Why Macroeconomic Forecasts Are Not an Investment Strategy

“All the macro forecasters were wrong” has become a tired refrain. It got another airing on September 15, 2026, when Derek Thompson devoted an episode of Plain English to the question, with Jason Furman as the guest and a title that put it bluntly: why experts keep getting the economy so wrong.¹

There is a stranger question hiding underneath it. If a handful of economists ought to be able to predict growth, inflation, recessions and stock returns years ahead, what exactly do we think markets are? Prices are supposed to be valuable because no single participant holds all the relevant information. They come out of millions of people reacting to their own circumstances, their own constraints and each other. Expecting a small group to anticipate the future state of that system, repeatedly, asks for something close to the opposite.

Forecasting is not central planning

The parallel above is worth stating carefully, because the strong version of it is wrong and a reader who spots that will stop reading.

A forecaster is estimating one thing: the distribution of GDP growth or inflation, given what is known today. A central planner has a much harder job, which is to set objectives, gather information on millions of products and preferences, predict how everyone will respond to the plan, allocate accordingly, then watch the responses and start over. The first is an estimation problem. The second is a control problem where the system being controlled reacts to the controller. Showing that economists can predict next year’s inflation reasonably well would tell you very little about whether a committee could allocate steel, labor and semiconductor capacity.

So believing in forecasts does not commit you to believing in central planning. The narrower point survives: heavy confidence in precise macro forecasts rests on assumptions about how much dispersed information can be gathered and processed from the top down, and those are the assumptions Hayek spent a career doubting.

What Hayek actually argued

The 1945 essay most often cited for “markets good, planning bad” says something more specific. Hayek’s framing was that the choice is not between planning and no planning at all. “Competition,” he wrote, “means decentralized planning by many separate persons.”² Which arrangement wins depends on which one makes fuller use of knowledge that already exists, scattered among the people who hold it.

His mechanism for that was the price. In the essay itself he described the price system as “a system of telecommunications which enables individual producers to watch merely the movement of a few pointers, as an engineer might watch the hands of a few dials, in order to adjust their activities to changes of which they may never know more than is reflected in the price movement.” A copper buyer does not need to know that a mine flooded. The price tells them enough to act.

Thirty years later, collecting his Nobel, Hayek turned the same argument on his own profession. The lecture was called “The Pretence of Knowledge,” and its complaint was that economists had borrowed the procedures of physics along with a precision the subject does not support. His conclusion is the sentence worth keeping: “I prefer true but imperfect knowledge, even if it leaves much indetermined and unpredictable, to a pretence of exact knowledge that is likely to be false.”³

How accurate are the forecasts?

Start with the event investors most want called. An, Jalles and Loungani assembled GDP forecasts for 63 countries from 1992 to 2014, covering 1,306 country-years that contained 153 recessions. In April of the year before each one, forecasters expected output to fall in 5 of the 153 cases. The average forecast error during those recessions was 5.85 percentage points, and was still about 2 points off in April of the recession year itself. Private-sector and IMF forecasts performed almost identically, so this is not a story about official optimism. Their summary: “What is rare is a recession that is forecast in advance.”⁴

Worth noting in the same table: false alarms are rare too, only 8 across 1,153 non-recession years. Forecasters do not cry wolf. They just do not see the wolf coming.

The Federal Reserve is unusually candid about this, and publishes the arithmetic itself. Alongside every set of projections it prints the historical error bands for forecasts of that horizon, measured against outcomes for 2006 through 2025. From the projections released on September 16, 2026:⁵

Average historical projection error ranges, percentage points, from Table 2 of the Fed’s September 16, 2026 projections. Computed as the root mean squared error of fall-released forecasts for 2006 through 2025 by various private and government forecasters. The Fed attaches roughly a 70% probability to outcomes landing inside these ranges.

A band of plus or minus 1.8 points on next-year growth is not a rounding error. The Fed’s own worked example spells it out: a participant projecting 3% growth is, on this record, covering a range of 1.2% to 4.8% for the following year. The accompanying note says the models behind the numbers “are necessarily imperfect descriptions of the real world.”

The most useful line in the whole document is easy to miss. In the Forecast Uncertainty box, the Fed observes that “the dispersion of the projections across participants is much smaller than the average forecast errors over the past 20 years.” Eighteen participants clustering around the same number tells you they read the same data, and says nothing about whether the number is right. Consensus can be tight and collectively wrong, and tightness is the thing that reads as confidence.

Where forecasters do earn their keep

The same literature contains the opposite finding, and it is the reason to keep reading forecasts at all.

Romer and Romer compared the Fed’s internal inflation forecasts against commercial ones and found the Fed “has considerable information about inflation beyond what is known to commercial forecasters.” Their regressions were stark enough that they concluded someone holding both forecasts should “put essentially no weight on the commercial forecast.” The reason is prosaic and matters: not inside information, but “the vast resources it devotes to forecasting.”⁶ A forecasting staff of that size buys a measurable edge over the commercial panel, and it is still an edge inside the error bands above.

The Philadelphia Fed’s Survey of Professional Forecasters, running since 1968, has been checked the same way. A 2023 research brief tested whether the panel gets worse when its members disagree, which is the intuitive guess, and found the reverse: “the SPF becomes the more relatively accurate forecast as forecaster disagreement rises.”⁷ Disagreement among forecasters is not a sign they have lost the thread.

Where they are reliably imperfect is speed. Coibion and Gorodnichenko found pervasive evidence that professional forecasts do not incorporate new information promptly; they underreact and adjust sluggishly.⁸ That is a real flaw, and it is a different flaw from the one people complain about.

There is also a scoring problem in the complaint itself. An economist who puts 30% on a recession that does not arrive was not wrong. Events given 30% should happen about 30% of the time, and judging a probabilistic forecast by whether its likeliest case materialized will mark down good forecasters and reward loud ones. Furman drew the related distinction in the episode excerpt, between the models and the people on television talking about them: “when you look at those models, you see things denominated in the tenths of a percent. When you see someone on TV, they sound much bigger than tenths of a percent.” His summary was that “economics, in summary, is a little bit better than economists on this set of questions.” He also conceded the case against the models where it holds: on the post-2021 inflation surge, “one place the models did go wrong was the inflation. There they were wrong by several percentage points.”

The price already contains everyone else’s forecast

Here is where this stops being a debate about economists and starts being about your portfolio.

Fama’s definition is the one to hold onto: “A market in which prices always ‘fully reflect’ available information is called ‘efficient.’”⁹ You do not have to accept the strong form of that to draw the practical conclusion. Grossman and Stiglitz showed the strong form cannot even be coherent when information costs money to gather, because nobody would pay for information that prices already gave away. What exists instead is what they called “an equilibrium degree of disequilibrium,” in which prices reflect informed traders’ information “only partially, so that those who expend resources to obtain information do receive compensation.”¹⁰

Markets need not be omniscient for that to bind, only competitive enough that a widely discussed macro view is already in the price by the time you hear it. Muth’s 1961 formulation of expectations put the same thing in one clause: “Information is scarce, and the economic system generally does not waste it.”¹¹

Financial forecasts have a property weather forecasts lack. Tomorrow does not read the forecast and change its mind. A valuable, publicly known prediction of high returns gets traded on, which lifts today’s price and lowers the return that was predicted. The forecast spends itself. What pays is knowing how the outcome differs from what today’s price already assumes. You can call a recession correctly and lose money because a worse one was priced, or call a boom correctly and lose money because a better one was priced.

Robert Lucas found the matching problem inside the models. Because the equations describe how people make decisions, and people change their decisions when the rules change, “any change in policy will systematically alter the structure of econometric models.”¹² Relationships fitted under one policy regime need not survive the next one. The object being measured moves when you act on the measurement.

None of this means markets are right. Shiller’s variance-bound work found that stock prices over the previous century moved “five to thirteen times too high” to be explained by news about the dividends that followed.¹³ Prices overshoot, bubbles happen, forced sellers move markets. But notice what that does and does not license. That markets are sometimes wrong is not evidence that you can tell when, in advance, in time to trade on it. Those are separate claims, and only the second one would pay.

Two adjacent bodies of evidence belong here in summary. On whether any variable predicts the equity premium out of sample, the answer is barely. Goyal, Welch and Zafirov’s update re-tested the predictors published since 2008 and found most had already lost their significance, though Campbell and Thompson showed that sensible restrictions leave predictive power that is small and “economically meaningful.”¹⁴ Our guide on the R² trap walks the statistics, and the CAPE evidence covers the one signal that does carry long-horizon information. And even a correct growth forecast would not hand you a return: Ritter found the cross-country correlation between real per-capita GDP growth and real equity returns over 1900 to 2002 was negative, at −0.37 across 16 countries, because growth funded by new capital and new firms need not reach existing shareholders.¹⁵ That argument gets its own guide, the stock market is not the economy.

How often would you have to be right?

Put numbers on it. Take the simplest possible version of acting on a macro view: each January you either hold stocks for the year or sit in Treasury bills, based on your call.

Using Damodaran’s annual nominal total returns for the S&P 500 and 3-month T-bills from 1928 through 2025, ninety-eight years, stocks beat cash in 66 of them. That single fact sets the bar. Someone who never forms a view, never reads a forecast and simply stays invested has, in effect, called it right 67% of the time. The starting line is not a coin flip.

In the 66 good years stocks compounded at 22.0% against 3.1% for cash. In the other 32, stocks lost 11.2% a year while cash made 3.9%. Buy and hold across all ninety-eight returned 10.02%; perfect foresight, switching correctly every single year, would have returned 15.79%. Solving for the accuracy a timer needs to merely match buy and hold:

Our calculation on Damodaran’s annual returns, 1928 to 2025, nominal and before tax. A “good” year is one in which the S&P 500 total return beat 3-month T-bills. Accuracy is assumed equal on good and bad years.

A forecaster right two-thirds of the time, which would be respectable, produces nothing at all here: at 67% accuracy and a 50 bp round trip the timer earns 9.45% against 10.02% for doing nothing. At 60% accuracy the gap widens to nearly two points a year. The reason is visible in the conditional returns above. The good years are much larger than the bad years are painful, so exiting wrongly costs more than exiting rightly saves, and a middling hit rate produces mostly the first kind of mistake.

This is not a new result, and arriving at it independently is some comfort that the arithmetic is sound. Sharpe ran the same experiment in 1975 on 1929 to 1972 data and concluded a manager should not try unless they could call the year right “at least seven times out of ten”; his tables put the threshold for beating buy and hold higher still, around 83%, with 74% the figure for beating a volatility-matched portfolio.¹⁶ A 2021 replication on 1929 to 2018 data landed back at seven out of ten.¹⁷ Vanguard, testing timing driven by anticipating economic surprises, put it at “correct at least 75% of the time to get a return only slightly higher” than a 60/40 portfolio.¹⁸

Set that hurdle against the record above. Professional forecasters, with staffs and models, saw 5 of 153 recessions coming a year out. The bar for a macro call to pay is around 70% accuracy, year after year, net of what trading and taxes take.

Try it with your own numbers

Move the accuracy slider to whatever you honestly believe your hit rate is, and set the cost to something like your real spread plus the tax you would owe on gains you realize to get out. The break-even number moves with the cost. The 67% that comes from staying invested does not.

What we recommend

Read forecasts, and treat each one as an input to a single question: how much uncertainty does my plan have to absorb? That is a different question from what to own next quarter, and it is the answerable one.

The practical difference shows up in how you phrase the question. Deciding to sell equities because a recession is coming in 2027 requires you to be right about the recession, its size, its timing, and about all of that relative to what is already priced. Treating a recession as one of the states of the world you should be able to live through requires none of that, and points instead at things you control: whether you hold enough cash and short bonds to avoid selling equities into a decline, whether your job income is correlated with the assets you own, whether a 40% drawdown breaks your withdrawal plan or merely annoys you.

Concretely, in rough order of how much they matter:

• Pick an allocation you can hold through a bad year, then test it against a range of outcomes rather than your favorite scenario. Summitward’s retirement tools run the distribution, and what Monte Carlo does and does not tell you covers how to read the output.
• Hold enough liquidity to avoid selling into a decline. Cash and short bonds cover an unforecast recession, and sizing them takes no forecast at all.
• Rebalance on a rule, not on a view. A band or a date both work. Either one makes you buy what fell without needing an opinion about why it fell.
• Separate decisions that contain a forecast from decisions that do not. Buying an asset because of the price on the screen today is a different act from buying it because you think the price will move. Locking in a TIPS yield works through that distinction on a live example.
• Write the plan down before the news cycle arrives, so the forecast you hear in a downturn is arguing with a document instead of with your nerves. Our guide to writing a financial plan has the structure.

You do not need to predict the economy to do well as an investor. That is the practical content of everything above.

Frequently asked questions

Should I sell stocks if economists are predicting a recession?

On the record above, a published recession call is both unlikely to be right and already reflected in prices if it is widely held. Selling on it asks you to be correct about the recession and about how much of it is priced. If a recession would genuinely break your plan, the fix is the allocation and the cash buffer, applied now and regardless of any forecast, rather than a trade timed off one.

Does this mean economists are bad at their jobs?

No, and the evidence cuts both ways. The Fed’s forecasts carry information commercial forecasters lack, and the SPF beats statistical benchmarks. What the record shows is that turning points are very hard, that error bands are much wider than point estimates suggest, and that the gap between a model output measured in tenths of a percentage point and a television segment about catastrophe is large.

If a recession does arrive, won’t stocks fall?

Often, but that is not enough to trade on. Equities have historically turned up well before a recession ends, the decline may be smaller than what was priced, and falling rates can lift valuations while profits shrink. Being right about the economy and wrong about the market at the same time is common.

Is there any forecast worth acting on?

Prices are forecasts you can read without predicting anything: a TIPS yield, a breakeven inflation rate, a bond’s yield to maturity. They tell you what a real return is available today. That is different in kind from someone’s estimate of where it will be next year.

How accurate would I have to be for timing to pay?

On 1928 to 2025 US data, about 70% of years called correctly at a 50 basis point round trip, against the 67% a buy-and-hold investor scores without forming a view at all. Published estimates land between roughly 70% and 83% depending on the comparison and the cost assumption.

Key takeaways

• Forecasters expected output to fall in 5 of 153 recessions across 63 countries as of April the prior year, and official and private forecasts were about equally unprepared.
• The Fed publishes its own historical error bands, plus or minus 1.8 points on next-year growth, and notes that the spread of opinion among its participants is much narrower than those errors.
• Forecasts still carry information. The Fed knows things commercial forecasters do not, and professional surveys beat statistical benchmarks, including when the panel disagrees.
• A forecast has to beat the price rather than the outcome, and a public, valuable forecast changes the price it was predicting.
• US stocks beat cash in 66 of the last 98 years, so never forecasting is already a 67% hit rate, and a yearly stocks-or-cash timer needs roughly 70% to break even after modest costs.
• That markets are sometimes wrong does not establish that you can identify when in advance.
• Build a plan that survives a range of outcomes instead of rebuilding the portfolio around whichever macro narrative is most persuasive this quarter.

Related guides

• The Stock Market Is Not the Economy takes the Ritter result apart link by link, with a calculator that walks GDP growth through to per-share return.
• The R² Trap explains why the charts showing valuation predicting returns look more certain than they are.
• Is Locking In Today’s TIPS Yields Market Timing? applies the forecast-versus-price distinction to a decision many retirees face right now.
• Missing the Market’s Best Days shows why the famous chart is a weaker argument for staying invested than the difficulty of the decision itself.
• Do Stock Valuations Still Matter? covers the one widely accepted case where prices carry long-horizon information about returns.
• Does the Fed Really Set Interest Rates? is the companion on unobservable quantities, including how far the FOMC’s own long-run estimate has drifted.
• The Stock Market Rewards Optimists, But Not the Naïve Ones covers the equity risk premium you cannot know precisely, and includes a plan stress-tester.
• Personal Finance Gurus vs. Economists looks at when simple rules outperform the textbook models under the same uncertainty.

Sources

1. Thompson, D., and Furman, J. (September 15, 2026). Why Experts Keep Getting the Economy So Wrong, Plain English With Derek Thompson, The Ringer. Quotations are from the edited excerpt published on the episode page, read September 16, 2026; no full transcript is posted.
2. Hayek, F. A. (1945). The Use of Knowledge in Society, American Economic Review 35(4), 519–530.
3. Hayek, F. A. (December 11, 1974). The Pretence of Knowledge, Nobel Memorial Prize lecture.
4. An, Z., Jalles, J. T., and Loungani, P. (2018). How Well Do Economists Forecast Recessions? IMF Working Paper 18/39. Published as International Finance 21(2), 100–121.
5. Federal Reserve (September 16, 2026). Summary of Economic Projections, Table 2 and the Forecast Uncertainty box. Error ranges follow Reifschneider and Tulip, FEDS 2017-020.
6. Romer, C. D., and Romer, D. H. (2000). Federal Reserve Information and the Behavior of Interest Rates, American Economic Review 90(3), 429–457.
7. Doelp, P. (December 2023). Forecast Accuracy and Forecaster Disagreement in the Survey of Professional Forecasters, Federal Reserve Bank of Philadelphia Research Brief. Sample 1985:Q1 to 2017:Q4.
8. Coibion, O., and Gorodnichenko, Y. (2015). Information Rigidity and the Expectations Formation Process, American Economic Review 105(8), 2644–2678.
9. Fama, E. F. (1970). Efficient Capital Markets: A Review of Theory and Empirical Work, Journal of Finance 25(2), 383–417.
10. Grossman, S. J., and Stiglitz, J. E. (1980). On the Impossibility of Informationally Efficient Markets, American Economic Review 70(3), 393–408.
11. Muth, J. F. (1961). Rational Expectations and the Theory of Price Movements, Econometrica 29(3), 315–335.
12. Lucas, R. E., Jr. (1976). Econometric Policy Evaluation: A Critique, Carnegie-Rochester Conference Series on Public Policy 1, 19–46.
13. Shiller, R. J. (1981). Do Stock Prices Move Too Much to Be Justified by Subsequent Changes in Dividends?, American Economic Review 71(3), 421–436.
14. Goyal, A., Welch, I., and Zafirov, A. (2024). A Comprehensive 2022 Look at the Empirical Performance of Equity Premium Prediction, Review of Financial Studies 37(11), 3490–3557. The “economically meaningful” phrase is from Campbell, J. Y., and Thompson, S. B. (2008), Predicting Excess Stock Returns Out of Sample, Review of Financial Studies 21(4), 1509–1531.
15. Ritter, J. R. (2005). Economic growth and equity returns, Pacific-Basin Finance Journal 13(5), 489–503.
16. Sharpe, W. F. (1975). Likely Gains from Market Timing, Financial Analysts Journal 31(2), 60–69. Paywalled; the thresholds quoted here are as reported in Buzzacchi and Ghezzi (2021).
17. Buzzacchi, L., and Ghezzi, L. (2021). The Odds of Profitable Market Timing, Journal of Risk and Financial Management 14(6), 250.
18. Vanguard (February 12, 2024, updated July 28, 2025). Adding value through a strategic approach.
19. Damodaran, A. (January 2026). Annual Returns on Stock, T.Bonds and T.Bills, NYU Stern. The source for every figure in the accuracy tables and the calculator.

Editor’s note

Educational content, not investment advice. The accuracy hurdle is our own calculation on Damodaran’s annual nominal total returns for the S&P 500 and 3-month T-bills, 1928 to 2025, defining a good year as one in which stocks beat cash, holding within-group returns at their full-sample averages and charging the stated cost whenever the position changes. It is one historical sample and it is before tax except for what you enter in the cost field. That it reaches 74.9% at the 2% switching cost Sharpe assumed, by a different route and on different data, is a check on the arithmetic rather than a replication of his study. Fed figures are from the projections released September 16, 2026, and will be superseded at the next meeting. Talk to a professional about your own situation.