Excess Savings and Twin Deficits: The Transmission of Fiscal Stimulus in Open Economies

The paper asks how debt-financed fiscal transfers are split, over time, between a rise in the deficit country’s own private saving and a deterioration in its current account, motivated by the sharp, simultaneous rise in fiscal deficits, private saving, and (for the U.S.) the current account deficit observed worldwide since the start of the Covid pandemic (Sec. 1, Fig. 1). The standard Ricardian view predicts households save 100% of a debt-financed transfer with no effect on the current account – but the paper notes this is inconsistent with the substantial MPCs documented out of pandemic stimulus checks, and with direct evidence (JPMorgan Chase Institute checking-account data, Fig. 2) that middle-class households built up and then rapidly “spent down” the excess balances from each of the three rounds of U.S. stimulus payments. The paper’s model instead features buffer-stock households who spend a windfall gradually according to their intertemporal MPCs, open economies with home bias in spending, and – unlike the authors’ own earlier small-open-economy work – a genuinely integrated world economy in which domestic fiscal policy can move the world interest rate.

Proposition 1 shows that, for a small open economy with a stable long-run asset demand function and an unchanged long-run level of government spending, a permanent increase in public debt ΔB eventually leaves private wealth completely unchanged (ΔA = 0) and is absorbed entirely by foreigners (Δnfa = −ΔB), for a long-run pass-through of exactly 1 (Sec. 3.1). The logic is that the country is too small to move the world interest rate, and at a steady-state interest rate of zero no local tax increase is needed to finance the new debt, so neither long-run interest rates nor after-tax income change – meaning the long-run level of private asset demand is pinned down independently of how the deficit was spent. The result holds “irrespective of how much fiscal policy affects private wealth in the short run,” irrespective of the monetary policy rule followed en route (as long as the economy returns to its natural allocation), and irrespective of whether the deficit financed government purchases or transfers, provided spending itself returns to steady state.

The speed of transition is governed jointly by the matrix of intertemporal MPCs (M) and the degree of home bias (α), with home bias playing the dominant quantitative role in slowing the transition (Sec. 3.2, Prop. 2-3). Without home bias (α → 1), a debt-financed transfer’s pass-through into the current account deficit exactly equals the iMPC matrix M applied to the fiscal deficit path (Proposition 2): with the paper’s calibrated impact MPC of M0,0 = 0.25, the impact current account deficit from a one-time transfer equals 0.25 dollar-for-dollar. With realistic home bias (the paper’s baseline is α = 0.16, calibrated to U.S. trade openness), only a fraction α of any round of spending falls on imports, so the short-run pass-through into the current account is “only slightly above α×M0,0” (equation 48) – much smaller than M0,0 itself – because the domestically-spent portion of the transfer recirculates as income and is partly re-saved before eventually leaking abroad. The paper summarizes this as “the combination of limited MPCs and home bias leads to slow transition dynamics in response to increases in public debt.”

Tracing the cross-section of who holds the newly-issued debt over time reveals three phases: first, both the middle class (bottom 80% of the wealth distribution) and the rich (top 20%) increase their savings; second, the middle class spends down its excess savings while the rich keep accumulating, partly aided by the demand boost the middle class’s own spending generates; and third, once middle-class excess savings are exhausted, essentially all of the marginal public debt ends up held abroad, as required by Proposition 1 (Sec. 3.3, Fig. 5). The paper notes this three-phase pattern is even more pronounced with realistic home bias than in the no-home-bias case, because the domestic demand boost from middle-class spending is what allows the rich to keep increasing their own savings for longer.

No – alternative monetary rules (a Taylor rule targeting home-goods inflation, and the “natural allocation” that eliminates domestic wage inflation entirely) generate very similar current-account and net-foreign-asset dynamics to the baseline constant-real-rate rule, because the exchange-rate appreciation induced by a monetary tightening and the resulting reduction in import demand roughly offset each other in the paper’s calibration (Sec. 3.4.1). Because the “natural allocation” is the fully flexible-wage limit of the model, the paper notes explicitly that “this shows that nominal rigidities are not important for our main results” – the slow-motion twin deficit is a real, not a nominal, phenomenon rooted in MPC heterogeneity and home bias.

A representative-agent (RANK) model implies Ricardian equivalence, with private wealth tracking public debt one-for-one forever and no twin deficit at all; a TANK model with hand-to-mouth agents does generate an immediate, contemporaneous twin deficit as in the classic 1980s hypothesis, but at realistic calibrations the implied pass-through is far too small to match the “rules of thumb” used by policy institutions; and the Blanchard (1985) perpetual-youth model, once matched to the same impact MPC and openness, behaves closely like the paper’s HANK model in the aggregate (Sec. 3.4.2, Fig. 7). For the TANK model, even with a generous hand-to-mouth share of μ = 0.25 and the paper’s U.S.-calibrated openness of α = 0.157, the implied pass-through is only about 5%, “much smaller than the 30%-50% range often assumed” by policymakers, and the current account deficit lasts only as long as the transfers themselves, unlike the HANK model’s persistent effect. The Blanchard model tracks the HANK model reasonably well because it too has a well-defined long-run asset demand function, though its net foreign assets decay faster because it lacks HANK’s “selection” effect, in which households who have not yet spent their transfer have systematically lower propensities to spend going forward.

No – the paper shows in two distinct exercises that a demand shock resembling the pandemic’s direct effect on spending cannot generate anywhere near the observed rise in aggregate saving, because in general equilibrium a shock that depresses spending also depresses income (Sec. 3.5). A shock to overall household discount factors, calibrated to match the actual 12% peak decline in U.S. output, generates a current account surplus (since foreign spending falls too) but only a peak cumulative excess-saving effect of about 1.5% of GDP – far short of the roughly 11% actually observed. A more targeted shock concentrated on home (services) spending specifically – closer to the pandemic’s actual sectoral pattern, and calibrated with a low intratemporal substitution elasticity (η = 0.5) so that it has real general-equilibrium bite – instead produces “excess dissavings” of about 3% of GDP at the peak, the opposite sign from the data. The paper concludes the pure Covid shock’s contribution to the observed rise in savings was likely small, whatever its exact incidence.

Proposition 4 shows that, in a symmetric-country calibration, the world-economy equilibrium decomposes exactly into an aggregate closed-economy response to the GDP-weighted world-average shock, plus each country’s small-open-economy response to its own deviation from that world average (Sec. 4.1). This is because the world collectively cannot run a current account deficit against itself or move its exchange rate relative to itself, so once variables are demeaned relative to the world aggregate, each country behaves exactly as the small open economy of section 3, holding the world interest rate and world import demand fixed. This decomposition licenses two applications: closed-form predictions for the world-average real interest rate response to a global debt increase (Corollary 1), and a simple cross-sectional regression prediction – a country’s excess private saving and current account should be linear in its fiscal deficit relative to the world average – that the paper takes directly to data (Corollary 3).

Applying Corollary 1 to realized Covid-era fiscal deficits, the model predicts a rise in the world real interest rate of about 71-72 basis points from a roughly 14.5%-of-initial-debt average increase in public debt, versus an estimated actual change of about 68 basis points (Table 2, Sec. 4.2) – a striking quantitative match given the simplicity of the calculation. The same corollary implies large cross-country dispersion in projected long-run net foreign asset positions depending on each country’s fiscal deficit relative to the world average: Denmark, which ran a comparatively small deficit, is projected to see its long-run NFA rise by around 11 percentage points of GDP, while the United States, which ran a comparatively large deficit, is projected to see its long-run NFA fall by around 2 percentage points (Fig. 9).

Regressing cumulative “excess” private savings and current account balances on cumulative “excess” fiscal deficits across 26 advanced economies as of five quarters after 2020Q1 delivers coefficients of 0.79 (savings) and -0.34 (current account) in the data, compared to the model’s calibrated predictions of 0.81 and -0.19 respectively (Sec. 5.2, Fig. 11) – close matches, especially for savings, though the current-account coefficient is not estimated precisely. The paper stresses that this quantitative match is “unique to our model with realistic home bias and MPCs”: a version of the same model without home bias predicts a savings pass-through of only 0.4 and a current-account pass-through of -0.6 after five quarters, “much too fast relative to the data.” Placebo-style regressions using each country’s lockdown stringency index or cumulative Covid deaths per capita in place of the fiscal deficit find essentially no association with excess savings (and even a wrong-signed one for Covid deaths), supporting the paper’s separate theoretical claim (Q7) that the pandemic shock alone had limited power to generate the observed excess savings, so that the simple fiscal-deficit regression is unlikely to be severely biased by omitted Covid-severity variation.

Relaxing the symmetric-country assumption and adding country-specific Covid demand shocks (inferred from each country’s actual consumption path net of the fiscal shock’s estimated effect) leaves the model’s cross-sectional pass-through coefficients within the data’s confidence interval, and shows that fiscal deficits – not the Covid shock – explain the “vast majority” of observed excess savings, with “almost no role” for the Covid shock (Sec. 6, Fig. 12-13). By contrast, the model has much more limited power to explain the cross-country variation in current accounts even after both shocks are included, which the paper attributes partly to the general difficulty theories have matching current-account data and partly to its own prediction that, if twin deficits genuinely unfold in “slow motion,” fiscal deficits should have limited explanatory power for current accounts specifically over short post-shock horizons.

The paper predicts that pandemic-era excess savings are likely to persist for a long time but will increasingly be held by the world’s wealthiest households, even as slow-motion twin deficits gradually pool the underlying debt across countries (Sec. 7, Fig. 14). This refines the original 1980s “twin deficit hypothesis” – popular after the Reagan tax cuts’ dollar appreciation and current account deterioration, but which fell out of favor after the Clinton-era “twin divergence” of a fiscal surplus alongside a current account deficit – by reconciling the mixed empirical record (Kim and Roubini 2008 versus Feyrer and Shambaugh 2012 and Guajardo, Leigh and Pescatori 2014) with a single mechanism: the paper’s “slow-motion” logic implies a causal twin-deficit relationship may be undetectable over short horizons, swamped by other shocks, yet should emerge more clearly as one considers longer horizons – consistent with the model’s prediction that the pass-through coefficient on current accounts continues to fall toward -1 only gradually, well beyond the paper’s five-quarter empirical window.