Why Economists Keep Getting Inflation Wrong
A short history of forecasting misses, and what humility would look like in a model.
In late 2021, most major central banks — the US Federal Reserve prominent among them — called rising prices "transitory." It was, by a wide margin, one of the more consequential forecasting misses of the last two decades. It also wasn't unusual. Forecasters have been confidently wrong about inflation on a roughly decade-long cycle since economists started trying to forecast it at all.
A short, embarrassing history
| Episode | Consensus forecast | What happened |
|---|---|---|
| 1970s oil shocks | Temporary supply disruption | A decade of double-digit inflation |
| Late 1990s | Persistent inflation risk from low unemployment | Inflation stayed low — the Phillips curve went quiet |
| 2021–2022 | "Transitory" supply-chain effect | Inflation peaked at multi-decade highs |
| 2023–2024 | Sticky, hard to bring down | Disinflation arrived faster than models projected |
Read down the table and the pattern isn't "forecasters are bad at their jobs." It's that inflation forecasting keeps failing in whichever direction the previous failure taught models to guard against. 1970s excess caution about persistence produced 1990s models tuned for stickiness, which underpredicted the 1990s disinflation. That miss, in turn, fed models more willing to call price spikes temporary — which is a large part of why "transitory" felt like a reasonable call in 2021.
The Phillips curve problem
Most inflation models still lean on some version of the Phillips curve: tight labor markets push wages up, wages push prices up. It's intuitive and it's been right often enough to stay in every textbook. But its predictive power has visibly weakened since the 1990s — unemployment has fallen without triggering the inflation the curve would predict, more than once, in more than one country. Models built on a relationship that only sometimes holds will only sometimes forecast well, and there's no reliable signal in advance for which regime you're in.
What actually moved 2021–2022, in hindsight
- Supply chains seized up in ways no post-war model had training data for.
- Fiscal stimulus during the pandemic was larger, more direct, and more front-loaded than prior recessions' responses.
- Energy prices spiked from a war that no economic model treats as a forecastable input.
None of these three drivers are things a standard inflation model is built to see coming. They're not inflation dynamics in the textbook sense — they're one-off shocks that happened to hit at the same time, and the models called "transitory" because historically, one-off shocks usually are.
"The core failure wasn't the models. It was treating 2021 as an ordinary business cycle when three unusual, simultaneous shocks made it very much not one." — a common retrospective read among central bank economists, 2023–2024
What humility would actually look like
The honest fix isn't a better point forecast — it's wider, more visibly communicated uncertainty bands, and a willingness to say "this could go either way" instead of a confident single number. Central banks have started doing more of this since 2022: the Fed's dot plots now get more airtime for their spread than their median, and forecast ranges get more prominent billing in policy statements than they used to.
That's a smaller, less satisfying answer than "here's the model that finally gets it right." It's also probably the correct one. Inflation forecasting keeps missing not because economists haven't found the right model yet, but because the thing being modeled — millions of individually rational price-setting decisions, reacting to shocks no dataset has seen before — may not be the kind of thing a single model was ever going to reliably predict.
Editor & Staff Writer
Final-year Economics major focused on monetary policy and currency markets. Edits The Gazette's markets desk.