How to Prepare for a Recession: What the Warning Signs Can and Can't Tell You

Indicator readings are as of October 23, 2026: unemployment 4.2% (September), real-time Sahm rule 0.00 (September), 10-year minus 3-month Treasury spread +0.78 points (October 23), NY Fed 12-month recession probability 12.8% and Cleveland Fed 11.4% (both from September data). Live rates are on the Summitward markets page.

Recessions are partly predictable. The Treasury yield curve has given a warning before every US recession since 1968, and the main recession models at the Federal Reserve banks are built on it. The same signal flashed for two years starting in late 2022, the New York Fed’s model put the odds above 50% for most of 2023 and 2024, and no recession came.¹

Recession preparation should therefore work whether the forecast is right or wrong. The historical record also shows where the strain on a household falls. In the 12 US recessions since 1948, stocks usually bottomed before the recession ended, and unemployment peaked at or after the end in every one. By the month unemployment peaked, the stock market had already climbed 20% to 56% off its low. Job losses arrive late in the cycle, after the market has started to recover, so the plan has to fund that stretch without selling investments.

What counts as a recession, and why you find out late

In the US, recessions are dated by the National Bureau of Economic Research’s Business Cycle Dating Committee. It looks for a significant decline in activity spread across the economy, using payrolls, income, spending and production. Two quarters of falling GDP is a common shortcut, but the committee does not use it as a rule. The NBER dates a recession from the month after a peak through the month of the trough.²

The committee waits until the data are clear, and that takes months. Since it began announcing turning points in 1980, it has named each peak 4 to 12 months after the fact and each trough 8 to 21 months after.³

Selected announcements. Source: NBER Business Cycle Dating Committee announcements.

The stock market does not wait. When the NBER announced the end of the 2007–09 recession in September 2010, the US market was 77% above its March 2009 low. When it announced the end of the 2020 recession in July 2021, the market had doubled from its low. Those figures are Summitward calculations from the Fama/French daily US market index, described in the methodology note below.

What the yield curve has gotten right

The yield curve compares long and short Treasury rates. The usual recession version subtracts the 3-month bill yield from the 10-year note yield. When the spread turns negative, bond investors are accepting less to lock up money for ten years than to roll it every three months. That usually means they expect the Federal Reserve to cut short rates, which it tends to do when the economy weakens. The long rate also contains a term premium, so the spread is a mix of rate expectations and compensation for holding duration, and the mix changes over time.

Arturo Estrella and Frederic Mishkin showed in 1996 that this spread beat other financial and macroeconomic indicators at predicting US recessions two to six quarters ahead.⁴ Their model turns the spread into a probability. A spread of +0.76 points maps to a 10% chance of recession four quarters later, −0.82 to 50%, and −2.40 to 90%. Their 1998 follow-up tested horizons of one to eight quarters out of sample and found the slope of the curve was the best single predictor beyond two quarters.⁵ Fed economists Eric Engstrom and Steven Sharpe later found that the information sits at the short end. In their work, a spread built from the expected 3-month rate six quarters ahead dominated the 10-year spread, and yields beyond 18 months added nothing.⁶

Professional forecasters have tended to give the curve less weight than it earned. Glenn Rudebusch and John Williams compared survey forecasts with a simple real-time yield-spread model and found the model did better at calling recessions a few quarters out.⁷

On the NY Fed’s monthly data, the spread turned negative before each of the eight recessions since 1968:¹

Monthly NY Fed 10-year minus 3-month spread, bond-equivalent basis. Counted to the first month of recession; the timeline below counts from the NBER peak month, one month earlier. Summitward calculation.

That record has three limits. The lead time ran from 6 to 17 months, which is too wide to time anything around. The 1989 inversion was only 0.16 points deep, so it depends on the measure used. And there were false alarms. The curve inverted in 1966, and the next recession began more than three years later. It inverted again from November 2022 through November 2024, reaching −1.71 points, and no recession has followed. (A two-month dip to −0.03 in March and April 2025 is too recent to judge.) The Cleveland Fed lists late 1966, a very flat curve in late 1998, and the 2022–24 inversion as its three notable false positives.⁸

A cleaner test is the record after the rule was published. Since Estrella and Mishkin’s 1996 paper, the curve called the 2001, 2008 and 2020 recessions. (The 2020 recession was caused by a pandemic the curve could not have foreseen.) It then sent the longest false signal in the NY Fed’s data, which start in 1959. The NY Fed’s probability peaked at 70.9% on May 2023 data and was above 50% in 20 of the 21 monthly readings from January 2023 through September 2024.¹ Three hits and one long miss is useful information, but it is too short a record to say how far a 30% or 70% reading can be trusted.

The Sahm rule tells you a recession has started

Economist Claudia Sahm designed her rule in 2019 as a trigger for automatic stimulus payments, a way to recognize a recession quickly. It fires when the three-month average unemployment rate rises 0.50 points or more above its lowest three-month average of the prior 12 months.⁹ On real-time data, which uses only the figures published at the time, it fired between two months before and four months after the NBER peak in each of the nine recessions since 1960. By the time it fires, the recession has usually begun.

It has two false readings in the real-time record: exactly 0.50 in November 1976, and 0.50 to 0.57 from July to September 2024. In July 2024, Sahm wrote that the rule was likely overstating labor-market weakness because immigration had swelled the labor supply.¹⁰ That September, Fed Governor Christopher Waller said that without a big negative shock, neither an inverted curve nor a Sahm trigger necessarily meant a recession, and that most of the rise in unemployment came from people entering the labor force.¹¹ The rule was published in 2019, so every trigger before that is a back-test of a rule fitted to that history. Since publication, it has one hit (2020) and one false alarm (2024).

Consensus forecasts catch recessions once they start

Professional forecasters do far worse a year ahead than once a recession is under way. Zidong An, João Tovar Jalles and Prakash Loungani studied 153 recessions in 63 countries from 1992 to 2014. In April of the year before, the private-sector consensus expected output to fall in only 5 of them. By October of the recession year, it expected a fall in 118.¹² Why Macroeconomic Forecasts Are Not an Investment Strategy covers that study and how accurate a market timer would need to be. By the time a recession is widely expected, stock prices already reflect the expectation.

Markets, the NBER and the job market usually turn in the same order

For each of the 12 recessions since 1948, we dated the stock market’s high and low, the first yield-curve inversion, the first Sahm trigger, the NBER’s announcements and the peak in unemployment, all relative to the NBER peak month. The market series is the Fama/French value-weighted US index, a daily total return including dividends, in nominal terms.¹³ Unemployment is the BLS monthly rate.¹⁴ Pick a recession to see its sequence.

Across the 12 recessions:

• Stocks fell 10% to 55% from high to low, with a median of about 21% on this total-return index. Five of the 12 declines exceeded 30%: 1969–70, 1973–75, 2001, 2007–09 and 2020.
• The market low came before the recession ended in 11 of 12. The exception was 2001. The recession ended in November 2001, but the bear market that began in 2000 kept going until October 2002, 11 months later.
• Unemployment peaked at or after the recession’s end in all 12, by 0 to 19 months. After the 1990–91 and 2001 recessions it kept rising for more than a year after the official end.
• By the month unemployment peaked, stocks were 20% to 56% above their low in every case. Deflating by CPI gives 21% to 53%, so the result does not depend on inflation.
• In the five deep declines, stocks were still 14% to 29% below their prior high when unemployment peaked.

Twelve episodes are a small sample, and the order has exceptions, 2001 most of all. Still, the pattern held in every recession on the unemployment side.

Why the order matters for a household

Your portfolio falls early. Your income is at risk late, and for a long time. Unemployment spells stretch out in a downturn. The median spell was 8.4 weeks in December 2007 and 25.2 weeks in June 2010, and 45.5% of unemployed workers had been looking for 27 weeks or more in April 2010.¹⁵ In September 2026 the median was 11.5 weeks.

Re-employment is slower in a recession, and the new job often pays less. In the BLS survey of long-tenured workers displaced from 2007 to 2009, 49% were working again in January 2010, and 45% of those back in full-time jobs earned as much as before. For workers displaced from 2023 to 2025, the figures in January 2026 were 66.1% and about 49%.¹⁶ Steven Davis and Till von Wachter estimated the long-run cost. Men with three or more years of tenure who lost jobs in mass layoffs lost the present value of 1.4 years of pre-layoff earnings when national unemployment was below 6%. When it was above 8%, they lost 2.8 years.¹⁷

Put the two timelines together. A household that loses its income near the unemployment peak and has to sell stocks to pay bills sells after the rebound has begun. In the five deep declines since 1948, that meant selling 14% to 29% below the old high, and every share sold misses the rest of the recovery. Yes, an Unrealized Loss Is Still a Loss shows how withdrawals during a decline leave a portfolio behind the index for years. The aim of preparation is to never be a forced seller during that stretch.

Income risk is also uneven across workers. Fatih Guvenen, Serdar Ozkan and Jae Song found that recessions mostly add to the chance of a large earnings drop, rather than raising volatility evenly.¹⁸ A follow-up paper on “worker betas” found that earnings move most with the economy at the bottom and the top of the earnings distribution.¹⁹ A finance or tech worker paid partly in employer stock can lose salary, bonus and portfolio value in the same downturn.

How to prepare, in order

1. Measure your runway. Add up essential monthly spending and divide it into cash plus any reliable income that survives a layoff. In the Federal Reserve’s 2025 household survey, 55% of adults had emergency savings covering three months of expenses.²⁰ Emergency Fund Sizing sizes the reserve to your job risk. The FU Money runway calculator tests a job loss and a market drop at the same time.
2. Hold the reserve somewhere that cannot fall. Bank deposits are FDIC-insured up to $250,000 per depositor, per bank, per ownership category.²¹ Treasury bills are backed by the US government, and their interest is exempt from state and local income tax.²² Where to Park Your Cash compares the options after tax.
3. Do not count a credit line as the reserve. Lenders can cut or freeze a home-equity line even if every payment has been on time.²³ In April 2008, about half of banks reported tightening terms on existing home-equity lines.²⁴ Total credit-card limits fell 28% from the third quarter of 2008 to the end of 2010, counting closed accounts.²⁵
4. Know your benefits before you need them. Unemployment insurance replaces about half of prior wages up to a state cap. Most states pay for up to 26 weeks; as of April 2024, 13 paid fewer, as few as 12 in Florida, Kentucky and North Carolina.²⁶ COBRA keeps employer health coverage for up to 18 months after a layoff, at up to 102% of the plan’s cost.²⁷ Losing coverage also opens a 60-day special enrollment window on the ACA marketplace.²⁸ Read your severance and equity terms now. Unvested RSUs usually stop vesting on the day you leave.
5. Separate your job from your portfolio. If your salary, bonus, unvested shares and vested shares all depend on one company or one industry, a single downturn hits all four. One Employer, Four Exposures and Human Capital Risk for Tech Workers cover how to measure and reduce that overlap.
6. Check what your bonds do in a downturn. Around the three recessions since 2000, the 10-year Treasury returned 15.1% in 2002, 20.1% in 2008 and 11.3% in 2020. The S&P 500 returned −22.0%, −36.6% and +18.0% in those years (the 2020 crash was over within five weeks). Treasuries do not protect against an inflation shock: in 2022 both lost about 18%.²⁹ Are Bonds Still Good Diversifiers? explains when each kind of downturn favors which asset.
7. Treat retirement accounts as the last resort. Early withdrawals are taxed as income and usually owe an extra 10%, with exceptions.³⁰ Since 2024, IRAs and employer plans that choose to offer it allow one penalty-free emergency withdrawal of up to $1,000 per calendar year, and no further one for three years unless the first is repaid.³¹
8. Plan the tax moves a downturn opens. A falling market creates tax-loss harvesting opportunities. Buying a substantially identical security within 30 days before or after the sale disallows the loss, and that includes purchases in your IRA.³² A year of low income after a layoff can also be a cheap year for a Roth conversion, provided you have cash to pay the tax and do not need the converted money for living costs.
9. If you are retired or close to it, protect the first years. Losses early in retirement, while you are withdrawing, do the most lasting damage. See Sequence of Returns Risk.
10. Write the downturn rules now. Decide in advance what triggers rebalancing, whether contributions continue after a layoff, and which spending gets cut first. A rule written in a calm month is easier to follow than a decision made at the low after months of bad news.

Using the indicators without timing the market

A rising recession probability is worth acting on when the action is cheap if the forecast turns out wrong. Building cash from new savings instead of selling stocks, updating a resume, putting off a large optional purchase, or paying down a variable-rate balance all cost little if the recession never comes.

Moving the portfolio out of stocks is a different bet. The economy can slow exactly as forecast while the market rises, because prices already reflect the forecast. Eugene Fama and Kenneth French tested US and international markets from 1975 to 2018 and found no evidence that an inverted curve predicted stocks would trail Treasury bills over the following one to five years.³³ (Both disclose that they are consultants to, board members of and shareholders in Dimensional Fund Advisors.) Cash also has a long-run cost. From 1928 to 2025, the S&P 500 compounded at 10.02% a year and Treasury bills at 3.37%.²⁹ An investor who sold at the November 2022 inversion and waited for the recession it predicted would have earned bill yields of about 5% in 2023 and 2024 while the S&P 500 returned 26.1% and 24.9%.

Some households should act on a warning. If you borrow against your portfolio, a market drop can force a margin sale at the worst price (see Personal Leverage). If you need a known amount within two or three years, such as a down payment or tuition, that money should already be out of stocks, recession or not. And if you know your job is at risk, raising cash is planning for your own situation, which you know better than any model does.

Key takeaways

• Recessions are partly predictable. The yield curve inverted 6 to 17 months before each of the eight US recessions since 1968. It also inverted from late 2022 to late 2024 with no recession.
• The Sahm rule confirms recessions; it rarely gives early warning. On real-time data it fired between two months before and four months after each NBER peak since 1960, and in 2024 it fired with no recession.
• The official call comes late. The NBER has announced peaks 4 to 12 months after they happened and troughs 8 to 21 months after.
• Job losses peak after stocks rebound. In all 12 recessions since 1948, unemployment peaked at or after the recession’s end, when stocks were already 20% to 56% off their low.
• Prepare the cash flow and leave the allocation alone. Hold a reserve that cannot fall in price, know your benefits, cut the overlap between your job and your investments, and write downturn rules in advance.

Frequently asked questions

Can anyone predict a recession?

In probabilities, partly. The yield curve has inverted before every US recession since 1968, and Fed models turn it into odds a year ahead. Predicting the month a recession starts, how deep it gets, or how stocks will react has a much weaker record. Forecasters expected output to fall a year ahead in 5 of 153 recessions across 63 countries.

Is the yield curve still inverted, and does it matter?

No. The 10-year minus 3-month spread was +0.78 points on October 23, 2026, and the NY Fed’s model put the odds of the economy being in recession in September 2027 at 12.8%. The 2022–24 inversion is the curve’s longest false signal, so treat any reading as one input among several.

Should I sell stocks before a recession?

The evidence argues against it for most investors. Stocks usually fall before the recession is confirmed and recover before it ends, and Fama and French found inversions did not predict stocks trailing T-bills. Sell only for reasons that hold even without a recession: a near-term spending need, a margin loan you could not meet, or an allocation that was too aggressive for you to begin with.

How much cash should I hold?

Enough to cover essential spending for the length of a realistic job search in your field, plus a margin for slower hiring in a recession. Single-income households, commission or bonus-heavy pay, and cyclical industries need more. Our emergency fund guide works through the factors.

Should I pay off debt before a recession?

Pay off high-rate and variable-rate debt first, but not at the expense of your cash reserve. A paid-off loan cannot pay your rent during a layoff, and credit lines you planned to fall back on can be cut. A low fixed-rate mortgage is usually the last debt worth rushing.

Do stocks always bottom before a recession ends?

Usually, but not always. In 11 of the 12 recessions since 1948 the market low came before the NBER trough. In 2001 the recession ended in November, and stocks kept falling until October 2002.

What is the Sahm rule?

A recession indicator created by economist Claudia Sahm. It triggers when the three-month average unemployment rate rises 0.50 points above its lowest three-month average of the previous year. It identifies recessions early in their course rather than predicting them in advance, and it fired falsely in mid-2024. The real-time reading for September 2026 was 0.00.

Related guides

• Why Macroeconomic Forecasts Are Not an Investment Strategy: how accurate a market timer would need to be.
• Emergency Fund Sizing: 3, 6, 12, or 24 Months?: sizing the reserve to your job risk.
• FU Money: a runway calculator that combines a job loss with a market drop.
• The Stock Market Is Not the Economy: why GDP growth and stock returns diverge.
• Are Bonds Still Good Diversifiers?: bonds in demand-driven and inflation-driven downturns.
• Yes, an Unrealized Loss Is Still a Loss: what withdrawals during a bear market cost.

Methodology

The timeline and the figures in the section on the order of turning points are Summitward calculations. For each NBER recession since 1948, the market low is the lowest daily level of the Fama/French value-weighted US market index from three months before the NBER peak to 12 months after the trough, stopping before the next recession. The market high is the highest level in the 24 months before that low. The unemployment peak is the highest monthly rate from the NBER peak to 36 months after the trough. The first inversion is the first negative month of the NY Fed spread after the previous recession and no more than 30 months before the peak. The Sahm trigger is the first real-time reading of 0.50 or more within six months of the recession. These windows were set before the results were computed. Returns are nominal total returns of a research index, not of a fund, and the unemployment rate is the current revised series.

Sources

1. Federal Reserve Bank of New York, “The Yield Curve as a Leading Indicator”: probit model of the 10-year minus 3-month spread (3-month on a bond-equivalent basis), monthly data and probabilities in allmonth.xls. Peak probability 70.85% on May 2023 data; 12.76% on September 2026 data. newyorkfed.org
2. National Bureau of Economic Research, US Business Cycle Expansions and Contractions. nber.org
3. National Bureau of Economic Research, Business Cycle Dating Committee Announcements. nber.org
4. Arturo Estrella and Frederic S. Mishkin, “The Yield Curve as a Predictor of U.S. Recessions,” Federal Reserve Bank of New York, Current Issues in Economics and Finance 2(7), June 1996. newyorkfed.org
5. Arturo Estrella and Frederic S. Mishkin, “Predicting U.S. Recessions: Financial Variables as Leading Indicators,” Review of Economics and Statistics 80(1): 45–61 (1998); NBER Working Paper 5379. nber.org
6. Eric C. Engstrom and Steven A. Sharpe, “The Near-Term Forward Yield Spread as a Leading Indicator: A Less Distorted Mirror,” Finance and Economics Discussion Series 2018-055, Federal Reserve Board. federalreserve.gov
7. Glenn D. Rudebusch and John C. Williams, “Forecasting Recessions: The Puzzle of the Enduring Power of the Yield Curve,” Journal of Business and Economic Statistics 27(4) (2009); FRBSF Working Paper 2007-16. frbsf.org
8. Federal Reserve Bank of Cleveland, “Yield Curve and Predicted GDP Growth,” September 2026 update: 11.4% probability of recession in one year; false positives in late 1966, late 1998 and late 2022 through late 2024. clevelandfed.org
9. Claudia Sahm, “Direct Stimulus Payments to Individuals,” The Hamilton Project, May 2019; and Federal Reserve Bank of St. Louis, FRED, Real-time Sahm Rule Recession Indicator (SAHMREALTIME), September 2026 reading 0.00. hamiltonproject.org, fred.stlouisfed.org
10. Claudia Sahm, “Sahm-thing more on the Sahm rule,” Stay-At-Home Macro, July 26, 2024. substack.com
11. Christopher J. Waller, “The Time Has Come,” speech at the University of Notre Dame, September 6, 2024. federalreserve.gov
12. Zidong An, João Tovar Jalles and Prakash Loungani, “How Well Do Economists Forecast Recessions?” IMF Working Paper 18/39 (2018), Table 1. imf.org
13. Kenneth R. French Data Library, Fama/French Research Data Factors (daily): market total return as Mkt-RF plus RF, through May 2026. dartmouth.edu
14. US Bureau of Labor Statistics, Unemployment Rate (UNRATE), via FRED. fred.stlouisfed.org
15. US Bureau of Labor Statistics, Current Population Survey: median duration of unemployment (LNS13008276) and share unemployed 27 weeks or more (LNS13025703); Employment Situation Table A-12 for September 2026. bls.gov
16. US Bureau of Labor Statistics, Worker Displacement news releases: 2023–2025 (August 27, 2026) and 2007–2009 (August 26, 2010). bls.gov (2026), bls.gov (2010)
17. Steven J. Davis and Till von Wachter, “Recessions and the Costs of Job Loss,” Brookings Papers on Economic Activity, Fall 2011; NBER Working Paper 17638. nber.org
18. Fatih Guvenen, Serdar Ozkan and Jae Song, “The Nature of Countercyclical Income Risk,” Journal of Political Economy 122(3) (2014); NBER Working Paper 18035. nber.org
19. Fatih Guvenen, Sam Schulhofer-Wohl, Jae Song and Motohiro Yogo, “Worker Betas: Five Facts about Systematic Earnings Risk,” American Economic Review 107(5): 398–403 (2017); NBER Working Paper 23163. nber.org
20. Board of Governors of the Federal Reserve System, Economic Well-Being of U.S. Households in 2025 (May 2026), Savings and Investments. federalreserve.gov
21. Federal Deposit Insurance Corporation, Understanding Deposit Insurance. fdic.gov
22. TreasuryDirect, Tax Forms and Tax Withholding for marketable securities. treasurydirect.gov
23. Board of Governors of the Federal Reserve System, consumer guidance on home equity line reductions and freezes, August 6, 2009. federalreserve.gov
24. Board of Governors of the Federal Reserve System, Senior Loan Officer Opinion Survey on Bank Lending Practices, April 2008. federalreserve.gov
25. Federal Reserve Bank of New York, Quarterly Report on Household Debt and Credit, 2026 Q2 data workbook: credit card limits of $3.70 trillion in 2008 Q3 and $2.66 trillion in 2010 Q4. newyorkfed.org
26. Center on Budget and Policy Priorities, “Policy Basics: How Many Weeks of Unemployment Compensation Are Available?” (revised April 16, 2024) and “Policy Basics: Unemployment Insurance.” cbpp.org (weeks), cbpp.org (overview)
27. US Department of Labor, COBRA continuation coverage (premium up to 102%) and COBRA FAQs for workers (18-month maximum after termination or reduced hours). dol.gov (premium), dol.gov (FAQ, 18 months)
28. HealthCare.gov, Special Enrollment Period. healthcare.gov
29. Aswath Damodaran, “Historical Returns on Stocks, Bonds and Bills: 1928–2025” (January 2026 release): S&P 500 with dividends, 10-year Treasury bond and 3-month T-bill annual returns. pages.stern.nyu.edu
30. Internal Revenue Service, Topic No. 557, Additional Tax on Early Distributions from Traditional and Roth IRAs. irs.gov
31. Internal Revenue Service, Notice 2024-55, emergency personal expense distributions under SECURE 2.0. irs.gov
32. Internal Revenue Service, Publication 550, Investment Income and Expenses (wash sales); Revenue Ruling 2008-5 (IRA purchases). irs.gov (Pub 550), irs.gov (Rev. Rul. 2008-5)
33. Eugene F. Fama and Kenneth R. French, “Inverted Yield Curves and Expected Stock Returns,” working paper, July 28, 2019. Both authors disclose consulting and board roles at Dimensional Fund Advisors. pwlcapital.com

Author disclosure

Educational content, not investment, tax or employment advice. I hold a diversified stock and bond portfolio and do not change it on recession forecasts. The historical timing figures are Summitward calculations from public data under the method above; different windows or a price-only index would move individual numbers. Indicator readings change monthly.