Wealth Redistribution After

By Andres Drenik, Gustavo Pereira and Diego J. Perez∗

A fraction of households in emerging of living (Cravino and Levchenko, 2017). economies save and borrow in foreign cur- Our findings suggest that the revaluation rency. By doing so they become exposed to of nominal net asset positions is an impor- revaluation effects on their wealth coming tant channel through which exchange rate from movements in the nominal exchange devaluations entail redistributive effects. rate. In this paper, we analyze the redis- tribution of liquid wealth across different I. Heterogeneity in Savings households after nominal exchange rate de- in Foreign valuations. The key channel we focus on is the revaluation of nominal net asset posi- tions denominated in different . Our analysis uses data from three dif- We begin our analysis by documenting ferent household surveys and covers the the fact that a significant fraction of house- following countries: Albania, Bosnia and holds in emerging economies have some as- Herzegovina, Bulgaria, Croatia, Czech Re- sets denominated in foreign currency, usu- public, Hungary, Macedonia, Peru, Poland, ally either in US Dollars or Euros. Our Romania, Serbia, Slovakia and Uruguay. analysis covers 13 countries from Eastern The first source of micro data is the Euro Europe and Latin America. We show that Survey carried out by the Austrian Central the likelihood of having assets in foreign Bank (OeNB) and covers all the mentioned currency is increasing in households’ in- European countries. This survey contains come. This fact is generalized across all biannual data from 2007 to 2015. While countries in our sample. Then, we ana- having wide coverage in terms of countries lyze the redistributive effects of a nominal and time span, this survey has limited infor- exchange rate . We compute mation regarding households’ holdings of fi- the magnitude of the revaluation of liquid nancial assets in foreign currency. The sec- wealth associated with a nominal devalua- ond source of data is a similar survey from tion of the local currency of 18% for dif- Peru conducted in 1997, for which a sim- ferent types of households. We find that ilar caveat applies. The last source is the there is an important degree of redistribu- household financial survey from Uruguay tion from households with low income to conducted in 2013 that contains detailed those with high income. information on households’ assets and lia- The existing literature has analyzed the bilities broken down by the currency of de- redistributive impact of nominal exchange nomination. The structure of this survey rate devaluations through its effects on the is similar to the structure of the Survey of labor market (Drenik, 2016) and the cost Consumer Finances conducted in the US. We first document the degree of hetero- ∗ Drenik: Department of Economics, Columbia geneity in the exposure of households to for- University, 420 W. 118th Street, New York, NY, eign currency. For this, we group house- 10027 (email: [email protected]); Pereira: De- holds into three terciles according to their partment of Economics, Columbia University, 420 level of income and compute the share of W. 118th Street, New York, NY, 10027 (email: [email protected]); Perez: Department of Eco- households that have some liquid asset de- nomics, New York University, 19 W. 4th Street, 6FL, nominated in foreign currency. For Peru New York, NY 10012 (email: [email protected]). We and Uruguay we consider cash holdings, thank Alberto Cavallo for useful comments and the Aus- time and savings bank deposits as all pos- trian and the Economics Department at Universidad de la Rep´ublica(Uruguay) for sharing their sible liquid assets. For the European coun- data. tries we consider cash holdings only due 1 2 PAPERS AND PROCEEDINGS MONTH YEAR to data limitations.1 Figure 1a shows the In Figure 1b, we show the fraction of share of households that have a positive households’ liquid wealth that is denomi- fraction of their wealth denominated in for- nated in foreign currency and also find that eign currency. The average share of house- it is increasing in households’ income in all holds exposed to foreign currency differs countries. This fact is mostly driven by significantly across countries, ranging be- the participation/non-participation margin tween 5% and 65%. Despite this hetero- documented above. geneity, within each country the likelihood Finally, we also document that these dif- of being exposed to foreign currency in- ferences in exposure to foreign currency creases with households’ income. The share persist over time. In the Online Appendix of households in the top income tercile ex- we show that in almost every point in time posed to foreign currency is higher than the the likelihood that households are exposed share for those in the bottom income ter- to foreign currency is increasing in income. cile in every country. These findings sug- This is due to the fact that although the gest that the market for foreign currency is share of households that are exposed to for- segmented, with more limited participation eign currency varies over time for each in- by low-income households. come tercile, it tends to co-move strongly across the income distribution. This persis- tence suggests that the redistributive effects

80 of devaluations should be present in most SRB devaluation episodes, regardless of when

ALB

60 they occur.

.MKD SVK CZE HRV II. Redistributive Effects of an 40 POL Exchange Rate Devaluation

.URU ROU

20 BGR Share of HHs with of with HHs Share Assets FC in HUN. BIH We investigate the redistributive effects PER of an exchange rate devaluation by ana-

0 lyzing how the value of liquid wealth de- Low Income Middle Income High Income nominated in local and foreign currency is (a) Share of Households with Assets in FC affected on-impact by an unexpected and permanent devaluation. The goal of this

SRB exercise is to determine how unequal is the 50 distribution of gains and losses experienced by households. We focus on Uruguay since 40 it is the country with the most granular

30 MKD . HRV data. We also measure effects on liquid wealth since changes in liquid wealth are

20 ALB more likely to translate into changes in con-

Share of Liquid Wealth of Liquid Share FCin SVK sumption in the short-run. This exercise is

10 . BIH BGR CZE. ROU POL similar to the one carried out by Doepke HUN and Schneider(2006) that focus on the re- 0 Low Income Middle Income High Income distributive effects of inflation in the US. Investments in local and foreign currency (b) Share of Liquid Wealth in FC in emerging economies can have different payoffs given the fluctuations of nominal Figure 1. Heterogeneity in Savings exchange rate. Figure2 shows the dis- in Foreign Currency tribution of the monthly devaluation rates that Uruguay experienced between 1995 and 2016. Large monthly devaluations 1In our Online Appendix we repeat the analysis for the European countries with bank deposits and reach (above 10%) are occasionally observed. Ad- similar conclusions. ditionally, there are also periods of con- VOL. VOL NO. ISSUE WEALTH REDISTRIBUTION AFTER EXCHANGE RATE DEVALUATIONS 3 tinued large devaluations, such as the one per US dollar) and NWFC (NWLC ) is the experienced between June and September net liquid wealth denominated in foreign 2002 when the nominal exchange rate in- (local) currency. creased by 89%. Finally, the distribution is Table1 presents summary statistics positively skewed, so large devaluations are about the composition of households’ nom- more common than large appreciations. inal portfolios.3 Statistics are presented for the overall population and by income ter- cile. First, notice that only a small frac-

25 tion of households have any type of nominal asset or liability. Only 11% of households 20 have assets in foreign currency, 9% have as- sets in local currency and a third of house- 15 holds have any kind of liability. However, Density the share of households with some assets 10 is increasing in household income. While

5 only 3% (5%) of households with low in- come have assets in foreign (local) currency, 0 -.1 0 .1 .2 .3 26% (19%) of households with high income Monthly devaluation rate have assets in foreign (local) currency. The share of households with any kind of liabil- Figure 2. Distribution of Devalua- ity is fairly stable across the income distri- tion Rates: Uruguay 1995-2016 bution. In terms of the intensive margin, on av- The first step in our analysis requires erage households have US$16,406 in foreign computing nominal net liquid asset posi- currency assets, US$7,324 in local currency tions by currency of denomination for each assets and US$3,081 in local currency lia- household in the survey. On the asset bilities (all conditional on having a positive side of the balance sheet we include: sav- amount). The average net liquid wealth is ings and checking accounts and cash.2 On around US$1,000. However, the distribu- the liability side we include: credit card tion of net liquid wealth is highly unequal: debt and non-mortgage/non-vehicle related the bottom decile is US$-2,864, the median debt. All these assets and liabilities are cat- is US$0 and the top decile is US$1,440. egorized as being denominated in foreign Mean holdings and net wealth are also in- (FC) and local currency (LC). Foreign creasing in income. While the average net currency assets/liabilities are almost com- liquid wealth is US$-637 among the low in- pletely denominated in US dollars. Thus, come households, it is equal to US$4,970 we define total net liquid wealth (TNW ) among high income households. measured in pesos and dollars as Next, we compute the revaluation gains and losses of these net liquid wealth posi- (1) TNWin Pesos = εNWFC + NWLC tions after a nominal exchange rate deval- uation. We measure the percentage change and in net liquid wealth measured in both cur- rencies (expressed in pesos and dollars) (2) TNWin Dollars = NWFC + NWLC /ε since they can provide bounds for real ef- fects. If the price index of the consumption respectively, where ε denotes the nominal bundle moves closely with the nominal ex- exchange rate (units of Uruguayan pesos change rate (full pass-through), one should focus on revaluation effects expressed in 2We do not include other types of liquid nominal fi- nancial assets (such as bonds), since data on holdings by currency of denomination is not available. However, 3Summary statistics on foreign currency liabilities most households savings are in the form of bank ac- are not included, since less than 0.1% of households de- counts and cash. clared having any. 4 PAPERS AND PROCEEDINGS MONTH YEAR

Table 1—Composition of Nominal Porftolio

Fraction Percentiles Positive 10th 50th 90th Mean A. All Households Assets FC 11% 800 6400 39200 16406 Assets LC 9% 440 2800 16000 7324 Liabilities LC 33% 160 1280 8000 3081 Net Wealth - -2864 0 1440 991 B. Low income Assets FC 3% 800 4000 8400 4858 Assets LC 5% 200 800 11200 4079 Liabilities LC 30% 156 960 4480 2913 Net Wealth - -1840 0 0 -637 C. Middle Income Assets FC 7% 720 5600 38400 12644 Assets LC 7% 400 2400 6400 3276 Liabilities LC 36% 160 1200 6400 2443 Net Wealth - -3160 0 320 38 D. High Income Assets FC 26% 1440 8000 40000 20500 Assets LC 19% 800 4000 28000 10472 Liabilities LC 31% 240 2400 9440 4225 Net Wealth - -4120 0 16000 4970

Notes: Figures in columns 2 to 5 are expressed in US Dollars. dollars to get an idea of real effects. The op- lute value of net wealth, since a fraction of posite is true if the price index does not co- households have negative net wealth. Note move with the nominal exchange rate (no that the revaluation effects in pesos and dol- pass-through). Thus, the realized effects lars for a given household can have the same will likely lie somewhere in between these sign if that household is a net debtor in one bounds. For this exercise we choose a nom- currency and a net creditor in the other cur- inal devaluation rate of 18%, which is equal rency. Results are presented in Table A1. to the magnitude of a devaluation episode When measured in pesos, net wealth in- that occurred while the survey was being creases after a nominal devaluation of 18% conducted. by 1.21% on average. The average effect is positive since most households have pos- Given the definitions of net wealth in (1) itive net wealth in foreign currency (as- and (2), the revaluation effects measured in sets in foreign currency exceed liabilities pesos and in dollars are defined as in foreign currency for most households). More importantly, revaluation gains are in- ∆TNW ∆ε εNW in Pesos = × FC creasing in households’ income. While the |TNWin Pesos| ε |TNWin Pesos| average increase in net wealth is 0.29% for households in the lowest income ter- and cile, it increases to 3% for households with ∆TNWin Dollars ∆ε NWLC high income. Also, revaluation gains are = − × , highly concentrated. While the top decile |TNWin Dollars| ε ε|TNWin Dollars| of the distribution of revaluation gains is respectively. Revaluation effects are mea- zero for low and middle income households, sured as percentage changes of the abso- it is 16.2% for high income households. VOL. VOL NO. ISSUE WEALTH REDISTRIBUTION AFTER EXCHANGE RATE DEVALUATIONS 5

Table 2—Revaluation Effects holds is higher than the gain experienced by low income households. Our findings rein- Percentiles force the results presented by Cravino and 10th 50th 90th Mean Levchenko(2017), who show that the cost A. All Households of living of poorer households increased by In Pesos 0.00 0.00 0.00 1.21 up to 1.62 times more than the cost of living In Dollars 0.00 0.00 18.00 4.79 of richer households after the 1994 Mexican B. Low income devaluation. In our case the redistribution In Pesos 0.00 0.00 0.00 0.29 operates through the revaluation of nomi- In Dollars 0.00 0.00 18.00 4.68 nal liquid assets in different currencies. C. Middle Income In Pesos 0.00 0.00 0.00 0.95 III. Conclusion In Dollars 0.00 0.00 18.00 5.72 D. High Income We have shown that a fraction of house- In Pesos 0.00 0.00 16.20 3.00 holds from many emerging countries have In Dollars -10.80 0.00 18.00 3.52 part of their wealth denominated in foreign currency. Within countries there is signifi- Notes: Figures expressed in percentage points of the cant heterogeneity in the likelihood of hav- absolute value of net wealth. ing assets in foreign currency. High-income households tend to have a larger fraction This indicates that the gains are concen- of their wealth in foreign currency and the trated among the richest households and differences across the income distribution that most of the redistribution operates are persistent over time. Using detailed through the participation margin. micro-data from Uruguay, we analyze the On the other hand, when measured in redistributive effects of nominal exchange dollars, net wealth increases by 4.79% on rate devaluations on households’ net nomi- average. In this case, the average effect is nal positions and show that those gains are also positive because most households have highly concentrated among households with negative net wealth in local currency (li- high-income. Thus, exchange rate policy abilities in local currency exceed assets in can have large impact on wealth inequality, local currency for most households). When and thus on consumption inequality as well. measured in dollars, average effects are not REFERENCES monotonic in household income anymore. The largest average gain corresponds to Cravino, Javier, and Andrei A Levchenko. 2017. middle income households, since the frac- “The Distributional Consequences of Large Devalu- tion of households with local currency lia- ations.” American Economic Review, 107(11): 3477– bilities is much larger than the fraction with 3509. local currency assets. Finally, a fraction Doepke, Matthias, and Martin Schneider. 2006. of high income households also experienced “Inflation and the redistribution of nominal wealth.” losses. These are the households that have Journal of Political Economy, 114(6): 1069–1097. assets primarily in local currency. Drenik, Andres. 2016. “Labor Market Dynamics Af- Our findings suggest sizable revaluation tern Nominal Devaluations.” Manuscript. effects of nominal exchange rate devalua- Lluberas, Rodrigo, and Juan Odriozola. 2015. “In- tions. These are greater in magnitude than flation, currency depreciation and households bal- those estimated by Lluberas and Odriozola ance sheet in Uruguay.” Documento de Trabajo del (2015), who analyze the effects of a nomi- Banco Central del Uruguay 2015/009. nal devaluation on total wealth (including real assets) and conclude that the amount of redistribution across households is small given that housing accounts for the main source of wealth. We also find that the net gain experienced by high income house- ONLINE APPENDIX

Additional Figures and Tables

HRV .2 .MKD SRB

ALB .15

SVK

.1 BGR

HUN BIH Share HHs w/ HHs FC Deposits Share . ROU . POL

.05 CZE 0 Low Income Middle Income High Income

Figure A.1. Share of Households Exposed to Deposits in FC

Table A1—Dynamic Behavior of Exposure to FC: Correlations Over Time

Low Inc.- High Inc.- Country Middle Inc. Middle Inc. Albania 58.2% 70.4% Bulgaria 61.2% 53.7% Bosnia Her. 42.8% 67.6% Czech Rep. 61.8% 61.0% Croatia 84.4% 79.8% Hungary 14.4% 35.4% Macedonia 90.0% 83.4% Poland 78.2% 72.9% Romania 34.2% 63.2% Serbia 79.6% 72.5%

Notes: First (second) column: correlation between share of households exposed to foreign currency in low (high) and middle income tercile over time. Sample period: 2007-2015. 2 PAPERS AND PROCEEDINGS MONTH YEAR

Albania Bosnia and Herzegovina Bulgaria Croatia .3 .6 .3 .6 .2 .4 .2 .4 .1 .2 .1 .2 0 0 0

Czech Republic FYR Macedonia Hungary Poland .3 .4 .5 .6 .4 .2 .4 .2 .3 .1 .2 .2 0 0 .1

2006h12008h12010h12012h12014h12016h1 2006h12008h12010h12012h12014h12016h1

Romania Serbia .8 .3 Share of HHs Exposed to FC .6 .2 .4 .1 .2 0

2006h12008h12010h12012h12014h12016h1 2006h12008h12010h12012h12014h12016h1

Low income Middle Income High Income

Figure A.2. Exposure to FC over time