Notes — Austan Goolsbee on Central Banking as a Data Dog

Notes on Austan Goolsbee in conversation with Tyler Cowen — Conversations with Tyler (https://conversationswithtyler.com/episodes/austan-goolsbee/), 25 June 2025 (recorded 3 March 2025).

Four questions [Adler frame]

Q1 — What is it about as a whole? A working central banker’s account of how he actually reasons about money, inflation, and financial stability — an empiricist’s creed set against theory-driven macroeconomics. Goolsbee, president of the Chicago Fed and a former Obama economic adviser, casts himself as a ‘data dog’ rather than a hawk or a dove: someone who sniffs the data before committing to a model. Cowen presses him across a sequence of contested questions — whether post-pandemic inflation was demand or supply, whether money still matters, whether stablecoins and shadow banking threaten stability, why housing keeps getting more expensive, whether the regional Fed system should survive, and how much AI will really change productivity and central banking itself.

Q2 — How is it argued? As a running duel between Cowen’s theory-first provocations and Goolsbee’s insistence on identification and evidence. Goolsbee repeatedly refuses to be talked into a clean position — a monetary rule, a pure-demand story of inflation, a Fed stablecoin — and instead reframes each as a question of whether the underlying shocks are stable enough to trust the model. His method is explicitly micro-economic (‘I bring a micro sentiment to the thinking about causality’) and analogical: the data dog, the LLM training sample, the dot-matrix printer, self-driving cars, believing or not believing in magic.

Q3 — Is it true, in whole or part? Well-grounded where he reasons from documented puzzles he has studied directly — the breakdown of the M2–inflation relationship, the negative measured productivity of construction (his paper with Chad Syverson), the simultaneity of post-pandemic inflation across domestically-driven economies. Appropriately hedged where he disclaims authority: he stresses repeatedly that the Chicago Fed does not set policy on CBDCs or crypto, and admits economists cannot fully explain why money matters at all, or why housing’s relative price has risen for decades. [?] His empirical claims (inflation soaring with unemployment above 6 per cent; the 2023 disinflation without recession) are asserted from memory, not sourced here.

Q4 — What of it? A transferable model of decision-making under model uncertainty: prefer robustness to shocks over commitment to any single rule, and treat every relationship as conditional on the shocks that generated it. It supplies the wiki’s fullest first-hand treatment of monetary-policy reasoning, shadow banking, and central-bank AI adoption, and connects to the wiki’s existing threads on productivity, general-purpose technologies, and the limits of large language models.

Glossary

Data dog — Goolsbee’s self-description: an economist who resolves questions by getting into the data rather than by ideological camp (hawk vs dove); ‘there’s a time for walking and a time for sniffing’. [§ On central banking as a data dog]

Supply shock vs demand shock — the two sources of a change in prices and output: a demand shock moves spending (and is what most central-banking machinery is built around); a supply shock moves the economy’s capacity to produce (labour supply, supply chains, productivity). Reading which one is driving inflation must come before any policy conclusion. [§ On central banking as a data dog]

Accounting identity (MV = PY) — a relationship true by definition, not by economic behaviour: the money stock (M) times how fast it circulates (velocity, V) equals prices (P) times real output (Y). Goolsbee refuses to make policy off it because an identity has ‘no economic content’ — if both M and V move, reading inflation off M alone misleads. [§ On central banking as a data dog]

Velocity of money — how fast a unit of money changes hands. Financial innovation (electronic payments, cards, stablecoins displacing cash and cheques) has made velocity unstable, which is why the old M2–inflation correlation broke down. [§ On central banking as a data dog]

Golden path — Goolsbee’s term for the 2023 outcome in which inflation fell sharply without the recession that every past episode, and the consensus forecast, said was required; his standing example of why a rule premised on past shocks can fail. [§ On whether post-pandemic inflation was mostly from demand or supply]

Shadow banking — lending and money-like deposit-taking done outside the formally regulated, deposit-insured banking sector (private equity, stablecoins, non-bank lenders); prone to runs, growing every year, and largely outside the Fed’s oversight and information. [§ On stablecoins, shadow banking, and financial stability]

Lender of last resort — the central bank’s role of supplying emergency liquidity to solvent institutions in a panic; its dilemma is what to do when the runs happen in the shadow sector it does not supervise. [§ On stablecoins, shadow banking, and financial stability]

Free banking era — the mid-19th-century US period (after the Second Bank of the United States) when almost anyone could start a bank and print money; Goolsbee cites its recurrent panics (1847, 1857) as the historical warning against unregulated money-like issuance. [§ On central bank digital currencies]

CBDC (central bank digital currency) — a digital form of central-bank money for public use; ‘retail’ means an ordinary person holds it directly (like a digital $10 bill), ‘wholesale’ means only banks do. Goolsbee is sceptical of the retail case (‘would Grandma call the Fed for her password?’). [§ On central bank digital currencies]

Training-sample problem — Goolsbee’s analogy for both LLMs and rule-bound central banking: a model is only as good as the past data it learned from, so it will confidently give the wrong answer (a ‘hallucination’) when the present is driven by shocks absent from that history — as in 2023. [§ On AI’s prospects for boosting productivity]

Construction productivity puzzle — the finding (Goolsbee and Chad Syverson) that measured productivity in home-building has been flat-to-negative for decades: ‘somehow, we’re forgetting’ how to build homes; a candidate explanation for housing’s rising relative price. [§ On housing and construction]

Key claims by section

On central banking as a data dog [§ On central banking as a data dog]

Goolsbee refuses the hawk/dove framing: he is a ‘data dog’, and academic economics’ discipline of getting into the data — plus its habits of causality and identification — is what he found most useful as a Fed president.
He starts every inflation question from basic supply-and-demand and a micro sense of causality, and insists on asking whether a move is a supply or a demand shock before concluding anything; most central-banking machinery is built around demand, which he thinks blinds it in supply-driven episodes.
The old M2-to-inflation relationship broke down in the data, he argues, because financial innovation (electronic payments, cards) radically changed the velocity of money — with both M and V moving, over-indexing on M misleads.
He will not pre-commit to a monetary-policy or nominal-GDP rule: rules are accounting identities with ‘no economic content’, and only work if future shocks resemble past ones. His 2023 counter-example — the ‘golden path’ — is that inflation fell almost as much as it ever has in a year without a recession, defying every demand-driven rule. [?]

On whether post-pandemic inflation was mostly from demand or supply [§ On whether post-pandemic inflation was mostly from demand or supply]

Goolsbee resists Cowen’s mostly-demand reading with three puzzles for a pure-demand story: inflation soared while US unemployment was above 6 per cent (output below potential); inflation rose simultaneously across countries that had far less fiscal and monetary stimulus than the US; and inflation did not fall when the stimulus rolled off but did fall in 2023 as supply chains healed. [?]
He rejects Cowen’s Japan/Switzerland counter — those are import-heavy and had low inflation for decades before COVID — and redirects to large, domestically-driven economies (Nigeria, India) that also saw high inflation at once.
His settled view: not 100 per cent supply, a serious demand component, but the demand-only account is unpersuasive.

On stablecoins, shadow banking, and financial stability [§ On stablecoins, shadow banking, and financial stability]

Pressed on why money matters at all (isn’t it a near-substitute for T-bills?), Goolsbee concedes economists cannot fully explain it — there is some ‘yield to safety’ markets price that theory does not capture — and warns this makes M2 an even shakier guide in a world of imperfect money-substitutes. [?]
On stablecoins fully backed by treasuries: he is not anti-crypto but is ‘nervous’ — money-like deposits redeemable on demand need either deposit insurance or serious restrictions on backing assets, or they end ‘in bank runs and tears’; ‘500 years of financial history’ warns of this.
Financial stability’s core conundrum: an overseen official banking sector alongside a growing, run-prone shadow sector the Fed neither controls nor has information on — formal banks are only about one-fifth of lending, private equity and non-banks the rest.
He partly agrees with Cowen (via Jeremy Stein) that squeezing official banks too hard drives activity into the shadows, but declines the stronger claim that raising capital on official banks is therefore useless — post-2023, the big banks were not the problem, which he credits to the capital that was raised after 2008.

On housing and construction [§ On housing and construction]

Across the Chicago district (including rural Iowa) the dominant complaint he hears is ‘despair’ about housing cost; in rural Iowa the binding constraint on attracting workers was that workers could not afford housing — so it is not purely a zoning/building-code story.
Housing’s relative price has risen roughly 4–5 per cent a year for a decade and a half (well before COVID), compounding into a large gap; he finds it genuinely hard to explain, which lends some weight to Cowen’s suggestion that 2008 was not really a nationwide bubble. [?]
His paper with Chad Syverson finds construction productivity flat-to-negative over long periods — ‘somehow, we’re forgetting’ how to build homes; candidate causes (Ed Glaeser’s land-use-regulation account, a lost move toward modular/prefab) are plausible but he found no smoking gun.

On the microeconomics of running a Reserve Bank [§ On the microeconomics of running a Reserve Bank]

Running the Chicago Fed is like running a corporation; a professional COO (the ‘first vice president’) handles operations, so much of the president’s apparent competence is delegation working well.
The most useful micro ideas for management are the theory of delegation and the distinction between cost, opportunity cost, and marginal cost: things with a marginal cost of zero (opening meetings, sharing information — knowledge as a public good) can be done freely and generously.

On reforming the Fed system [§ On reforming the Fed system]

Asked whether fiscal pressure (and DOGE-style targeting of visible costs) might one day cut the 12 regional research staffs, Goolsbee defends the 1913 design’s ‘genius’: 7 politically-appointed governors plus 12 Reserve Bank presidents chosen by regional boards, so the FOMC is not just New York and Washington.
He calls the FOMC ‘the world’s greatest deliberative body’ and argues the regional structure guards against monoculture and groupthink — ‘let some flowers bloom’; centralising to one data-driven staff, however efficient it sounds, would be dangerous.
On AI in the Fed’s five core functions (monetary policy, the payment system, bank supervision, financial services to banks, community/regional expertise): he sees a plausible role in flagging supervision risks, but notes a current rule forbids general AI-tool use over confidential-information leakage. He is sceptical that an on-premises AI would beat humans at predicting bank failures — ‘you’re only as good as the training sample’, which in 2023 would have demanded a recession.

On central bank digital currencies [§ On central bank digital currencies]

Cowen frames the dilemma: do nothing and stablecoins proliferate beyond control; issue a retail CBDC and risk disintermediating community banks, plus public hostility. Goolsbee repeatedly stresses the Chicago Fed does not set this policy — Congress and the Board of Governors do.
He is unpersuaded by the retail case (‘Grandma calling the Fed for her password’) and more open to wholesale, but questions how a wholesale CBDC differs from existing wire transfers.
His deeper worry is a return to free-banking-era conditions (recurrent panics: 1847, 1857) if anyone can issue money-like instruments; attaching risky, run-prone things to the payment system is where he draws the line, invoking Paul Volcker’s question of what sits inside vs outside the rescue safety net, and Eric Budish’s paper on the 51-per-cent attack risk in decentralised blockchains. Tether, chartered abroad, he concedes the US can do little about beyond pressing for genuine backing assets. [?source — attributes specific claims to Budish’s and Volcker’s views]
On whether a well-run Fed stablecoin would dominate the market, he will say only ‘maybe’ — the Fed is ‘behind-the-scenes plumbing’, and (via Edmund Burke) there is a line, however hard to locate, that a central bank should not cross: ‘I like safe.‘

On AI’s prospects for boosting productivity [§ On AI’s prospects for boosting productivity]

Post-COVID US productivity has run above the pre-COVID trend, not merely recovered to it; most economists’ explanations (hybrid work, the Great Resignation’s labour reallocation, a jump in new-firm creation) are one-time level effects, not durable growth.
Only a genuine new technology can raise growth persistently by diffusing sector to sector; if AI is a general-purpose technology like electricity or computing, the gains play out over decades — first in the producing sector, later in the using sectors (his computers-to-Walmart-inventory analogy). Chicago Fed work finds the current surge concentrated in AI-intensive industries, consistent with an early general-purpose-technology pattern, but adoption is still too low to explain a large aggregate effect.
He flags an over-anticipation risk: if equity values and AI-infrastructure investment surge ahead of realised productivity, the economy can hit capacity constraints and overheat.

On o1 Pro and hallucination [§ On o1 Pro and hallucination]

On whether o1 Pro would beat new assistant professors on an economics quiz, Goolsbee distinguishes look-up questions (AI wins) from judgement questions with no correct answer (he is sceptical); hallucination is ‘an extremely damaging problem’ precisely where a mistake is an ‘unforgivable error’ — you must change the penalty function.
He resists the ‘just use the better/paid model’ rebuttal with the dot-matrix-printer analogy: advocates always claim the latest version is indistinguishable from the real thing when it is not, and steering him to a pricier tier only shows he is too cheap, not that hallucination is solved.
His self-driving-car warning: 15 years ago rapid improvement led people to predict no professional drivers within five years; instead the rate of improvement slowed sharply. Much recent AI gain comes from more data and compute, which hit diminishing returns fast, so he does the thought experiment of what jobs today’s AI (not extrapolated AI) would actually replace. He splits people into those who ‘believe in magic’ and those who do not, placing himself firmly among the sceptics.

On MBA education and debating Ted Cruz [§ On MBA education and debating Ted Cruz]

On MBA reform: he taught platform competition to Chicago Booth MBAs, found them data-minded and curious, and judges the MBA a well-functioning, valuable market under pressure mainly on the opportunity-cost (programme-length) side.
The closing anecdote: Goolsbee, a year older than Ted Cruz and part of the national team of the year, beat Cruz (the runner-up) by exploiting Cruz’s weakness at thinking on his feet — teasing him to the judge until Cruz lost his temper.