US or Domestic Monetary Policy: Which Matters More for Financial Stability?

Large enough that “the US dollar financial system outside of the USA is bigger than the domestic US banking system” (Introduction, p. 37). Before any estimation the paper assembles the scale figures that make the spillover question plausible. On trade invoicing: “Gopinath (2015) reports that about 40% of imports is invoiced in dollars, based on a sample of 43 countries. Since the USA accounts for roughly 10% of their imports, this means that roughly 30% of imports involve non-US entities and are priced in dollars.” On finance: the BIS reports 2018 US dollar liabilities of banks located outside the United States “on the order of US$15 trillion,” which Borio et al. (2017) suggest foreign exchange swaps would roughly double; the US Treasury reports foreigners held 45 percent of outstanding Treasury securities in 2018, an amount over $6 trillion; and total assets of US depository institutions are around $17 trillion. The authors also set out a baseline against which any financial-friction channel has to be judged: in a simple Mundell-Fleming model, US easing boosts output and depreciates the dollar, but “the overall impact on a non-US economy and its exports is ambiguous, depending on the degree of product substitution, and the non-US monetary policy response,” so controlling for domestic growth, growth prospects and interest rates is what isolates the friction-based channels (pp. 37-38).

The number of consecutive quarters in which the trend component of the 2-year sovereign yield falls – a measure of sustained easing, chosen because leverage responds to persistent conditions rather than to surprises (Section 2.1, pp. 40-41). The duration counter increments whenever the eight-quarter moving average of the nominal 2-year yield is lower than in the previous quarter and resets to zero otherwise. Two justifications are given for the 2-year yield. First, the effective lower bound: “while the short-term interest rate is one of the most widely used indicators of monetary policy, during the recent period of unconventional monetary policy, it displays little movement near the effective lower bound. In many countries such as the USA, policy easing was provided by purchasing government bonds, which lowered longer-term rates, but not the short-term policy rate.” Second, expectations: the 2-year rate has “the added advantage of capturing expected monetary policy easing, which is likely to be more relevant for financial firms’ leverage decisions than the current short-term policy rates.” The authors also explain why they do not join the surprise-identification literature: that approach “identifies what is arguably an exogenous shock and can thus investigate a causal relationship with other fast-moving variables such as asset prices. However, firm leverage is likely to be a slow-moving variable reacting more to monetary policy expectations than to very small monetary policy surprises which may offset each other over subsequent monetary policy announcements” (fn. 7, p. 40). They also explicitly avoid unobservable benchmarks: “We explicitly focus on readily observable market rates, as opposed to deviations from Taylor rules, the natural interest rate, or other benchmarks based on unobservable variables that require strong model-based assumptions to compute.”

Moderately – correlations with alternative filters range from 0.2 to 0.8 (fn. 8, p. 40). The baseline eight-quarter moving average is “positively correlated with alternative measures with correlation equal to 0.4, 0.6, 0.8, and 0.2” for a four-quarter moving average, a twelve-quarter moving average, a one-sided HP filter with smoothing parameter 1600, and Hamilton’s (2018) filter with four lags respectively, all significant at 1 percent. The lowest of those – 0.2 against the Hamilton filter – is a reminder that “duration of easing” is a construct whose measurement matters, even though the paper reports robustness to a continuous alternative that cumulates actual rate changes rather than counting quarters.

Because book equity is unreliable for financial firms – with a z-score and a risk-adjusted return on equity as alternatives (Section 2.1, p. 41; Section 2.3, p. 46). Leverage is “the market value of equity plus the book value of liabilities divided by the market value of equity,” specifically “to avoid problems associated with accounting measures of the book value of equity.” The robustness checks substitute the z-score – “a measure of the probability a firm will become insolvent, equals the average return on assets (ROA) plus the average ratio of equity to assets, divided by the standard deviation of ROA over the past 2 years” – and a risk-adjusted ROE, “the level of the ROE divided by its standard deviation over the previous 2 years,” and find the results hold.

988 publicly listed financial firms across 21 countries and six industries, with leverage varying enormously by industry (Section 2.1, p. 41, Table 1). The countries are 18 advanced economies plus Brazil, Mexico and South Africa, over 1998Q1-2014Q4. Median leverage runs from 1.5 for asset managers and 1.9 for real estate firms up to 4.5 for investment banks, 6.5 for insurance and 10.5 for banks, with wide interquartile ranges (banks 6.4 to 19.8; investment banks 3.2 to 12.2). Firm counts are skewed: 369 real estate firms and 241 banks, but only 47 investment banks and 85 in the “other” category. Industry percentiles are computed from firm-level medians “to avoid over-representation from firms with more observations.”

Log leverage on the duration counter plus lagged macro controls and firm fixed effects, estimated separately for each of the six industries (Section 2.2, pp. 41-42). The coefficient is a semi-elasticity – “the percentage change in financial firms’ leverage for each additional one-quarter of monetary policy easing” – and is converted to a marginal effect in levels at the median so the numbers are comparable with raw leverage. Controls are lagged “to avoid endogeneity biases” and cover real GDP growth year-on-year “to capture changes in income and confidence,” stock index growth “to control for the cost of equity financing as well as growth prospects and the automatic valuation effect on leverage from stock prices,” a volatility index for financial-market uncertainty, and a sovereign bond rating for sovereign risk. Firm fixed effects absorb structural differences in business models, domestic regulation and accounting practice, and since “firms do not change countries, firm fixed effects control for both country- and firm-specific factors.” Standard errors follow Driscoll and Kraay (1998), robust to heteroskedasticity and to cross-sectional as well as temporal dependence. The authors also pre-empt one specific confound raised by Christiansen and Ranaldo (2007): they checked the stock-bond correlation in their sample and found it “about zero,” so the duration estimates are “not driven by a systematic bond-stock correlation” (fn. 19, p. 42).

One quarter of easing raises median banking leverage by 0.191 (10.5 to 10.7), and eight quarters take it from 10.5 to 12.0 (Section 2.2, pp. 42-43, Table 2 and Fig. 3). The bank estimate is significant at 1 percent. Insurance moves from 6.5 to 6.6 on one quarter of easing (0.081, 1 percent). Investment banks show 0.080 at the 5 percent level; asset management 0.005 at the 10 percent level and over two years rises “from a very modest 1.48 to 1.52”; real estate (0.002) and the “other” category (0.018) are not significantly different from zero. The conclusion is limited to where significance holds: “domestic policy easing increases domestic leverage for banks, insurance, asset management, and investment banks.” One caveat the authors flag themselves: “if we substitute four-quarter-lagged duration for the contemporaneous duration in Eq. (2), the results are somewhat weaker” (fn. 24, p. 42). The interpretive framing is even-handed: “easing does have its intended short-term impact, but over time, risks in the system can build up.”

US easing raises non-US financial firms’ leverage, in most sectors by more than their own country’s easing does (Section 2.3, pp. 43-46, Table 3 and Fig. 4). Adding the US duration to the specification and dropping US firms, the domestic coefficient barely moves (banks 0.223), while the US coefficient is 0.361 for banks, 0.103 for insurance, 0.023 for real estate, 0.008 for asset management, 0.099 for investment banks and 0.036 for the residual category – “significantly different from zero at the 1% level for all sectors, except for investment banks, where it is still significant at the 5% level.” Over two years, “leverage of banks rises from roughly 16.7 to 19.6, of insurance companies from 8.2 to 9.0, and of investment banks from 5.8 to 6.6.” The ranking the authors emphasize: “we find that the effects of US monetary policy easing are either equal to those of domestic monetary policy easing (for investment banks and asset managers), greater (for banks), or substantially greater (for insurance, real estate and other financial firms).” Note that the point of the raw-data figure and the regression differ in size – the unconditional plot shows bank leverage going from 16.7 to 24.8 after two years of US easing, against 19.6 once controls are applied – so the controls matter, and the smaller conditional number is the one the paper leans on.

Yes, and if anything tightening moves leverage more (Section 2.3, p. 46). “We find that prolonged monetary policy tightening drives leverage down in the same way as prolonged easing pushes it up.” Concretely, for banks the marginal effect of a quarter of domestic easing is 0.223 against 0.398 for a quarter of tightening, and for US policy 0.361 for easing against a fall of 0.567 for tightening. “This same pattern holds for other parts of the financial system,” and the symmetry survives the continuous measure of policy stance. This matters for the interpretation: the result is not a one-way ratchet in which easing builds leverage that tightening cannot unwind.

No – euro area easing has essentially no effect outside the euro area (Section 2.4, pp. 46-48, Table 4 and Fig. 5). Replacing US duration with a euro area duration built from the 2-year German Bund yield and dropping euro area firms, “with the possible exception of insurance firms, spillover effects from euro area monetary policy are both economically and statistically very close to zero. And, even in the case of insurance firms, the estimates suggest an impact that is one-third of that of comparable US policy easing.” Domestic easing continues to work as before in the non-euro-area sample (banks 0.156, insurance 0.071, investment banks 0.065, all at 1 percent), so the null is about the euro’s reach, not about the method. The explanation offered is the currency’s international role rather than the size of the economy: “the fraction of invoicing in euro roughly matches the import and export share of the euro. And, in finance, while non-US banks issue $15 trillion in liabilities, non-euro-area banks only issue €4 trillion in liabilities” – a contrast made sharper by the footnote that euro area banks’ assets in 2017 were “US$34.8 trillion, nearly 50 percent higher than US banks’ assets” (fn. 29, p. 48). Japan could not be examined “because the lack of interest rate variation over our sample period makes this difficult” (fn. 28, p. 46).

Larger with financial development; smaller with trade openness and with gross dollar liabilities (Section 3.1, pp. 49-53, Table 5). Three interactions are added, each entering at a four-quarter lag because the data are annual: a financial development index from Sahay et al. (2015) based on depth, access and efficiency (sample median 0.788); trade openness as exports plus imports over GDP (median 55.2 percent); and the stock of bank and nonbank US dollar liabilities to banks located outside the country, relative to total credit in the country (median 6.6 percent). Financial development raises the responsiveness of bank, insurance and real estate leverage – “likely a consequence of the fact that deeper financial markets are more prone to transmit monetary policy across borders, as well as support more interconnected institutions” – while asset managers and investment banks are less affected, which the authors attribute to those firms’ sophistication and market access letting them “hedge their exposure to changes in US policy more effectively.” Trade openness lowers the response for banks, insurance and real estate, and the authors are careful about what that means, since openness and dollar liabilities correlate at 0.45 across country medians: “Most likely, US dollar invoicing is so common, the more a firm trades, the greater access it has to US dollar borrowing. Consequently, trade openness is just another measure of US dollar liabilities.” Dollar liabilities themselves also lower the response for banks, insurance and real estate, significantly at 1 or 5 percent. One sharp secondary finding: “net claims have virtually no impact. Consistent with work of Obstfeld (2012) and others, we find that gross stocks are what matters” (fn. 38, p. 53). Throughout, the total US effect stays large and significant: “Without exception, the impact of US policy easing on non-US financial firm leverage is large and significantly different from zero. Furthermore, for all but investment banks, they are uniformly larger than the impact of domestic policy easing.”

Substantially, but almost always upward – only the two most dollar-indebted countries in the sample are exceptions (Section 3, p. 49 and Section 3.1, p. 53, Fig. 6). Estimating the spillover coefficient separately by country-industry group gives estimates ranging “from zero, or slightly negative, to more than three times the full sample average,” though the authors decline to report them: “a paucity of data makes the country-level spillover estimates unreliable, so we do not report them here. The number of firms in each industry group in a country is in many cases in the single digits” (fn. 31, p. 49). What they do report, from the interaction specification evaluated at each country’s own medians, is that “with the exception of countries with very high levels of US dollar liabilities – specifically, Ireland (IRL) and Mexico (MEX) – US monetary policy easing increases banking system leverage substantially. In roughly three-quarters of the cases, the increase after 2 years of consecutive easing in the USA is more than 10% from the baseline. For Italy, that means a rise from 17.7 to 21.1; for Canada, from 10.8 to 12.7; and for Korea from 18.3 to 21.6.”

Because there are two opposing effects, and the mechanical valuation effect works against the credit-supply effect (Introduction, pp. 37-38; Section 3.2, pp. 54-56, Table 6). The mechanism is set out before the estimation. Push: “the supply of credit by US-based lenders might increase as the Fed cuts rates if lenders search for yield abroad,” raising credit across foreign firms and countries. Pull: “Firms that have dollar liabilities (or firms in countries with high aggregate dollar liabilities) will see their balance sheets strengthen if the dollar depreciates as the Fed cuts rates. Indeed, the local currency value of their dollar debt would decrease, thereby lowering leverage in a mechanical fashion.” Following Adrian and Shin (2009, 2014), the firms then re-optimize: “they can act passively and allow leverage to decline; they can increase borrowing proportionally to keep leverage unchanged; or they can procyclically increase borrowing to drive up their leverage.” Adding the bilateral dollar exchange rate and a triple interaction with dollar liabilities to the specification yields three findings: “total effects are unchanged”; “the push factors, as captured by the common effect … remain the dominant source of the spillover effects”; and the triple-interaction coefficient “is negative in all instances,” so “domestic currency appreciation tends to lower leverage by more in places with higher US dollar liabilities.” But that is only the partial effect – the sum of the triple interaction and the exchange rate interaction “is generally positive,” and the authors read that plainly: “Most firms, therefore, tend to further increase leverage as a result of the ‘windfall’ from domestic currency appreciation. In this sense, leverage appears to be pro-cyclical.”

By lagged controls, a stated direction for the residual bias, and a GMM check – not by an instrument for policy (Section 2.5, p. 49; Introduction, p. 39). The Introduction concedes the design choice openly: “we do not explicitly distinguish between systematic and unexpected monetary policy. The central question of this paper is whether the duration of policy easing has an impact on financial stability, irrespective of what led to the easing or whether it was expected. Our approach is to control for factors that might have led to the interest rate cut, and might affect financial vulnerabilities independently.” Two endogeneity sources are then addressed. On dynamic-panel bias from fixed effects with lagged regressors, “given the long sample, this bias is likely to be small.” On policy responding to leverage, the direction is spelled out: “monetary policymakers ease (tighten) in response to lower (higher) aggregate leverage, endogeneity would lead us to underestimate the effect of duration” – and an Arellano-Bover (1995) and Blundell-Bond (1998) GMM estimator, using lagged differences and second lags of duration as instruments, gives results “nearly unchanged relative to the baseline.”

A simple easing indicator, an interest-rate-level interaction, a quadratic term, a crisis-excluded sample, and a continuous stance measure (Section 2.3, p. 46; Section 2.5, pp. 48-49). Replacing the duration counter with a 0/1 easing indicator “forces the entire impact of easing on leverage to occur at once, rather than gradually,” and yet “results remain broadly similar: domestic and US easing increase leverage by roughly equal amounts.” Interacting the pre-easing interest rate level with duration finds “some evidence that the impact of easing is higher when interest rates are lower, both domestic and USA,” attributed to the nonlinearity of discounting profits by one plus the interest rate. Adding squared duration terms to test whether side effects compound over time yields “only … modest evidence of nonlinearity in a few cases,” and the authors decline to resolve it: “Either there is no strong nonlinearity, or there is insufficient variation in the data to address this issue definitively.” Excluding the 2007Q2-2009Q2 crisis window – dated from Bear Stearns’ subprime hedge fund disclosure to Bernanke’s July 2009 remark that “the extreme risk aversion of last fall has eased somewhat” – leaves results “largely unchanged,” so “identification does not rely on variation during the crisis period.”

That prudential policy is the right first line of defence, but only if it covers the whole financial sector – and that monetary retaliation is self-defeating (Section 4, pp. 56-57). On coverage: “Prudential policies would seem an appropriate first line of defense against the effects of domestic monetary policy on leverage. Importantly, however, policies need to be deployed across the entire financial sector, including the various nonbank intermediaries who tend to increase their leverage in the wake of a policy easing. Our results are further evidence for the view that focusing prudential policies on bank capital requirements alone misses the wider financial stability risks inherent in monetary policy expansion.” On the cross-border dimension, the authors identify a trap: “If easing of monetary policy in the USA increases leverage elsewhere, these other countries could respond by tightening domestic monetary policy. However, this creates its own problems as higher domestic interest rates could intensify capital inflows, driving leverage up even further, defeating the purpose of the response.” They then catalogue the three classes of prudential response countries actually use – capital requirements on financial firms; borrower-side limits such as loan-to-value or debt-service-to-income restrictions on households and leverage limits on firms; and restrictions on cross-border financial flows including capital flow management measures, alongside foreign exchange intervention during volatile episodes – with the caveat that “it is important that the framework be comprehensive, or activities will escape the regulatory perimeter putting the financial system at ri[sk].” The broader point on the interaction of the two policies is also carefully stated: “when business cycles are correlated, domestic and US monetary policies will work together to amplify the swings in financial sector vulnerability. And, when countries are at different stages of the business cycle, domestic authorities may feel the need to counter the impact of US policy.”