Exchange Rate Dynamics and Monetary Spillovers with Imperfect Financial Markets

Abstract

We develop a quantitative model with imperfections in domestic and international financial markets that generates strong effects of U.S. monetary policy on emerging markets (EMs). Financial imperfections prevent arbitrage both between local EM lending and borrowing rates, and between local-currency and dollar borrowing rates. An adverse feedback effect between financial health and external conditions amplifies the domestic “financial accelerator,” leading to large cross-border spillovers of U.S. monetary policy shocks. The model implies a link between uncovered interest parity violations and local credit spreads, a prediction we show the data strongly supports.

Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

The effects on foreign economies of monetary policy shifts in the United States, often referred to as “spillovers,” are the subject of increasing attention. The financial media regularly publishes stories highlighting global reverberations from Federal Reserve decisions, and foreign policy makers often express concern about the impact of U.S. monetary policy on their own economies.¹ A fast-growing empirical literature quantifies the effects of U.S. monetary shocks on foreign countries’ financial and economic developments. A common finding in this literature is that these spillovers can be substantial, particularly on emerging market economies (EMs henceforth).²

One prominent theme in the empirical literature, highlighted in influential work by Miranda-Agrippino and Rey (2020), is that a major channel of spillovers is through financial conditions. In this view, monetary policy actions in the United States exert powerful effects on asset prices and financial intermediaries around the globe. When the Federal Reserve tightens policy, global asset prices decline, foreign currencies depreciate sharply against the dollar, and financial constraints tighten in foreign economies. These developments occur in reverse when the Fed eases. A detailed analysis of Turkish data by Giovanni et al. (2022) provides additional evidence on these patterns, by showing that these global effects feed into emerging markets’ local credit conditions, ultimately affecting the cost of credit facing local nonfinancial borrowers. These authors also document systematic, time-varying violations of uncovered interest parity (UIP): the premium on the cost of credit in local currency relative to the cost of dollar credit, which is typically positive, tends to be larger when overall financial conditions are tight.

In this paper we develop a two-country quantitative macroeconomic model that can account for substantial spillovers of U.S. monetary policy, in which imperfections in both domestic and international financial markets play a key role. We show that the model can generate strong effects of U.S. monetary policy shocks on (a) the prices of assets in EMs; (b) the tightness in EM financial conditions, as measured by the credit spread facing local nonfinancial borrowers; (c) large movements in the local currency relative to the dollar, driven in part by an endogenous deviation from UIP that widens when the Fed tightens; and (d) considerable effects on EM gross domestic product (GDP). Our model can thus capture the salient facts uncovered by the papers referenced above, including the response of non-U.S. asset prices, credit conditions, and UIP premiums to U.S. monetary policy shifts. In addition, our framework is able to capture the implications for real economic activity of these financial developments.

A key aspect of our theory is a failure of UIP. This failure arises because financial imperfections limit the ability of EM borrowers to obtain foreign-currency denominated financing.³ More specifically, EM banks (meant to capture financial intermediaries broadly defined) borrow in both domestic and international capital markets to finance domestic investments. The domestic capital market consists of local currency-denominated loans, and the international market operates in dollar-denominated loans. Both forms of external financing are subject to agency frictions. Further, these frictions are more severe for funds of foreign origin, effectively making the local asset less valuable as collateral against dollar loans than against domestic-currency loans. The two forms of financing are thus imperfect substitutes, a property that leads to a failure of UIP. The failure of UIP takes a specific form: we show that the model predicts that the UIP premium (defined as the premium on the local safe rate relative to the dollar safe rate, inclusive of expected dollar appreciation vis-á-vis the local currency) is positively linked to the expected yield on the local risky asset in excess of the local safe rate, that is, the local credit spread. In the empirical section of the paper, we test this model prediction extensively using EM data, and find strong support for it. In general equilibrium, we show that both the credit spread and the UIP premium are inversely linked to EM banks’ net worth.

We embed the financial imperfection just described within a conventional New Keynesian production economy and use our framework to study the spillover effects of U.S. monetary policy shocks on the EM. We find that the model predicts considerable effects of U.S. policy shifts on emerging market GDP, consistent with the evidence, largely because of the presence of the financial imperfection. The main mechanism works as follows. A tightening of U.S. monetary policy triggers losses in EM borrowers’ balance sheets. This happens for two reasons. First, given the presence of some balance sheet mismatch on the part of EM banks (as their assets are denominated in local currency, while some of their debt is in dollars) the local currency depreciation triggered by the tightening raises the real burden of the dollar-denominated debt, reducing banks’ net worth. Second, the depreciation also creates some increase in local real interest rates, because total expected inflation falls as the exchange rate is expected to appreciate, also creating some drag in investment spending and in local asset prices.

Weaker local balance sheets then initiate powerful feedback effects. Balance sheet deterioration raises the agency costs of external finance. The local lending spread increases as a result, making credit more expensive for local borrowers and triggering declines in investment and in the price of capital (or Tobin’s q), and ultimately slowing activity. These developments then feed back into borrowers’ financial positions, weakening them further. These feedback effects operating through domestic conditions are well-known in the literature, and usually referred to as the “financial accelerator” following Bernanke, Gertler, and Gilchrist (1999).

Our model adds a second set of feedback effects, based on the interaction between balance sheets and external conditions, that amplifies the domestic-based financial accelerator. A weakening of local balance sheets widens the UIP premium on the local currency, which is accommodated via a depreciation of the latter against the dollar. Because local balance sheets are partly mismatched, a weaker local currency then feeds into balance sheet health, further weakening it, and once again initiating both rounds of feedback. The end result is sharply amplified declines in local investment, asset prices, and exchange rates, and ultimately GDP (through a large contraction in investment demand). Given the strength of these feedback mechanisms, the amount of amplification of U.S. monetary shocks is considerable despite a relatively modest degree of balance sheet mismatch (for the typical borrower in the model, the majority of debt is still denominated in local currency, consistent with the available data from EMs).

A key prediction of our model that lies at the core of the aforementioned financial amplification channel is that the UIP premium is tied to the local credit spread. We test this prediction using monthly data for several EMs for the period from 2000 to 2022. To do so, we estimate a version of the forward-looking exchange rate equation predicted by our model, in which we include proxies for the local credit premium as well as interest rate differentials. We first estimate a version of these regressions by pooling data for all EMs in our sample. We find strong support for the prediction that a measure of domestic credit constraints is tightly linked with the real exchange in these countries.

Our model also predicts that the strength of the association between the UIP premium and the local credit spread is governed by a parameter, γ, that measures the extent of frictions in the international credit market facing borrowers in a given country. Next, we exploit the panel dimension of our data set to test this prediction. While the value of γ is not directly observable, the model suggests it should be positively linked to a country’s overall macroeconomic vulnerability. Accordingly, we use the vulnerability index constructed by Ahmed, Coulibaly, and Zlate (2017) as an indirect proxy of γ. Ahmed, Coulibaly, and Zlate (2017) show that the value of this indicator before the onset of global financial stress periods is a very good predictor of subsequent pullback by international investors, thereby confirming that the indicator is a good proxy for γ.

We first split our sample based on the value of the vulnerability index, and estimate the exchange rate equations separately for each group. We find that countries with weaker fundamentals (proxying for higher γ) see a much tighter association between the UIP premium and the domestic credit premium, as predicted by the model. We then complement the split-sample approach by estimating the empirical UIP specification country by country. We show that the estimated coefficient for the domestic credit premium for each country lines up cross-sectionally with the country’s respective vulnerability measure, in line with the predictions of our model.

In our final empirical analysis, we estimate a vector autoregression (VAR) model similar to Christiano, Trabandt, and Walentin (2010), whose approach has been widely used to assess the effects of U.S. monetary shocks on the U.S. economy. We augment empirical analysis in Christiano, Trabandt, and Walentin (2010) to include data for EMs. We show that our model’s predictions on the spillover effects of a U.S. monetary policy shock on EM activity are broadly in line with the VAR-implied ones, that also demonstrate large effects. This is in contrast with the predictions of a standard New Keynesian model without credit market frictions, in which the effect on EM output is fairly small under realistic parameterizations.

Our paper builds on a large literature that develops open-economy New Keynesian macroeconomic models (e.g., Corsetti and Pesenti 2001; Gali and Monacelli 2005; Erceg, Gust, and Lopez-Salido 2007; Farhi and Werning 2014; Corsetti, Dedola, and Leduc 2018). This literature is based on the seminal work by Obstfeld and Rogoff (1995), who study the effects of monetary and fiscal policies in open economies. The models in this literature generally feature frictionless domestic and international financial markets,⁴ while we depart by introducing financial market frictions following Gertler and Kiyotaki (2010).⁵

This paper also relates to a lengthy literature that was developed in response to the EM crises of the 1990s, which in several cases highlighted the balance-sheet channel of exchange rate fluctuations. Well-known examples include Krugman (1999), Céspedes, Chang, and Velasco (2004), and Gertler, Gilchrist, and Natalucci (2007).⁶ Our model also features balance-sheet effects of exchange rates but is otherwise quite different from the models in this literature. In our setting financial imperfections are microfounded via an explicit agency problem, with the share of assets financed by each type of debt determined endogenously as the solution to an optimal portfolio problem. This setup leads to the interaction between domestic and external feedback effects described earlier, a finding that we believe is novel, and implies greater financial amplification than in existing models. Also different from the existing literature, our paper focuses on quantifying the cross-border effects of U.S. monetary policy.

Also related to our work is a set of papers that focuses on the role of financial market imperfections in exchange rate determination (e.g., Hau and Rey 2006; Bruno and Shin 2015b; Gabaix and Maggiori 2015; Itskhoki and Mukhin 2021). Different from the setup in these papers, in our setup EM intermediaries seek funding in different currencies to fund a local productive asset, and imperfect arbitrage arises because enforcement frictions have greater severity for funds of foreign origin. This leads to the distinct prediction that the currency premium is tied to the local lending spread, which we find has strong support in the data, and which underpins the strong financial amplification effects that are ultimately responsible for the large spillover effects of U.S. monetary policy.

The focus of our paper is related to Gourinchas (2018), who quantifies the different channels of spillovers from U.S. monetary shocks. The details of both modeling frameworks are quite different however, with ours devoting attention to endogenizing both the EM’s local lending spread and the currency premium. Aoki, Benigno, and Kiyotaki (2016) develop a small open economy model with financing frictions to study monetary and financial policies in EMs, which shares several similarities with our model. Our work differs both in terms of focus—we study spillovers from U.S. monetary policy within an asymmetric two-country model—and in terms of modeling features, for example, we highlight the importance of allowing for dollar invoicing of international trade in EMs. In addition, our paper emphasizes the critical role of endogenous deviations from UIP in shaping dynamics; we study this issue within a simplified setting that allows for some analytical results, and also provide evidence supporting the model-implied UIP deviations. These considerations also differentiate our work from other related papers, including Banerjee, Devereux, and Lombardo (2016), who focus only on domestic financial frictions in accounting for cross-border spillovers, and Fornaro (2015) and Devereux, Young, and Yu (2019), who focus on capital controls and exchange rate policy during sudden stops in the context of a small open economy framework with an occasionally binding collateral constraint.⁷

1 A Two-Country Model with Imperfect Financial Markets

Time is discrete and runs to infinity: $t=0,1,2,…$⁠. There are two countries, a small EM (home) and the United States (foreign). Financial intermediaries finance the acquisition of EM physical capital with funds borrowed both from home households and from U.S. households. Each type of financing is denominated in the source country’s currency. Because of limited enforcement friction, intermediaries may face limits in their ability to borrow. The friction affects the two types of borrowing (domestic and foreign) asymmetrically: enforcement problems are more severe for funds of foreign origin. This asymmetry leads to a failure of UIP, as we will show. The four types of agents in the home economy are households, bankers, firms, and the central bank. We describe each of these agents in turn, and then briefly describe the foreign economy.

1.1 Households

A continuum of identical households of measure $N$ live in the home economy. Each household has two types of members: workers and bankers, with measures $1−f$ and f, respectively. Workers supply labor and return the wages they earn to the household. Each banker manages a financial intermediary and also transfers his or her earnings to the household. There is perfect consumption insurance between the two types of household members.

The representative household chooses a consumption index C_t, labor supply L_t, consumption of a domestically produced (⁠$CD,t$⁠) and imported (⁠$MC,t$⁠) basket, deposits with domestic financial intermediaries D_t (denominated in terms of the consumption index), and holdings of a one period nominal risk-free bond, B_t (in zero aggregate net supply). Its optimization problem consists in choosing a state-contingent sequence ${Ct+j,Lt+j,CD,t+j,MC,t+j,Dt+j,Bt+j}j=0∞$ to maximize expected discounted lifetime utility

where U(C, L) is the period utility function. The consumption index is a CES aggregate of the domestically produced and imported baskets:

where $1−ω>1/2$ is the home bias and $η>1$ is the elasticity of substitution between goods of domestic and foreign origin. The maximization of (1) is subject to a sequence of budget constraints of the form

for all t, with

Here, P_t is the consumer price index (CPI); $PD,t$ and $PM,t$ are the prices of the domestically produced and imported basket, respectively; W_t is the nominal wage; R_t is the (real) gross interest rate on deposits; $Rtn$ is the nominal interest rate; and Π_t are profits from firms and bankers rebated to the household.

Letting $UC,t≡∂U(Ct,Lt)∂Ct$ and $UL,t≡∂U(Ct,Lt)∂Lt$⁠, the first-order conditions for $Dt,Bt,Lt,CD,t,$ and $MC,t$ associated with the problem above are, respectively:

The baskets $CD,t$ and $MC,t$ are themselves CES composites of a continuum of good varieties produced at home and abroad, respectively, each of measure unity. Thus, $CD,t=(∫01CD,j,tϵ−1ϵdj)ϵϵ−1$⁠, where $CD,j,t$ is consumption of domestic variety $j∈[0,1]$⁠. The associated first-order condition yields a standard demand curve for domestic variety j from the domestic household:

where $PD,j,t$ is the price set by domestic producer j, and where

is the domestic price index (i.e., an index of prices of the goods produced domestically).

An analogous set of conditions holds for the imported consumption good basket $MC,t$⁠, with the same parameter ϵ governing the elasticity of substitution between individual varieties.

1.2 Bankers

Each banker in the household operates a financial intermediary. Bankers exit randomly: any banker operating in period t continues into period t + 1 with exogenous probability σ. With the complementary probability, the banker exits, rebates his or her earnings to the household, and begins a career as a worker. At the same time, workers in the household become bankers with probability $(1−σ)f1−f$⁠, so a measure $(1−σ)f$ of new bankers enter each period and exactly offset the number that have exited. Entrant bankers receive a small equity endowment from the household so they can start operations.

Banker i chooses assets $Ai,t$⁠, deposits issued to domestic households in the local currency $Di,t$⁠, and deposits issued to U.S. households in dollars, $Di,t*$⁠.⁸ (Throughout, we use * to refer to foreign variables.) The assets $Ai,t$ consist of claims on the EM’s physical capital. The optimization problem facing banker i is to choose a state-contingent sequence ${Ai,t+j,Di,t+j,Di,t+j*}j=0∞$ to maximize

where

is the household’s stochastic discount factor (SDF) between period t and period t + j, and $Ni,t+j$ is terminal net worth if the banker exits at t + j. The banker’s objective function is the expected value of its payout to the household, evaluated using the household’s discount factor. (Note that the probability of exiting in period t + j for a banker alive in period t is $(1−σ)σj$⁠.)

Maximization of (12) is subject to two constraints. The first is the banker’s budget constraint, given by

for each t. Here Q_t is the (real) market price of a claim on a unit of capital; Z_t is the nominal payoff generated by a unit of asset holdings; δ is capital’s depreciation rate; $Rt*$ is the real interest rate in the foreign currency; and $St$ is the real exchange rate, expressed as the price of the home consumption index in terms of the foreign index. Note that a real depreciation of the home currency is captured by a decrease in $St$⁠.

The balance sheet identity,

states that the value of the banker’s assets equals the value of its liabilities (consisting of the sum of the value of deposits issued in domestic and foreign currency and the value of net worth, all expressed here in terms of the home basket). This identity can be combined with (14) to yield

where $RK,t≡ZtPt+(1−δ)QtQt−1$ is the return on bankers’ assets. Equation (16) can be equivalently used in place of (14) as the banker’s balance sheet constraint.

The second constraint arises due to moral hazard. After borrowing funds, the banker may decide to divert assets for personal gain, rather than honoring obligations with creditors. Diverting means selling a fraction $Θt∈(0,1)$ of assets secretly in secondary markets. The remaining assets are then seized by the banker’s creditors in bankruptcy proceedings. We assume that Θ_t depends on the composition of the banker’s liability portfolio:

where $Θ(·)$ is a function satisfying $Θ′>0$⁠. Thus, the banker is able to divert more assets when the fraction of his or her assets financed by foreign liabilities (equal to $St−1Di,t*$⁠, once expressed in terms of the home basket) is larger.

The assumption that $Θ′>0$ captures the notion that the legal and institutional environment in EMs, as well as the nature of the EM capital which serves as collateral, effectively make it more difficult for foreign creditors to recover assets from a defaulting borrower, compared with domestic depositors.⁹

The banker’s decision then consists in comparing the continuation value $Vi,t$⁠, which measures the present discounted value of future payouts from operating “honestly,” with the gain from diverting funds, $ΘtQtAi,t$⁠. Therefore, the banker’s portfolio choice must also satisfy the incentive constraint

which requires the banker’s continuation value to be no smaller than the value of diverting funds. If (18) were not satisfied, no rational creditor would be willing to lend to the banker, in recognition of the latter’s incentive to default. This form of constraint, first introduced by Gertler and Kiyotaki (2010), has been widely used in recent literature as a way to endogenize limits to bankers’ ability to attract funds. Our setting differs from the closed-economy variants by allowing the degree of frictions to depend on the extent of foreign-currency borrowing.

We now proceed to solving the banker’s problem using the method of undetermined coefficients. We guess that the continuation value $Vi,t$ satisfies $Vi,t=ΨtNi,t$⁠, where $Ψt$ is a time-varying coefficient that is independent of banker-specific variables. We also define the following ratios:

The variable $ϕi,t$ is the ratio of assets to net worth, the banker’s leverage. The variable $xi,t$ is the ratio of foreign financing to assets. Both ratios turn out to be independent of banker-specific variables. In anticipation of that result we write $ϕi,t=ϕt$ and $xt=xi,t$⁠.

From (16), the evolution of the banker’s net worth satisfies

Given this condition, along with (12) and (18), we may express the banker’s problem as

subject to

where

with

The variable μ_t is the excess marginal value to the banker of assets over deposits; $ϱt$ is the excess marginal cost of domestic relative to foreign funding (equivalently, the marginal value of foreign relative to domestic funding); and ν_t is the marginal cost of domestic funding. Note that the banker uses the discount factor $Λt,t+1Ωt+1$ to evaluate payoffs, which weighs the household’s SDF with the prospective value of a unit of net worth to the banker (given by $Ωt+1$⁠). Condition (23) makes clear that the incentive constraint places a restriction on the maximum leverage $ϕt$ the banker can take on. The first-order condition associated with the choice of x_t is

This condition equates the marginal value of foreign relative to domestic funding, $ϱt$⁠, to its marginal cost, which relates to the fact that a larger x_t tightens the incentive constraint (18).

As long as the total excess return $μt+ϱtxt$ satisfies

the incentive constraint binds. We will assume this to always be the case. Then from (23), the banker’s leverage $ϕt$ is given by

We will restrict attention to cases in which x_t can be solved for uniquely from (28) as a function of μ_t and $ϱt$ (which are independent of any bank-specific variables). Then, from (29)$ϕt$ is also independent of bank-specific variables, confirming our earlier statement.

We now turn to aggregation across bankers. Let $At=∫0fAi,tdi, Dt*=∫0fDi,t*di$⁠, and $Nt=∫0fNi,tdi$⁠. Given that $ϕt$ and x_t are independent of bank-specific factors, these aggregates satisfy

If banker i is a new entrant, we assume he or she receives an equity endowment $Ni,t=1fWt$⁠, where $Wt$ is a function of aggregate quantities to be specified later. If banker i is instead a continuing banker, his or her net worth is given by (21). Aggregating across all bankers yields the evolution of aggregate net worth:

1.3 Firms

The two types of firms are capital goods producers that create new capital using final goods and existing capital and final goods producers that employ labor and capital to produce final output. We describe each in turn.

1.3.1 Capital goods producers

We introduce capital producers to capture convex costs of adjusting the capital stock in a way that facilitates aggregation. This results in a “Tobin’s q” behavior of investment, with the rate of investment increasing in the replacement cost of capital.

A domestic representative capital goods producer combines investment, I_t, and rented capital, K_t, to produce new capital. Investment is itself an aggregate of domestic and foreign-produced goods. The capital goods producer’s activity is subject to adjustment costs: producing I_t units of new capital requires $It+ψI2(ItKt−δ)2Kt$ units of goods, where δ is capital’s depreciation rate. Thus, adjusting the capital stock (beyond depreciation) entails quadratic costs.

The capital producer then sells new capital at the market price Q_t. Its maximization problem can be expressed as¹⁰

The associated first-order condition is

which links the price of capital, Q_t, positively to the investment-capital ratio.

The aggregate capital stock then evolves as

with K₀ given. New capital I_t is created using a mix of domestic (⁠$ID,t$⁠) and imported (⁠$MI,t$⁠) goods, combined by means of an aggregator analogous to (2):

From cost minimization, capital goods’ producers demand for domestic and imported goods are

Like the case of consumption goods, $ID,t$ and $MI,t$ are themselves CES composites of the domestic and foreign varieties, respectively: $ID,t=(∫01ID,j,tϵ−1ϵdj)ϵϵ−1$⁠, where $ID,j,t$ is the amount of domestic variety $j∈[0,1]$ used in production of new capital goods. From cost minimization, the resultant demand curve for variety j from capital producers is

with $PD,t$ given in (11). Similar conditions hold for the imported aggregate $MI,t$⁠.

1.3.2 Final goods producers

A continuum of mass unity of differentiated final goods producers sell in domestic and foreign markets and are subject to pricing frictions. Let the $j∈[0,1]$ subindex denote the producer of domestic variety j. This producer employs capital, $Kj,t$⁠, and labor, $Lj,t$ inputs to produce $Yj,t$ units of variety j, by means of the production function

Cost minimization yields $WtZt=(1−α)αKj,tLj,t$ for any j (Z_t is the nominal rental rate of capital), implying

with $Kt=∫01Kj,tdj, Lt=∫01Lj,tdj$⁠. Cost minimization also yields the following expression for (nominal) marginal cost, denoted as MC_t:

Let $Yt≡(∫01Yj,tϵ−1ϵdj)ϵϵ−1$ represent an index for domestic aggregate output. For future reference, we note that up to a first order, the previous index relates to aggregate employment and capital in a manner analogous to (40):¹¹

The pricing friction follows the Calvo formulation, whereby firm j is randomly allowed to reset its price with probability $1−ξp$⁠. We assume that firms in the domestic economy set prices in the local currency: they set the price of goods sold in the domestic market in the domestic currency, and that of goods sold in the foreign market in the foreign currency. (This contrasts with producer currency pricing, in which prices are set in the producer’s currency, regardless of the destination). We will assume that exporters in the foreign country (the United States) set prices in dollars. Thus, our pricing assumptions are consistent with the dominant currency paradigm (Gopinath et al. 2020), in which export prices in any country are set in a “dominant” currency (the dollar in our case). These assumptions are motivated by evidence suggesting that a large fraction of international trade is invoiced in a small number of dominant currencies, with the U.S. dollar playing an outsized role (see, e.g., Goldberg and Tille 2008; Gopinath et al. 2020).

Let $P¯D,t$ denote the domestic price chosen by producer j if it can optimally reset its price in period t. (This optimal price is the same for any j, so we omit the j subindex). Producer j sets $P¯D,t$ to maximize the present value of profits generated while that price remains effective, subject to demand by domestic consumers and capital goods producers (Equations (10) and (39), respectively). The reset price $P¯D,t$ satisfies the standard optimality condition

Given that fraction ξ_p of producers set price $P¯D,t$⁠, with the remaining producers leaving their price unchanged, the domestic price index (11) satisfies

Turning to home’s export price, let $P¯M,t*$ be the price set by producer j in the foreign market (denominated in the foreign currency, i.e., in dollars) in case the producer can reset its price in period t, and let $PM,t*=(∫01PM,j,t*1−ϵdj)11−ϵ$ denote home’s export price index. The price $P¯M,t*$ is chosen to maximize the present value of profits generated from foreign sales. The resultant optimality condition is

where $Et$ is the nominal exchange rate (i.e., the price of the EM’s currency in dollars). The evolution of home’s export price index is given by

The fact that home producers set export prices in dollars implies a low pass-through of exchange rate changes into home’s export prices. To see this, suppose that prices are very rigid (ξ_p is close to one), and consider a depreciation of the home currency against the dollar (a decrease in $Et$⁠). Under our pricing assumption, the home’s export price index in dollars, given in (47), will remain unchanged despite the lower $Et$⁠. This stands in contrast to what would occur under producer currency pricing, in which case the (dollar) export price index would satisfy $PM,t*pcp=PD,tEt$⁠. Thus, with a near-fixed $PD,t$⁠, home’s export price index $PM,t*pcp$ would decline one-for-one with a depreciation of the home currency.

The relation between the real exchange rate $St$⁠, introduced earlier, and the nominal exchange rate $Et$ is

whereby the real exchange rate equals the ratio of the home to foreign CPI.

1.4 Central bank

As a baseline, we assume the central bank in the EM sets the short-term nominal rate $Rt+1n$ according to the Taylor rule that targets only domestic inflation:

We will later consider alternative policy rules. The rule (49) does not include an output gap term. Our motivation for that assumption is the evidence in Kaminsky, Reinhart, and Végh (2004) indicating that monetary policy in emerging market economies generally does not feature strong countercyclicality, unlike in advanced economies. We also allow for a domestic monetary disturbance $εr,t$⁠, which is assumed to follow an exogenous first-order autoregressive process with persistence parameter ρ_m.

1.5 The foreign economy

The foreign country is analogous to home, with two differences: there are no financial frictions, and exporting firms practice producer currency pricing. The absence of financial frictions implies that the foreign household can be viewed as directly holding physical capital. Foreign households can also deposit dollar funds directly in EM banks, subject to the friction described above. Aside from these differences, foreign agents’ decision problems are similar to those just described, so for the sake of brevity we report here the resultant equilibrium conditions.

A measure $N*$ of households live in the foreign economy. The foreign household’s Euler equations for nominal dollar bonds, real dollar bonds deposited in EM banks, and capital are the following:

Optimization by U.S. household also requires a standard set of transversality conditions. Labor supply satisfies

The foreign household’s demands for foreign-produced and imported bundles satisfy, respectively,

The foreign CPI is

The conditions for investment, the evolution capital, and goods demand by capital producers are

Foreign nominal marginal cost, the capital-labor ratio, and aggregate output satisfy

Foreign firms set prices in the foreign currency and let the export price adjust with changes in the value of the currency (i.e., satisfy the law of one price). The optimal reset price $P¯D,t*$ satisfies

and the producer price index obeys

The law of one price for the foreign-produced basket implies

Finally, the foreign central bank follows a Taylor rule that targets producer inflation and deviations of output from its steady-state level $Y*$⁠, and includes a foreign monetary policy shock $εr,t*$⁠:

The shock $εr,t*$ follows an AR(1) with persistence ρ_m.

1.6 Market clearing and equilibrium

Market-clearing aggregate assets held by home bankers, A_t, implies that the latter must equal the aggregate value of capital in period t (including period-t investment):

The home good’s market clearing condition is

whereby aggregate home output is either consumed domestically or exported. The term $N*N$ reflects the foreign population size is $N*$⁠, while the home population size is $N$ (note that all variables are expressed on a per-household basis).

A similar market-clearing condition holds for foreign aggregate output:

Finally, the balance of payments, obtained by aggregating the budget constraints of home households, bankers, and firms, is given by

This condition states that the EM’s net accumulation of foreign liabilities (the left-hand side), expressed in terms of the foreign basket (i.e., in real dollars), equals the value of the EM’s imports minus the value of its exports (the right-hand side), also expressed in terms of the foreign basket (and adjusted by relative population sizes).

An equilibrium consists of thirteen home aggregate quantities ${Yt,Ct,Lt,CD,t,MC,t,At,Dt*,Nt,It,Kt+1,ID,t,MI,t,MCt}$⁠, nine home prices ${Wt,Zt,Pt,PD,t,P¯D,t,PM,t,Qt,Rt+1n,Rt+1}$⁠, five banker coefficients ${μt,ϱt,νt,xt,ϕt}$⁠, 10 foreign quantities ${Yt*,Ct*,Lt*,CD,t*,MC,t*,It*,Kt+1*,ID,t*,MI,t*,MCt*}$⁠, nine foreign prices ${Wt*,Zt*,Pt*,PD,t*,PM,t*,P¯D,t*,P¯M,t*,Qt*,Rt+1n*}$⁠, and the nominal and real exchange rates ${Et,St}$⁠, satisfying the 48 Equations (4)-(9), (24)-(32), (34)-(35), (37)-(38), and (41)-(71).

1.7 UIP premiums and credit spreads

There is a violation of UIP in this economy. In this section we highlight its fundamental source. UIP fails because the agency friction limits intermediaries’ ability to arbitrage between riskless debt denominated in different currencies. The model makes a stark prediction on the size of the UIP violation: UIP premiums are larger whenever the domestic credit spread (the domestic expected return on capital net of the risk-free rate) is elevated.

A number of simplifications of the model help convey the basic intuition. Thus, for the remainder of the section we make use of the following assumption.

  

The intuition underlying condition (72) is straightforward. Suppose the intermediary marginally reduces its domestic borrowing, and marginally increases its foreign borrowing. The benefit of this operation is given by the left-hand side of (72) multiplied by β, which is the funding cost of domestic loans minus the (expected ex post) cost of domestic loans. The cost of the operation is that marginally raising $xi,t$ tightens the leverage constraint by γ, implying a foregone excess return of $γμt=γβEt[RK,t+1−Rt+1]$⁠. If the banker’s portfolio is optimal in the first place, the benefit of the operation must equal its cost.

Let $z≡ log(Z)$ for any variable Z, and assume that in steady state $RK≈R≈R*$ (i.e., steady-state differences between gross returns are small). Then to a first order, Equation (72) is

which can be solved forward to yield

The real exchange rate depends positively on the expected sum of short-term real interest rate differentials, as in standard macro models; but different from them, as long as $γ>0$ it also depends inversely on the infinite sum of expected excess returns on capital (or credit spreads). The presence of this additional, nonconventional term is an important feature of economic transmission in this economy, as we illustrate in Section 2. In Section 3, we test condition (74) in the data.¹²

1.8 Effects of reductions in intermediary net worth

Fluctuations in aggregate intermediary net worth N_t play a key role in shaping model dynamics, and are key to the model’s implications for the cross-border effects of U.S. monetary policy. In this section we make several simplifying assumptions that yield a special case in which the dynamic effects of exogenous shocks to net worth can be derived analytically up to a first-order approximation. To be clear, in the complete model N_t is fully endogenous (including to shifts in U.S. monetary policy), but considering an exogenous-N economy yields useful insight into the model’s transmission mechanism.

To that end, we make the following additional simplifying assumption.

    

To gain intuition, consider first a simple case in which the parameter composite $ε≃0$ (which occurs when w is close to θ). From (81) this means that the effect on the foreign funding ratio, $x^t$⁠, on the premium $μ˜t$ is negligible. The coefficients in (87) and (86) simplify to $Ψw≃γθ1−ρw$ and $Ψd≃0$⁠. A decrease in $w^t$ lowers the exchange rate $s^t$ with elasticity $γθ1−ρw$ on impact, and lowers the price of capital $q^t$ with elasticity $θ1−ρw$⁠. This occurs because lower net worth must be matched by a higher domestic premium, due to bankers’ binding incentive constraint. This is accommodated by a lower price of capital today. In turn, a higher domestic premium is associated with a larger UIP premium on the local currency, triggering a depreciation. The stock of dollar debt $d^t*$ declines permanently.

When $ε>0$⁠, a second effect kicks in. Now, everything else equal, lower $d^t*$ implies a lower premium $μ˜t$⁠: a reduced stock of foreign funds eases the financing friction. (This effect explains why $d^t−1*$ enters (86) with a negative sign). Because of this effect, in the wake of a currency depreciation and the associated lower $d^t$⁠, a second force, that grows over time, exerts some upward pressure on the currency. This makes $d^t$ mean-revert instead of experiencing a permanent drop.¹³

Overall, despite its minimal nature due to the stark simplifications made in this section, our model’s predictions on the effects of a decrease in net worth are well-aligned with the phenomena observed in open economies experiencing financial stress. Thus, in the wake of a negative shift in intermediaries’ net worth, the model endogenously generates an increase in the domestic credit premium, a drop in domestic asset prices, a wider UIP premium on the domestic currency vis-á-vis the dollar, a depreciation of the local currency, and an outflow of dollar funding.

2 Quantitative Analysis

We now turn to a quantitative analysis of the model. We focus on the model’s predictions of the effects of U.S. monetary policy shocks, $εr,t$⁠, on the EM. We now drop the simplifying assumptions made in Sections 1.7 and 1.8, and instead study a calibrated version of the complete model as laid out in Sections 1.1 through 1.6.

2.1 Functional forms and parameter values

We assume the following functional forms for the period utility function U and for the enforcement friction Θ:

U(C, L) is a standard utility function. The function $Θ(x)$ has features similar to the linear case studied earlier, but has the advantage of implying an interior solution for the portfolio variable x. In steady state, the ratio x is positively related to the difference between discount factors β and $β*$ and inversely related to the parameter γ. This allows us to use data on foreign-currency-denominated debt shares in EMs as a way to discipline the parameter γ.

The U.S. openness parameter, $ω*$⁠, is assumed to be arbitrarily small: $ω*→0$⁠. We also assume that the U.S. is arbitrarily large relative to the EM, $N*/N→∞$⁠, and we let $ω*N*/N→ω$⁠, implying that trade is balanced in a steady state with unit relative prices. The rationale for these assumptions is that an EM is very small in size relative to the United States, and which therefore has negligible weight in U.S. consumption and investment baskets. As we will see, this implies that there are no “spillbacks” from the EM onto the U.S. economy. Put differently, the U.S. behaves effectively like a closed economy. This makes the model analysis simpler and more transparent.

Table 1 shows the remaining parameter values. We set the U.S. discount factor $β*$ to 0.9950, implying a steady-state real interest rate of 2% per year. This choice is consistent with several recent studies (e.g., Reifschneider 2016) and is motivated by estimates of the U.S. natural rate (e.g., Holston, Laubach, and Williams 2017). To calibrate the EM’s discount factor, we rely on estimates of Mexico’s long-run natural rate from Carrillo et al. (2017) of about 3%, and accordingly calibrate β to 0.9925.¹⁴ The preference and production parameters $χ,ω,η,ξp,θp,α,δ$ are set to conventional values. The elasticty of Tobin’s q to the investment-capital ratio is set to 0.25, following Bernanke, Gertler, and Gilchrist (1999). The Taylor rule parameters $γπ,γy$ are set to standard values. The monetary shock’s autoregressive parameter ρ_m is 0.5, as in textbook analyses (e.g., Gali 2015), consistent with a moderate amount of persistence.

Turning to the parameters related to the financial market friction, we set the survival rate σ_b to 0.95, implying an expected horizon of bankers of 6 years. This value is around the midpoint of the range found in related work.¹⁵ We set the remaining three parameters to hit three steady-state targets: a credit spread of 200 basis points, a leverage ratio of 5, and a ratio of foreign-currency debt to domestic debt (⁠$D*/SD$⁠) of 30%. The target for the credit spread reflects the average value of 5-year BBB corporate bond spreads in major Asian and Latin American emerging market economies over the period 1999–2017 (excluding the global financial crisis period). The target leverage ratio is a rough average across different sectors. Leverage ratios in the banking sector are typically greater than five,¹⁶ but the nonfinancial corporate sector generally has lower asset-equity ratios (between two and three in emerging markets).¹⁷ Our target of five reflects a compromise between these two values. Finally, evidence in Hahm, Shin, and Shin (2013) on ratios of foreign-currency deposits to domestic deposits in EMs suggests an average of about 30%. This value is also consistent with evidence presented in Chui, Shin, and Shin (2016), showing that average private-sector foreign currency debt across EMs (for the period 2006-2014) as a percent of total (i.e. domestic- plus foreign-currency denominated) debt is a little over 20%. These targets imply $θ=0.41,ξb=0.067$⁠, and $γ=2.58$⁠. The implied value for the steady-state ratio of foreign liabilities to assets is x = 0.18 (note that x follows from our targets for $ϕ$ and $D*/SD$⁠, via the balance sheet identity (15)).

2.2 Cross-border spillovers of monetary policy

Figure 1 shows the effects of an unanticipated increase in $εr,t*$ of 25 basis points. This shock would imply, in the absence of any endogenous response of U.S. inflation or the output gap, an increase of 100 basis points in the annualized U.S. nominal interest rate. The figure shows the effects in our baseline model with financial market frictions (blue line with circles) and also includes the effects in a “standard New Keynesian” model: an economy with complete financial markets and producer currency pricing (i.e., a two-country version of Gali and Monacelli 2016, augmented with endogenous investment).

Our baseline model predicts large effects on EM GDP of the U.S. monetary disturbance. Aggregate activity in the EM falls by 0.25% on impact, a decline of about two-thirds as a large as the drop in U.S. GDP itself. One quarter after the shock, the reductions in EM and in U.S. GDP are roughly the same size. An important reason for the sizable effect on EM GDP is a sharp reduction in investment spending, which falls by nearly 1.5%, which is even more than in the United States, and about six times as much in the standard New Keynesian model.

The presence of strong feedback effects due to the financial market friction is key to the large decline in EM investment following the U.S. rate hike. Two types of feedback effects are at work. On the one hand, the standard “financial accelerator” (Bernanke, Gertler, and Gilchrist 1999, BGG henceforth) is present: lower intermediary net worth weakens investment spending and Tobin’s q, which work to depress net worth further. What our model adds is a “second-round” feedback effect operating through the exchange rate and intermediaries’ dollar-denominated debt. As shown in Section 1.8, lower intermediary net worth is associated with a wider UIP premium and, accordingly, with a depreciated exchange rate. Given a lower exchange rate, the dollar-denominated debt in banks’ balance sheets constitutes a greater burden, which affects net worth negatively and starts another round of feedback. Put differently, in our model there is feedback between N_t and Q_t, for a given $St$ (as in BGG); but there is also feedback between N_t and $St$⁠, for a given Q_t. In general equilibrium, fluctuations in all three variables influence and reinforce each other. These three-way feedback effects between N_t, Q_t, and $St$ are responsible for the roughly six-fold amplification of the investment decline compared to a case in which the credit frictions are absent.

The presence of intermediary frictions also adds persistence to real variables: investment in the EM recovers more slowly than it does in the United States, and more slowly than the shock itself (which has persistence $ρm=0.5$⁠). The reason is that rebuilding net worth takes time, as seen in the first panel. The response of investment inherits the sluggishness in the recovery of net worth. Finally, our model also implies a sizable real exchange rate depreciation in the wake of the shock—twice as large as in the standard NK model—once again driven by the powerful feedback effects between net worth and the exchange rate.

In stark contrast, the effects on EM activity of the U.S. monetary shock in the standard New Keynesian model are overall quite modest. An important reason behind this result is the presence of expenditure-reducing and expenditure-switching effects that affect EM activity in opposite directions. On the one hand, lower U.S. aggregate demand is associated with lower demand for EM goods, working to depress EM GDP. On the other hand, the EM’s exchange rate depreciates when the U.S. tightens, making EM goods cheaper and leading households and firms in both countries to switch expenditure toward EM-produced goods, boosting EM GDP. The net result is a very modest decline in EM activity. In addition the standard NK model cannot, by construction, generate any effects of U.S. monetary policy on local credit spreads or UIP premiums.¹⁸

Our model departs from the above in two main ways. First, the financial constraints-driven three-way feedback effects work to greatly enhance the effect on domestic absorption, via amplified fluctuations in investment spending. Second, the presence of dominant currency pricing dampens expenditure-switching effects. These two channels together imply a sizable hit to EM activity from a tightening of U.S. monetary policy, consistent with much evidence.

To clarify the role of intermediary balance sheets, in Figure 2 we report the effects of a one-time redistribution of wealth between bankers and households, sized to induce the same effect on net worth as in the previous experiment.¹⁹ This shock has no effects on the standard NK model, as it is just a redistribution of resources within the representative household. With capital market frictions, in contrast, lower net worth feeds into higher credit and UIP premiums, lower asset prices, and a depreciated home exchange rate, all of which are qualitatively illustrated in the simplified setting of Section 1.8. There are two key differences between that simpler setting and the complete model studied here. First, given that bankers are long-lived, the evolution of net worth is now endogenous. This opens the door to the feedback effects explained above, whereby aggregate net worth affects the price of capital and the exchange rate, but is also affected by them. Second, with endogenous production the decline in Tobin’s q triggers lower investment demand and, given nominal rigidities, lower aggregate GDP.

While an analysis of optimal monetary policy is beyond the scope of this paper, the previous discussion offers some insights into how the credit frictions complicate the task of the EM central bank. Avoiding the kind of financial tightening effects shown in Figure 2 would require stabilizing intermediaries’ net worth N_t. But this is likely not possible with just one policy instrument (the nominal rate $Rtn$⁠), because N_t depends on both the domestic asset price (Q_t) and the exchange rate (⁠$St$⁠), and moderating the effects of $εr,t*$ on these two variables requires adjusting the nominal rate in opposite directions—that is, leaning against the fall in Q_t calls for lowering $Rtn$⁠, but fighting the depreciation calls for increasing it. Thus, it will likely not be feasible for the EM central bank to completely insulate intermediaries’ net worth from U.S. monetary shocks.²⁰

2.3 Domestic monetary policy and the UIP puzzle

Here, we turn to the implications of our mechanism for the transmission of domestic monetary policy. We begin by considering a domestic monetary policy shock and highlight how the effects of this shock differ in our model compared to the standard NK setup.

Figure 3 reports the effects of a 25-basis-point rise in $εr,t$⁠. Two observations stand out. First, the exchange rate dynamics following the domestic rate hike are very different from those predicted by the standard NK model. In particular, the response of both the nominal and real exchange rates is muted relative to the effect implied by the NK model. The reason is that tighter domestic policy, by depressing domestic asset prices and slowing the economy, creates a drag on intermediary balance sheets, as shown in the first panel. The associated rise in the currency premium, which is greatest in the initial periods, is responsible for the milder appreciation of the domestic currency upon impact. This same mechanism implies the absence of exchange rate overshooting in our model, consistent with recent evidence, and different from the NK model.²¹

Second, the effects on activity of the domestic monetary contraction are also larger in our setting with credit frictions: the ensuing output contraction is uniformly larger. This occurs because the decline in net worth, and the associated rise in the domestic credit spread, induce a grater slowdown in investment spending, explaining an overall larger output downturn. We highlight, however, that the degree of amplification of the foreign monetary shock is much larger than that of the domestic shock: the output decline is almost four times larger in the latter case, and only about a third larger in the former. This finding is consistent with the conventional view EMs are particularly vulnerable to external shocks.

The “dampening” of the exchange rate response to domestic monetary policy due to the credit friction has two interesting corollaries. The first is that an exchange rate management policy is difficult to maintain in the presence of credit frictions. This is seen most clearly by considering an exchange rate peg. Absent the credit friction, UIP holds. Therefore, maintaining a peg calls for raising the domestic policy rate one-for-one with any increase in the foreign rate. In contrast, with the credit friction, raising the domestic rate strains the balance sheets of intermediaries, limiting their ability to arbitrage and thus raising the UIP premium, with a resultant lower effect on the currency than if the UIP premium remained constant. As a result, the domestic policy rate needs to be adjusted more than one-for-one with the foreign rate if the central bank wishes to keep the exchange rate constant.

We illustrate this result in Figure 4. The figure shows the effects of a U.S. monetary shock under a nominal exchange rate peg, in the standard NK model (green dash-dotted line) and in our model with credit frictions (blue circled line). The NK model features an increase in the domestic nominal rate of the same size as the foreign rate (about 25 basis points), inducing a modest output contraction. In our baseline model, the upward adjustment in the domestic nominal rate is much larger, and the output drop is accordingly more severe.

The second corollary is that our model can account for the uncovered interest parity puzzle (Engel 2014). Consider the third row of Figure 3. In the NK model, UIP holds, and so the nominal interest rate differential between the EM and the U.S. following the rise in domestic rates exactly matches the expected nominal depreciation of the EM currency. This is not the case in our model: in the short run, the higher domestic nominal rate is now associated with an expected appreciation—resulting from the fact that higher nominal rates impair intermediary balance sheets, widening the UIP premium.

To illustrate the ability of the model to account for the UIP puzzle, we follow equation (6) in Engel (2014) and run the regression

on model-simulated data, with the two monetary disturbances (domestic and foreign) as driving forces. Table 2 shows the results, for different ratios of the standard deviation of domestic monetary shocks (σ) to the standard deviation of foreign monetary shocks (⁠$σ*$⁠). In the NK model, the coefficients are always a = 0 and b = 1, as expected. In our model, as long as domestic monetary shocks are large enough (i.e., more important than foreign ones), the coefficient b is below unity, and is inversely linked with the relative shock size $σ/σ*$⁠.²² A large $σ/σ*$ seems plausible for EMs, where the volatility of nominal short-term rates is much higher than in the United States.

3 Evidence on UIP Regressions

Our theory has an important testable prediction: unlike conventional open economy macroeconomic models, such as Gali and Monacelli (2005) and subsequent literature, our model features endogenous deviations from UIP, with the currency premium moving in tandem with the domestic credit spread. In this section, we test this model prediction in the data by estimating versions of the forward-looking exchange rate Equation (74), as often done in the empirical literature on the determinants of exchange rates.²³ We start with examining the connection between the domestic credit spread and the real exchange rate in a panel-data setting where we pool data from several emerging economies. We find strong support for the mechanism proposed in our model. Next, we explore the cross-country heterogeneity of our sample. We show that the domestic credit spread is more strongly associated with the real exchange rate in emerging economies that have more fragile domestic macroeconomic conditions, consistent with the predictions of our model.

3.1 Empirical UIP regression equation

Starting from (74), we let $xt≡∑j=1TEt{rk,t+j−rt+j}$ and $rtdiff≡∑j=1TEt{rt+j−rt+j*}$⁠, and taking first differences we can write

where $ϵt≡Et[st+T+1]−Et−1[st+T]$⁠, that is, the forecast as of this period of the real exchange rate T + 1 periods ahead minus the forecast as of the previous period of the exchange rate T periods ahead. We assume that if T is large enough, both forecasts are approximately equal, and therefore $ϵt≈0$⁠.²⁴

Equation (91) forms the basis for our UIP regression analysis. Our baseline estimation uses monthly data from several emerging economies. The number of countries included in the regressions is mostly governed by data availability, as we explain in detail below. We measure s_t by the (log) bilateral real exchange rate against the dollar, calculated by multiplying the nominal exchange rate by the ratio of the local to the U.S. CPI.

To approximate x_t, we use two different measures of spreads, defined as the difference between the rate at which corporations in emerging economies borrow and the rate on government bonds of the same maturity and the same currency. The first measure is yields on local currency corporate bonds minus yields on domestic government bonds of the same maturity. The resultant corporate borrowing spread is a widely used proxy for the “external finance premium” (Bernanke, Gertler, and Gilchrist 1999) arising due to the presence of financial market frictions.²⁵ Thus, we measure x_t as

where $rtcorp$ is the local currency corporate bond yield (in annual terms), $rtgov$ is the yield on long-term domestic government bonds, and T = 60 months.

The second measure we use to approximate x_t is dollar-denominated corporate bond spreads. This measure has greatest coverage in terms of number of countries. Specifically, we measure x_t by the spread between 5-year dollar-denominated corporate bonds and U.S. Treasury bonds of the same maturity.²⁶²⁷

Finally, we construct $rtdiff$ as

where $rtgov*$ is the (real) long run U.S. government yield. In (93), real yields are constructed by subtracting from nominal yields the expected inflation rate in each month, calculated as the average inflation rate over the past year.²⁸ These calculations make the simplifying assumption that the expected sum of one-period yields differentials in (91) are well approximated by the T-month maturity bond yields.²⁹

The final panel regression equation we arrive at after incorporating all these definitions is

where j denotes countries, t is months, and η_k denotes country fixed effects. We also include changes in the VIX index to control for global risk aversion.³⁰

3.2 UIP premium: Baseline panel regressions

We estimate (94) by pooling monthly data from several emerging economies.³¹ The panel is unbalanced, with most countries starting in the 2000s, and runs until the second month of 2022. Comparing our theoretical and empirical UIP equations, (74) and (94), one would expect $βx=−γ<0, βr=1$⁠, and $βv<0$⁠.

Table 3 reports the panel regression results. Columns 1–3 display the estimated coefficients when the credit premium is proxied for by credit spreads on dollar-denominated corporate bonds. Columns 4–6 show the corresponding results when the premium is measured using local-currency corporate bond spreads. Note that we include country fixed effects in all the regressions, and compute t-statistics using Driscoll and Kraay (1998) standard errors, the panel data analog of Newey and West (1987) time-series standard errors.

Column 2 shows our baseline specification with both the interest differential and the corporate bond spread. The coefficient of interest is β_x, and a negative value on this coefficient indicates a negative association between the EM’s real exchange rate and credit premium, as predicted in our model. As shown in column 2, the coefficient for the spread is negative, as expected, and is highly statistically significant. Moreover, the presence of the spread improves the equation fit considerably: R² rises from nearly zero to around 0.25. Column 3 performs a robustness check by including the VIX. The spread continues to be significant even when this variable is included. Column 4 reports the corresponding results for local-currency bonds. The finding that the corporate spread is highly significant reemerges here, as does the fact that the presence of the spread adds considerable explanatory power relative to a regression with the interest differential only. The only meaningful difference is that the magnitudes of β_x for local-denominated bonds are larger in absolute value than that of dollar-denominated bonds. The VIX has a negative sign, as before, but is now significant. Nonetheless, the spread continues to be significant.

3.3 UIP premium: Cross-country heterogeneity

Our theory implies the coefficient for the spread in the regression is γ, the parameter governing the degree of frictions in cross-border borrowing. Thus, the model implies that in countries with larger γ, one should observe a larger (in absolute value) coefficient for the spread in the regression. In this section we explore the panel dimension of our data set to test this prediction.

The value of γ, however, is not directly observable. As an indirect proxy, we use the vulnerability index constructed by Ahmed, Coulibaly, and Zlate (2017). This index is based on six macroeconomic variables reflecting the overall strength of macroeconomic fundamentals. The reasons we choose to rely on this index are as follows. First, our model implies that a larger γ leads to greater vulnerability to external shocks, which should manifest itself in worse macroeconomic fundamentals. Second, Ahmed, Coulibaly, and Zlate (2017) show that the index does an excellent job at predicting which countries suffered greater pullback from investors during periods of global stress (such as the taper tantrum of 2013). As such, the index captures well how wary investors feel about lending to a particular EM, consistent with the interpretation of our parameter γ. Finally, our analysis requires an indicator that is consistent across countries in the sample, a requirement that this index satisfies.³²

We proceed as follows. We first group EMs in our sample as either Vulnerable or Nonvulnerable EMs. We classify Vulnerable (Nonvulnerable) EMs as those countries whose vulnerability index is higher (lower) than the sample median.³³ We then rerun the baseline empirical UIP regression, Equation (94), separately for these two groups of countries.

Table 4 shows the estimation results for vulnerable and nonvulnerable EMs. Our theory predicts that the coefficient β_x should be larger in absolute value for the Vulnerable EMs. Columns 1–3 display the estimated coefficients, β_r, β_x and $βν$⁠, for the group of vulnerable emerging economies. Columns 4–6 show the corresponding results for countries in the nonvulnerable group. We also estimate the baseline UIP regression for a group of advanced countries, shown in columns 7–9, which should have an even lower γ than the less-vulnerable EMs. The spread variable in all these regressions corresponds to the dollar-denominated corporate bond spread. As before, we include country fixed effects in all the regressions, and compute t-statistics using Driscoll and Kraay (1998) standard errors.

Three key results are worth emphasizing from the table. First, the coefficient of interest, β_x, is negative for all three group of countries. But importantly, the magnitude of the coefficient is largest in the vulnerable EMs (the estimated β_x is –0.65 in the vulnerable group versus -0.33 in the less-vulnerable group). Second, the spread adds meaningful explanatory power only for the vulnerable emerging economies: R² rises quite significantly in the vulnerable group, from around zero to 30%, while in the nonvulnerable group the increase in explanatory power is about a third as large. Last, but not least, the credit spread plays a minor role in explaining the real exchange rate in advanced economies. As shown in column 9, the coefficient β_x is barely significant, and R² increases only 1.5 percentage points from 3% to 4.5% when credit spreads are included in the regressions (comparing columns 7 and 8). We take this finding as a suggestive evidence that countries with weaker fundamentals (proxying for higher γ) see a much tighter association between the UIP premium and the domestic credit premium, consistent with the model.

We complement these panel regression results by estimating the empirical UIP specification in Equation (91) country by country. We then exploit the cross-country heterogeneity by examining whether the coefficient β_x lines up cross-sectionally with the vulnerability measure.

Figure 5 presents our results. The y-axis in the figure shows the estimated coefficient for the first-differenced credit spread, β_x, for each country in our sample. The estimated coefficients for all countries in our sample are negative, as expected. Moreover, for the majority of countries the estimated coefficient is statistically different than zero (as indicated by the blue color in the figure). The x-axis displays the corresponding vulnerability ranking of the country, as computed in Ahmed, Coulibaly, and Zlate (2017). The dark-gray solid line represents the fitted regression for the relationship between the estimated coefficient value and the average vulnerability ranking. This fitted-regression line is negatively sloped, indicating that in more-vulnerable countries the coefficient β_x is more negative, consistent with the predictions of our theory. Note that countries in the fragile-five group appear at the bottom-right of the figure, as expected, supporting the prediction that the relation between exchange rates and credit spreads is especially tight in these countries.³⁴

3.4 UIP premium: Robustness

This section runs two robustness checks of the baseline results presented in Table 3. The first estimates the UIP regressions in levels instead of in first differences. The second uses bank lending-deposit spreads data to proxy the extent of domestic financial market imperfections, instead of corporate bond spreads.³⁵

Turning to our first robustness check, we start from (74), impose the definition of x_t and $rtdiff$ as before, and assume that $st=ft+s^t$⁠, where f_t is a deterministic time trend and $s^t$ is the long-run real exchange rate. We assume the real exchange rate is stationary around the trend, so that if T is large enough $Et{s^t+T+1}≈0$⁠.³⁶ The equation then becomes

Table 5 shows our panel estimation results for the level specification. Columns 1–3 display the estimated coefficients when domestic market frictions are proxied for by credit spreads on dollar-denominated corporate bonds, and columns 4–6 show the corresponding results when the degree of financial frictions is measured using local-currency corporate bond spreads. We include country fixed effects and a linear or quadratic trend for each country in all the regressions. Focusing on columns 3 and 6, we see that our baseline results remain unchanged: the coefficient for the spread x_t remains negative, as expected, and highly significant, especially for the regressions using local-currency spreads.

The second robustness test is that we use bank lending spread data to proxy for the tightness of domestic financial constraints. More specifically, we rerun the regression shown in Equations (94) and (95) (i.e., the first difference and the level specifications, respectively) using the bank lending-deposit spreads to measure x_t. One reason that the bank lending-deposit spreads also may be a good proxy for the tightness of domestic financial market frictions is that corporations in many of these economies quite heavily depend on direct bank lending in addition to debt issuance to obtain financing. However, it is important to note that an advantage of corporate bond spreads (a market-determined measure) relative to bank lending rates is that bank loan contracts likely contain nonprice terms that are difficult to quantify and that may also capture the kind of impediments to external finance that the theory emphasizes (an argument emphasized by Gertler and Lown 1999).

Our results for both the first-differences and the level specifications for the real exchange rate regressions are shown in Tables 6a and 6b, respectively. The main results remain robust to using the alternative measure of financial markets imperfections, as the estimated coefficient for spread for both the first-differences and the level specification remain negative and significant.

4 Effects of U.S. Monetary Shock: VAR versus Model

Our model’s predictions for the cross-border spillovers from a U.S. monetary policy shock are consistent with those implied by VAR-based estimates. To show this, we augment the VAR model of Christiano, Trabandt, and Walentin (2010) to include quarterly aggregate GDP, investment, and the real exchange rate from a set of emerging economies. We focus on EMs that are not on fixed exchange rate regimes, consistent with our model.³⁷ More specifically, we start from the original specification of Christiano, Trabandt, and Walentin (2010) that includes U.S. GDP, inflation, unemployment, capacity utilization, consumption, investment, and the federal funds rate. We then estimate their model after incorporating data for EM GDP, investment, and the exchange rate for the period 1978:I–2008:IV. We use a Cholesky shock identification which relies on ordering the federal funds in the next-to-last position, with the last variable being the EM exchange rate. This ordering embeds the standard assumption that the only real variable that the U.S. monetary policy shock affects contemporaneously is the federal funds rate, while allowing the U.S. monetary shock to affect the EM exchange rate contemporaneously.

Figure 6 compares the VAR-implied effects of a monetary shock that raises the fed funds rate by 1 percentage point with those predicted by an extended version of our model. The extended model augments the setup of Section 1 with nominal wage stickiness, habit persistence in consumption, and adjustment costs in trade flows, among other features. These features help produce more empirically realistic dynamics, as the literature has widely emphasized (e.g., Christiano, Eichenbaum, and Evans 2005).³⁸ The dashed lines in the figure represent the effects predicted by a “standard” DSGE model, which includes the real frictions just mentioned, but assumes frictionless financial markets and producer currency pricing.

Starting with the United States, the model captures the dynamic response of U.S. output and investment very well. The shock induces output to fall around 0.50% at the trough, very close in magnitude to the decline implied by our model. U.S. GDP displays a slow and hump-shaped response to a shock, peaking a little over one year after the monetary shock hits, which the model also captures quite well. Lastly, while the VAR-implied effect of the monetary shock on the fed funds rate is roughly gone after a year, the U.S. economy continues to respond well after that. The model predicts that U.S. GDP remains below its path absent the shock for about 3 years, consistent with the evidence, even though the effect of the shock on the interest rate dies out after a year and a half. The model-implied response of U.S. investment over the first two to 3 years after the shock is closely aligned to the VAR-implied effects as well. Overall, this suggests that our model’s internal propagation channels allow it to capture the dynamic effects of a monetary policy shock.

Next, we compare our model’s predictions for the effects of a U.S. monetary policy shock on the EM with the VAR-implied effects. The VAR implies an effect on EM GDP at a horizon of 1 to 2 years that is comparable in size to the effects on U.S. GDP itself (a decline of about 0.5%), a result that the model replicates quite well. Thereafter, the VAR implies a large degree of persistence in EM GDP, larger than in the model, though the confidence bands are quite wide. Overall, the model replicates the VAR-implied response reasonably well, with the model-predicted GDP path mostly inside the VAR-implied confidence bands.

Turning to EM investment, here again we find support in the data for a key channel in the model, working through investment spending. The magnitude of the EM investment decline in the data is larger than the decline in U.S. investment, consistent with our model. As with GDP though, there are some differences in the dynamic pattern of the effects, with the model-implied response more front-loaded than the VAR-implied one.

Finally, the large near-term response of the EM exchange rate implied by our model also finds support in the data, with the magnitudes of the depreciation in the first few quarters being broadly comparable between model and data. Thereafter, however, the predictions differ: the model implies a relatively quick return of the real exchange rate back to its no-shock path, while the VAR responses feature a much more persistent decline.

Overall, we conclude that the model’s predictions on the spillover effects of a U.S. monetary policy shock on EM activity, investment, and the exchange rate are broadly consistent with the VAR-implied ones (with the largest discrepancy being the high persistence in the empirical real exchange rate response). We highlight that this is not the case in the standard DSGE model, which predicts effects on EM output, investment, and the exchange rate that are very far from their empirical counterparts. We have also found that the model-implied effects on the EM UIP premium of the U.S. monetary shock are consistent with the estimates by Kalemli-Özcan (2019), who use estimates from Kalemli-Özcan and Varela (2021) of UIP premiums, which rely on survey data of exchange rate expectations. See the Internet Appendix for a comparison.

5 Conclusion

In this paper we develop a two-country New Keynesian model with imperfect domestic and international financial markets to study the cross-border spillovers from U.S. monetary policy. The model features strong financial amplification due to the powerful interaction between internal and external feedback effects. Consistent with the estimates we obtain from a VAR model, this mechanism leads to large spillovers from U.S. monetary shocks to EMs. We believe our model is better tailored than existing macroeconomic models to some of the specific features of EMs, which are often seen as being particularly vulnerable to volatile capital flows and other external pressures.

Despite strong amplification working in part through exchange rate volatility, the model calls into question the common view that monetary policy should be used to mitigate exchange rate fluctuations. The reason is that the endogenous currency premium partly offsets the conventional effect of a change in the domestic policy rate on the exchange rate. The resultant “disconnect” between the exchange rate and the domestic policy rate implies that much larger domestic macroeconomic volatility is necessary for a given reduction in exchange rate instability.

Looking forward, it would be useful to employ a version of our model to consider optimal policy and how it can be implemented in the context of interest rate policy and foreign exchange market inventions on the part of EM central banks. Given the endogenous deviation from UIP in the model, there may be a role for interventions in foreign exchange over and above conventional interest rate policy. This extension is left for future research.

Acknowledgement

We thank Chris Erceg for many fruitful discussions that inspired much of this work, as well as Gianluca Benigno, Jordi Galí, Sebnem Kalemli-Ozcan, and Paolo Pesenti for very useful comments. Special thanks to our discussants Giancarlo Corsetti, Tommaso Monacelli, Jenny Tang, Luca Fornaro, Emine Boz, Ambrogio Cesa-Bianchi, Oleg Itskhoki, and Stephanie Schmitt-Grohé for very helpful suggestions. We also thank seminar participants at various institutions for their comments. Mike McHenry, Serra Pelin, and Mikael Scaramucci provided outstanding research assistance. The views expressed in this paper are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York, the Board of Governors of the Federal Reserve, or the Federal Reserve System. Supplementary data can be found on The Review of Financial Studies web site.

Footnotes

References