John Cochrane’s Lesson on Inflation
Contents
This past week, on 1 October, I had the privilege of attending Mario Draghi’s Karl Brunner Distinguished Lecture. I will write about that at a later time. While looking for the recording, I came across John Cochrane’s inflation lecture in the same series.
Cochrane starts with inflation and ends up asking how governments finance their promises. That is a useful route into a subject usually discussed through the next central bank meeting. We spend a great deal of time debating whether interest rates should move by 25 basis points. We spend less time asking what makes the liabilities being issued at those rates worth holding.
I found the lecture compelling, although I would separate its central insight from its claim to explain recent inflation. Treating government debt as an asset with an underlying source of value is powerful. Establishing which mechanism drove a particular inflation episode requires more evidence. Cochrane is also unusually explicit about where the models still struggle, which makes the lecture more interesting than a retrospective victory lap.
Fiscal theory of the price level: what backs a government promise?
Suppose a company borrows money. Eventually, investors expect cash flows sufficient to support that borrowing. The company can refinance a maturing bond, but refinancing works because someone believes there is something worth financing.
A government has a different source of cash flow: its ability to tax, less what it spends before paying interest. Economists call that difference the primary surplus. Cochrane’s fiscal theory of the price level applies an asset-pricing relationship to the government: the real value of its nominal debt must be supported by the expected present value of future primary surpluses.
The word expected does considerable work. A government can run a large deficit during an emergency without generating inflation if people expect sufficient fiscal backing afterwards. Conversely, confidence in that backing can deteriorate before the government borrows another cent. A promise about future taxes or spending has changed, so the value of an existing claim changes today.
Here is a deliberately simple version. Imagine nominal government claims of 100, backed by expected future surpluses worth 100 in purchasing power. Set the initial price level to one. Now issue another 10 of claims without increasing their backing. Holding output, discount rates and the other relevant assumptions fixed, a price level of 1.1 restores the relationship: 110 divided by 1.1 is still 100.
Nobody has missed a nominal payment. The claims have instead lost purchasing power. The government has honoured the number printed on the promise while the number buys less.
This is a valuation illustration, not a rule that a 10% increase in public debt produces 10% inflation. Future surpluses, growth, interest rates and demand for government liabilities all move. Governments can also roll debt over indefinitely under appropriate conditions; the argument does not require every bond to be retired and public debt to fall to zero.
What I like about this framing is that it forces the question of who ultimately supplies the resources. Borrowing can move a tax burden through time. Inflation can shift part of the burden onto holders of nominal claims. The spending itself still uses real resources, whatever the financing arrangement is called.
There is an important qualification. A government budget constraint is not unique to fiscal theory. Other macroeconomic models have to satisfy it too. The distinctive claim concerns what adjusts when the constraint is disturbed. Do future taxes and spending respond to support the debt at the prevailing price level? Or is fiscal policy sufficiently fixed that the price level must adjust? The equation alone cannot decide which description fits a particular government.
Why lower inflation does not mean cheaper groceries
Cochrane’s account of the pandemic is that governments issued liabilities and transferred purchasing power without a corresponding expectation of future repayment through fiscal adjustment. Prices rose to reduce the real value of those claims.
His interpretation explains something that continues to confuse the public discussion: inflation can fall back towards normal while the price level stays high. A temporary inflation burst need not reverse itself.
If a household’s weekly shop rises from 100 to 120 and then to 122.40, inflation has fallen to 2% in the final year. The household is still paying 22.40 more than at the start. A central banker can accurately describe progress while the household accurately describes a continuing loss of purchasing power, unless its income has caught up.
In Cochrane’s model, a one-time loss of fiscal backing produces a rise in the price level spread over time by sticky prices and the structure of government debt. Once that adjustment is complete, inflation can ease without prices returning to their old level. Lower inflation does not, by itself, reimburse the bondholder or restore the household budget.
I find that distinction more persuasive than treating the entire pandemic episode as a clean test of fiscal theory. There were also closed factories, disrupted shipping, energy shocks and a sudden change in what people wanted to buy. Bernanke and Blanchard’s analysis of US pandemic inflation gives substantial weight to commodity prices, supply bottlenecks and the shift from services towards goods. Fiscal demand contributed to that pressure; labour-market tightness became more important for persistence later.
These explanations overlap. A transfer creates purchasing power; a supply constraint affects what that purchasing power can buy. The empirical question is how much inflation each mechanism explains, under which policy response. Pointing out that governments spent a lot does not settle it, any more than pointing to a shortage settles the question of aggregate demand.
Timing also needs care. Cochrane stresses that inflation eased without the kind of recession many expected. That challenges some forecasts, but it does not show that monetary tightening was unnecessary. The Fed’s first rate increase was in March 2022; US twelve-month CPI inflation reached 9.1% in June. An observed decline after that cannot tell us what would have happened without tightening. Expectations of future policy further complicate any attempt to read causality directly from the chart.
What the zero-interest-rate years tell us
An equally interesting part of the lecture concerns the inflation that failed to arrive. After the financial crisis, central banks created enormous quantities of reserves. Interest rates stayed near zero for years. Some predictions called for accelerating inflation; others warned of unstable deflation. The long period of relatively quiet inflation is evidence that those predictions need explaining.
Fiscal theory draws a distinction between changing the composition of government liabilities and adding liabilities without additional backing. When a central bank buys a Treasury bond and issues reserves, it exchanges one government liability for another. A deficit-financed transfer changes the consolidated position differently. Calling both operations “printing money” loses information that matters.
There is a useful corporate-finance instinct here: before explaining the consequences of a transaction, work out what actually changed on the balance sheet.
That still leaves room for quantitative easing to matter. Swapping long-duration bonds for reserves changes the risk and maturity characteristics of what investors hold. The Fed’s own research on portfolio purchases examines effects on mortgage markets through such channels. An asset swap can change yields and financial conditions without adding net government liabilities.
Nor were the zero-rate years a controlled experiment. Investors could expect future policy changes, and central banks used other tools. I take Cochrane’s charts as a serious challenge to simple stories about reserves and inflation. Choosing among more complete theories requires specifying the other policies and expectations operating alongside them.
Higher interest rates and the government’s interest bill
Cochrane does not conclude that central banks should respond to inflation by doing nothing. He explicitly supports raising rates in response to inflation. His question is what the models assume about the fiscal consequences.
Higher rates make financing more expensive for households and companies. They also make it more expensive for the government. For a country with debt equal to annual GDP, a one-percentage-point rise in the average interest rate on the entire stock would add roughly 1% of GDP to annual interest costs. That is an eventual scale calculation: fixed-rate debt reprices as it matures, and the timing depends on maturity structure and other features of the public balance sheet.
The larger bill has to be accommodated somehow. A conventional disinflation story may assume that future primary surpluses adjust to cover it. Cochrane asks what happens when fiscal policy does not provide that support.
With long-term debt, a rate increase can initially reduce the market value of outstanding bonds. That revaluation helps the model generate an initial fall in inflation. But a permanently higher interest-rate path, without fiscal adjustment, creates different longer-run effects. Much of the later lecture is devoted to why obtaining a convincing, persistent disinflation response is harder than the standard verbal explanation suggests.
I appreciated that Cochrane leaves the difficulty visible. He explores changes to the way expectations enter price-setting, but presents them as research in progress. The lecture’s confident fiscal interpretation sits beside an admission that the monetary transmission model still needs work.
For me, the practical implication is to read monetary and fiscal announcements together. A central bank can change the cost of government financing. Whether the government responds with taxes, spending restraint, further borrowing or a credible plan for growth influences what that policy achieves. Institutional independence matters, but it cannot remove the government’s financing constraint.
Where AI investment fits
This brings me back to a question I have been working through in AI Capex Arms Race: Who Blinks First?: how much future income supports the investment being made today, and who bears the loss if that income disappoints?
The connection needs limits. A hyperscaler borrowing to build a data centre is undertaking a private investment, with assets and revenues that may support the financing. If the project fails, shareholders and creditors can lose money. That does not automatically create an inflation problem for the government.
Still, there are two places where the lecture helps organise the question.
The first is timing. Construction, chips and electricity infrastructure require resources now. The productivity gains arrive later, if they arrive. The BIS’s July 2026 bulletin on AI and the global economy describes an increasingly debt-financed investment surge alongside uncertain and uneven productivity benefits. That combination can put demand pressure and future supply improvements on different schedules. It makes the central bank’s job harder without giving us a simple rule that AI capex is inflationary.
The second is the distinction between a good technology, a good investment and a good source of government revenue. Those outcomes can diverge.
As I argued in Maybe Meta Is Right About AI, useful models can become cheaper while the benefits accrue to businesses that use them. A model provider can lose pricing power even as its customers become more productive. Some infrastructure investors could earn poor returns during a period of substantial economic progress. There is no accounting rule requiring the company that funded a useful technology to capture its social value.
For fiscal backing, the relevant question is what happens to taxable income and spending over time. A productivity gain could broaden the tax base and make existing debt easier to support. But its size, timing and distribution matter. So do changes in public spending and the discount rate applied to future surpluses. “AI will produce growth” is several assumptions short of a fiscal plan.
Now add a hypothetical public guarantee. Suppose a government decides that some privately financed computing infrastructure is strategically indispensable. If it guarantees the borrowing, private investors have acquired a claim on public resources in the bad outcome. The project may still be sensible, but its financing should be evaluated with that contingent liability included.
A particularly awkward case would combine disappointing productivity, lower-than-expected tax receipts and demands for public support. The same adverse outcome would weaken the expected revenues backing government debt and increase the claims on them. This is my extension of Cochrane’s argument, not a prediction he makes about AI, nor a claim that such a rescue is inevitable.
It changes the questions I would ask about industrial policy. What exactly has the state promised? Can investors lose money without interrupting an essential service? Can the infrastructure remain useful under new ownership? A data centre can keep operating after its original equity holders have had a very bad experience. Preserving productive capacity and protecting the investors who financed it are separate policy decisions.
Growth helps, but the promises still matter
Cochrane closes by emphasising growth and fiscal resilience. A government wants borrowing capacity when the next crisis arrives. Looking only at whether it can service today’s debt misses how much more it might need to issue in an emergency.
I would resist using the lecture to declare either that central banks control everything or that they control very little. Its value is in making the financing assumptions visible. The same current deficit can mean different things under different expectations of future policy. The same rate increase can arrive with different fiscal responses. The same investment boom can expand future productive capacity while distributing losses in ways nobody agreed to explicitly.
What I will take from Cochrane’s inflation lecture is a habit of asking what supports a promise before deciding what it is worth. For government debt, that means a plausible account of future revenues and spending. For AI infrastructure, it means revenues, ownership and a clear allocation of downside risk. In both cases, I want to know what happens to the financing plan when the optimistic forecast fails. That is usually when the assumptions become expensive.
Slides screenshots from John H. Cochrane’s 2025 from the Swiss National Bank’s Karl Brunner Distinguished Lecture at ETH Zurich.