Monetary Policy: Asymmetries, Lags, and the Case for Purposefulness

Introduction

The textbook story of monetary policy is deceptively simple: raise rates to cool inflation, cut rates to stimulate demand. In practice, the transmission mechanism is neither symmetric nor uniform. Different channels operate at different speeds, rate hikes bite harder and faster than rate cuts heal, and the labour market – the variable central bankers care most about – is the slowest to respond. Understanding this structure explains both why central banks should be deliberate in their actions, and why a rapidly falling unemployment rate during above-band inflation is one of the most worrying signals in macroeconomics.

Hikes Are Fast, Cuts Are Slow

When the RBA raises rates, the aggregate demand impact is near-mechanical for cash-flow-constrained households, which dominate marginal consumption adjustments. Around 70% of outstanding Australian mortgages are variable rate. When rates go up, repayments increase within a billing cycle or two. The debt servicing ratio jumps – that's cash straight out of household budgets. Discretionary spending falls almost automatically. This isn't a confidence story; it's arithmetic. Housing wealth declines too, reinforcing the effect through a negative wealth channel.

The impact is not perfectly uniform – many households are ahead on mortgage buffers, and the marginal propensity to consume out of cash-flow changes is heterogeneous. But the aggregate effect is fast. The stock of fixed-rate loans has grown since the pre-2020 era, introducing rollover lags for that portion of the market. Still, in Australia the dominance of variable-rate exposure makes the cash-flow channel comparatively rapid relative to economies like the United States.

Rate cuts work differently. Mortgage repayments fall, but households don't immediately spend the difference. They save it – paying down debt, rebuilding precautionary buffers depleted during the tightening. Spending follows only when confidence returns, which depends on the labour market, the housing market, and the broader narrative. The wealth effect operates in reverse too: house prices respond to cuts, but spending out of wealth gains is slower than retrenchment from wealth losses. And credit supply doesn't ease symmetrically – banks tighten lending standards on the way up and take their time relaxing them on the way down.

Standard linear IS specifications estimate a single transmission parameter that averages across both directions. This undersells the speed of hikes and oversells the speed of cuts. The asymmetry matters for policy: when you tighten, the pain arrives before the payoff. When you ease, the stimulus takes longer to materialise than you'd expect.

Two Sources of Inflation, One Instrument

This asymmetry is compounded by a fundamental constraint: the central bank has one instrument – the cash rate – but inflation has two distinct sources.

Demand-pull inflation works through the textbook channel. Excess demand tightens the labour market, pushes wages up, and feeds into prices. The rate instrument addresses this directly: tighten demand, open the unemployment gap, and inflation falls.

Supply-push inflation – energy shocks, supply chain disruptions, import prices, weather events – is different. Rates can't fix a supply shortage. Hiking into a supply shock destroys demand without addressing the source. You get the unemployment cost without the inflation payoff.

So central banks try to "look through" supply-side inflation. Let it pass through, hold rates, trust that supply normalises and the price level effect is one-off rather than persistent. This is sound strategy – most of the time. The risk is that if supply-side inflation persists long enough, it shifts expectations. Once backward-looking wage bargaining replaces forward-looking target anchoring, supply-side inflation becomes demand-side inflation. The "look through" window closes.

This is why the inflation anchor matters so much. The anchor holds as long as people believe it holds. The central bank's job is partly to maintain that belief – which is itself an argument for purposeful, credible policy-making rather than erratic responses.

The post-COVID experience illustrates the difficulty. The 2022–2023 inflation surge was heavily supply-driven – energy prices, global supply chains, the pandemic hangover. The RBA correctly tried to look through much of it. But it persisted long enough that the relationship between the unemployment gap and inflation appears to have reawakened. Expectations may have partially shifted. And distinguishing residual supply-side inflation from demand-pull inflation in real time is fiendishly difficult.

The Lag Chain: Three Speeds of Transmission

Once a rate change is made, its effects ripple through the economy at different speeds. The economy has at least three gears:

Financial markets and aggregate demand: fast (quarters). The exchange rate, bond yields, and asset prices respond almost immediately. The output gap responds within two to four quarters via the IS curve. This is the channel that moves first.

Inflation: medium (follows the output gap with a lag). The Phillips curve picks up changes in the unemployment gap, but with a delay. Today's inflation partly reflects labour market conditions from two to four quarters ago. This creates a crucial information problem: the inflation you're observing now is a lagged signal, not a current one.

The labour market: slow – until it isn't. This is where the story gets interesting. Unemployment is dominated by its own persistence. Empirically, around 75% of last quarter's unemployment gap carries forward into this quarter. Only a fraction of the current output gap feeds through. This means the labour market inherits the demand signal, but heavily filtered – like watching the ocean through frosted glass.

Over a typical forecast horizon of four quarters, a positive unemployment gap (above NAIRU) barely closes even if the output gap has already turned. The output gap can flip sign quickly because rates act on it directly. The unemployment gap drags, carrying three-quarters of its previous value forward each period.

This is empirically realistic. Labour markets are sticky. Hiring takes time; firing takes time; matching takes time. But it creates a critical asymmetry in information value.

When Persistence Breaks: The Alarm Signal

If unemployment is converging gradually toward NAIRU, that's expected – just the persistence term doing its work. Low signal. The model predicted it.

If unemployment moves suddenly – falling fast below the NAIRU while inflation is above the band – that's high signal. Something the model didn't anticipate. And the implications are serious.

A fast drop in unemployment implies the output gap is strongly positive and widening. Demand is accelerating despite restrictive policy – the IS curve either isn't biting or something is overwhelming it. But here the speed mismatch matters: the output gap can turn quickly, unemployment adjusts slowly via Okun's Law, and inflation depends on the unemployment gap via the Phillips curve. So by the time unemployment stabilises, you've already locked in a period of above-target inflation that hasn't materialised yet. You're stacking inflationary impulses: the inflation you're seeing now reflects conditions from quarters ago, and the inflation consequences of today's tight labour market won't arrive for quarters to come.

That's the scenario where the usual persistence – the 75% carry-forward – stops being your friend. It normally buys you time: one meeting's delay doesn't materially change where unemployment ends up. But when the persistence breaks down, waiting means the inflation pipeline is loading faster than you realised.

The Australian Episode: 2025–2026

The current Australian experience is a clean illustration of these dynamics.

Through 2024 and into early 2025, the data supported easing. Unemployment was trending up toward 4.3%. Inflation was falling back toward the band. The RBA cut – measured, deliberate, and if anything criticised for cutting too slowly. The data at the time justified the moves.

Then the labour market turned. Unemployment peaked around mid-2025 and rolled over sharply – from ~4.3% to 4.08% and falling. Not the slow Okun-style convergence. A decisive turn. Simultaneously, trimmed mean and weighted median inflation bottomed around 2.8–3.0% in mid-2025 and re-accelerated – back to 3.3–3.5%, above the band. Both channels pushing the wrong way at once.

If the unemployment decline instead reflected a positive supply shock – productivity gains or labour supply expansion – the inflation implications would be different. But the simultaneous re-acceleration in trimmed-mean inflation suggests demand is the dominant force.

With hindsight, the final cut in 2025 looks mistimed – though the data at the time did not clearly signal the labour market turn. The turning point was essentially unforecastable in real time. The persistence parameter says unemployment should move slowly. It didn't. That's a structural break in the dynamic, not a policy error. The model would have supported the cut too.

The RBA's response has been purposeful. The December 2025 meeting flagged concerns, preparing markets for a potential shift. The February 2026 hike – back to 3.85% – was delivered as foreshadowed: no surprise, minimal disruption. Markets had already priced it. That's a central bank leading, not chasing.

The Case for Purposefulness

The debate now is whether the RBA should hike again in March or wait until May. The case for March is the speed of the unemployment decline: if demand is stronger than the model thinks, waiting locks in more inflation pipeline. The case for May is purposefulness.

Central banks deliberately pace themselves outside of crisis (the GFC and COVID being the obvious crisis examples) because they are less likely to jump at shadows and make short-run mistakes that way. The threshold for action should be asymmetric: move when you're roughly 80% confident that not moving would be worse. Not 51%. The cost of an unnecessary hold is one meeting's delay. The cost of an unnecessary move is credibility erosion, market disorder, and a harder-to-read reaction function.

Consider the sequence the RBA has built:

1. Cuts in 2025: Data-justified, delivered at a measured pace, resisted external pressure to go faster.
2. December meeting: Flagged concerns, prepared markets for a potential shift – no surprise.
3. February 2026 hike: Delivered what was foreshadowed – markets already priced it, minimal disruption.
4. March hold (if they take it): Consistent with the same deliberate approach – act on clear signals, observe, signal first.
5. May hike: Markets already expect a further hike in May – and given the data to date, that looks quite probable. The expectations channel is already doing the work.

A March hike without similar groundwork would break the pattern. It risks looking like the February hike was an admission of error on the cuts, and now they're scrambling. Back-to-back moves after a policy reversal communicates panic. A measured pace – hike, observe, hike if needed – communicates control.

The irony is that the same Okun persistence that makes fast unemployment moves so informative also justifies the deliberate pace. One meeting's delay doesn't materially change where unemployment ends up, but it materially changes whether the institution looks like it knows what it's doing.

And foreshadowing does real work. It tightens financial conditions through the expectations channel before the rate move even happens. It brings households, markets, and the commentariat along – everyone arrives at the same conclusion at roughly the same time. An expected hike tightens conditions smoothly. An unexpected hike tightens through shock – disorderly, harder to calibrate, and burning credibility you'll need later.

The Model's Value at Turning Points

If there's a broader lesson, it's about the role of macroeconomic models at moments like this. The linear Okun coefficient is estimated from the 95% of the time where labour markets adjust gradually. The 5% where they snap – post-COVID recovery, the current episode – dominates the policy-relevant moments but barely moves the statistical estimate.

You could try regime-switching, threshold effects, or state-dependent persistence. But you'd be estimating nonlinearity from a handful of episodes in a 40-year sample. The results would be too wide to be useful, and you'd be overfitting to events that are each structurally different.

The pragmatic approach is to use the linear model as a baseline but interpret it knowing that the persistence breaks down at turning points. The model tells you the steady-state dynamics. Judgement tells you when you're not in steady state.

That's arguably what the RBA got right with the foreshadowing. They weren't waiting for the model to confirm the turn. They were reading the leading indicators and applying judgement on top. The model said "unemployment should still be drifting up slowly." The data said otherwise. They moved.

The model's value in those moments isn't the point forecast – it's the framework. It tells you what to worry about (the Phillips curve consequences of a fast unemployment drop) even if it can't tell you when the break will happen.

Updates

Update (27 February 2026): On X/Twitter, Peter Tulip (formerly RBA, now Chief Economist at the Centre for Independent Studies) rightly points out that the consumption research suggests the cash-flow channel via mortgage repayments is less important to the aggregate consumption response than I implied. The wealth and intertemporal substitution channels likely do more of the heavy lifting. The framing in the "Hikes Are Fast" section leaned too heavily on the mortgage repayment story – the media-friendly version rather than the research-supported one. That said, the timing asymmetry between hikes and cuts – which is the core argument of the piece – likely still holds: the cash-flow hit is immediate, while wealth and substitution effects take longer to build. The speed point is somewhat separable from the level point. But the mechanism deserves a more careful treatment than I gave it here.