced because we exclude countries with zero values or full dollarization (as this may be driven by legal requirements for instance) and some countries that have missing control variables7.

The baseline estimation model is given by:

Dollari,t=αi+β1CurrSubsi,t+β2Portfolioi,t+β3MktDevi,t+β4Accessi,t+β5Insti,t+β6Controlsi,t+εi,t

where Dollari,t is the measure of dollarization given by deposits or loans in foreign currency as percent of the corresponding total for both households and firms. CurrSubsi,t groups the variables capturing the currency substitution dimension—the negative relation between the demand for local currency and inflation—proxied by the inflation rate and nominal exchange rate depreciation against the US dollar. Overall, inflation and nominal exchange rate depreciation are expected to positively impact deposit dollarization, though the bearing on loan dollarization is ambiguous and context specific. Household’s and firm’s preference to take on local or FX loans will depend on the conditions (rates, duration) with the relative attractiveness depending on the local circumstances. Portfolioi,t represents the portfolio optimization considerations—expected return of holding foreign exchange—estimated by the deposit or loan interest rate spreads to the US dollar and relevant real interest rates. Per this model, if the domestic deposit yields higher returns than a corresponding dollar deposit, deposit dollarization is lower, while loan dollarization should be higher (and vice-versa).

Similarly, higher domestic real deposit rates should be associated with lower levels of deposit dollarization, and a higher level of loan dollarization. MktDevi,t proxies for market developments and externalities that lead to risk mispricing affecting dollarization. These are captured by the Chinn-Ito index of capital account restrictions, external debt to GDP and M2 to GDP ratios, as well as GDP per capita. Lower capital account openness should go along with restrictions on the use of FX and discourage both deposit and loan dollarization. Higher external debt to GDP is likely to lead to higher balances in foreign exchange by the respective entities which owe this debt8. Accessi,t represents variables reflecting ease of access to foreign exchange—such as availability of export earning industries—and it is proxied by net exports to GDP. This variable has an ambiguous impact on deposit dollarization. While it could be expected that the ability to keep money overseas should reduce domestic deposit dollarization, the ability (of banks) to mobilize resources in foreign exchange may increase their appetite for passing on the exchange rate risk while making use of (potentially cheaper) foreign funding. lnsti,t stands for institutional characteristics that influence dollarization, including the exchange rate regime and the political freedom (Honig, 2009).

Fixed exchange rates—if credible—could lead to indifference between holding deposits in domestic currency or FX and thus likely lead to lower dollarization, given the often more cumbersome procedures and higher fees with FX holdings. The effect of a floating exchange rate is ambiguous: if purchasing power is largely determined by import prices and the domestic currency fluctuates significantly, then there is a higher risk of keeping domestic currency. The opposite also applies—if purchasing power is mostly a function of goods prices in local currency, FX holdings become riskier. Controlsi,t are relevant control variables such as the population size. αi are potential country fixed effects (controlling for time invariant crosscountry differences), and εi,t is an independent identically distributed (iid) error term. Variables are either measured in percent (inflation, exchange rate depreciation, real interest rates and interest rate spreads), percent of GDP (external debt, M2, and net exports), or reflect either an index (KA index, polity) or a dummy (income level). GDP per capita is measured in thousands of US dollars and population in millions of inhabitants. Therefore, coefficient estimates are interpretable as the change in the share of dollarization in response to a one unit change in the respective explanatory variable.

b. Model Selection

Previous studies on dollarization have discussed some potential issues on the econometric analysis of this topic. Generally, dollarization is affected by its persistence and endogeneity of some explanatory variables. In order to circumvent these problems, some authors have made used of cross-sectional OLS regressions and lagged variables (De Nicolo et al., 2003 and Levy-Yeyati, 2006). Additionally, to account for possible autocorrelation in the error term, other authors have employed annual panel data regressions and standard errors, or have modeled it explicitly (Neanidis and Savva, 2009 and Basso et al., 2011). With the purpose of deriving a baseline model that considers these potential concerns, we estimate various models with all regressors lagged by one period to minimize potential endogeneity.

Tables 3 to 6 show the results for of deposit and loan dollarization for household and firms, and for each we consider eight models to ensure the robustness of results. Models 1 through 5 are pooled OLS regressions with time dummy variables (model 2), a trend variable (model 3), a dummy for income level (model 4) and an interaction term between income level and trend (model 5). Model 6 performs an Arellano-Bond estimation and includes the lag of the specific case of dollarization as regressor, which, as expected, turns to be positive and significant, while most of the other variables are greatly reduced in their explanatory power. This implies that slow-moving factors that affect the long-run level of dollarization tend to be washed out, as they are captured by the lagged dependent variable. Given the heterogeneity across countries, and the results of Breusch-Pagan LM test strongly rejecting the poolability of the data, we also estimate random (model 7) and fixed effects models (model 8) correcting for the presence of heteroskedasticity and autocorrelation in the standard errors9. In addition, potential collinearity of the independent variables is a concern. To better understand whether this might have any repercussions for the coefficient estimates we make use of two test statistics. First, we compute variance inflation factors, a measure of multicollinearity commonly used. Variance inflation factors remain always below 2.5 for all variables. Second, the condition number test for the dollarization regression yields an index value of 5.30 for deposit dollarization and 5.98 for loan dollarization, suggesting that there is no support for multicollinearity (see Annex 4). We also assess the variation in the coefficient estimates by dropping one variable at a time from the baseline regression. While there are variations, the coefficients remain relatively stable and there is no instance of a swing from a positive significant to a negative significant coefficient value, or vice versa. This suggests that there are likely limited implications from collinearity.

Table 3.

Deposits Dollarization: Firms

article image
Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 4.

Deposit Dollarization: Households

article image
Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 5.

Loans Dollarization: Firms

article image
Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 6.

Loans Dollarization: Households

article image
Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Overall, the goodness of fit measures when available are relatively low. This means that while the factors in the econometric model can account for a significant fraction of the dollarization, there is still a non-negligible portion of dollarization yet to be explained, suggesting that there are some unique features which are not fully captured by traditional explanatory variables. As we saw from the literature review, micro-data variables such as an economic agents’ financial sophistication, age, etc. which matters for deposit and loan dollarization are not captured in our model.

Deposit dollarization: Households vs. Firms10

The results on deposit dollarization show great similarities between households and firms and one interesting difference (Table 3 and 4). In general, across model specifications, market development variables appear to be the most crucial factors driving deposit dollarization for both households and firms. Financial deepening, measured by M2 to GDP ratio appears highly significant, suggesting that, on average, an increase of one standard deviation in this variable (i.e. 21.7 percent) is associated with a reduction on both households and firms’ deposit dollarization of about 13 percentage points (pps). This is also what was found in previous studies such as Corrales et al. (2015). As financial sector development takes hold and more financial products in domestic currency are offered, dollarization declines. Financial sector deepening not only captures the diversity of savings products available, but also provides a more stable environment requiring less dollarization. Capital account openness appears particularly important for firms across all specifications, implying that less restrictions are associated with larger levels of dollarization. As previously found in the literature (e.g. Levy-Yeyati, 2006), results show that larger levels of external debt as percent of GDP are related to greater levels of deposit dollarization. On average, an increase of one standard deviation in this variable (i.e. 31.8 percent) increases deposit dollarization for households and firms in about 3.2 pps. Net exports as percent of GDP, a proxy for access to foreign exchange, appears as an important factor reducing deposit dollarization for both households and firms. Increasing net exports as share of GDP by one standard deviation (i.e. 16.8 percent) reduces dollarization by 1.7 pps.

For the two groups, the spread between domestic and US deposit rates appears significant in the non-Pooled OLS models. The results suggest that a one standard deviation increase in the spread (i.e. 11.5 percent) decreases deposit dollarization by 1.4 pps, on average across the two groups, supporting the portfolio view described earlier. Currency substitution considerations also appear relevant in random and fixed effects models. Currency substitution model does not appear to be an import explanatory factor in explain deposit dollarization of households and firms. For firms, depreciation against the US dollar increase deposit dollarization significantly, while for households the relevant variable seems to be inflation, but results are not statistically strong. This could be due to the period covered which overall, including in developing countries, was a disinflationary period, with inflation as a result having less of an impact on deposit dollarization. Finally, another explanation is that individuals hedge themselves through other means than just depositing their money in FX, but by buying real assets such as houses, cars, etc.

There is one interesting difference between the two set of regressions (tables 3 and 5). For households, better institutions are consistently, across models, associated with lower levels of dollarization, while this variable seems irrelevant for firms’. This result suggests that for households, which are less sophisticated than firms and typically less resilient, stronger institutions, by reducing risks may discourage dollarization (see Honig, 2009 for similar findings). Firms, however, having both the sophistication and the necessary skill set to hedge themselves from certain risks, are less dependent on strong institutions. This result could suggest that financial market development variables also captures relevant institutional conditions for firms.

Loan dollarization: Household vs. Firms

Contrary to deposit dollarization, the results for the determinants of households’ and firms’ loan dollarization show significant differences (Table 5 and 6). One limitation of our analysis is that borrowing is often constraint by banking regulations that is not directly captured in our regressions due to lack of data. In some countries, banking regulations for households differ from firms when it comes to FX borrowing (see Cayazzo et al., 2006). Variables capturing the currency substitution view are all statistically insignificant. This suggests that borrowing by households or firms in foreign currency is not directly determined by the inflationary environment, or (risks from) exchange rate depreciation. Regarding the portfolio view, the variables are largely insignificant. In other words, the relative cost of borrowing in foreign exchange cannot explain why households or firms borrow in FX (see also Mengesha and Holmes, 2013).

For households, M2 to GDP ratio is the only variable consistently significant and with a negative sign across almost all model specifications, suggesting that an increase in this ratio by one standard deviation (i.e. 21.7 percent) decreases households’ loan dollarization by 3.4 pps. In other words, having access to a deep and diverse banking system tends to reduce FX loan dollarization, and encourages borrowing in local currency. In contrast, this variable seems to play no significant role as determinant of firms’ loan dollarization. Within market development variables, only capital account openness exhibits some explanatory power for firms’ loan dollarization, with an increase in one standard deviation (i.e. 1.4 units) representing, on average, an increase of 4.7 pps in firms’ loan dollarization.

An open capital account encourages borrowing in FX for firms, though this impact is much weaker for households, where it is barely statistically significant. This could be explained by the ease with which firms can borrow abroad, which is not the case for households that face a higher hurdle rate. Similarly, net exports as percent of GDP does appear significant and with the expected negative sign, explaining firms’ loan dollarization across models, while for households, this sign is mostly not significant. This suggest that for firms, the ability to earn FX reduces the incentives to borrow in FX, while for households, which presumably earn very little in FX (or even nothing), this is not a significant factor in explaining loan dollarization. The exchange rate policy of a country has no statistically significant impact on the borrowing in FX by firms, which may be a reflection of firms being sophisticated. For households, the floating exchange rate reduces, as expected, the borrowing in FX, as a floating exchange rate increases the risks of borrowing in FX (see Garcia Pascual et al., 2006 for micropudential measures of how to reduce the FX related-risks. In this context, the analysis of Cerutti, Claessens and Laeven (2015), looking at the period 2001-2013 study corroborates our findings. Their analysis that the usage of macroprudential policies, especially foreign exchange related macroprudential ones, are common among emerging and to a lesser extent lower income country corroborates our findings. In AE, where dollarization is less of a concern, borrower-based macro-prudential policies are instead preferred. The quality of institutions, however, has neither an impact on the FX borrowing of firms or households.

Deposit dollarization: Comparing higher versus lower income countries

The regressions on deposit dollarization for households and firms do not show a significant coefficient for the income level dummy (Table 7 and 9). However, the sign is negative in both cases, reflecting the fact that deposit dollarization is lower in higher income countries. Does the level of development of the country—proxied by income per capita—impact the level of deposit dollarization for households or firms? To deepen the analysis, we explore differences in the determinants of deposit dollarization by further disaggregating regressions into higher and lower income countries for both households and firms. We document these differences using the results for pooled OLS regressions, and we also report results for random and fixed effects models (Tables 7 and 8).

Table 7.

Deposits Dollarization: Firms. High versus Low Income Countries

article image
Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 8.

Deposits Dollarization: Households. High versus Low Income Countries

article image
Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1