Exchange rates and the transmission of global liquidity

The net exports channel and the financial channel, and they work in opposite directions, so the exchange rate concept that matters differs between them (Introduction, p. 3). The real effects operating through net exports “are well-known and are standard in open economy macro models,” and there activity picks up when the domestic currency depreciates. “By contrast, the financial channel operates through the liabilities side of the balance sheet of domestic borrowers, so that it is when the domestic currency appreciates that balance sheets strengthen and economic activity picks up” (p. 3). The consequence the paper draws out is methodological: the trade-weighted effective exchange rate is the right object for the net exports channel, but “the relevant exchange rate for the risk-taking channel is that with the international funding currency – almost invariably the US dollar, but also increasingly the euro” (p. 4, citing Hofmann, Shim and Shin 2016). That wedge between the two exchange rate concepts “provides a window for a reconciliation of the risk-taking channel with the net exports channel, and permits an empirical investigation that disentangles the two channels.”

The paper sets out both a demand-side and a supply-side reason, and its empirical design is built to tell them apart (Introduction, pp. 3-4). On the demand side, a borrower with dollar liabilities and domestic-currency assets sees its balance sheet strengthen when the dollar depreciates; separately, an exporter with dollar receivables or an asset manager with dollar assets but domestic-currency obligations “would hedge currency risk more aggressively when the dollar is expected to depreciate further,” and incurring dollar liabilities or the equivalent off-balance-sheet transaction is how that hedge is executed. On the supply side – the risk-taking channel named by Bruno and Shin (2015b) – “a weaker dollar flatters the balance sheets of dollar borrowers, whose liabilities fall relative to assets,” and from the creditor’s side “the stronger credit position of the borrowers reduces tail risk in the credit portfolio and creates spare capacity for additional credit extension even with a fixed exposure limit through a value-at-risk (VaR) constraint or economic capital (EC) constraint.” The mapping to data follows directly: bilateral rates for borrower net worth and hence demand, the broad currency index for global banks’ portfolio VaR and hence supply.

BIS Locational Banking Statistics, quarterly, with the currency composition of banks’ balance sheets and country and sector breakdowns of counterparties (Section 4, pp. 10-11). The locational statistics “capture outstanding claims and liabilities of banks located in BIS reporting countries, including intragroup positions between offices of the same banking group,” are compiled on balance-of-payments-consistent principles, and yield break- and exchange-rate-adjusted changes in amounts outstanding as an approximation for flows. Lending covers loans and holdings of debt securities. Samples: US dollar, Q1 2002 to Q3 2015, 106 borrower countries; Japanese yen, the same window, 114 countries; euro, Q1 2010 to Q3 2015, 93 countries. In each case “the set of counterparty countries excludes the country of the currency denomination itself, and all countries whose exchange rate is pegged to the respective currency denomination” (Table 1 notes, p. 11), and growth rates are winsorized at the 5 percent level in each tail. Two auxiliary series: the broad BIS nominal effective exchange rate indices, geometric trade-weighted averages covering 61 economies; and, because policy rates were at the zero lower bound for much of the window, changes in the Krippner (2015) two-factor shadow policy rate as the monetary stance measure.

Periods of funding-currency weakness line up visibly with faster growth in lending denominated in that currency, most clearly for interbank flows (Section 2, pp. 6-7). Graph 2 plots cumulative cross-border lending flows in each currency to borrowers outside the respective currency area, shading quarters in which that currency depreciated in trade-weighted terms; for the dollar “those dark shaded bars … tend to be associated with steeper growth in cross-border lending in US dollars,” a pattern holding for both interbank and non-bank lending but “much more pronounced for the former category, suggesting that bank-to-bank flows are most sensitive to exchange rates” (p. 6). The yen shows a similar picture with interbank again most sensitive. The euro is the outlier: euro-denominated cross-border lending outside the euro area “grew steadily during the pre-crisis period” and “this growth appeared to be uncorrelated with the strength of the euro, which actually appreciated during most of that period,” with a negative relationship emerging only post-crisis. The descriptive statistics add scale: average quarterly growth of cross-border lending was 2.2 percent in dollars, about 1.1 percent in yen and about 1.5 percent in euros, with yen lending much the most volatile, and with lending to banks both faster-growing and more volatile than lending to non-banks in all three currencies (Table 1, pp. 10-11).

A 1 percent dollar depreciation is associated with roughly a 0.63 percentage point contemporaneous increase in the quarterly growth rate of dollar-denominated cross-border lending to borrowers outside the United States (Section 2, pp. 7-8). The fitted coefficients are -0.629 (p = 0.000) for the dollar and -0.61 (p = 0.000) for the yen over Q1 2002 to Q3 2015, and -0.636 (p = 0.011) for the euro over Q1 2010 to Q3 2015 – so “the magnitude of the estimated impact for the Japanese yen … is similar to the one for the US dollar,” and the euro’s post-crisis estimate “is also in the same range of magnitude” (p. 8). Over the full window the relationship holds for the dollar and yen “but not for the euro,” which the authors read as “early evidence for the emergence of the euro as a major global funding currency” only later.

Yes – in borrowing-country-specific regressions most coefficients are negative for both bank and non-bank lending, and statistically significant for one or both sectors in most countries (Section 5.1, pp. 12-13). Regressing each country’s quarterly growth of dollar-denominated cross-border lending on the relevant exchange rate change, “the majority of the borrowing country-specific coefficients are situated in the lower left-hand quadrant of Graph 4, indicating a negative relationship for lending to both bank and non-bank borrowers,” with significance at the 10 percent level for banks, non-banks or both across most of the sample (p. 12).

That the credit-supply component is the stronger one: the broad dollar index coefficient is about -0.50, against -0.22 for the bilateral rate, and -0.15 for the bilateral rate’s orthogonal component once the index is included (Section 5.1, pp. 13-15, Table 2). In the full sample of 106 countries with borrower-country fixed effects and standard errors clustered at the borrower-country level, an appreciation of the dollar against a given country’s currency is associated with a statistically significant decline in cross-border lending to that country (-0.224, significant at 1 percent), a rise in the broad dollar index likewise (-0.496, 1 percent), and both remain significant when entered together (-0.498 for the index, -0.148 for the orthogonalized bilateral rate). The authors’ own framing of the comparison: “the impact of a one percentage point increase in the broad US dollar index on cross-border bank lending (50 basis points) is more than twice as large as the respective impact of a one percentage point rise in the bilateral exchange rate value of the US dollar against the currency of the borrowing country (22 basis points),” rising to more than threefold in the joint specification (p. 15). This pattern “persists across both borrowing country groups (AEs, EMEs, OFCs) and borrowing sectors (banks and non-banks).” Worth noting for calibration: the R-squared values in these panels are low – around 0.03 to 0.05 in the full sample – so the exchange rate is a statistically robust but not a dominant driver of quarter-to-quarter lending growth.

Yes, and it is read as evidence for a core-periphery structure of international banking (Section 5.1, p. 15). “In most cases, the estimated coefficients in the regressions for bank borrowers tend to be larger than the respective coefficients for non-bank borrowers” – in the full sample, -0.664 on the broad index for bank borrowers against -0.295 for non-banks. The authors attribute this to “the greater procyclicality of bank balance sheets,” consistent with the “double-decker” model of Bruno and Shin (2015b) “in which large global banks in the core of the system provide cross-border funding to periphery banks, which in turn lend to local non-bank borrowers.”

Yes, and Latin America is the documented exception (Section 5.1, pp. 15-16, Table 3). For emerging Asia and emerging Europe both the broad dollar index and the bilateral rate are negative and strongly significant, with the index coefficient the larger; for Africa and the Middle East the index is significant while the bilateral rate is not. “In contrast to the other EME regions, Latin America appears to be more affected by the credit demand component than by the credit supply component of the financial channel of exchange rates. The estimated coefficients on the bilateral exchange rate and on its orthogonal component are statistically significant, while those on the broad dollar index are insignificant” (p. 15). The sector breakdown localizes it further: this pattern “is driven by lending to non-banks rather than by interbank lending,” which the paper reads as fitting “the intuition that the former should be more directly affected by a deterioration in the creditworthiness of currency-mismatched borrowers triggered by a depreciation of the local currency against the US dollar.” Offshore financial centres, which the authors note “typically [do] not conform to patterns that are present for other types of lending,” in fact show the largest coefficients of any group (-0.543 bilateral, -0.651 index).

By a Cholesky decomposition in which cross-border lending is ordered ahead of the exchange rate, so FX shocks have no contemporaneous effect on lending – an ordering the authors say works against their own hypothesis (Section 3, pp. 9-10; Section 5.1, p. 15). The four endogenous variables are the change in the shadow policy rate, the growth of cross-border lending, log VIX, and the exchange rate (bilateral or index), following the variable ordering in Bruno and Shin (2015a). “Most importantly, in all structural panel VAR specifications that we explore, the cross-border lending variable is ordered ahead of the FX variable. This rules out any contemporaneous effects of the FX rate on cross-border lending, thus tilting the odds against us finding the results predicted by the theoretical model of Bruno and Shin (2015b)” (p. 15). Because lagged dependent variables in a dynamic panel make OLS biased (Nickell 1981), the system is rewritten in first differences; impulse responses follow Lütkepohl (2007), with standard errors computed nonparametrically from 1,000 simulation replications over a 10-quarter horizon and 95 percent confidence bands. This is an identifying assumption, not a natural experiment: the negative responses are what the model implies given that recursive ordering.

Negative, statistically significant and persistent responses, with the broad-index shock both deeper and longer-lasting than the bilateral shock (Section 5.1, pp. 15-19, Graphs 5-7). The negative impact holds for lending to all sectors, to banks, and to non-banks, and “the estimated negative impact is quite persistent, remaining statistically significant for six to eight quarters after the occurrence of the shock in the case of the bilateral exchange rate and over ten quarters in the case of the broad dollar index” (p. 15). The supply-over-demand ranking recurs: “the contractions in cross-border lending triggered by an increase in the broad dollar index are both deeper and more persistent than those caused by an appreciation of the US dollar bilateral exchange rate,” and a dollar appreciation again hits interbank lending harder than non-bank lending. The pattern survives disaggregation: significant and persistent contractions for advanced economies, emerging markets and offshore centres, with the supply component dominant in all three, and “virtually all aspects of our main results are present in all four major EME regions” (p. 19).

Similarly in the panel regressions, but with smaller and much less persistent dynamic effects, concentrated in emerging Asia and offshore centres (Section 5.2, pp. 20-23). The yen panels show a strong negative relationship at the global level (-0.222 bilateral, -0.411 index, both at 1 percent) and for advanced economies, emerging markets and offshore centres alike. By region, emerging Asia (-0.530 bilateral, -0.622 index) and Latin America (-0.222, -0.425) are negative and significant, while “the estimated impact of fluctuation in the exchange rate value of the yen on cross-border lending to emerging Europe and Arica and the Middle East is insignificant” (p. 22). In the SPVAR, full-sample impacts are negative and significant for both exchange rate measures, but “the estimated impacts are somewhat smaller and considerably less persistent than their US counterparts” (p. 21), and the yen’s funding-currency role “is most evident in the case of bank lending to borrowers in emerging Asia and in OFCs” – unsurprising, the authors note, “given the relatively high share of yen-denominated lending to borrower in those two country groups.” Their summary judgment: “although the yen exhibits many of the properties of an international funding currency, it has more of a regional flavour relative to the US dollar, which is unique in its role as the preeminent global funding currency” (p. 22).

Only after the crisis, only partially, and mainly within Europe (Section 5.3, pp. 23-25). Twenty-quarter rolling-window regressions of euro-denominated cross-border lending growth on the broad euro index show the coefficient “was positive at the beginning of our time window, which saw periods of euro strength coupled with a rapid expansion of euro-denominated cross-border lending,” then moving “slowly, but steadily” into negative territory and becoming statistically significant in the post-crisis period 2010-2015 – driven mainly by interbank rather than non-bank lending (p. 23). In post-crisis panel regressions on 93 borrower countries the relationship is not significant at the global level (bilateral 0.041, index 0.116, both insignificant), but it is significant and negative for advanced European countries outside the euro area (index coefficient -1.014, significant at 5 percent; bilateral -0.669, also at 5 percent) and for emerging Europe through the orthogonalized bilateral rate (-0.116, 5 percent). (The accompanying sentence in the text describes the non-euro-area Europe result as a stronger euro being associated with an “increase” in lending; the tabulated coefficient is negative, so the table’s sign – more euro strength, less euro lending – is the one consistent with the paper’s own thesis and the rest of its results.) The euro SPVARs are weaker still: insignificant for the full sample and most country groups, significant and negative only for emerging Europe and Africa and the Middle East, “for which the share of euro-denominated cross-border lending is relatively high,” and there with the bilateral rate mattering more than the broad index. The paper’s own bounded statement: “during the post-crisis period, the euro has emerged as an international funding currency, at least at the regional (European) level” (p. 24). Notably, the full-sample (2002-2015) euro panel and SPVAR results “largely yield insignificant results” and are not reported in the paper (fn. 6, p. 23).

It supplies the main axis of a four-way “diamond” whose other edges were established elsewhere (Introduction, pp. 4-5, Graph 1). Avdjiev, Du, Koch and Shin (2016) document the triangle formed by dollar strength, cross-border dollar bank lending, and deviations from covered interest parity; Avdjiev, Bruno, Koch and Shin (2017) document another triangle among dollar strength, cross-border bank lending, and real investment. Taken together, the authors argue, these amount to “a diamond-like relationship between four key macroeconomic and financial variables: (i) the strength of the US dollar, (ii) cross-border bank lending in dollars, (iii) deviations from covered interest parity (CIP) and (iv) real investment,” and “the main contribution of our paper is that it provides robust empirical evidence for the existence of the main axis in the above ‘diamond’ – the axis that links exchange rate fluctuations in a given currency and cross-border bank lending denominated in that currency” (p. 4). The paper also situates itself against the global financial cycle literature (Miranda-Agrippino and Rey; Rey 2015), the evidence that capital flow types co-move and correlate negatively with the VIX (Forbes and Warnock 2012), and the international bank shock-transmission literature going back to Peek and Rosengren (1997, 2000).

It claims a robust, replicated negative association that it interprets as the risk-taking channel; it does not claim a causal experiment (Introduction, p. 4; Conclusion, pp. 25-26). The strongest statement in the paper is about robustness across methods: “The results obtained in all of the above empirical settings conclusively point to a robust negative relationship between the value of a given funding currency and cross-border bank flows denominated in that currency.” The inferential step is then flagged as an interpretation: “We interpret these findings as evidence for the existence of a risk-taking channel of currency fluctuations” (p. 4). The conclusion restates the core finding in association language – “exchange rate fluctuations of the main international funding currencies are closely tied to fluctuations in cross-border bank lending denominated in those currencies” – and says the results “corroborate the existence of the risk-taking channel of currency appreciation in the spirit of Rey (2015) and Bruno and Shin (2015b)” (p. 26). The supply-versus-demand decomposition rests on the interpretation of the two exchange rate measures rather than on separate instruments for supply and demand, and the dynamic evidence rests on the recursive SPVAR ordering described in Q10.