# Four Shocks That Didn't Break the U.S. Economy — Notes on Macro Musings with Ben Harris > Notes after listening to the 2026-08-24 Macro Musings episode with Ben Harris: why the expected recession never arrived, why the real oil bottleneck is refining rather than crude, and whether AI can grow the U.S. out of its fiscal hole. Educational only; no investment advice, no stock recommendations. Published: 2026-08-24 Locale: en Tags: macro, fiscal deficit, AI productivity, energy markets, judgment ![A refinery complex at dawn, pipelines running far into the distance, one unit dark and shut down while a flare still burns beyond it](/covers/macromusings-2026-08-24-ben-harris-on-ai-fiscal-sustainability-and-the-res-cover.png) > Accumulated stores are the great destiny of the realm. If grain is plentiful and wealth to spare, what undertaking could fail? > > — Jia Yi, *Memorial on Accumulating Stores* (Western Han, c. 178 BCE; translated by the author) ## What this episode is about In the 2026-08-24 episode of Macro Musings, David Beckworth brings back Ben Harris of the Brookings Institution. The opening is unusual: the host starts by admitting he was wrong. He had been confident, more than a year ago, that a particular policy agenda would wreck the economy. It didn't. The rest follows from that admission. The first half dissects why nothing broke. The middle detours into oil and sanctions design. The second half covers Harris's new paper on whether AI can grow the United States out of its fiscal hole. An episode containing both "I got this wrong" and "here is my model" is worth an hour. ## The main points **1. A hundred economists in a cave.** Harris lists four shocks from 2025: trade policy (the average trade-weighted tariff going from roughly 2.5% to the high twenties and back down), net immigration falling from about a million a year to roughly zero or negative, a signature bill adding some $5 trillion in new debt, and an unprecedented campaign against Federal Reserve independence. Lock a hundred economists in a cave, describe only those four things, ask them to forecast GDP, and he thinks "everyone would have a negative in front of it." Nobody did. **2. Count the dollars collected, not the rate announced.** This is the most portable idea in the episode. On paper the tariff rate rose tenfold. In practice roughly $250 billion was collected — about $200 billion more than a normal year — and around $170 billion of that looks likely to be refunded. Net it out and you get a tenth of a percentage point of GDP. Add transshipment (goods that used to come from China arriving via Vietnam) and understated invoices, and Harris's conclusion is that the measurement source has to change: stop reading the schedule, read the receipts. **3. Offsets, genuine errors, and lags.** Beyond the shocks being smaller than advertised, data center construction added perhaps half a point to growth, and the same bill that piled on $5 trillion in debt also pushed roughly half a point into disposable income through refunds — a medium-term headwind acting as a near-term tailwind. Some things economists simply got wrong: with a million fewer workers, construction wages should have spiked and didn't, and no labor economist he asked had a good explanation. And some things are merely late. Construction wages are now starting to outpace typical wages; the risk premium on Treasuries is creeping up. To borrow the host's Friedman paraphrase, these shocks operate with long and variable lags. **4. The oil problem is refining, not crude.** Harris runs the numbers: about 15 million barrels a day leave the Strait of Hormuz, of which 6 to 6.5 million can be routed through Saudi and Emirati pipelines, plus a couple million that keep coming out anyway thanks to what he calls "some really brave ship captains." The shortfall is far smaller than headlines suggest. Applying conventional elasticities, he and Robin Brooks get a worst case of $125 to $150 Brent — painful, not automatically a global recession. The real damage is that roughly 10% of global refining capacity has been destroyed, much of it Russian, by Ukrainian drone strikes. Crack spreads are through the roof. Diesel is the acute problem because it sits underneath trucking, groceries and retail; jet fuel is why airfares jumped. Refineries take years to rebuild, so this lives on a timescale of quarters to years. He estimates the whole episode added about 0.6 points to headline inflation in 2026 and 0.2 to core — a pass-through of roughly a third — and thinks the worst is already behind us. **5. The fiscal arithmetic is one line long.** Deficits run about 6% of GDP. Interest alone is 4%. Strip that out and the primary deficit is about 2%, projected to stay there for a decade. To have any chance of growing out of the debt, that primary deficit needs to be zero or better, ideally a 1% surplus. The big bill roughly doubled the medium-term problem — eighteen months ago he'd have said 5% and 1%. The country now pays more on its debt than on defense. **6. If AI rescues the budget, halve your optimism.** The base case is a late-1990s internet shock: faster growth, real productivity gains, deficits falling toward 2% and a primary deficit near zero. But Harris and his co-authors give five AI-specific reasons to discount that. First, better healthcare means longer lives — in their most optimistic longevity scenario the 65-and-over population goes from about 74 to 76 million, meaning two million more people drawing Social Security and Medicare. As he puts it, what is wonderful news for humanity is terrible news for the federal budget. Second, and largest, the equilibrium interest rate rises: with capital rushing into data centers and chips, Treasuries must pay more to compete. Their extreme scenario lifts the whole yield curve by 35%. Third, displaced workers land on income support. Fourth, income shifts from labor to capital, and capital is taxed more lightly, so revenue falls. Fifth, an AI arms race — pure waste, modeled at $35 billion a year, and he says he's become *more* convinced of this one since drafting. The rule of thumb is almost a jingle: if you think AI takes deficits from 6% to 2%, expect 4%; if you think 6% to 4%, expect 5%. **7. Rather than the government owning stakes, let households own them.** Both host and guest are visibly uncomfortable with the government taking equity in tech companies — the entity that collects your taxes also holding your shares, with the obvious question of whether generosity buys favorable treatment. Harris says it starts to feel a lot like China and not much like capitalism. His alternative is almost anticlimactically plain: the Secure 2.0 Act contains a little-noticed Saver's Match, giving low- and middle-income workers fifty cents on every dollar they put into a 401(k). Raise it to one-for-one. That is a one-line piece of legislation, from a bill both parties already passed. For context, the U.S. hands out roughly $250 billion a year in retirement tax breaks, and almost all of it goes to the top 30%. **8. Sanctions can cap the price instead of the quantity.** This was his portfolio at Treasury. The Iran approach targeted quantity — drive exports toward zero. With Russia, with a fragile post-COVID macroeconomy and oil above $120, that wasn't safe. So the design switched to capping the price Russia receives, enforced by controlling the services a barrel needs: tankers, insurance, financing, flagging. The Western coalition controlled roughly 90% of those services. It worked for a while, opened a wide discount on Russian crude, and the $285-a-barrel forecasts from reputable banks never happened. Then Russia bought hundreds of tankers and built a shadow fleet. He notes the scoreboard: 216 tankers sanctioned by the last day of the previous administration, zero since; the rest of the G7 has lowered the cap, the U.S. still sits at $60. ## Going further ### "I called the direction right and still lost money" This is the most retail-relevant thread in the episode, and the person asking is the host himself. He got the direction right: tariffs went up, immigration went down, debt went up. What he got wrong was the consequence. It comes apart into three layers, each a habit worth keeping. The first is **the gap between announced and collected**. Policy publishes a rate; the economy feels cash. In between sit exemptions, refunds, transshipment, evasion and companies planning ahead. So the next time a headline announces a percentage, don't jump to industry effects. Go find a number that has landed: how much was actually collected this quarter, who paid it, how much came back. If that number doesn't exist yet, the news is still just a headline. The second is **offsets**. More than one thing happens at a time. Data center capex and tax refunds were pushing up while tariffs and immigration pulled down. Deriving a macro outcome from one variable is close to guaranteed to fail — which is why "right about the event, wrong about the money" is the normal case, not the exception. The third is **lag**. Harris stresses that some expected effects are arriving now: construction wages outpacing, term premia creeping up. The investing implication is slightly counterintuitive. When a call doesn't pay off immediately, there are two live possibilities — you were wrong, or it isn't due yet. The only way to tell them apart is to write the falsification condition at the moment you make the call: if by date X indicator Y still hasn't moved my way, I was wrong. A view with no falsification condition can always be rationalized, and therefore can never teach you anything. ### "AI is this powerful — why halve the optimism?" The common chain of reasoning is: productivity jumps, growth accelerates, the problem solves itself. What Harris's paper contributes is the missing link in the middle. Fiscal language makes it clearest. Debt sustainability is a ratio. AI grows the denominator — true. But the same investment boom lifts interest rates, because a finite pool of dollars is competing across uses, and when data centers look better and better, Treasuries have to pay more to hold anyone's attention. The numerator grows too. Their extreme case raises the entire yield curve by 35%. This isn't a story where growth solves everything; it's a race between the numerator and the denominator. Move that structure onto an individual valuation and it is identical. You get optimistic about a growth story and raise your future cash flows — but the capital boom underwriting that story also raises the discount rate. One assumption, two sides of the model. Raising growth while leaving the discount rate untouched is cheating on your own spreadsheet. It also explains something else: why genuinely productive eras don't automatically produce good asset returns. That isn't irrationality. That's the denominator losing the race. ### "So why isn't the scarcity where I assumed it was?" The crude-versus-refining segment is the best structural lesson in the hour. When conflict erupts, every eye goes to the most upstream, most quotable thing: crude. It's priced daily and written about hourly. But Harris's arithmetic says the crude shortfall is far smaller than assumed — pipelines route around it, brave ships keep sailing, and with refining capacity destroyed, demand for crude fell too. That gap partly closes itself. The genuine scarcity is one layer down: the ability to turn crude into diesel and jet fuel. It's scarce because it can't be replicated quickly — refineries take years — and because a tenth of it was physically destroyed. The crack spread is the receipt showing which layer collects the rent. That is a complete worked example of finding the bottleneck. The test for who really holds pricing power along a chain is never who sits furthest upstream or who generates the most news. It is **which layer breaks first when demand arrives, and can't be rebuilt in the near term**. Being short of something is not the same as being able to price it; scarcity you can replicate isn't scarcity. The mirror image also holds: when good news lands and the price doesn't move, it's often because the good news landed on a replaceable link. One line worth sitting with: discussing energy diversification, Harris quotes his old boss at Treasury, Janet Yellen — Putin doesn't control the wind and the sun. That sentence quietly relocates renewables from a climate argument to a national security one. ## Worth looking up - Macro Musings with David Beckworth, 2026-08-24 episode, guest Ben Harris - The Brookings paper "Can AI restore fiscal sustainability in the US?" (Harris with Neil Mehrotra and William Overcash) - Harris's short piece "Four Reasons Trump's Economic Agenda Hasn't Tanked the Economy" - His work with Robin Brooks estimating supply elasticities around the Strait of Hormuz - The Saver's Match provisions in the Secure 2.0 Act — bipartisan, already law - Jim Hamilton's long-running research on oil shocks and recessions, as a counterweight to the "oil matters less now" claim ## The one thing to take away **One idea: the size of a shock is measured in money that actually moves, not in the number that was announced.** The strongest line in the episode is Harris saying we have to switch the source we use to size tariffs — stop reading the rate, count the dollars collected. The rate rose tenfold; the net effect was a tenth of a percentage point. That gap isn't an anomaly. It's the rule. Everything that gets announced has to travel from proclamation to cash, and along the way it passes through exemptions, delays, detours, refunds and human adaptation. Most forecasting failures aren't failures of direction. They're failures to distinguish an announcement from a payment. **One exercise you can do today: pick the announcement you worried about most — or looked forward to most — in the past month. A change of policy at work, a rent increase notice, a number on a medical report, a new rule at your child's school. It does not have to involve investing. Take a sheet of paper and draw two columns. In the left, write the headline number. In the right, write one thing only: the specific amount and date on which it will actually leave or enter your account, your calendar, your day.** If the right column comes out blank, what you spent the month worrying about was a headline. Keep the sheet. In thirty days, take it out and check your answer.