Quantitative Tightening and the Macroeconomic Reality
When the System Tightens: What Quantitative Tightening Actually Does to Your Risk Models I was sitting across a conference table from the risk team at a mid-sized asset manager in late 2022. Good people, experienced people, the kind of shop that had weathered 2008, navigated the COVID volatility, and built their frameworks carefully over two decades of hard lessons. We were reviewing their liquidity stress scenarios, the kind of exercise that, in most years, is professionally useful without being professionally urgent. Their models looked fine, their buffers looked more than adequate. I ran the numbers three times, not because anything was wrong, but because the results seemed too comfortable for the environment we were sitting in. The Bank of England had been raising rates, the Fed was well into its tightening cycle. Quantitative tightening, the deliberate reduction of central bank balance sheets after years of extraordinary expansion, had been underway for months. And yet everything on those spreadsheets looked orderly. I should have trusted the discomfort more than the spreadsheets. The Situation The problem with QT is that it is not a single event I can model. It is an atmospheric change. Between 2009 and 2021, central banks globally expanded their balance sheets to an almost incomprehensible degree, the Fed alone went from roughly $900bn to over $8 trillion. That capital had to go somewhere, it found its way into asset valuations, into compressed credit spreads, into the quiet assumption baked into almost every risk model that liquidity was ambient, that it was simply there, like oxygen. What QT does is reduce the oxygen concentration, slowly, incrementally. And because the reduction is gradual, the feedback loop is delayed, which is precisely what makes it dangerous. By Q1 2023, the asset manager I had been advising was seeing things that their models had not flagged as probable. Refinancing costs had jumped in ways that ate into assumptions built during a period of structurally different rates. Counterparty appetite had thinned, not dramatically, not catastrophically, but perceptibly. Two positions they had classified as liquid turned out to be liquid in theory and illiquid in practice. There was no single headline event, no Lehman moment, nothing had broken, everything had just become slightly harder, simultaneously, across every dimension that mattered at once. I remember one of the senior risk managers saying, with a kind of tired accuracy: “We planned for the doors to get narrower, we did not plan for all of them to get narrower at the same time.” That line stayed with me, because it was exactly right, and it described something that conventional stress testing, built around individual shock scenarios, is structurally ill-equipped to capture. Three Things QT Does That the Rates Headline Doesn’t Tell You **First: it changes what “liquid” means.** In a QE environment, markets are deep because central bank purchases create a persistent buyer with no return requirement. Remove that buyer and liquidity becomes conditional, it exists when sentiment is stable and disappears precisely when you need it most, which is to say, when sentiment isn’t. Assets that traded freely in 2020 and 2021 carried liquidity assumptions that were products of that specific environment, those assumptions did not automatically update when the environment changed. The models were not wrong, they were answering a question that no longer reflected reality. **Second: the transmission lag is long enough to be genuinely deceptive.** Rate rises are felt quickly, in mortgage costs, in corporate debt service, in consumer spending data. Balance sheet reduction works over a longer cycle, through the gradual withdrawal of reserve balances from the banking system and the slow repricing of risk appetite throughout the credit chain. This means institutions can operate inside deteriorating conditions for months before anything manifests in their numbers, the system tightens before the data tells you the system is tightening. By the time the evidence is visible, the adjustment window has narrowed considerably. **Third: QT interacts with everything else at once.** The rates shock was the headline, the balance sheet reduction was the mechanism running underneath it, invisibly. Taken alone, either would have been manageable for most well-run institutions, together, they changed the physics of the environment. Duration risk repriced, collateral values shifted, the correlation assumptions in multi-asset portfolios, correlations built during a decade of suppressed volatility, began to behave unexpectedly. Risk managers who had stress-tested each factor individually found themselves in a world where the factors had become entangled. What This Means for Your Organisation Right Now The question most risk functions are asking is whether they can survive a rate shock, that is the right question, but it is the second question. The first question is whether your liquidity assumptions, your correlation assumptions, and your counterparty models were calibrated in a world that no longer exists. For most institutions, the honest answer is: partially. The models were updated, but the underlying assumptions about how markets behave, how liquidity moves, and how correlated risk manifests were formed in an extended period of exceptional monetary accommodation. QT is not just a policy reversal, it is the removal of the conditions under which modern risk management frameworks were largely built and refined. Running those frameworks forward without interrogating their foundations is not risk management, it is institutional memory applied to a changed environment and called discipline. The asset manager I mentioned did not fail, they adapted, but later than they should have, and at greater cost than necessary. The lesson was not that their risk team was inadequate, the lesson was that the environment had changed its operating assumptions and nobody had formally updated theirs to match. Closing The market does not care when your models were last calibrated, and it is completely indifferent to the decade in which your assumptions were formed.