The global capital flows cycle: structural drivers and transmission channels

A tight link between a global risk factor and a global capital flows cycle; a structural decomposition showing financial shocks outweigh US monetary policy in driving that risk; and a test of how openness and the exchange rate regime shape transmission to different flow types. The authors list them in order: “First, we show that a measure of global risk that summarizes the co-movement of stock market returns in 63 economies (a Global Stock Market Factor) is tightly connected to a cycle in global capital flows… Second, we investigate the structural drivers of this global risk measure and, crucially, find that financial shocks, which can be interpreted as exogenous changes in the risk bearing capacity of the financial sector, matter more than US monetary policy shocks in driving global risk. Other shocks, such as those driven by geopolitical or economic policy uncertainty or by the US demand are not particularly [relevant]… Third, we study how capital account openness and the exchange rate regime influence the transmission of global risk to different types of capital flows” (Section 1). The claim the paper positions itself against is the dilemma reading: under a global financial cycle “the policy choice is restricted between an independent monetary policy and capital account openness, whereas the exchange rate regime is irrelevant, see Rey (2015) and Passari and Rey (2015).” It also engages the sceptical side directly, noting Cerutti, Claessens and Rose (2017), “whose findings indicate that global factors do not explain more that 25 per cent of capital flows variation across countries” (Section 2).

50 countries – 18 advanced, 32 emerging – quarterly from 1990Q1 to 2017Q4, with a set of financial centres excluded and gross rather than net inflows used. The exclusions are stated and justified: “A number of financial centres including Cyprus, Ireland, Hong Kong, Luxembourg, Malta, as well as Belgium and the Netherlands have been excluded as their cross-border capital flows record extremely large values with respect to GDP and are very volatile” (fn. 4). The United States, United Kingdom and Switzerland are kept in the baseline on substantive grounds – “we believe it is important to keep these large financial centres in the sample as their global banks have been the main conduit of the global financial cycle” – but results excluding them are also reported (fn. 7). The flow concept is gross inflows, that is “net purchases by non-residents of securities issued by domestic residents of a country,” chosen “to capture common trends across countries, not on ’net’ flows that can offset each [other]” (fn. 5). Because gross flows are “large and very volatile,” in some cases peaking above 100 percent of GDP in a quarter, dependent variables are winsorised at the 1 percent level (Section 3). Capital account openness is measured either by the updated Chinn-Ito de jure index or by a de facto measure based on total external liabilities from the extended Lane-Milesi-Ferretti dataset; exchange rate regimes come from Obstfeld, Shambaugh and Taylor (2010) or Ilzetzki, Reinhart and Rogoff (2017), and the two differ materially in how much flexibility they see – “the Ilzetzki et al. (2017) classification includes a relatively low share of flexible exchange rates compared to the one by Obstfeld et al. (2010), 14% against 40% respectively.” Both policy variables are annual and interpolated to quarterly with a cubic spline (fn. 6).

It is the first principal component of country-average stock returns for 63 economies, and the paper’s own regressions show it is the only risk proxy that moves every category of capital flows. The authors emphasise parsimony as a feature: the factor “can be computed with a fraction of the data, around 60 time series, as opposed to the over eight hundred series used by Miranda-Agrippino and Rey (2015), and with a simpler econometric procedure (a simple principal component analysis as opposed to a hierarchical dynamic factor model)” (Section 4). Its validity is argued by co-movement with three other published global-cycle measures (Miranda-Agrippino and Rey, Bonciani and Ricci, Scheubel et al.), with financial risk proxies (the Gilchrist-Zakrajsek Excess Bond Premium, VIX/VXO, VSTOXX) and with uncertainty and geopolitical measures (Baker, Bloom and Davis; Caldara and Iacoviello). The narrative fit is described in terms of two long risk-appetite episodes: “two long periods of elevated risk-appetite, namely the second half of the Nineties and the period between the 2001 recession and the great financial crisis, both followed by large spikes in risk aversion” (Section 4.1). The empirical case for preferring it is decisive in the paper’s own terms: replacing it with the VIX leaves “a statistically significant impact only in the case of portfolio (equity and debt) flows, not on other categories or total capital flows,” and the Excess Bond Premium and the Geopolitical Risk Index behave similarly, so “our Global Stock Market Factor or similar global factors such as the one by Miranda-Agrippino and Rey (2015) or Bonciani and Ricci (2018) are the only proxies of risk impacting all categories of capital flows” (fn. 16).

One by external instrument, three by sign restrictions, in a seven-variable Bayesian VAR. The information set is “four US variables (the interest rate on the one-year Treasury bill, the log of the Consumer Price Index, the log of the S&P500 index and of the US dollar index) and three global variables (the yield of an US dollar High-Yield Corporate Bonds index, the log price of oil and the Global Stock Market Factor)” (Section 4.2). The monetary policy shock uses the Jarocinski-Karadi and Gertler-Karadi logic: “changes in short-term interest rates (so called interest rate surprises) in a short window around US monetary policy announcements are correlated with monetary policy shocks but uncorrelated with other shocks,” so instrumenting the interest-rate equation residual with those surprises recovers the shock. The other three follow Rubio-Ramirez, Waggoner and Arias (2016) sign restrictions. A US demand shock lowers short rates and stock prices, inflation and oil prices, weakens the dollar, and raises global risk. A financial shock – “an exogenous tightening of financial conditions independent of monetary policy,” following Cesa-Bianchi and Sokol – lowers short rates (via higher bond prices) and equity valuations, softens inflation and oil, raises risky bond yields, and appreciates the dollar as a safe haven. A geopolitical risk shock raises global risk and oil and consumer prices while equity valuations and interest rates fall and the dollar appreciates, giving it “a stagflationary flavour that distinguishes it from both demand and financial shocks,” with the authors noting these signs are compatible with narrative-identified uncertainty shocks in Piffer and Podstawski (fn. 12).

About a fifth of global risk variance is US monetary policy, but financial shocks account for more, and the gap widens once model uncertainty is taken into account. The paper first notes the question is new: prior structural work “convincingly shown that US monetary policy is transmitted to the global economy also by affecting global risk, but they have stopped short of providing a quantification of how much monetary policy really matters for global risk when compared to other potential disturbances” (Section 4.2). At the twelve-month horizon, with results reported as the mean plus the 15th and 85th percentiles of the Bayesian distribution: “around one fifth of the fluctuations in global risk at medium-term horizons are indeed due to US monetary policy,” and “financial shocks actually matter more than US monetary policy for global risk fluctuations. This is all the more evident if one looks not only at the mean effect (23% as opposed to 19%) but properly takes into account model uncertainty and considers other percentiles as well. At the 85th percentile, financial shocks can account for up to around 70% of the forecast error variance of global risk, whereas US monetary policy cannot explain more than around 30%.” Geopolitical risk shocks explain “13 percent on average, and up to 26 percent once we move to higher percentiles,” while US demand shocks “explain less than 10% of its overall [variance].” Results for longer horizons are reported as very similar (fn. 13).

An episode in which monetary policy was pushing global risk the other way – which is the paper’s sharpest argument that the cycle is not simply made by the Fed. “Quite strikingly, there are some instances in which movements in the Global Stock Market Factor are not happening because of US monetary policy shocks, but rather despite monetary policy pushing global risk in a different direction. Consider, for instance, the 2003-2008 period. After loosening the monetary policy stance in response to the brief recession that followed the collapse of the stock market bubble, the Fed embarked in a tightening cycle that was interrupted only by the inception of the Great Financial Crisis. During this period, despite tightening monetary policy, global risk fell (i.e. global appetite for risk increased) and capital flows actually surged. The fall in Global Risk before the crisis as well as its spike during the crisis, is accounted for in our framework by a financial shock, i.e. exogenous changes in the appetite for risk unrelated to monetary policy. After the crisis erupted, US monetary policy was quickly and substantially loosened, counteracting the spike in Global Risk triggered by the financial shock” (Section 4.2). The exercise is explicitly framed as adjudicating a live dispute – “we would like to know how much of the decline in global risk before the Great Recession was actually due to US monetary policy being too loose, as some commentators have argued, or driven by a genuine appetite for risk.”

A pooled fixed-effects panel with four lags of the dependent variable, domestic pull factors and global push factors, estimated with Driscoll-Kraay standard errors – and deliberately without time fixed effects. Pull factors are real GDP growth and inflation; push factors are lagged world GDP growth, global risk, and US monetary policy measured by Gertler-Karadi surprises (Section 5.1). The omission is a design necessity rather than an oversight: “Time (quarter/year) fixed-effects are not included, since they would preclude the identification of push factors, in particular our proxy of global risk, which do no not vary across groups” (fn. 15) – and when the paper later turns to interaction terms, which do vary across countries, it does add time dummies as a robustness check. The choice of monetary policy measure is defended empirically against three alternatives – the level and change in the effective fed funds rate extended with the Wu-Xia shadow rate, and the Baker et al. monetary policy uncertainty sub-index: “monetary policy surprises proved to be the most robust regressor for capital flows among these different measures. In particular, we did not find a robust connection between the level or the change in the US Fed Funds Rate and capital flows. Possibly, interest rates, even in the United States, are endogenously determined by global financial conditions and it is therefore necessary to isolate exogenous monetary policy shocks. Monetary policy surprises have the additional advantage of being clearly exogenous” (fn. 14).

Modest, and the paper says so plainly before arguing the average conceals what matters. “For one standard deviation shock in global risk, the decline in gross inflows ranges from 0.1% of GDP in the case of equity to 0.8% of GDP in the case of other investment. Overall, gross capital inflows decline by 1.7% of GDP on average across our panel of countries against a global risk shock. These numbers are rather small when compared to the volatility (14%) of capital flows” (Section 5.1). The authors immediately flag why the average is the wrong statistic for their purpose: “This is naturally the average impact across a number of economies with different characteristics that may influence the transmission of risk shocks, such as capital account openness or the exchange rate regime.” They also calibrate the shock size against history: one standard deviation “corresponds broadly to the change in our Global Stock Market Factor during the Russian sovereign debt crisis and the ensuing collapse of Long-Term Capital Management in 1998Q3, the trough of the US bear stock market in 2002Q3, the Bear Stearns bail-out in 2008Q1, the global financial crisis in 2008Q3 and Q4 and the euro area sovereign debt crisis in 2011Q3” (fn. 17). The endogeneity of the risk factor is addressed with a two-step system GMM estimator treating the factor as endogenous and controls as predetermined, which yields results “very similar to our benchmark fixed-effects estimates” – though notably “the estimated impact of a standard deviation shock in global risk on total capital flows using the system-GMM estimator is much larger (2.9% of GDP).”

It shows up in portfolio flows but not in bank-type flows or totals – and the paper argues this is consistent with, not contrary to, the global-banking literature. “US monetary policy surprises have a negative impact on total capital flows, in particular statistically significant for portfolio flows, as one would expect from the theoretical literature on the global financial cycle… However, we do not find a statistically significant impact of US monetary policy surprises on ‘other investment’, which includes cross-border loans, to a large channeled through banks, and on ’total flows’. This is only apparently in contrast with the findings of Bruno and Shin (2015a). As shown in the next subsection, US monetary policy surprises may still influence cross-border loans through their impact on global risk” (Section 5.1). Replacing the factor with the structural shocks themselves confirms the ordering: “financial shocks have the strongest impact on capital flows, in general robust across different categories of flows,” while US monetary policy shocks are significant “in particular for portfolio flows” (Section 5.2). The indirect route is where US policy reappears: using the shocks’ historical contributions to global risk rather than the shocks directly, US monetary policy “becomes relevant for direct investment and total capital flows. In other words, US monetary policy can affect these flows only to the extent that it induces significant shifts in global risk.” The introduction states the resulting claim with its hedges intact: “global risk is also driven by other shocks, in particular financial shocks, and has a large idiosyncratic component, so that US monetary policy may be considered neither as the unique nor as the main factor behind the global financial cycle, at least as regards capital flows.”

By interacting the risk factor and the monetary policy surprise with openness and peg dummies, with the sign of the interaction coefficient distinguishing trilemma from dilemma – and with two explicit corrections for regime endogeneity. The decision rule is stated precisely: a negative interaction coefficient “would signal that those economies with that particular feature, i.e. capital account openness or a rigid exchange rate, are more sensitive to risk or monetary policy shocks compared to the rest of the sample (trilemma). If, instead, the coefficient theta is negative for capital account openness, but not statistically different from zero for the exchange rate regime, this means that the latter does not matter for the transmission of shocks (dilemma between capital account openness and monetary autonomy)” (Section 6.1). On endogeneity: “There is an issue of potential endogeneity of the policy variables with respect to monetary policy and risk shocks, which may force an adjustment in the prevailing policy regime. To deal with these endogeneity concerns, following Obstfeld et al. (2018), we take the measures of capital account openness lagged by four quarters… and we exclude from the sample all the episodes of currency, banking and sovereign debt crises, as classified by Laeven and Valencia (2013)… when changes in the policy regime are more [likely].” The theoretical prior being tested is spelled out first: a shock changing financial conditions in the centre country “would force those countries with an open capital account and a fixed exchange rate regime to follow the monetary and financial conditions of the centre economy, otherwise capital flows would force a readjustment. Instead, flexible exchange rate regimes would be shielded as the exchange rate would absorb the divergence in interest rates or risk premia.”

Openness and the regime both matter for risk shocks but neither matters for US monetary policy surprises, and the regime effect is concentrated in other investment and direct investment rather than portfolio flows. “Table 9 provides two clear policy messages. First, capital account openness and the exchange rate regime do matter for the transmission of global risk shocks to capital flows, but do not matter for the transmission of US monetary policy shocks. The interaction terms between our global risk factor and the policy variables are negative and statistically significant… This is not the case for the interaction terms between US monetary policy surprises and policy variables. Second, more rigid exchange rate regimes, in particular ‘strict pegs’ are associated with a stronger transmission of risk shocks to ‘other investment’ and ‘direct investment’, not necessarily to portfolio flows. The absolute value of the coefficient of the interaction term between risk shocks and the dummy controlling for strict pegs is particularly large in the case of ‘other investment’, suggesting that this particular category, which includes cross-border bank loans, is behind the trilemma” (Section 6.1). An anomaly is reported rather than suppressed: “the coefficient associated with the Global Stock Market Factor is sometimes positive and significant, suggesting that for sufficiently low values of the Chinn-Ito index… and flexible exchange rate regimes… risk shocks may be even associated with a ‘rise’ in capital flows.” The authors check this directly and report that “when re-estimating the baseline model… within subsamples including only countries with low values of the Chinn-Ito index and flexible exchange rate arrangements, the coefficient associated with global risk is not statistically different from zero or, again, negative” (fn. 19).

Robust across eight separate cuts, including the one that removes all invariant push factors in favour of time dummies. Substituting the de facto openness measure – total external liabilities to GDP – for the Chinn-Ito index gives results that “are virtually identical.” Adding direct exposure to US portfolio investment, built from TIC bilateral data, produces a coefficient “not statistically different from zero, suggesting that it is financial openness per se and not the particular exposure to the United States that matters for the transmission of risk shocks” (Section 6.2). Splitting advanced from emerging economies “does not weaken the support for the policy trilemma”; excluding euro area economies, which the de facto classifications label as pegged, leaves it present; excluding the United States, United Kingdom and Switzerland does not affect the conclusions; excluding 2008-09 does not; and reintroducing the banking, currency and sovereign debt crisis observations “brings no substantial changes.” The most demanding version drops the invariant push factors entirely and substitutes time dummies to absorb any omitted global factor, under which “the coefficients of the interaction terms between risk and capital account openness and between risk and strict pegs are again negative and statistically significant.” Swapping the Ilzetzki et al. classification for Obstfeld et al. gives the same answer, “showing that the trilemma is robust to the choice of this particular classification.”

Large enough to look like a sudden stop for open, pegged emerging markets – which is the paper’s answer to the sceptics. “Indeed, we noted that the estimated average impact of one standard deviation global risk shock is rather modest compared to the size and volatility of capital flows. However, this average effect masks important differences across economies that come to the surface once we identify the channels of transmission and isolate the policy trilemma” (Section 7). The escalation is stepwise: “controlling for capital account openness the average impact of risk shocks on total capital inflows increases by one percentage point of GDP from 1.7% to 2.8% of GDP. For those economies that have liberalised the capital account and adopt an exchange rate target, the impact augments by more than one percentage point, up to 4.0% of GDP. This impact becomes economically significant, when compared to the volatility of capital inflows in the sample (13.8% of GDP).” For emerging markets the figures are starker: “In emerging markets that are open and fixing the exchange rate, a one standard deviation risk shock leads to a decline in total gross capital inflows by more than 4% of GDP, four times larger than the average impact across all emerging markets (1% of GDP). Considering that an emerging market would typically receive an inflow of almost 7% of GDP on average (the sample mean), a risk shock would lead to a ‘sudden stop’ by foreign investors.” In both cases “‘other investment’ is the category explaining the quantitative impact of the trilemma.”

Through time variation in the policy regimes themselves: as openness and pegging both rose, the aggregate sensitivity of flows to risk rose with them. The authors pose the tension explicitly – “one may wonder whether there is a contradiction between the tight link between our Global Stock Market Factor and gross capital inflows, lending support to the presence of a policy dilemma (Rey, 2015), and the results… finding evidence of a classical trilemma. Not necessarily, as the policy regimes may change through time, reinforcing or weakening the link between global risk and capital flows” (Section 7). The evidence offered is that average de jure liberalisation and the share of countries on strict pegs “both tend to increase over time” across the 1990s, the 2000s and the post-2010 period, and “the higher these two measures, the stronger the expected sensitivity of capital flows to global risk according to the trilemma. Indeed, this is exactly what we find once we re-estimate our model across these three different periods… the impact of global risk on capital flows since 2010 is three times as large as the impact back in the 1990s.”

Because for portfolio investors the adjustment to a risk shock happens in prices rather than quantities, and because currency risk is either second-order or hedged. The paper gives four reasons in sequence (Section 7). “For an equity investor, the volatility of the exchange rate is a factor of second-order importance compared to the volatility of equity prices, therefore it is not surprising that the exchange rate regime would not matter.” For a local-currency bond investor exchange rate volatility does matter, but “this is however a typical investment for sophisticated portfolio investors, who may recur to derivative instruments to hedge the currency risk.” More generally, bond investors avoid currency risk by holding their own currency – “the ‘home-currency bias’ identified by Maggiori, Neiman and Schreger (2018).” And finally the arbitrage argument: “under financial integration, international arbitrage leads to a rapid adjustment of prices and returns across internationally traded assets, equalising borrowing costs without the need to generate large adjustments in capital flows (Dedola and Lombardo).” The converse explanation for why loans carry the trilemma cites Basso, Calvo-Gonzalez and Jurgilas on loan dollarisation: “they show that lending rates between two different currencies may diverge even for countries with hard exchange rate pegs, in violation of the uncovered interest parity. In this case, domestic residents may underestimate the implicit exchange rate risk of borrowing in foreign currency in tranquil times, leading to credit booms that turn into busts and capital flows reversals when volatility rises.”

That US monetary policy must be traced through global risk rather than modelled as acting directly on flows, and that the shift from bank to market-based finance may make risk shocks propagate faster. On the modelling point: “we claim that it is important to isolate the contribution of US monetary policy shocks to global risk to understand its international [transmission]. It may be difficult to establish a direct link between US monetary policy and capital flows, without ‘passing through’ global risk” (Non-technical summary; Section 8). On domestic policy: “we show that domestic monetary and exchange rate policies may influence the transmission of global risk to capital flows. This is especially true for loans, which are particularly sensitive to deviations in the uncovered interest parity and whose nominal value is not affected by risk shocks.” The financial stability implication is forward-looking and expressed as a warning rather than a finding: “Since the role of market-based finance is on the rise – also among emerging markets – at the expenses of that of global banks, our results call for a careful assessment of the financial stability implications of global risk shocks. As the composition of global liquidity shifts away from bank loans to other sources of financing, such as equity and bonds, sudden shifts in investors’ risk attitude can in fact propagate faster than in the past.”