Macroprudential FX regulations: Shifting the snowbanks of FX vulnerability?

Because FX exposure kept rising even as concern about it grew, and because the tool itself was hard to evaluate for both data and identification reasons (Section I, pp. 1-3). The authors note that while countercyclical capital buffers, reserve ratios, leverage ratios and loan-to-value limits have all been analyzed, FX regulations “received less attention, despite the long-standing research documenting the vulnerabilities associated with currency mismatch” (p. 1). The scale of the underlying exposure grew over the sample: “over our sample period from the mid-1990s through the end of 2014, total FX borrowing in international debt securities and bank loans more than tripled to about $12 trillion USD,” and cross-border FX borrowing rose by around $2.5 trillion between 2009 and 2015 – “a sharp contrast to almost no change in comparable cross-border borrowing in local currency over the same period” (pp. 1-2). Two obstacles had blocked evaluation: “insufficient data linked to the limited experience with these tools until recently,” and “identification challenges related to macroprudential regulations being introduced as a response to financial and macroeconomic developments” (p. 2). The authors also state the ambition explicitly: a proper evaluation must cover “not only the direct effects of these measures on the intended sector of the economy (such as banks), but also any spillovers or leakages.”

A screening asymmetry between banks and markets that interacts with currency choice, so that regulating banks’ FX exposure reallocates FX credit toward lenders who cannot screen (Section II, pp. 24-44 of the working paper; summarized on pp. 2-3). “Domestic firms seek funding from lenders, but have private information about their productivity. Banks can screen firms at a cost and identify unproductive, low-productivity, and high-productivity firms, while market investors can only lend indiscriminately. Funding in foreign currency is cheaper than in domestic currency, but subject to exchange rate risk. When the domestic currency depreciates, low-productivity firms and their associated banks default” (p. 2). Regulation enters through one of two prices: it “increases banks’ cost of funding in foreign currency (if the regulation is a liability-side measure) or the equilibrium lending rate to firms (if an asset-side measure).” The consequence is a reallocation, not simply a contraction: “Total factor productivity declines, as FX lending shifts to investors (who cannot screen) and a share of FX lending shifts to less productive firms.”

No – it says the welfare effect is ambiguous, and the paper does not resolve it (Section II, p. 3; Conclusions, p. 39). “The overall impact on welfare is ambiguous and reflects a trade-off between two forces: FX regulation provides the benefit of reducing the social cost of bank failure after depreciations at the cost of reducing output due to the less efficient allocation of FX lending” (p. 3). The conclusion is equally direct that the empirical work does not close the gap: “the analysis in this paper does not provide a full cost-benefit calculation of the impact of macroprudential regulations – either in the theoretical model or the empirical analysis,” and it names further costs left out, “such as on the distortions created as firms, banks, and individuals find other ways to reduce the impact of the regulations (perhaps by shifting business to other countries with a different regulatory framework)” (p. 39).

Nothing about the main predictions, except that exporters default less often after a depreciation (Section II.E, pp. 44-46). Because exporters tend to be larger and more productive (Bernard et al. 2003; Melitz 2003), the extension makes some high-productivity firms exporters with a fraction of output priced in foreign currency – a natural hedge – and makes exporter status publicly observable, so investors can infer their productivity. That makes market funding in foreign currency cheaper for exporters than bank funding in either currency, so “exporters prefer to obtain funding from investors in F, irrespective of macroprudential FX regulation.” The stated result: “All of the key implications from the model developed above continue to hold, except that now exporters default less often after a depreciation of the domestic currency,” partly from the natural hedge and partly from cheaper funding.

132 regulatory changes across 48 countries from 1995 to 2014 at quarterly frequency, built by merging four existing sources, with capital controls excluded by construction (Section III.A, pp. 47-55). The definition is currency-based: “regulations that discriminate based on the currency denomination of a capital transaction,” which “do not include capital controls – which discriminate by the residency of the parties involved in the transaction – although there is substantial overlap in these two types of measures given that transactions between residents and non-residents are more likely to involve FX” (p. 15). The four sources are Shim et al. (2013) on housing-related policy events for 60 countries monthly 1990-2012; Vandenbussche et al. (2015) on 16 emerging European countries 1997-2010; Cerutti et al. (2017) on 12 macroprudential policies in 64 countries 2000-14; and Reinhardt and Sowerbutts (2018) on 60 countries from 1995. The sample excludes reserve-issuing countries – “long-standing members of the Euro Area, the US, Switzerland, and Japan” – “to focus on countries more vulnerable to currency mismatches and the global financial cycle,” and excludes BIS-defined offshore centres apart from Singapore and Hong Kong, leaving 48 countries: 17 advanced and 31 emerging, with emerging markets accounting for “the vast majority of FX regulatory actions.” Coverage is good for Asia, Europe and South America and “more limited for the Middle East and Africa” (fn. 19, p. 16). By the end of the sample the cumulated counts are 30 liability-side and 37 asset-side measures. Two candour notes worth carrying: the measure is a -1/0/+1 dummy, “the standard treatment in the literature, because it is impossible to compare discrete changes in different types of regulations,” and the paper declines to disaggregate further because “the sample size becomes too small to yield meaningful results.”

A quarterly country panel of gross cross-border inflows scaled by annual GDP on the contemporaneous regulation dummy plus three lags, with country and global-time fixed effects, evaluated as the sum of the four coefficients (Section IV.A, pp. 62-68). The scaling choice is deliberate: inflows are divided by a four-quarter moving average of annual GDP “because the sum of the contemporaneous coefficient and three lags on fxm reported in the regression tables can then be read as the effect on capital flows to annual GDP over a one-year period” (fn. 23, p. 20). All independent variables are lagged one quarter (GDP growth one year) “to reduce endogeneity concerns,” and all variables are winsorized at 2.5 percent. The global-time effects carry a methodological argument: global factors are known to drive capital flows (Forbes and Warnock 2012; Rey 2013; Avdjiev et al. 2016a) “but there are different views on which factors are most important,” and “by controlling for a global-time fixed effect, we do not need to take a stance on exactly which global factors are important.” As a check, specifications substituting explicit global variables – global volatility, global growth, changes in US interest rates – “have no meaningful impact on the key results,” with the expected signs (lower volatility, higher global growth and lower US rates going with stronger bank FX borrowing). Six controls are included, chosen for time variation so they are not absorbed by country effects: changes in non-FX macroprudential regulation, real GDP growth, exchange rate volatility, the interest rate differential, the change in sovereign rating, and Chinn-Ito financial openness. An innovation flagged by the authors is that cross-country exposures are weighted by financial rather than trade exposure, following Lane and Shambaugh (2010) and Bénétrix et al. (2015), since many emerging markets are “more exposed to US dollar movements than predicted based purely on trade patterns” – though a sensitivity test shows the weighting “does not impact the key results.”

Cross-border FX loans to banks fall by 0.50 to 0.66 percent of GDP over the following year – roughly a third of the median – with no significant change in non-FX borrowing (Section IV.B, pp. 68-72). The authors put the magnitude in context twice. Against the sample: “FX loans are around 1.9% of GDP at the median of our sample (across quarters when inflows were positive), suggesting that tighter FX regulations correspond to a decline in FX cross-border loans by banks of about one-third.” Against individual countries: “In both [Brazil and Indonesia], FX loans to banks are a little less than 1% of GDP, suggesting that an increase in macroprudential FX regulations corresponds to a reduction in FX loans to banks by over half” (p. 22). The non-result is load-bearing for the argument: “Banks do not significantly increase their borrowing in local currency to compensate for their reduced borrowing in FX,” and the effect on total international bank borrowing is “weakly negative – as expected – but only significant at the 10% level,” implying the FX reduction is not fully offset. Note the verb the results section uses is associational (“are correlated with a significant decrease”), even though the abstract and non-technical summary say “causes.”

Corporate international FX debt issuance rises by 0.05 to 0.06 percent of GDP, about 10 percent of the median, with no significant effect on non-FX issuance (Section IV.B, pp. 72-74). The authors call this “moderate, given that net FX debt issuance is around 0.6% of GDP (at the sample median when net FX debt issuance was positive),” but much larger for specific countries: “in Brazil and Indonesia, FX debt issuance is 0.26% and 0.36% of GDP, respectively, suggesting that tighter FX regulations correspond to roughly a 15% to 20% increase in this issuance” (p. 23). The “shifting” statistic follows from dividing one by the other: “after an increase in FX regulations, about 10% of the decline in FX exposure in banks shifts to corporate debt issuance (and thereby to investors and other non-bank financial institutions).” A footnote raises this to about 13 percent by adjusting for the fact that, on enhanced BIS data, “on, average, 62% of FX loans from banks to non-banks are lent to the corporate sector,” while noting those data exist for only seven countries (fn. 27, p. 24). The authors’ own reading of the net effect: “there is still a meaningful net reduction in aggregate FX borrowing in the economy.”

They function as the paper’s main defence against omitted variables (Section IV.B, pp. 74-77). The additional tests find no significant effect of FX regulations on cross-border loans to non-banks, and no significant effect at the 5 percent level on international debt issuance by banks in any currency – “sharply contrast[ing] with the results for corporate debt issuance.” A supplementary test on domestic lending, using a newer BIS series with acknowledged limits (it “only begin[s] in 2012Q3,” covers “only 30% of the countries in our sample,” and is available only for lending to non-banks rather than corporates), finds a significant reduction in domestic FX lending by banks with no significant effect on non-FX lending. The inferential use is spelled out: “The fact that corporates do not simultaneously increase international debt issuance in domestic currency, and that banks do not significantly increase debt issuance in any currency, also suggests that these results are not capturing some type of omitted variable that would lead to a general increase in international borrowing or debt issuance in foreign currency” (p. 25).

Yes, and the difference is in the leakage rather than the direct effect: only liability-side measures significantly push corporates into FX debt issuance (Section IV.C, pp. 81-88). Both types are associated with a significant decrease in banks’ FX borrowing, with the asset-side coefficient estimated larger but “only significant at the 10% level in column (8).” For corporate FX issuance, “this effect is only estimated to be significant (at either the 5% or 10% level) for liability-side regulations,” and “the magnitude of the coefficient on FX debt issuance is also estimated to be about three times larger for liability-side than asset-side regulations” (p. 26). Redoing the shifting calculation, “an increase in liability-side FX measures causes FX debt issuance by corporates to increase by 16% of the reduction in FX loans by banks (instead of 10% when all FX regulations are aggregated),” or 26 percent with the 62-percent adjustment (fn. 33, p. 27). The policy conclusion is stated as a conditional ranking: “asset-side regulations may better improve a country’s resilience to currency movements, as they decrease bank exposure to currency risk but simultaneously generate less shifting of this risk to other sectors.” Three candidate explanations are offered without being tested against each other: liability-side measures “affect all forms of bank funding in all states of the world” while asset-side measures often reach only segments of lending such as mortgages; asset-side measures in countries like Hungary and Poland are largely FX mortgage LTV and DTI rules, so bite on households rather than corporates; and liability-side measures target shorter-maturity flows, which macroprudential measures burden relatively more. The section opens with its own caveat: “These results should be interpreted cautiously, however, as these finer divisions of different forms of macroprudential FX regulations imply that there are more limited degrees of freedom for the analysis.”

Partially, using two databases with intensity information, at the price of a much narrower sample (Section IV.C, pp. 88-91). Cerutti et al. (2017) carry intensity for FX reserve requirement changes on a sample similar to the paper’s; Vandenbussche et al. (2015) carry intensity for FX liquidity requirements, the maximum ratio of FX loans to own funds, and risk weights on FX mortgage, consumer and corporate loans – but those data “end in 2010 (thereby missing a period of active use of macroprudential FX regulations)” and are “only available for Eastern Europe (thereby severely limiting country coverage).” With those caveats, the estimates are: a large increase in FX reserve requirements (quantitative index value 5) lowers FX loans to banks by 0.46 percent of GDP and raises FX international debt issuance by 0.15 percent; a 10 percentage point rise in the FX liquidity requirement lowers FX loans to banks by 2.9 percent of GDP and raises corporate FX issuance by 0.36 percent; tightening the maximum ratio of FX loans to own funds from 400 to 200 percent lowers FX loans to banks by around 1 percent of GDP and raises corporate FX issuance by 0.05 percent; and raising the risk weight on FX mortgage, consumer or corporate loans by 50 percentage points above local-currency weights lowers FX inflows to banks by 0.26, 0.46 and 1.83 percent of GDP respectively, each with about a 0.07 percent of GDP increase in corporate FX issuance. “All of these estimated effects of specific changes in FX regulations have the expected sign, and several of the estimates (such as for foreign currency liquidity requirements) are consistently significant despite the limited sample size” (p. 28).

By purging the regulation variable in a first stage and re-running the baseline on the residuals, following Auerbach and Gorodnichenko (2013) and Furceri et al. (2016) (Section IV.D, pp. 91-99). The concern is stated concretely: “increased corporate debt issuance in FX could heighten concerns about risks to domestic financial stability related to aggregate FX exposure, causing policymakers to tighten regulations on banks’ exposure to FX” (p. 29). The first stage regresses FX regulation on GDP growth, domestic credit growth, house price growth, financial openness, use of non-FX regulations, expected GDP growth, exchange rate appreciation, and in two specifications lagged values of the baseline dependent variables. The authors report candidly how weak the prediction is: these variables “usually [have] the expected sign, but most are not individually significant. This captures the well-known challenge of predicting exactly when macroprudential regulations are adjusted. The one exception is the coefficient on exchange rate appreciation, which suggests that larger appreciations consistently correspond to increases in macroprudential FX regulations.” Using the residuals as the explanatory variable, “the results support the key conclusions from the main analysis” and are “usually robust across the six different specifications,” the exception being the specification including house price growth, whose limited data shrink the sample. (The working paper’s sentence describing this table states the signs the other way round from every other result in the paper – an apparent transposition; the direction established throughout is that tighter regulation lowers bank FX loans and raises corporate FX issuance.) The further robustness battery – dropping the offshore centres, dropping 2008Q3-2009Q2, using only tightenings, dropping the financial weighting, dropping one country at a time, and adding controls for the current account, institutional quality and aggregate financial exposure – leaves “the main results discussed above … unchanged.”

By letting the sensitivity of stock returns to the exchange rate depend on the cumulated stance of FX regulation, through an interaction term (Section V.A, pp. 101-106). Quarterly stock index returns for a country’s financial sector (proxying banks) or broad market (proxying corporates) are regressed on the growth of a financially weighted exchange rate, the cumulated FX regulation measure over the current and three preceding quarters, their interaction, and controls chosen following Baele et al. (2010) for macro, liquidity, risk-premium and global-volatility channels, with country fixed effects. The marginal effect of the exchange rate is then the sum of the level coefficient and the interaction times the regulatory stance. The predicted signs come from the model: domestic-currency appreciation raises ex-post profits so the level coefficient should be positive, the interaction should be negative if regulation reduces exposure, and “we would expect the coefficient µ to be more negative for banks than for corporates” because corporates can switch to market-based FX borrowing. The proxy limitation is stated up front: “We do not have precise measures of returns for just banks or just corporates,” so financial returns stand in for banks and the overall index – which “includes corporates, banks, and non-bank financial institutions” – stands in for corporates.

A one percentage point depreciation cuts financial-sector returns by 1.46 percentage points at a neutral regulatory stance but only 0.67 points after tightening; for the broad market the fall goes from 1.18 to 0.75 points, and only the bank interaction is significant at 5 percent (Section V.B, pp. 106-115). The sample is up to 24 named countries over 2000Q1-2014Q4, with standard errors clustered by country and the country set limited by availability of financial-sector return data. The exchange rate coefficient is positive and significant throughout, as predicted. The interaction is negative in all four main columns, “but this coefficient … is only negative and significant at the 5% level for bank returns …, and the estimated magnitude of the coefficient is over 50% larger for banks than corporates in each case” (p. 34). The authors’ summary of the magnitudes is that “tighter macroprudential FX regulations reduce the sensitivity of stock returns to exchange rate shocks for both banks and the broader economy, but the effect is almost twice as large for banks (and insignificant for corporates).” When a constructed corporate return series is used instead – the residual from regressing the broad index on the financial index – the interaction is “insignificant and positive,” and the authors decline to read the sign: “Since the coefficient estimate is insignificant, we are cautious about interpreting the sign of the effect on corporates – but instead can conclude that this result suggests any effect of macroprudential regulations on the sensitivity of corporates to exchange rate movements is small and insignificant.” The pattern strengthens where theory says it should: restricting to emerging markets, and separately to large exchange rate moves (below the 10th and above the 90th percentile), leaves signs and significance unchanged “but the estimated magnitudes of the coefficients are all larger.”

Yes, and the authors flag the reason: small firms drop out of the measurement (fn. 47, p. 35). “Smaller firms are more likely to rely on banks for funding, and if FX regulations cause banks to reduce their FX lending to these smaller firms, these firms may be unable to issue debt on international markets. These smaller firms would therefore be forced to reduce their FX borrowing and exposure – whether by shifting to local currency borrowing or not borrowing at all. These effects would not be captured in the empirical analysis, as these smaller firms are also less likely to be included as part of the main equity index.” This cuts against the leakage reading, and the authors put it forward themselves rather than leaving it to a critic.

Higher lending rates and a lower non-performing loan share, with the borrower-count channel present but insignificant – and all three offered as conditional correlations, not tests (Section V.C, pp. 115-118). The model points to three channels: banks’ lending rates rise because of the regulatory tax; the number of firms borrowing from banks falls as higher costs push low-productivity firms toward investors; and the share of non-performing loans falls because banks shed low-quality borrowers. On annual data with country and time fixed effects, and again with real GDP growth and credit growth added as business- and financial-cycle controls, the lending rate coefficients are positive and significant at least at the 10 percent level, the borrower-count coefficients are negative but insignificant (becoming significant at 5 percent when only asset-side measures are used), and the NPL-share coefficients are negative and significant at least at 10 percent. The authors’ calibration of the evidence is explicit: “These results are not definitive empirical tests, however, and should simply be interpreted as conditional correlations which provide supporting evidence for the key channels highlighted throughout this paper” (p. 37).

Three measurement gaps and one missing exercise – and the authors argue the measurement gaps push estimates toward zero (Conclusions, pp. 118-121). The data “include limited information on firm or bank exposure to foreign currency that occurs without crossing borders (such as if a local household makes a bank deposit in foreign currency)”; they “do not incorporate any transactions or changes in exposure that occur entirely through trading or lending in a third country (as often occurs in financial centers)”; and “the analysis does not include information on hedging – whether natural or in financial markets – which could reduce an entity’s vulnerability to currency movements even if it has large gross FX positions” (p. 38). On direction: “Many of these data challenges, however, might be expected to bias estimates toward zero, thereby suggesting some of the effects estimated in the paper could actually be larger if better data existed. For example, if firms respond to tighter FX regulations at home by issuing FX debt abroad and selling it to a foreigner (with the entire transaction in London), this leakage would not be captured in our analysis.” The fourth gap is the cost-benefit calculation (see Q3).

That FX regulations are a usable substitute for capital controls where controls are unavailable, and that the regulatory perimeter is where the residual risk sits (Section I, pp. 4-5; Conclusions, pp. 121 onward). On substitution: vulnerabilities from FX borrowing “have prompted some countries to consider the use of capital controls. Our results suggest that any such countries could instead consider macroprudential FX regulations – especially countries for which capital controls (but not macroprudential FX regulations) are illegal, such as in the European Economic Area and in some trade agreements” (p. 4). On the perimeter, the snowplough metaphor is worked out precisely: moving snow off the road “makes the road system safer,” but “just as the snowplow inevitably pushes a portion of the snow to block your driveway, macroprudential FX regulations can also shift some vulnerability to other sectors that are outside the regulatory perimeter. These other institutions may be harder to monitor and less well informed than banks, less able to screen for the risks inherent in corporate borrowing in FX, and less able to handle subsequent losses after a depreciation” (p. 5). The overall verdict is calibrated to the asymmetry the paper measures: the regulations “can substantially improve the resilience of the banking sector to the global financial cycle,” but “the benefits to the broader economy may be more moderate.”