U.S. Monetary Policy and the Global Financial Cycle

It asks how financial globalization changes the international transmission of US monetary policy through financial intermediation and global asset prices – “an area that has been largely neglected by the literature.” The framing is that while the growth of cross-border flows and holdings is well documented, “what has not been explored as much… are the consequences of financial globalization for the workings of national financial markets, and for the transmission of US monetary policy beyond the national borders” (Section 1). The paper places itself among several established transmission channels rather than against them – the standard Keynesian demand channel, the credit channel of Bernanke and Gertler (1995), and the risk-taking channel of Borio and Zhu (2012), Bruno and Shin (2015a) and Coimbra and Rey (2017) – stating that “these channels are complementary.” Its methodological claim against prior work on US monetary policy, leverage and the VIX is specific: earlier studies “rely on limited-information VARs (four to seven variables) and on Cholesky identification schemes,” so “it is therefore unclear whether their results survive a more robust identification of monetary policy shocks,” with omitted variables an additional concern in small systems (fn. 8). The contribution is thus framed as enabling, “we believe for the first time, the joint analysis of financial, monetary and real variables, in the US and abroad.”

A single common component estimated from a large heterogeneous panel of risky asset prices; its shape and turning points are informative but its level is not. The benchmark panel is 858 monthly price series for 1990-2012; the authors formally test the number of factors and report that the data support one common global factor accounting for over 20 percent of the common variation (Section 2, Table B.2). Because they prefer cross-sectional breadth but are conscious that a short span limits the subsequent VAR, they repeat the extraction on a 303-series panel covering only the US, Europe, Japan and commodities back to 1975, where one factor accounts for about 60 percent. Two technical caveats are stated rather than buried. Factors are obtained by cumulating estimates from stationary first-differenced log price series, so they are “consistently estimated only up to a scale and an initial value”; consequently “positive and negative values displayed in the chart do not convey any specific information per se. Rather, it is the overall shape and the turning points that are of interest” (fn. 12). And the sign is a normalisation: the factor is rotated to correlate positively with the major stock market indices, so an increase means higher global asset prices. The factor’s narrative fit is offered as a plausibility check – it declines with every NBER recession, rises sharply from the early 1990s through 1997-98 when the Russian default, the LTCM bailout and the East Asian crisis reverse the dot-com build-up, rises again from early 2003 to the third quarter of 2007, then plunges.

By projecting the factor on realized global variance and treating the inverse of the centred residual as a risk-aversion proxy – a decomposition the paper justifies from its own model and explicitly flags as impure. The theoretical warrant comes from Section 4.1: in a large class of asset pricing models, including the paper’s own, “the common component of risky asset prices reflects aggregate volatility scaled by the aggregate degree of effective risk aversion in the market” (Section 2). The construction follows Bekaert, Hoerova and Duca (2013), who estimate variance risk premia as the gap between implied variance and expected physical variance: the authors first estimate realized monthly global volatility from daily returns on the global MSCI index, then take the inverse of the centred residuals from projecting the factor on that realized variance. Two honest qualifications accompany it. The realized-variance step assumes “that monthly realized variances calculated summing over daily returns provide a sufficiently accurate proxy of realized variance at monthly frequency” (fn. 16). And the residual is not pure risk aversion: “our empirical proxy for aggregate risk aversion could reflect also other factors such as expected cash flow growth or risk free rates,” which the authors say are controlled for in the VARs through the inclusion of interest rates and industrial production (fn. 17). As supporting evidence for the factor’s interpretation, the paper shows it co-moves closely, with the expected negative sign, with the VIX, VSTOXX, VFTSE and VNKY, and notes that comparison with the Gilchrist-Zakrajsek spread and the Baa-Aaa spread shows commonalities but “the synchronicity is slightly less obvious” (fn. 14).

That risk-neutral global banks became large enough to price global risky assets, an interpretation supported by the timing of banking flows but offered as one possible reading. The recovered risk aversion “is in continuous decline between 2003 and the beginning of 2007, and it decreases to very low levels, at a time where volatility was uniformly low, global banks were prevalent, and may have been the ‘marginal buyers’ in international financial markets” (Section 2). The authors link this to Shin’s (2012) documentation of the rising share of banking flows in aggregate capital flows through 2008, note that “the timing of the surge of banking flows coincides with the decline in global risk aversion in 2003,” and observe that estimated risk aversion rises during 2007, jumps with Lehman’s bankruptcy, and stays persistently high as global banks’ importance declines. In Section 4.1 the reading is restated with the model’s vocabulary and appropriate hedging: “One possible interpretation of the decline in the measure of aggregate risk aversion observed between 2003 and 2007… is therefore that it was driven by risk-neutral global banks becoming large and important for the pricing of risky assets, sustaining an increase in risky asset prices on a global scale.”

Because an external instrument lets the contemporaneous transmission coefficients be estimated from the projection of VAR residuals on the instrument, dispensing with “often implausible timing restrictions”; on first-stage statistics the high-frequency instrument holds up in open-economy VARs where the narrative alternative does not. The identification logic is stated compactly: “if the instrument correlates with the VAR innovations only via the contemporaneous monetary policy shocks, a projection of the VAR innovations on the instrument isolates variations in the variables which are solely due to the shock” (Section 3.1). The reported first-stage F statistics, against a weak-instrument threshold of 10, are the key comparison: for the domestic VAR, 17.9 for FF4 and 10.9 for an extension of the Romer-Romer narrative series; for the global VAR with world aggregates, 14.8 versus 2.3; for the floaters VAR, 14.9 versus 2.8 (Table 1). The authors conclude that “the relevance of the narrative series, and the plausibility of the impulse response functions identified with this instrument however, deteriorate dramatically in both open economy global VARs. In contrast, the first stage IV statistics associated to the high-frequency based identification are only marginally altered.” The choice of policy variable is also substantive: using the one-year government bond rate together with the roughly three-month horizon in the instrument means the estimated shock captures “standard monetary policy shocks that affect the fed funds rate, but also implicit and explicit Fed communication and actions that affect interest rates at longer maturities” (Section 3.2).

Because a market-based surprise is a clean shock only if markets can instantly strip out the systematic component of policy, and the rich information set is the paper’s remedy for the failure of that condition. The concern is the Fed information effect of Nakamura and Steinsson (2017) – also the signalling channel of Melosi (2017) and the Delphic component of forward guidance in Campbell et al. (2012): “In the presence of information asymmetries, the high-frequency surprises are also a function of the information about economic fundamentals that the central bank implicitly discloses at the time of the policy announcements. Failure to account for this effect hinders the correct identification of the shocks, resulting in severe price and real activity puzzles particularly in small VARs” (Section 3.1). The authors’ response is that “the information sets in our VARs controls for a wealth of other shocks, both domestic and international, to which the Fed endogenously reacts, and allows identification of monetary policy shocks above and beyond what is expected by market participants.” Two further arguments support the medium scale: a single specification lets the global responses be judged “against the background of the response of the domestic business cycle,” acting “as a disciplining device to ensure that the identified shock is in fact inducing responses that do not deviate from the standard channels of domestic monetary transmission”; and Banbura et al. (2010) show that a medium-scale VAR of comparable size recovers the shocks and reproduces theoretical responses (fn. 20). The payoff is visible in the domestic results: the narrative instrument “recovers responses that display a pronounced price puzzle,” while the high-frequency identification delivers prices that “continue to slide into negative territory, in a manner consistent with the presence of price rigidities” (Section 3.2).

A conventional contraction with a strong financial amplification signature, which serves as the internal validity check for the global results. Following a shock that raises the policy rate by 1 percent, production and capacity utilisation contract, housing starts fall, unemployment rises significantly and prices adjust downward, with the effects building over the horizon rather than arriving suddenly (Figure 5, Section 3.2). On the financial side the one-year rate’s response dies out quickly, and “comparing the responses of the policy rate and the slope of the yield curve we note that the fast rebound of the policy rate is not sufficient to generate the implied 40bps increase in the 10-year rate,” so the term spread narrows. The Gilchrist-Zakrajsek excess bond premium – corporate spreads net of default considerations – rises suddenly, which the authors read as both heightened perceived risk aversion and higher corporate funding costs, “evidence of a powerful financial amplification mechanism of monetary policy shocks that operates at the domestic level.” Equity prices drop strongly and immediately, house prices fall, mortgage spreads widen very significantly, and the dollar appreciates against a basket of currencies.

The global factor falls about 40 percent on impact and measured risk aversion rises over 50 percent above its average trend; the factor’s magnitude is translated into stock-market terms only under a stated equal-loading assumption. Because the factor has no meaningful measurement unit, the authors quantify it through its contribution to index fluctuations: “The factor explains about 20% of the common variation in our panel of international asset prices. If we assume that all asset prices load equally on the factor, the 40% impact fall would roughly translate into a 8% impact decrease in the local stock market” – a number they check against the estimated US, UK and German equity responses (Section 3.2, Figures 5 and 10). On risk aversion they are candid that “quantifying the rise in risk aversion is less straightforward; but the shock substantially raises it by over 50% above its average trend.” For credit and flows the paper does not report a single elasticity but characterises the scale: “The decline in credit, both domestic and cross-border, whether we look at flows to banks or to non-banks, is in the order of several percentage points and thus economically significant,” and the leverage responses are likewise “in the order of several percentage points, and hence economically relevant.”

It is included as a control, and it does not move on impact – which is what allows the credit results to be read as financial rather than demand-driven. “The US monetary policy contraction leaves global growth unchanged on impact. The inclusion of global growth here serves two purposes. First, it allows us to consider changes in global financial conditions once we have controlled for economic activity on a global scale. Second, it helps ensure that we are not confounding the effects of a US monetary policy shock with other global shocks that affect credit through their effects on growth” (Section 3.2). The paper adds a second discriminating check on the credit response: the contraction in global domestic credit “is not driven by US domestic credit,” which is shown by the separate response of global credit excluding the US.

Leverage contracts strongly and quickly for the intermediaries with capital-market operations and only slowly and weakly for the broader banking sector – a distinction the paper draws deliberately on risk-taking grounds. The authors “separate brokers/dealers and G-SIBs from the aggregate banking sector, due to their different risk-taking behavior,” using Flow of Funds data for US security brokers and dealers and bank-level balance sheets for euro-area and UK global systemically important banks (Section 3.2). The estimated pattern is “consistent with increased funding costs, heightened levels of risk aversion and declining asset prices that alter the valuation of banks’ balance sheets.” For the aggregates, “while no appreciable change is visible for the US, the leverage of European banks contracts significantly to bottom about a year after the shock hits,” with the reduction “more modest and more delayed when compared to that of the G-SIB and the Broker-Dealers,” which the authors attribute to domestically oriented retail banks taking longer to adjust. A data caveat applies throughout this block: credit, international inflows and leverage are originally quarterly and converted to monthly by interpolation, with results reported to be robust to starting the estimation sample in January 1990.

That the credit and capital-flow contraction among independently floating countries is quantitatively very similar to the world aggregate – which the authors read as the exchange rate regime failing to be a fully protective shield, not as regimes being equivalent. The exercise substitutes floater aggregates for world aggregates in the same VAR, using the IMF’s de-facto classification to isolate 32 countries as independently floating from a 54-country panel, and again controlling for global economic activity (Section 3.2, fn. 23 and 26). The result is stated with its own limits attached in the same breath: “the magnitude of the contraction in the credit variables is very similar to that obtained over the full sample. It should be clear that these results do not imply that exchange rate regimes are equivalent. However, they do indicate that the exchange rate regime may not be successful in providing a fully protective shield against US monetary policy shocks.” The conclusion restates the interpretation as an echo of Rey (2013) – “the Mundellian trilemma may have really morphed into a dilemma: as long as capital flows across borders are free, and macroprudential tools are not used, monetary conditions in any country, even one with a flexible exchange rate, are partly dictated by the monetary policy of the hegemon (the US)” – immediately followed by “This of course does not mean that exchange rate regimes do not matter, as Klein and Shambaugh (2013) and Obstfeld (2015) rightly point out” (Section 5).

They rule out the possibility that local financial tightening reflects local monetary policy, because both central banks ease. Looking at local equity indices, bilateral dollar rates, corporate spreads and policy rates for two large floating currency areas, the responses “are remarkably similar” across the two: equity indices plummet on impact consistent with the global factor’s fall, the dollar appreciates significantly against both with the appreciation “relatively short-lived in both cases,” reverting “in the span of one to four quarters,” and corporate bond spreads rise very significantly on impact (Figure 10, Section 3.2). The decisive piece is the policy-rate response, which implies “an endogenous monetary easing of about 30bps” in both areas, estimated with more uncertainty for the UK: “This also implies that the tightening of financial conditions in the UK and the Euro Area cannot be ascribed to a domestic monetary policy tightening, and are instead a consequence of the US monetary policy spillovers.”

It is explicitly a stylized interpretive device, building on Zigrand, Danielsson and Shin (2010), that rationalises a time-varying aggregate risk aversion from the wealth distribution between two investor types. Global banks are leveraged, fund themselves at the US risk-free rate in dollars, and are risk-neutral but bound by a Value-at-Risk constraint; the authors justify the risk neutrality by implicit bailout guarantees and note that “whatever the microfoundations, the crisis has provided ample evidence that global banks have taken on large amounts of risk and that this risk was not priced by creditors” (fn. 30). Asset managers are mean-variance investors with constant risk aversion who also hold non-traded regional assets. Because the bank’s binding VaR constraint enters its first-order condition exactly as a risk-aversion term would, market clearing yields expected excess returns as the sum of a global and a regional component, with the global component equal to “the aggregate variance scaled by the aggregate degree of effective risk aversion” – itself the wealth-weighted average of the two types’ effective risk aversions (Proposition 1, Section 4.1). The paper is clear about which ingredient is load-bearing: “The fact that only asset managers, and not the global banks, have a regional portfolio is non essential… The asymmetry in risk aversion (risk neutral banks with VaR constraint and risk averse asset managers), however, is important for the results” (fn. 31). It points to Coimbra and Rey (2017) for “a more realistic dynamic stochastic general equilibrium model,” and the conclusion concedes the open question: whether “open economy extensions of these models would be able to generate a Global Financial Cycle whose features would match the empirical regularities uncovered in this paper” is something that “still remains to be seen.”

That procyclical leverage extends beyond the US, and that the banks loading most on global risk before the crisis earned the highest pre-crisis returns and suffered the largest post-crisis losses. Using a panel of 166 financial institutions in 20 countries from 2000 to 2010, of which 21 are classified as globally systemically important, the authors find “the positive association between leverage growth and balance sheet growth goes well beyond US borders,” extending Adrian and Shin’s (2010) US finding (Section 4.2, fn. 35). The procyclicality “tends to be a stronger feature of the behavior of financial institutions that engage in global capital markets operations,” in particular the former stand-alone investment banks and the large European universal banks “whose investment departments have played a central role in channelling US dollar liquidity worldwide in the years immediately preceding the financial crisis.” The empirical counterpart of the model’s bank-return equation plots each bank’s average return against its loading on the global factor, splitting at August 2007. In the pre-crisis sample the association is positive, and the systemically important banks “tend to have both higher average betas, and larger returns,” which “suggests that global banks were systematically loading more on world risk in the run-up to the financial crisis, and that their behaviour was delivering larger average returns, compared to the average bank.” Sorting on pre-crisis betas but plotting post-crisis returns shows that “the institutions that were loading more on global risk pre crisis suffered the largest losses after the systemic meltdown began.”