Taper Tantrums: Quantitative Easing, Its Aftermath, and Emerging Market Capital Flows

Quantification. Earlier work using event dummies or period indicators could establish the direction of US monetary policy spillovers; extracting the magnitude of the surprise from futures prices lets this paper attach dollar figures and distributional predictions to them. The framing is that “previous studies that use period indicator variables or alternative approaches to examine U.S. monetary policy spillovers are only able to make qualitative statements about the direction of impact. In contrast, using derivatives price changes we can quantify the impact of U.S. monetary policy on emerging market capital flows depending on the magnitudes, signs, and dispersion of the extracted monetary policy shocks” (Section 1). The authors set out three specific defects of dummy-based identification: dummies “may fail to include dates that are not widely-recognized as surprises or may miss dates that contain a surprise insofar as rates did not change”; using raw Treasury yield changes “may lead to an attenuated estimated monetary policy effect if the lack of any change is itself a surprise”; and “using dummies to identify a monetary policy shock obscures the magnitude of the shock” (Section 3). They locate themselves among three identification traditions – panel estimation with announcement dummies, structural VARs, and high-frequency identification – and note that “the literature has not converged on a unifying set of assumptions to identify exogenous shocks to monetary policy (even for the pre-crisis period).”

Because the target rate essentially stopped moving after December 2008 while quantitative easing explicitly targeted longer-term rates, so the informative surprise migrated out the curve. The authors note that “since December 2008… there have been almost no changes in the target federal funds rate, and, until recently, FOMC statements for the most part worked to maintain the perception that the policy rate would continue near zero. In this setting, and in light of QE’s explicit goal of influencing longer-term interest rates, we use a measure of monetary policy shocks at the ZLB as in Rogers, Scotti and Wright (2014)” (Section 3). The underlying justification for any market-based measure is Kuttner’s (2001) point that while “expectations of Fed policy actions are not directly observable, futures prices are associated with a traded derivative contract, providing a ‘market-based proxy’ for these expectations.” The paper is candid about the cost of the choice, and addresses it in robustness: “a concern with the five-year Treasury futures contract is that these contracts may not be as liquid as contracts of shorter maturity such as the one-month ahead and two month ahead contracts that are the most heavily traded contracts” (Section 5.3).

Pre-crisis from March 1994, QE from December 2008 to April 2013, tapering from May 2013 – with the Lehman-to-QE interval deliberately dropped. The QE start date is chosen because “it is at this point that the Fed can no longer undertake conventional simulative monetary policy by lowering the interest rate,” and the taper start because May 2013 is when Bernanke first mentioned the possibility of tapering purchases (fn. 9). The excluded window is justified on substantive grounds: “The period between the collapse of Lehman Brothers and the beginning of QE was marked by global ‘flight to safety’ and its inclusion in either neighboring sub-period muddles the analysis in the sense both that it is a period of extraordinary uncertainty and that it truly belongs to neither classification.” One internal inconsistency in the working-paper text is worth noting for a careful reader: Section 3.1 describes the pre-crisis period as March 1994 to August 2008, while footnote 9 defines the pre-crisis dummy as equal to one from March 1994 to July 2008.

That the shock moves both the expected short rate and the term premium in conventional times, but that in the unconventional periods the term premium dominates and its effect grows with maturity – a claim the authors state with visible hedging. In the conventional period a positive shock significantly moves yields at all maturities, with the effect diminishing at ten years; the short-rate-expectation effect “diminishes sharply with maturity,” while “the risk compensation effects are relatively stable across different maturities” (Table 2, Section 3.1). In the QE and taper periods the yield effects are larger and, unlike the conventional period, “monotonically increase over time” across maturities; “despite a role for the FF5 shocks during the unconventional QE and tapering periods altering the expected path of future short rates, the largest effects are associated with sizeable and statistically significant revisions in term premia. Further, these risk premia effects monotonically increase with maturity.” The magnitudes make the contrast concrete: a one-standard-deviation shock raises the ten-year yield by 4.73 basis points on a one-day window pre-crisis (4.81 on two days) against unconditional standard deviations of 5.76 and 8.31 basis points, but by 12.2 basis points during QE (13.6 on two days) against 7.10 and 9.99. The authors’ conclusion is framed as a conjecture rather than a demonstration: “Since the effect of a shock on the expected path of future short rates is likely to be relatively short-lived over the life of a long-term maturity bond, we can speculate on the manner in which the FF5 shocks map into revisions in the compensation for interest rate risk,” and “during the period of unconventional monetary policy, it may be the case that our measured FF5 shock has more to do with variation in required risk compensation relative to variation in the expected path of the short rate.” They also flag the mild irony that this holds “despite the fact that high frequency variation in futures contracts are employed in the construction of the FF5 shocks in the first place.”

Portfolio balance and signaling both predict a negative coefficient on the shock; the confidence channel predicts a positive one; the liquidity channel operates through credit rather than flows directly. The portfolio balance channel works through the reduced private supply of long-duration assets: prices rise, yields fall, and because the Fed has removed duration risk “investors should require a smaller premium to hold the reduced quantity of long-term securities,” so “if the portfolio channel is in operation we expect that monetary policy shocks will be inversely correlated with emerging market flows and valuations” (Section 2.1). The signaling channel works through the expected-short-rate term: taken as a commitment to keep rates lower for longer, it implies persistent advanced-emerging interest differentials that “may trigger a carry trade, resulting in sizeable capital flows into emerging markets,” so again a negative coefficient. The confidence channel is the sign-flipper: “an announcement of tapering might serve as a signal that the FOMC is feeling sanguine about global economic prospects, lowering relative risk aversion and… increasing capital flows to emerging markets,” so “we would thus expect a loosening monetary policy shock dominated by the confidence channel to drive capital outflows from emerging markets.” The liquidity channel runs through bank reserves: large-scale purchases credit reserves that “are more easily traded in secondary markets than are long-term securities,” so the liquidity premium falls and “liquidity-constrained banks can extend credit to borrowers, resulting in decreased borrowing costs and elevated lending levels.” Because the channels carry opposite sign predictions, the paper reads channel dominance off coefficient signs rather than identifying channels separately.

Monthly interpolations of annual TIC survey holdings, decomposed into flows, valuation changes and a residual gap – with the gap and a financial-centre attribution bias both acknowledged. The underlying estimates are Bertaut and Tryon’s (2007) method through December 2011, which cumulates monthly TIC S net transactions onto annual survey benchmarks with monthly valuation adjustments, and Bertaut and Judson’s (2014) method from January 2012 onward, which uses the TIC SLT form introduced in 2011 and so reports market-value positions directly, meaning “valuation adjustments are reported directly and the approach to distributing the gap between flows, valuations and positions over the year requires fewer assumptions” (Sections 4.1.1-4.1.2). The residual gap is inherent: extrapolating a survey position forward on flow data leaves “a substantial gap between the cumulation-implied holdings at the time of the next survey and the value of reporting holdings in that month,” attributable to approximation and measurement error in the prices used for valuation adjustments and to transaction costs included in reported transactions but not in holdings surveys; it is distributed across months in proportion to each month’s share of net transactions. The attribution problem is stated plainly: “By construction, the transaction data are recorded by country of first cross-border counterparty, rather than actual end buyer or seller of the security. Thus, estimates calculated in the above fashion will tend to overestimate holdings by residents of financial center locations and underestimate holdings by residents other countries.” The authors’ own handling choice is disclosed and motivated: they combine the gap with flows rather than with valuations “on the assumption that the error attributable to flow mis-measurement is like to be higher than that attributable to prices.” For emerging market debt the return index used is “the average of USD EMBI+ and the local currency bond index weighted by the currency composition of U.S. resident positions,” reflecting the increasing local-currency share. The panel covers Argentina, Brazil, Chile, Colombia, India, Indonesia, Korea, Malaysia, Mexico, Peru, the Philippines, Russia, South Africa, Thailand and Turkey.

A random-effects panel with a lagged dependent variable, contemporaneous shock terms interacted with period dummies, and a large set of push and lagged pull controls; random effects is chosen on a stated substantive assumption and corroborated by a Hausman test. The dependent variables are debt and equity positions, monthly net flows and monthly valuation changes, the latter two scaled by GDP, plus the monthly change in the bilateral dollar exchange rate expressed so that an increase is a local-currency appreciation (Section 5). The lagged dependent variable is included “to account for the strong autocorrelation we observe in the flows and holdings time-series,” and country-specific controls “are included with a lag to rule out simultaneity.” On the effects specification: “We use the random effects model instead of fixed effects under the assumption that our sample is sufficiently long that country-level unobserved heterogeneity cannot be considered immutable. A Hausman test corroborates our choice, rejecting fixed effects.” Standard errors are White-corrected and clustered at the country level. Notably, the paper also reports that results are robust to running the estimations with destination-country fixed effects, and to scaling by lagged holdings instead of GDP, though those tables are “not reported but available from the authors” (fn. 14).

That between policy sub-periods, changes in overall emerging market positions were driven mainly by monetary-policy-induced revaluation rather than by money actually moving. The claim is stated in the same form in the introduction and conclusion: “the most robust finding is that valuation changes for both debt and equity played a key role in the change in overall positions observed between sub-periods. That is, in nearly every specification, the effect of monetary policy shocks on asset returns is larger than that for physical flows” (Section 1); and “the evolution in overall emerging market debt and equity positions between various policy sub-periods appear to be largely driven by U.S. monetary policy induced valuation changes” (Section 6). The authors tie this to their term structure result: “This finding is consistent with the notion that our shocks may capture a revision in required risk compensation across financial markets.” Within the QE period the pattern is sharp – “in the QE period, changes in positions are attributable entirely to valuation changes” – while “in the unwinding period, the flows become a statistically significant contributor to position changes.”

An order of magnitude during QE and a factor of two to three during tapering. “During the QE period, for instance, the coefficient on equity valuations in response to a monetary policy shock is ten times higher than the coefficient on debt valuations” (Section 1; Section 5.1). During tapering, “the effect of monetary policy shocks on equity measures is double or even triple the magnitude of the effect for debt.” A related asymmetry appears in how valuations translate into positions: “for equity flows in the QE period, while valuation changes translate to a statistically significant effect on positions, this is not the case for debt since positions are not statistically significantly affected by monetary policy shocks.”

Consistency and scale: the coefficients turn inversely signed and significant at the 1 percent level across essentially all variables, and an order of magnitude larger than in either earlier period for several of them. “Across alternative measures of debt and equity flows and valuation changes, we observe inverse and statistically significant coefficients suggesting that the period of taper talk and the actual unwinding was associated with significant outflows from emerging markets. It is also noteworthy that across the board the coefficients associated with the unwinding period are higher than both the pre-crisis and the QE periods and for some specifications an order of magnitude higher for debt positions, debt valuations, and equity flows. Moreover, the levels of statistical significance across all specifications in the taper period are consistently at the 1% level” (Section 5.1). The specific contrast with QE on the debt side is that “the effects of monetary policy on debt flows during the taper period are much higher and statistically significant than during the QE period.” The authors’ reading is stated as an interpretation of what the market inferred rather than as an identified mechanism: “the market interpreted the unwinding of unconditional monetary policy as a signal that normalcy was being restored to the U.S. economy and, consistent with both the signaling and portfolio balance channels, expected monetary tightening in the U.S. both in the near-term and ongoing in the future led to a massive retrenchment from emerging markets.”

They offer two alternative readings and do not choose between them. “While it may seem a little puzzling that the flow measures do not exhibit statistical significance for either debt or equity, one explanation may be that there is more noise in the measured flows data during periods of higher volatility. On the other hand, it is plausible that there was indeed no detectable increase in flows because the positions inflated due to improved expectations for the valuation of emerging market firms” (Section 5.1). The second reading is developed: “Many large emerging markets appeared to weather the crisis well. Optimism about emerging markets would contribute to increased valuations for emerging market firms. Given that our flow variables of interest are specific to the United States, the valuation changes could be attributed to increased domestic investment or increased investment from other locales.” The honest summary they give is that “during the QE period there appears to be a significant and consistent relationship between FF5 surprises and for both emerging market debt and equity valuations, rather than flows.”

Pre-crisis, a tightening surprise coincides with emerging market currency appreciation – the opposite of the textbook prediction – while during the unconventional periods, and especially during tapering, the sign flips to the expected one. “During the pre-crisis period, an unexpected policy shock is, on average, associated with a move in emerging market currencies in the same direction (relative to the USD)… While one might expect interest rate differentials to play a textbook role in currency determination going forward, we nevertheless observe the opposite, possibly suggesting a role for a confidence channel whereby U.S. policy tightening is, for instance, correlated with expectations of global economic expansion” (Section 5.1). “In sharp contrast, we observe that a U.S. policy shock is associated with emerging market currency fluctuations in the opposite direction during UMP periods. This is particularly true during the later tapering period,” where “positive (tightening) U.S. policy shocks are associated with large and significant emerging market currency depreciations.” The authors are careful not to oversell the size: in the taper period the exchange rate response ranges from a 0.52 percent appreciation for a mean-minus-one-standard-deviation shock to a 0.75 percent depreciation for a mean-plus-one-standard-deviation shock, against an unconditional monthly bilateral exchange rate standard deviation of 3.57 percent, so “while the impact on bilateral exchange rates is statistically significant, the magnitudes are not large from an economic perspective” (Section 5.2). One number in this passage of the working paper does not sit with the others: the mean QE-period shock is said to lead to “a 1% monthly appreciation,” while the surrounding one-standard-deviation range for the same period is given as a 0.16 percent depreciation to a 0.24 percent appreciation, which suggests the mean figure should be read as roughly 0.1 percent.

Roughly $150 million a month per country at the mean shock, with one-standard-deviation swings approaching $1 billion – sizeable, the authors argue, relative to the small and illiquid markets involved. The worked example uses equity positions, where the coefficient on the shock is minus 0.99 percent of annual GDP in the QE period and minus 0.97 in the taper period, with mean shocks of minus 0.02 and plus 0.016 and average country GDP of $791.29 billion and $911.88 billion respectively. This gives “a monthly increase of $153.5M in emerging-market equity positions” at the mean QE shock and “monthly outflows of $144.1M” at the mean taper shock (Section 5.2). The dispersion matters more than the mean: one standard deviation either side of the mean corresponds to monthly equity position changes ranging from minus $672.26 million to plus $979.26 million during QE, and from minus $970.45 million to plus $682.35 million during tapering. The authors’ justification for calling these large is comparative rather than absolute: “Given that these are simple one-standard deviation shocks in either direction and that these local markets tend to be relatively small and illiquid, position changes of these magnitudes are quite sizeable.” They also acknowledge that the scaled ratios look small by construction, since “the numerator is a monthly USD flow or monthly USD valuation change, whereas the denominator is the GDP level for the previous year” (fn. 13).

Mostly as the literature predicts for liquidity and political risk, but against expectation for the S&P return and US growth, which the authors attribute to a wealth effect. Global liquidity behaves as expected: “in all but one specification the TED spread, our measure of global liquidity, is inversely correlated with capital flows,” so widening spreads coincide with reduced holdings and flows (Section 5.1.1). The VIX is less clean – positively correlated with debt positions and valuations but not debt flows, which “could arise due to an increase in the risk premium on emerging-market debt,” and inversely correlated with equity positions only at the 15 percent level with no significant effect on equity flows or valuations. The S&P 500 return runs the wrong way relative to a substitution story: the literature “suggest[s] that the return on advanced economy equities should evince a negative relationship with emerging market equity flows,” yet the estimate is positive and significant, which the authors read as “a wealth effect of the U.S. return on capital flows,” noting “this result, however, is not without precedent, as Forbes and Warnock (2012) find a similar pattern.” US real GDP growth is treated the same way, with the two countervailing forces named explicitly before the positive coefficient is reported. Advanced-economy interest rates are inversely related to emerging debt positions and flows but positively related to debt and equity valuations. On the pull side, current account, fiscal balance and gross government debt are each inversely correlated with debt positions and flows but positively with equity measures – a split the authors explain by the dual nature of these variables: they signal investor confidence but also “indicate increased financing needs, generating a mechanical relationship with debt flows in particular,” and “for debt flows, the effect of financing needs appears dominant.” The ICRG political risk index enters positively for debt positions and flows, “consistent with the prediction that capital flows to a country increase as political risk declines,” and real exchange rate appreciation is positively correlated with equity flows – which the authors note is close to tautological, “since real exchange rate appreciation is often itself used as a measure of increased capital flows.”

Four alternative shock measures, two alternative scalings, fixed effects, and unscaled data; the short-maturity fed funds futures measures change the pre-crisis picture substantially while leaving the taper conclusions intact. Replacing FF5 with Kuttner’s current-month measure (MP1) and Gurkaynak’s path measure (MP2), entered together, gives a different conventional-period result: “In the pre-crisis period, there appears to be no effect of MP1 on any of our bond flow measures. In contrast, the coefficient on the MP2 monetary surprise measure is positive and statistically significant for debt positions and flows” – a sign the authors interpret as markets reading a tightening as a sanguine signal, that is, as the confidence channel. This is a real contrast with the baseline: “Recall in comparison that the coefficients on the longer five-year Treasury futures measure (FF5) measure was negative and significant in nearly every specification” (Section 5.3). During QE the two measures also diverge, with MP1 positively correlated with total debt positions and MP2 inversely correlated with debt positions, valuations and flows. The taper-period conclusion, however, survives: “Across both monetary surprise measures and alternative measures of debt and equity flows and valuations, we see inverse and statistically significant coefficients… once again the coefficients associated with the unwinding period are an order of magnitude larger than both the pre-crisis and the QE periods.” Longer-horizon contracts (MP3, MP4) “did not enter the regressions with consistent statistical significance,” so the paper reports only MP1 and MP2. A two-year Treasury yield change measure gives results that “remain robust especially for the taper/unwinding period,” and decomposing that yield shows changes on FOMC dates “tend to result in larger part from changes in the term premium, although the decomposition is frequently close to an even contribution.” Scaling by lagged holdings, adding destination-country fixed effects, and dropping the scaling entirely all leave the pattern intact.

A large run-up in positions and flows during QE and a sharp valuation reversal during tapering, against a backdrop in which several push factors were moving the other way. Total holdings across debt and equity rose on average by 165 percent between the pre-crisis and QE periods and by a further 27 percent between QE and taper, with the difference in means statistically significant; average monthly flows rose 421 percent from pre-crisis to QE and fell 49 percent from QE to taper, though “the decline is not statistically significant” (Table 3, Section 4.3). Monthly changes in both bond and equity valuations “exhibit significant declines, on average across countries, between the QE and the Taper periods,” and the EMBI bond and MSCI emerging equity indices fell on average by 80 percent and 76 percent respectively in the taper period. The push factors complicate any simple risk-off story: the VIX “actually declines significantly between the QE and the Taper periods to a value below that which prevailed during the pre-crisis average,” the TED spread falls in every consecutive sub-period, and the S&P’s average annual return in the taper period is “very high relative to the QE and pre-crisis periods” – which is part of why the paper controls for these separately from the identified shock. On the pull side, average real GDP growth, the fiscal balance and the public debt ratio deteriorated across successive periods, while ICRG political risk improved significantly in each.