Central Banks as Dollar Lenders of Last Resort: Implications for Regulation and Reserve Holdings

Layer 1: Overview

This paper investigates why non-U.S. central banks accumulate large holdings of dollar-denominated foreign exchange reserves, focusing on a previously under-emphasized motive: the currency mismatch of private-sector non-financial firms. When domestic firms borrow heavily in dollars despite having predominantly local operating revenues, the central bank faces potential liability as a dollar lender of last resort (DOLLR) in the event of a banking crisis coinciding with a dollar appreciation. The paper combines motivating empirical evidence with a formal theoretical model to analyze the optimal policy mix between ex ante financial regulation (bank capital requirements) and ex post reserve accumulation, and then extends the model to characterize global externalities arising from decentralized reserve-holding decisions.

The empirical work uses an unbalanced panel of 52 non-U.S., non-Eurozone countries (excluding Hong Kong as an extreme outlier) with 357 observations covering 2013-2020. The sample includes 12 advanced economies, 29 emerging economies, and 11 developing economies. The key dependent variable is central bank dollar reserves as a share of GDP; the key right-hand-side variable is cross-border dollar-denominated bank loans to non-financial corporations (NFC), also as a share of GDP, drawn from BIS Locational Banking Statistics. Because banks tightly offset their own currency exposures (dollar assets and liabilities correlate at 0.965 in the panel), the relevant mismatch resides on NFC balance sheets, not bank balance sheets. Cross-border NFC dollar lending proxies for total NFC dollar lending, with correlations of 0.66 overall, 0.89 for advanced economies, and 0.73 for emerging economies in the 21-country subsample where total data are available.

In the full 53-country univariate regression including Hong Kong, the R-squared is 0.53 and the slope coefficient is 5.3 (t-statistic 7.6): a one-percentage-point increase in NFC dollar loans to GDP is associated with a 5.3-percentage-point increase in dollar reserves to GDP. Excluding Hong Kong, the R-squared falls to 0.083 and the slope to 1.3 (t-statistic 2.5). Splitting by income group, the relationship holds for advanced economies (coefficient 3.7, t-statistic 2.2, R-squared 0.31) and emerging economies (coefficient 2.4, t-statistic 2.5, R-squared 0.18) but is absent and wrongly signed for developing economies. Panel regressions with standard reserve-accumulation controls (M2/GDP, financial openness, bilateral trade with the U.S., GDP per capita, log population) and country fixed effects leave the key coefficient broadly stable and significant at the 5% level for both advanced and emerging economies.

The theoretical framework models a two-period small open economy in which households have an exogenous preference for dollar-denominated safe assets (capturing the dollar’s special status), banks intermediate between these households and a fixed investment project, and banking crises occur with probability q. When the home currency depreciates, currency-mismatched NFC borrowers incur liquidity costs that are quadratic in the share of dollar funding; these costs flow through to the banking system. The central bank can respond with two instruments: (i) accumulate dollar reserves R$ at a carrying cost equal to the dollar-domestic interest rate spread S; (ii) impose capital requirements, which crowd out home-currency deposits but cannot directly control dollar deposits (since mismatch resides off the bank balance sheet in the NFC sector). The optimal level of dollar reserves is decreasing in S and increasing in the fraction of failing banks’ dollar liabilities (pB$). When banking crises and exchange rate depreciations are correlated — as is empirically documented — dollar reserves serve an additional hedging function, because the central bank is more likely to need dollar liquidity precisely when the dollar is strong.

The paper’s primary normative contribution is to show that decentralized central banks over-accumulate reserves relative to a global planner’s optimum. Each central bank, acting as a price-taker in the market for safe dollar assets, ignores that its own reserve hoarding reduces the global supply of dollar-denominated safe assets, driving down the dollar interest rate. A lower dollar rate, in turn, widens the dollar-domestic rate spread S and makes dollar borrowing more attractive to NFCs, amplifying the very mismatch the reserves are supposed to hedge. A global planner internalizes this feedback and therefore prefers lower reserve accumulation combined with tighter capital requirements. This result (Proposition 1) holds for all values of the households’ discount factor beta above a threshold that is shown to be below zero under the natural condition that reserve holdings do not exceed the supply of safe dollar assets — meaning the proposition holds robustly for any realistic calibration, including in extensive numerical experimentation where the threshold never exceeds 0.5. In the paper’s global numerical example, the global planner’s equilibrium has dollar reserves fall from 54.62 to 27.99, capital requirements rise from K=7.61 to K=23.77, dollar borrowing B$ fall from 59.99 to 42.98, and the interest-rate spread S narrow by approximately one percentage point, relative to the decentralized outcome. The welfare decomposition shows that bank profits decline but are more than offset by gains in household utility from dollar deposits and reductions in carrying costs, taxation deadweight costs, and liquidity costs from mismatch.

A further extension examines global risk-sharing. When banking crises are imperfectly correlated across countries, a supranational pooling of reserves (e.g., through the IMF) allows reserves to be reallocated ex post to countries in crisis, reducing total required reserve holdings. This risk-sharing motive reinforces the case for international coordination but raises additional institutional challenges around moral hazard and monitoring. The paper concludes that, analogously to the Basel process for capital regulation, an international coordination mechanism for reserve holdings would be globally welfare-improving, but this potential benefit is less widely recognized.

Layer 2: Deep Dive

What is the paper’s core empirical identification strategy and what are the main limitations?

The empirical strategy is correlational: the paper regresses central bank dollar reserves (as a share of GDP) on cross-border NFC dollar loans (as a share of GDP) in a panel of 52 countries over 2013-2020, progressively adding controls (M2/GDP, financial openness, bilateral trade with the U.S., GDP per capita, log population, nominal exchange rate) and country fixed effects. The authors are explicit that the regressions cannot establish causality and should be interpreted as suggestive motivating patterns rather than tight causal tests. The main data limitation is that the BIS only provides complete cross-border NFC dollar lending data, not total (cross-border plus local) NFC dollar lending; total data are available for only 21 countries (10 advanced, 11 emerging), and the correlation between the two measures is 0.66 overall (0.89 advanced, 0.73 emerging). Additionally, dollar-denominated bond-market borrowing by NFCs is excluded. The paper also cannot cleanly separate dollar borrowing by exporters (who are naturally hedged) from dollar borrowing by purely domestic non-tradable firms (who are genuinely mismatched).

What is the mechanism through which reserve accumulation creates a global externality?

Central banks collectively purchase large quantities of dollar-denominated safe assets (e.g., U.S. Treasuries). Each individual central bank takes the dollar interest rate as given (price-taking assumption) and does not account for the effect of its own purchases on the aggregate supply of dollar safe assets in global markets. In the global equilibrium, however, central bank reserve accumulation reduces the net supply of dollar safe assets available to private households, pushing up dollar asset prices and lowering the dollar interest rate. A lower dollar interest rate narrows the dollar-domestic rate spread S, making dollar borrowing cheaper for NFCs, and therefore encouraging greater currency mismatch of private-sector liabilities. This increased mismatch is the very risk that motivated reserve accumulation in the first place, creating a self-defeating dynamic: decentralized reserve hoarding amplifies the aggregate fragility it seeks to hedge. The global planner internalizes this feedback and prefers less reserve accumulation to let the dollar interest rate remain higher, which discourages NFC dollar borrowing even without direct regulatory control over the NFC funding mix.

What roles do capital requirements and funding-mix regulation play in the model, and how do they differ?

Capital requirements (equity capital mandates) act by crowding out home-currency bank deposits; they do not directly affect dollar deposits because the interior optimum for dollar borrowing by banks is independent of total deposit funding in the baseline model without crisis-exchange rate correlation. Thus in the baseline model, capital requirements do not change dollar borrowing and do not change optimal reserve holdings. When banking crises and exchange rate depreciations are positively correlated, however, capital requirements that reduce total deposits (both home-currency and dollar) do reduce optimal reserve holdings, because holding dollar reserves hedges the need to bail out both types of deposits when crises concentrate in strong-dollar states. Funding-mix regulation (direct control over the proportion of dollar versus home-currency deposits) more directly reduces dollar mismatch and allows the central bank to cut reserves substantially further. In the numerical example with capital-only regulation, reserves fall from 56.9 to 54.6; with both capital and funding-mix regulation, reserves fall to 38.5. The paper notes, however, that funding-mix regulation is unlikely to be empirically relevant because currency mismatch resides predominantly on NFC balance sheets outside the regulatory perimeter, not on bank balance sheets.

Under what conditions does the global planner prefer more reserves than the decentralized outcome (the ‘wrong-way’ effect)?

There is one channel through which a global planner might want more reserves than individual central banks: by holding more reserves, the planner would depress the dollar interest rate and thereby increase bank profitability (banks can borrow cheaply in dollars and earn the spread). This ‘wrong-way’ bank-profit effect is captured by the term (Q$ - beta) in the global planner’s first-order condition and grows when the spread between the cost of equity capital and the dollar deposit rate is large — i.e., when beta (the discount factor, or equivalently the inverse of the gross cost of equity) is very low. Proposition 1 establishes that the global planner prefers fewer reserves than the decentralized outcome for all beta above a threshold beta-hat. Under the natural constraint that reserves cannot exceed the total supply of dollar Treasury securities, beta-hat is shown to be negative, meaning the global-planner-prefers-fewer-reserves result holds for all positive values of beta. In extensive numerical experimentation, the threshold was never found to exceed 0.5, implying that the wrong-way effect would only dominate if the cost of equity capital exceeded 100% — an implausible calibration.

How does the paper handle the correlation between banking crises and exchange rate depreciations?

The baseline model assumes crisis probability is independent of the exchange rate. The paper then extends to allow a positive correlation: the probability of a banking crisis rises to (q + h) when the home currency depreciates (dollar strengthens) and falls to (q - h) when it appreciates. This setup nests the baseline as h = 0. With h > 0, two new effects arise. First, dollar borrowing by banks increases because their effective cost of dollar debt is reduced by the implicit put option they have when the dollar appreciates: they default more in the appreciation state, and dollar depositors bear losses. Second, the central bank’s optimal reserve holdings increase substantially, because holding dollars hedges not only future dollar-denominated bailout costs but also home-currency-denominated bailout costs (since crises cluster in dollar-appreciation states where home-currency deposits are worth less in dollars). The formula for optimal reserves gains an additional term proportional to (ph/qz)(Bh + B$) — meaning total bank deposits, not just dollar deposits, now motivate reserve holdings. In this richer environment, any capital regulation that reduces total bank deposits will also reduce optimal reserve holdings, which was not true in the baseline.

What does the risk-sharing extension (Section 5) contribute?

Section 5 asks what happens when banking crises are imperfectly correlated across countries, creating scope for risk-pooling. The paper reverts to h = 0 (no exchange rate-crisis correlation) and an inelastic dollar safe asset supply (theta_$2 = 0) to isolate the risk-sharing effect. If a mass q of countries experience crises independently each period, and a supranational institution (like the IMF) can hold a common pool of reserves and allocate them to countries in crisis, then each dollar of pooled reserves provides 1/q times the crisis coverage of a dollar held at the individual-country level. This multiplier means the total required pool of reserves is dramatically smaller: optimal pooled reserves scale with pqB$ rather than pB$. However, the carrying-cost term in the FOC is also reduced by q^2, which partly offsets the coverage multiplier. For empirically relevant small values of the interest-rate spread S, the coverage effect dominates and pooled reserves are substantially lower than individual-country reserves. The extension reinforces the paper’s main message — international coordination reduces required reserve holdings — but also highlights additional institutional challenges: pooling requires the supranational institution to be able to reallocate reserves away from countries not currently in crisis, raising serious moral hazard and monitoring issues.

How does this paper relate to Bocola and Lorenzoni (2020), and what is the key theoretical distinction?

Bocola and Lorenzoni (2020) is the closest antecedent: it also models reserve accumulation as driven by currency mismatch in the private sector and the central bank’s role as a dollar lender of last resort. The current paper’s key additions are: (i) it explicitly introduces financial regulation (capital requirements, and hypothetically funding-mix regulation) as an alternative or complementary tool to reserve accumulation, showing how the optimal mix depends on the carrying cost of reserves relative to the welfare cost of stringent regulation; (ii) it develops the global externality argument — that decentralized reserve accumulation depresses the dollar rate and thereby endogenously exacerbates the mismatch the reserves are intended to hedge — and shows that a global planner prefers a different mix (more regulation, fewer reserves); and (iii) it provides explicit cross-country empirical evidence linking central bank dollar reserve holdings to NFC dollar borrowing to motivate the mechanism.

How does this paper relate to the literature on ‘mercantilist’ versus ‘precautionary’ motives for reserve accumulation?

The paper classifies its motive as falling within the broad ‘precautionary’ view, alongside the sudden-stops literature and the banking-system flight-to-dollar-assets literature (Obstfeld, Shambaugh and Taylor 2010, who use M2/GDP as their key proxy). The paper differs from M2-based frameworks by focusing specifically on corporate-sector dollar mismatch rather than the risk of domestic depositor flight. The paper distinguishes itself from the mercantilist view (Dooley et al. 2003; Aizenman and Lee 2010; Benigno and Fornaro 2012), which attributes reserve accumulation to exchange rate management and trade surplus recycling. The normative contribution also relates to Fanelli and Straub (2021), who find that individual countries over-accumulate reserves relative to a global planner; however, that paper’s mechanism is mercantilist (exchange rate stabilization) whereas this paper’s is precautionary (dollar LOLR).

How does this paper connect to the literature on international coordination of financial regulation?

The paper shares with Clayton and Schaab (2022) the conclusion that countries acting individually impose insufficiently stringent capital requirements relative to the global optimum, motivating the Basel Process of international regulatory cooperation. However, the paper argues that even if capital regulation is fully coordinated internationally, this is not sufficient to achieve the global optimum — there additionally needs to be a separate mechanism to restrain reserve accumulation, because excess reserve holding depresses the dollar interest rate and exacerbates corporate dollar mismatch through a general-equilibrium channel that capital regulation alone cannot offset. The paper thus identifies reserve coordination as a distinct policy dimension that has received less policy attention than capital coordination.

Why are Eurozone countries excluded from the empirical sample?

Eurozone member countries benefit from either explicit or implicit ECB support in dollar markets. Measuring dollar reserve holdings at the individual country level (e.g., on the Bank of Italy’s balance sheet) and relating them to that country’s corporate-sector dollar borrowing would be conceptually misleading, because the relevant backstop is the ECB at the union level rather than the national central bank. The relevant LOLR function is pooled across Eurozone members. Including them would therefore introduce a systematic bias in the proxy for the dollar LOLR motive.

What are the scope conditions on the empirical results?

The significant positive association between NFC dollar borrowing and central bank dollar reserve holdings holds for advanced economies (coefficient 3.7, t-statistic 2.2) and emerging economies (coefficient 2.4, t-statistic 2.5) but is absent and correctly (negatively) signed but insignificant for developing economies. The authors note that for advanced economies, the result for the subsample is sensitive to removing both Hong Kong (already excluded from the baseline) and Switzerland, given the small number of countries. The results are presented as suggestive correlations rather than causal estimates; missing data on local-currency NFC dollar lending (available for only 21 countries) and on dollar bond-market borrowing are acknowledged as limitations. The theoretical results apply most cleanly when the interest-rate spread S is not too large (so that the small-S configuration is empirically relevant) and when the discount factor beta is above a threshold that is never found to exceed 0.5 in calibrations.

What is the model’s treatment of the dollar interest rate and safe asset scarcity?

In the small open economy version, the dollar interest rate (equivalently, the price of dollar safe assets Q$) is exogenously given, consistent with the small-country price-taking assumption. In the global model, Q$ is endogenized: households have a quadratic extra utility from holding dollar safe assets, so Q$ = beta + theta_d + theta_$1 - theta_$2 * D$, where theta_$2 governs the sensitivity of the dollar rate to the total supply of dollar assets (D$). The spread S = Q$/Q_h - 1 becomes endogenous and falls when central banks absorb dollar assets (reserves R$), since this reduces the net supply available to private households. The externality is zero when theta_$2 = 0 (perfectly elastic supply), and increasing in theta_$2. The paper thus situates the externality squarely in the ‘global safe asset scarcity’ framework originating with Caballero, Farhi and Gourinchas (2008) and Bernanke (2005).

What is the welfare decomposition from the global numerical example?

Table 5 normalizes total welfare in the no-regulation, no-reserve benchmark to 100. Moving from no-regulation to the local-planner outcome (with capital requirements and reserves) raises total welfare from 100 to 113.4, driven largely by a reduction in the deadweight costs of taxation (from -131.9 to -70.7) as reserves substitute for costly fiscal bailouts, despite increased carrying costs of reserves (-18.6) and higher liquidity costs due to unchanged dollar borrowing. Moving from the local-planner to the global-planner outcome raises welfare further to 120.4. This additional gain comes from: a large reduction in carrying costs of reserves (from -18.6 to -5.8), reduced deadweight taxation costs (from -70.7 to -61.3), reduced liquidity costs from mismatch (from -13.8 to -7.1), and increased household utility from dollar deposits (55.8 vs. 43.9) — all more than offsetting a decline in bank profits (138.8 vs. 172.6).

What policy implications does the paper draw, and how are they scoped?

First, international coordination of reserve holdings — analogous to the Basel Process for capital regulation — would improve global welfare by internalizing the safe-asset-scarcity externality. The paper frames itself as initiating a conversation about what such a coordination process might look like; it does not propose a specific mechanism. Second, tighter capital regulation combined with reduced reserve accumulation is the globally optimal policy mix, but individual central banks will not choose this combination unilaterally because they do not internalize the general-equilibrium impact of their reserve holdings on global dollar rates. Third, the risk-sharing extension implies that pooled supranational reserve management (e.g., through the IMF) could substantially reduce the total quantity of reserves needed globally, but this requires the supranational institution to have significant powers to reallocate reserves across countries mid-crisis, raising governance challenges around moral hazard and monitoring. Fourth, the paper does not advocate for coordinating away all reserve holdings — it acknowledges other legitimate reserve motives (sudden stops, domestic bank runs, exchange rate management) not modeled here.

Key Concepts

Dollar lender of last resort (DOLLR): A central bank that stands ready to supply dollar liquidity to its domestic banking system during a crisis in which currency-mismatched borrowers face distress because the home currency has depreciated against the dollar. The DOLLR role motivates holding dollar reserves in advance.

Currency mismatch: A situation in which non-financial corporations (and, by extension, the banking sector that lends to them) have liabilities denominated in dollars while their revenues and assets are predominantly in home currency, creating exposure to losses when the home currency depreciates. In this paper’s framework, mismatch is measured by the ratio of cross-border NFC dollar bank borrowing to GDP.

Carrying cost of reserves: The expected negative return earned by the central bank on its dollar reserve holdings, equal to the spread S between the domestic interest rate (what the central bank pays on the government bonds it issues to finance reserve purchases) and the dollar interest rate (what the reserves earn). A higher S makes reserves more costly to hold and tilts the optimal policy toward financial regulation.

Safe dollar asset scarcity externality: The general-equilibrium feedback by which individual central banks’ reserve accumulation reduces the net supply of dollar-denominated safe assets available to private households, lowers the dollar interest rate, and thereby makes dollar borrowing cheaper for NFCs — amplifying the currency mismatch that motivated reserve accumulation in the first place. Individual price-taking central banks do not internalize this externality.

Decentralized vs. global-planner equilibrium: The decentralized equilibrium is one where each country’s central bank sets capital requirements and reserve holdings to maximize own-country welfare, taking the dollar interest rate as given. The global-planner equilibrium internalizes the impact of aggregate reserve accumulation on the endogenous dollar interest rate. The paper establishes (Proposition 1) that the global planner chooses strictly fewer dollar reserves and strictly higher capital requirements than the decentralized equilibrium, for all empirically plausible parameter values.

Precautionary reserve motive: The class of explanations for foreign exchange reserve holdings based on self-insurance against adverse future shocks, including sudden stops, domestic depositor flight, and (in this paper) the need to serve as dollar lender of last resort when corporate currency mismatch generates systemic banking distress. Contrasted with the ‘mercantilist’ motive based on exchange rate management and trade surplus recycling.

Risk-sharing (pooled reserves): The efficiency gain achievable when banking crises are imperfectly correlated across countries and a supranational institution holds reserves centrally and redistributes them to countries experiencing crises. Each dollar of pooled reserves provides 1/q times the crisis coverage of a dollar held by an individual country, where q is the fraction of countries in crisis at any given time, enabling total reserve requirements to be substantially smaller.