Quantitative Tightening Around the Globe: What Have We Learned?

Before 2020 only one country — the US from 2017–2019 — had made meaningful progress reducing central bank securities holdings, so the authors describe theirs as the first cross-country assessment of QT. They explain that after the 2008 crisis the slow recovery, sub-par inflation and uncertainty about QT’s effects caused most central banks to delay unwinding; in the US, balance sheet reduction initially appeared to have minimal effect but was then blamed for a sharp tightening in funding conditions that forced a return to balance sheet expansion sooner than expected. The situation changed abruptly in 2021: as activity rebounded and inflation spiked, central banks needed to tighten faster than expected, which created the opportunity to start QT even under substantial uncertainty, because contractionary effects would no longer constrain the ability to raise rates meaningfully above zero. The Bank of England was first, starting QT in February 2022 when its policy rate reached 50bps; the RBNZ, Riksbank, Bank of Canada, RBA, Federal Reserve and ECB all announced QT starts during 2022.

The paper documents substantial progress through end-2023 — about a 40% reduction in aggregate security holdings in Canada and Sweden, 25% in New Zealand, and 15% in the US and UK. Peak positions differed considerably: focusing on government securities relative to GDP, the BoE held the largest position at 37%, more than double the comparable 8%–16% ratios for the RBA, BoC, RBNZ and Riksbank; measured against the size of the domestic government securities market, the RBNZ peaked at about 40%, well above the US peak of about 28%. Projections under existing program parameters show wide variation in eventual normalization: the BoC, BoE, RBNZ and Riksbank are on track to unwind around 80% or more of their QE purchases by end-2025, while the Fed and RBA would unwind only about 50% of the increase in holdings since the pandemic, and the ECB even less. Programs also differ in design — whether passive run-off is slowed by caps (the US, and initially Sweden), whether bonds are actively sold (New Zealand, Sweden, the UK), and whether corporate and other non-government securities are also unwound.

Pooling all QT events other than wind-downs, the paper estimates that a QT announcement corresponds to an increase of about 4 bps in the two-day change in government bond yields at the 1- through 30-year horizons, while the very short-term impact on 3-month yields is about zero. The point estimates in the baseline table are 0.005 at 3 months (insignificant), 0.048 at 1 year, 0.037 at 2 years, 0.039 at 5 years, 0.042 at 10 years and 0.040 percentage points at 30 years, significant at the 10%–1% level from one year out. Control variables behave as expected: a 1pp surprise increase in the policy rate significantly raises yields from 3 months through 5 years, peaking at the one-year horizon, and positive economic data surprises significantly raise yields at all horizons of a year and longer. The authors read the yield response as consistent with QT tightening financial conditions through expectations that the private sector must absorb more debt securities (a portfolio balancing effect), or with QT being interpreted as signalling a tighter stance of policy via the policy rate.

Estimates with a separate dummy for each of the 39 individual QT events suggest substantial heterogeneity: 59% of the coefficients are positive — particularly for the UK, Australia and Canada — while about 41% are negative, particularly in the US. The authors note that if the US is excluded, two-thirds of the coefficients are positive, and attribute the muted US response to the extensive preliminary discussion by the Fed in advance of QT. Individual estimates are usually small but occasionally large: Sweden’s announcement of the start of QT on 28 April 2022 corresponded to a 42bps increase in 1-year yields. In the summary of results, individual announcement effects range from no impact up to 17bps (Australia), and up to 69bps for the UK when individual announcement effects are aggregated by economy since 2021. The authors caution that these coefficients capture changes in bond yields on QT event days that could occur for any reason, so they may simply be capturing other, non-QT news occurring simultaneously.

In contrast to the results for yields, QT announcements appear to have no consistently significant impact across a range of other financial variables, though the signs are in the expected direction. Testing a broad stock index, a broad corporate bond index, the exchange rate versus the US dollar, the Goldman Sachs Financial Conditions Index, inflation compensation at 3- and 5-year horizons, and the government bond convenience yield (the 10-year swap spread), the authors find that QT announcements generate a fall in the stock index, corporate bond index, inflation compensation at each horizon and convenience yield, alongside an exchange rate appreciation and tightening in financial conditions. Only two of these are significant at the 5% level: the corporate bond index (-0.002) and the convenience yield (-0.020). Estimated effects for individual countries remain heterogeneous.

Part of the effect corresponds to changes in expectations about the policy interest rate over the next 3–6 months, which the authors read as consistent with QT announcements being interpreted as a signal of stronger central bank commitment to tightening — a channel that operates even when the announcements were widely expected. They are explicit that this signalling relationship may also reflect other forward guidance occurring at the same time as the QT announcement, including shifts to more assertive language about commitment to tightening and price stability, and that as such the actual impact of QT may be somewhat smaller than the baseline estimates; controlling for changes in policy expectations reduces the estimated effects at shorter horizons but has minimal impact on longer-duration yields or on active QT. QT also appears to work by increasing duration risk and steepening the yield curve, especially for active QT, which has a larger impact on longer-term yields, while passive QT has a larger impact on shorter-term yields and appears to work primarily through the signalling channel and a flattening of the curve. The authors report no measurable portfolio rebalancing effect for equities of the kind found for QE, and only very small cross-country spillovers.

The authors state that comparison is difficult and offer only “very rough” comparisons, but conclude that the estimated effects of QT are still meaningfully smaller than for QE on two-day estimates, though the US QT2 estimate falls in the range of estimates for QE3 — the QE program conducted under minimal market stress. They cite a survey of pre-pandemic QE studies reporting effects on 10-year yields ranging from -16 to -107bps, and note that the majority of those episodes came from periods of heightened market stress during the Global Financial Crisis or the euro crisis, with estimated QE impacts declining meaningfully as market conditions stabilize. Two further asymmetries are flagged: QE programs are typically announced during market stress and so are likely to have much larger effects through the liquidity channel, and pre-2020 QE announcements usually specified a fixed quantity whereas QT announcements have generally not specified a target for the aggregate reduction or an end date.

The paper finds no significant pricing effects for government bonds around the narrow QT implementation dates, whether QT occurs through passive run-off or active sales. Changes in government bond yields are not significantly different on QT dates than on non-QT dates; and when countries sell bonds through active QT, there is no significant difference in yield dynamics between securities that are actively sold and similar securities not sold by the central bank on the same date. The authors conclude that, for the narrow implementation windows, they find no evidence that QT has distorted pricing of government bonds in the secondary market.

Over time, QT implementation has resulted in significant reductions in central bank total assets and in banking-sector liquidity balances, and this decline in liquidity balances is generally associated with a significant rise in the spread between overnight funding rates and the central bank deposit rate. Post-pandemic QT in the US is the exception so far: the decline in the Fed’s securities holdings has corresponded to a decline in the overnight reverse repo balance of nonbanks, so reserve balances at banks have been little changed and the spread between the federal funds rate and the interest rate on reserves has been steady, even as US repo rates have trended up. Separately, the convenience yield of government bonds — measured as the swap spread between the interest rate swap rate and the government bond yield of the same maturity — has declined since QT began; in panel regressions this swap spread decreases meaningfully when the share of outstanding government bonds held by central banks falls and when the supply of government bonds increases, which the authors take to suggest government bonds became less “convenient” under QT. They state explicitly that it is impossible to isolate how much of these changes is directly caused by QT versus other, potentially related, changes in the economic and financial environment.

No: the authors find a deterioration in government bond liquidity recently, but report that this deterioration is more tightly correlated with elevated interest rate volatility than with QT, and they find that bid-to-cover ratios at government auctions have either stayed unchanged or slightly increased during QT. Liquidity is measured with the Bloomberg government bond liquidity index, based on yield-curve fitting errors of individual government securities, and auction demand with the bid-to-cover ratio. On this evidence they conclude QT has not weakened demand at government bond auctions.

Panel regressions on IMF International Financial Statistics data find that nonbank domestic investors’ securities holdings are more sensitive to changes in central bank holdings during QT periods than in non-QT periods, and that these investors play an important role in absorbing the drawdown in central bank balance sheets. In the US this change in behaviour for domestic nonbanks was stronger for the post-pandemic QT episode (QT2) than for QT1 in 2017–2019. Evidence on other investor types is more heterogeneous across economies, although foreigners — including foreign official-sector holders as well as banks and nonbanks — have played an important substituting role in several countries. The authors emphasise that their baseline regression estimates the correlation between changes in central bank holdings and each investor group’s holdings, and does not identify a causal link.

In the US, the flow-of-funds category “households” — which includes hedge funds — has been a particularly important replacement for the Fed’s securities unwind during QT, especially during QT2, while patterns are less conclusive in other economies. A counterfactual analysis for the US confirms that households purchase more government securities in response to a reduction in the Fed’s share during QT than would be predicted from non-QT periods, while foreign investors — most likely foreign central banks — tend to reduce holdings alongside the Fed. The authors qualify this: the heightened sensitivity of households/hedge funds during QT2 was likely exacerbated by factors tied to the aggressive tightening of the overall policy stance, noting that the deeply inverted yield curve appears to have decreased demand from real money investors while increasing demand from institutions motivated by the Treasury basis trade.

The authors list several caveats and are explicit that their results may not generalise. First, almost all QT events in the analysis occurred during the unusual post-pandemic recovery, so the relationships may not apply during standard recoveries or periods of slow growth — for example the strong nonbank demand that absorbed the reduction in central bank holdings may have stemmed from large post-Covid stimulus and may not be available in future. Second, although the paper draws on seven central banks, the number of observations is still limited, particularly for announcement effects, and there is substantial heterogeneity across country experiences. Third, the estimates could understate the impact if markets incorporated changes before the narrow event dates or if effects take longer to play out than the short estimation windows — both of which the authors consider likely. Fourth, the analysis does not control for changes in the supply of government debt, including changes in the duration or outstanding stock of issuance that occurred at the same time as QT. Finally, they warn that the relatively smooth adjustment to date could suddenly generate sharper market movements as balance sheets shrink — as occurred in the US in 2019 — and that it is impossible to estimate with any precision what level of reserves would correspond to such a “tipping point.”

Their overall conclusion is that QT programs have been successful so far: central banks have made meaningful progress unwinding securities holdings with only a very modest tightening in financial conditions and no meaningful disruption to market functioning. They characterise QT as working as central banks intended — largely “in the background,” not seen as the active tool for adjusting policy, but providing a small degree of support to efforts to tighten financial conditions. Because some of the impact appears to work through signalling tighter monetary policy, they suggest the discussion and guidance around QT could itself be important in determining its effects. With the caveats above, they argue the analysis should give central banks more confidence to unwind asset purchases in future than they had in 2021, while flagging that the challenge is assessing when the smooth adjustment to date could suddenly transition to a liquidity crunch.