When 'Overheating' No Longer Looks Like Low Unemployment

We dare not fight the tiger bare-handed, dare not ford the river without a boat. Men know that one danger, and no other. So we go trembling and careful, as if standing at the edge of a deep gorge, as if walking on thin ice.

— Book of Songs, “Xiao Min” (pre-Qin China; translated by the author)

What this episode is about

On 28 August 2026, Bloomberg’s Odd Lots released an interview recorded at Jackson Hole with Chicago Fed President Austan Goolsbee. The title is blunt: he’s worried the economy is overheating.

The interesting part isn’t whether he’s bullish or bearish. It’s that he spends most of the hour taking apart a set of words we all use casually. Is policy restrictive? Where is the neutral rate? Is the economy overheating? He takes each one and explains why the question, as posed, can’t be answered — and only then tells you what he actually watches.

For anyone who doesn’t do macro for a living and just wants a way to think about rates, the value here isn’t the conclusion. It’s watching someone demonstrate how to make disciplined judgements about something you cannot observe.

The main points

1. “Is policy tight enough?” hides your own assumption inside the answer. The first question was whether policy is currently restrictive. Goolsbee’s answer: depends what you think underlying inflation is. What matters is the real rate — nominal minus expected inflation. If inflation is heading back to 2%, current settings are restrictive; if inflation is really running north of 3%, the real rate is far lower and policy isn’t tight at all. He offers a rough long-run landing spot: 3% nominal, 2% inflation, 1% real. The same nominal number supports opposite conclusions depending on the inflation assumption you plug in — so when someone tells you rates are too high, ask what inflation number they’re carrying in their head.

2. The neutral rate is a Sasquatch. He was an academic for thirty years, and yet he says R-star is close to useless for deciding what to do next: “I always called our R-star our Sasquatch — you can never see it until after it’s left, and you go, well, here was a footprint.” He doesn’t deny the concept exists. He refuses to steer by something only observable in hindsight. Delivered with the Tetons as a backdrop, it’s a good joke; it’s also the methodological spine of the episode.

3. Good news about AI may, in the short run, be an argument for higher rates. This is the most counterintuitive stretch. If AI genuinely lifts productivity growth, the long-run neutral rate goes up — faster growth requires a higher steady-state interest rate. What happens in the short run depends on whether the windfall is expected: if it lands on you unexpectedly and behaviour hasn’t adjusted, inflation falls and rates can come down. But the bigger the hype — the more everyone believes a giant bounty is coming and spends today against it — the easier it is to overheat the economy now and be forced to raise rates. He notes we lived through exactly this in the mid-to-late 1990s. So the arrow everyone silently draws from “AI is working” to “rates will fall” isn’t there.

4. Sectors competing for resources isn’t the same as an overheating economy — but he doesn’t think we’re far off. He draws the line carefully: data centres bidding up land, electricians and construction crews is direct inter-industry competition. To count as overheating, it has to get out of its lane and push up wages and prices outside that lane, because if one sector rises while another falls, the national unemployment rate and GDP growth rate still tell you honestly where you are in aggregate. He jokes that the data centre has become the all-purpose boogeyman: “I half expect people to say, I need to go to the dentist, they can’t see me for three weeks — thanks a lot, data centers.” Then he turns: go to Cedar Rapids, Iowa, ask what the biggest problem is, and you hear that data centres are buying up all the land, driving prices up, nobody can do any construction, you can’t get an HVAC person. “It doesn’t feel like we’re far from the sort of traditional excess demand and output above potential.” And if that happens in the aggregate, back to question one: then policy isn’t restrictive enough.

5. He is waiting for a specific kind of evidence, not a number. He splits inflation three ways: tariffs, oil, and services. He’s hopeful about the first two as temporary — though he admits saying “transitory” gives him hives. The services piece isn’t caused by tariffs and isn’t caused by oil prices; that’s the deeper concern. His standard is unsentimental: you can’t say every quarter, “ah yes, it’s about to go away, just not yet.” That’s why he dissented at the last meeting’s cut — he wasn’t comfortable front-loading rate cuts while counting on inflation to be transitory, especially with the government shut down and no data to look at. He’s explicit that he isn’t against cutting; he wants evidence it’s actually heading back to 2% first. He also warns that productivity growth has been pretty crummy for six months running — don’t all conclude we’ve had a change of era before it’s actually manifested.

6. On the 2022–23 disinflation, his split is roughly two-thirds supply, one-third demand. But he refuses to let people have it both ways — blaming the run-up on fiscal and monetary stimulus while crediting the decline entirely to supply-chain healing. The Fed’s real achievement, he argues, was not losing the anchor: even with CPI pushing toward 10%, inflation compensation in TIPS stayed steady around 2.3% CPI, which historically corresponds to about 2.0% PCE — exactly the target. He offers a sharp image: feed all of history into an AI and ask what to do when inflation is double or triple target, and it says jack rates to 20% and induce a huge recession, because that’s the only way it’s ever been done. Recognising that part of this shock wasn’t permanent, and that the much-mocked immaculate disinflation was in fact possible, deserves credit. On the other side, he concedes flatly that they were slow out of the gate: “You can’t look back and say they weren’t.”

7. Forward guidance and the reaction function are different things, and conflating them causes damage. What he opposes is forward guidance in the literal sense — here is what I intend to do with rates at the next meeting — because it ties your hands and adds volatility. What he fully supports is letting the world understand what he’s looking at and why. He points out a structural flaw in the dot plot along the way: the median inflation forecast and the median rate path don’t necessarily come from the same person, so the chart doesn’t actually function as a reaction function. On whether the bond market should tell the Fed what to do, he quotes Paul Volcker: “Our job is to act and the market’s job is to react — let’s not get the order mixed up.” Gathering information from markets, he’s entirely for. Treating market prices as instructions, he isn’t.

Going further

”The news says cuts are coming — should I get positioned?”

Almost everyone has had this thought. What this episode offers isn’t “stop reading the news.” It’s a different question to ask.

Goolsbee is clear: he won’t tell you how he’ll vote next meeting, but he will tell you what he’s watching — whether services inflation persists, whether it’s genuinely converging on 2%. In investing terms, the difference is this: the first is a forecast, the second is a criterion.

Forecasts expire; criteria don’t. When you’re handed a forecast, all you can do is believe it or not, and re-pick a side with every new headline. When you’re handed a criterion, you can check the answer yourself. How did services inflation print this month? Closer to 2%, or further? You don’t need the person to speak again to know which way their position has moved.

This transfers straight to reading companies. A note with a price target is a forecast. A note saying “the thing that matters here is whether gross margin holds above a certain level, and when that second line reaches full utilisation” is a criterion. You can verify the second one yourself in three months; with the first, you can only wait for them to change their mind. A useful filter: after reading anyone’s view, find the part of it that could be proven wrong. If there isn’t one, the piece contains less information than you thought.

”If AI is this hot, why do some people say overheating and others say slowdown?”

Because they’re talking about two different levels, and this episode draws that line sharply: inside the lane, and outside it.

Data centres taking the electricians, bidding up land, stalling everyone else’s construction — that’s industries competing for the same physical resources. It makes some people miserable and others rich, but it can net out in aggregate: one goes up, another comes down, and national unemployment and GDP growth need not move. Real economy-wide overheating requires the heat to spill over and lift wages and prices in unrelated sectors.

That line is practically useful. When you see a story about a sector expanding furiously and suppliers running tight, ask two questions. First: has the heat left the lane? Second, and within the lane: is the tight layer actually capturing pricing power? Scarcity and bargaining power are not the same thing — a full order book doesn’t guarantee the margin holds; you may simply be the one squeezed in the middle. Goolsbee is candid that he hears plenty of this from executives, and that in some ways he feels he hears too much of it, because most of those complaints are still inside the lane.

There’s a further twist. This episode says that if everyone believes the AI bounty is coming and spends against it today, the short-run result is overheating and higher rates. Which means: the more fully the market believes the story, the higher the odds that the cost of capital rises — and a higher cost of capital is least kind to exactly the long-duration growth names the story is about. The good news isn’t the problem. The problem is how much of the good news has already been pulled forward. That’s not bearishness; it’s a reminder that the same fact can carry opposite signs on different horizons.

”I saw this number and panicked”

There’s a small passage in here that’s easy to skim past and is methodologically the prettiest thing in the episode.

When he arrived at the Fed in early 2023, the dominant public argument was that inflation couldn’t come down because wage growth was too fast, and wages are the lion’s share of services costs. He says that gets the dynamics backwards: wages are stickier than prices. When a shock hits, prices move first and wages follow; on the way down, prices fall first and wages follow. So “wage growth is still high” simply doesn’t imply “inflation can’t come down.” The queue hasn’t finished moving.

He runs the same mental move on the bond market. Long yields rising has at least three possible causes: expected inflation, an expected higher path for policy rates, or simply more issuance and more competition for buyers. He’s least persuaded by the “the world is panicking about US creditworthiness” story, on mechanical grounds: if you truly believed a country was going to default, the rate wouldn’t be sitting at five and a quarter, which is historically pretty normal.

Put together, that’s a habit worth stealing: when a number makes your pulse jump, force yourself to write down two opposite causes for it, then go find the third piece of evidence that separates them. High wages could mean inflation is getting away, or it could mean wages always lag. Rising long yields could be panic, or just supply. Most people’s error isn’t arithmetic — it’s that only one story surfaces, and every subsequent fact gets filed into it.

Worth looking at

• Bloomberg Odd Lots, “Austan Goolsbee Is Worried the Economy Is Overheating,” 28 August 2026, recorded at Jackson Hole
• The Fed’s own documentation of the SEP and the dot plot (federalreserve.gov) — worth seeing for yourself that the median inflation and median rate paths are tallied separately
• TIPS breakeven inflation rates on FRED (T5YIE, T10YIE) — this is the line he means when he talks about whether the anchor held
• BLS quarterly labour productivity data — the source of “six months in a row of pretty crummy productivity growth”
• If you want to understand why supply shocks leave central banks without a playbook, start from the observation that unemployment and inflation move together under them

One thing to take away

What you need from someone isn’t their forecast. It’s their reaction function.

The whole episode demonstrates one idea: whether a person will tell you what evidence would change their mind determines whether their opinion is useful to you. Goolsbee refuses to say how he’ll vote, but lays out the criteria — services inflation, genuine convergence on 2%, whether productivity turns. That’s worth far more than a confident prediction, because you can check it yourself, and you will know when he’s wrong.

Here’s the exercise, and it has nothing to do with investing:

Pick something you’re currently arguing about with someone, or can’t settle in your own head — whether to change jobs, whether to finally say the thing to a particular person, whether to intervene in a habit your kid has. Then write one sentence: what specific, observable thing would make you admit your original judgement was wrong. Specific enough that someone else could check it. Not “if things get worse,” but “if he doesn’t reach out to me once in the next three months.”

If you can write it, you now hold a criterion, and every new piece of information moves you a defined notch. If you can’t, that usually means the thing isn’t a judgement for you anymore — it’s a position. And positions don’t move in response to evidence; they move in response to cost. Telling those two apart is worth considerably more than guessing the next step right.

Raw Markdown

This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.