Recession Signal: Sell the S&P 500 or Sit Still? 25 Years Tested

Selling the S&P 500 at the first recession signal and buying back once price turned up again turned $100,000 from April 2001 into $1,002,801 after tax by the end of 2025. Just holding the index turned it into $789,844, and on the way it fell 55% from top to bottom. The signal rule never fell more than 34%. Over the full 69 years I could test, back to 1957, it kept the lead: 11.74% a year against 10.71%. And it was not only the S&P 500: on six funds, from the Nasdaq-100 and small caps to financials and developed markets outside the US, the same signal beat holding on every one from 2000 to 2026, and cut the worst fall on every one.

The recession signal I used was not a headline or a pundit. It was the leading-indicators composite inside my own Bull/Bear Index, the same blend of the Weekly Economic Index, real retail sales, inventories to sales, the CFNAI and the OECD leading indicator that I wrote up in the leading economic indicators guide, rebuilt so that every month only uses data that had actually been published by then. One detail decided everything: how you come back. The same exit with the popular “back in when the economy looks healthy again” re-entry turned the same $100,000 into only $561,580.

So this article answers five questions for index investors, not stock pickers. Does getting out at the first warning beat sitting still once you count the rallies you miss? If you do get out, what gets you back in? Can the leading dial mark a moment that avoids a large drawdown, or is it mostly noise? Does the same signal help on other funds, not just the S&P 500? And how does the whole idea compare with the ATH staircase momentum strategy I already published, on the same window, capital and tax?

How I tested the recession signal

I wrote the rules down before running anything, including the threshold and the definition of a false alarm, and I did not change them after seeing 2020 or 2022. Two clarifications went into the same file before the first run, and one sensitivity test (the full Bull/Bear index) was added before it was run. After the first run I found that the indicator history itself let some data in too early, so I rebuilt the data and ran everything again. The original file also had a one-month delay on the signal, as a guard against exactly that problem. With a history that only contains published data the guard is built in, so I dropped it: the rule now acts on what was known at each month-end.

The money side copies the ATH staircase momentum post so the two can sit in one table: $100,000 at the start, 19% tax on every closed holding period (losses credited back the same way), 0.10% cost per switch, and buy-and-hold taxed once at 19% at the very end. The index is the S&P 500 total return series (dividends in). The window is April 2001 to December 2025. A false alarm is an exit with no NBER recession already under way or starting within 12 months, using the official NBER business cycle dates.

The rule runs continuously from 1957, so it carries its own position into the window. On 2 April 2001 every version was already in cash: the composite had crossed below zero at the end of December 2000, when the index was 13% below its peak. Buy-and-hold starts fully invested on the same day. That is the fair comparison, because a reader who followed the rule would have been in cash at that point.

Does selling at the first recession signal beat sitting still?

Yes, if price decides when you come back. No, if you wait for the economy to give you permission. The second version is the one most people have in mind when they say “I’ll get out before the recession and back in when it’s over,” and it was the worst of the four.

Three things stand out. First, the price re-entry rule won on both counts at once: more money at the end and a much smaller hole on the way. Every version of the exit cut the worst drawdown hard, from -55% to somewhere between -25% and -41%. If the thing you fear most is watching half your index money disappear, the recession signal did that job.

Second, the wait-for-the-all-clear version paid for that protection with 1.5 percentage points a year. Over 25 years that is almost $230,000 less on a $100,000 start. In my time in the market test a plain 200-day rule showed the same pattern, a smaller hole bought with a smaller pile, and the recession dial with a slow re-entry is no different.

Third, the control row matters. A 10-month moving-average rule on its own, with no macro input at all, made 7.12% a year. The dial plus the average (R2) made 9.76%. So the combination did something neither part did alone: the dial picked when to leave, which cut the number of exits from 17 to 10, and price picked when to come back, which kept the time out short.

Which re-entry rule works, and how much of the rebound you miss

The exit is the part people argue about. The re-entry is the part that decided the result.

“Exits that cost money” are the ones where the index was higher on the day I bought back than on the day I sold. “Rebound missed” is how far the index had already climbed from its lowest month-end close during the cash spell by the time the rule bought back in.

The economy recovers after the market does. That is the whole problem with R1. The composite is built from activity data, and activity bottoms months after prices. In 2009 the S&P 500 bottomed in March, and the composite only turned positive at the end of the year. By the time R1 bought back, at the end of December 2009, the index had already risen 68% from its low. R1 bought back 9% below where it had sold, which still looks like a win, until you see that buy-and-hold owners had been collecting that rally for nine months.

Price does not wait for the data. R2 came back in when the index itself had turned, which meant missing a median 3% of the rebound instead of 9%. Letting either signal bring you back (R4) sounds like the best of both, but it did worse than price alone: brief recoveries in the dial pulled R4 back in during June 2002, just before the next leg of the bear market, and again for a month in mid-2008, with an extra round of costs and tax each time. The fixed six months (R3) was worst on this measure, at 17%: no thinking required, and also no edge worth the trouble.

Event by event: drawdowns avoided vs false alarms

The averages hide how uneven this was. Here is every stress period the plan asked me to look at, scored separately.

Where the recession signal earned its keep

2001 and 2008, both real recessions. At the end of December 2000 the index was 13% below its peak, and the 2001-02 bear market still had most of its damage to do. R2 stayed in cash until March 2002, bought back 12% below where it had sold, and finished the 2001-02 window down 2.5% while buy-and-hold lost 29%. In 2008 the exit came at the end of April, when the S&P 500 was 10.5% off its peak, and R2 did not buy back until June 2009, 32% lower. That kept the strategy’s drawdown to -18% while the index fell 55% from top to bottom. Those two episodes are most of the case for the idea.

COVID: right for the wrong reason

The composite went negative in August 2019 and again at the end of October 2019, months before anyone was pricing a pandemic. It was reacting to a manufacturing and trade slowdown, not predicting a virus. R1 happened to be in cash for the crash and kept its drawdown to -9.5%. R2 had already bought back in November 2019, when price was still above its average, and took the full -34%. I would not count COVID as evidence that the dial sees shocks coming. It was a coincidence that happened to look like foresight.

2018 and 2022: late or silent

The Q4 2018 drop of almost 20% happened with the composite still positive. It only crossed in February 2019, after the market had largely recovered. 2022 was worse. The index fell about 25% to its October low on inflation and rate hikes while activity held up, and the composite, using only data published at the time, stayed positive the whole way down, sliding to +0.06 by the October low. It slipped below zero only in mid-December, two months after the low, and the rule sold at the end of December 2022, when the index was already 8% off the bottom. R2 bought back a month later, 6% higher. R1 sat out until October 2023 and bought back 11% higher. A slowdown in the economy and a bear market in stocks are different events, and 2022 was a bear market without a recession.

False alarms that cost a little each

For the price re-entry rule, seven of ten exits were false alarms: 2002, after the recession had ended, 2012, the 2015-16 industrial slump, 2016, February 2019, December 2022 and a soft patch in spring 2024. Each cost a few percent at most, not a disaster, but they add up. The 2025 tariff scare was a near miss rather than a signal: the composite sank to +0.05 in June 2025 and +0.03 in July but never crossed, so the rule stayed invested. For 2026 so far there has been no exit either: the composite has stayed positive all year, and its normalized score hit the top of its range in July before easing to about +0.46 in early October.

Where the edge came from

When I split the window in two, the picture is blunt. These are compound rates on the running equity, before the final tax bill on buy-and-hold, so they flatter buy-and-hold a little.

The advantage was built in the first nine years, when the index went nowhere through two deep bear markets and the rule turned that lost decade into 7.19% a year. In the sixteen years since, every timing version lagged buy-and-hold, R2 by about two and a half points a year. That is the shape of any insurance: long stretches of paying premiums, and a few years when it pays out. Over the whole 25 years, and over the whole 69, the payouts were bigger than the premiums.

Why the smaller drawdown matters as much as the return

The extra point a year is the headline. The part that decides whether a real person sticks with a plan is the hole in the middle.

A 55% fall needs a 124% gain just to get back to where you started, and in 2008-09 that took until 2012. Very few people sit through that without selling somewhere near the bottom, and the investor who sells at the bottom does far worse than either line in this article. The signal rule’s worst moment was the COVID crash, which was over in weeks and recovered within five months. In 2008, when holding meant watching more than half the money disappear, the rule had been in cash since the end of April.

So the rule wins twice: a little more money at the end, and a path that is much easier to live with. The second win is the one I would weigh more, because a strategy only works if you are still following it when it matters.

Robustness: 69 years, a stricter signal, the full index

A 25-year window with two big bears is a small sample. So I ran the same rules from July 1957 to September 2026, using the S&P 500 price index plus dividends from Robert Shiller’s dataset, cash at the 3-month T-bill rate from FRED, and no tax, since nobody kept one tax regime for 69 years.

This is the strongest evidence in the article, because it covers ten NBER recessions instead of three, from 1957-58 to COVID, including the deep bear markets of 1969-70, 1973-74 and the early 1980s. Letting price bring you back beat holding by about one point a year over 69 years, with the same -34% worst drawdown against -55%. Waiting for the dial to clear lost about two and a half points a year, the same pattern as in the shorter window. A stricter version of the signal fired less often but arrived later, which removed the edge.

One more lever is what the cash earns while you are out. The tests above use the actual 3-month T-bill rate, which averaged only 1.7% from 2001 to 2025 and sat near zero in 2009-15 and 2020-21. If your cash earns at least 3% (a deposit, a short bond fund), the edge grows a little:

The effect is modest because the rule is in cash only about 15% of the time, but it all goes the same way. Over the long sample a flat 3% is actually less than T-bills paid on average (4.3%), which is why that cell is lower. Even with the 3% floor, waiting for the dial to clear (R1) still lost to holding, by about a point a year from 2001 and more than two over the long sample.

The last row of the robustness table matters for anyone who reads my Bull/Bear dashboard. The full index reacts to credit spreads, volatility, momentum and sentiment as well as the economy, so it flips far more often. Used as an in-or-out switch on the S&P 500, it gave 22 exits in 25 years and lost more than two points a year to holding. The dial is built to describe the regime. It is not a sell button, and this test is the clearest reason I can give for not using it as one.

Does the recession signal work on other ETFs?

Most readers do not hold the S&P 500 total return index itself. They hold a Nasdaq fund, a small-cap fund, a sector or a world fund. So after the main results were in, I ran the same rule on six ETFs with history back to about 2000, which means both recession bear markets of this century are inside. One change makes it practical: every fund comes back in on the same date, when the S&P 500 closes a month above its 10-month average, and exits on the same date, when the composite crosses below zero. So there is one signal for everything you own. Each test is a single lump sum, dividends in, cash at 2% while out, 0.10% per switch, no tax.

Six out of six, and the gains are largest where the recessions hurt most. Financials fell 83% in 2007-09 if you held them, and the signal had the money out from April 2008, so its worst fall was about half that. The Nasdaq-100 lost 83% from the dot-com peak, and the signal was out from December 2000. Funds outside the US followed the same pattern, which makes sense: US recessions drag the rest of the world down with them. The full-period numbers for the S&P 500 are higher than in the main test because this run starts in January 2000, ten months before the first exit, and has no tax.

A start near the 2000 peak flatters any exit rule, so the fairer question is how often you would have come out ahead if you had started in any month. I checked every monthly start from 2000 onwards:

On average the signal came out ahead at every horizon and on every fund, and the worst fall in any window was smaller every time. The win rate is the honest part: over five years you came out ahead in 56% to 69% of starting months, over eight years in 60% to 78%. The losing windows are the ones that sat inside the long bull market after 2010 and only collected false alarms. The winning ones that contained a recession won by a lot, which is why the average edge is so much bigger than the median one. Over a single year the most common result was a draw: in about half of all 12-month windows the signal never fired, so both versions held the same fund all year.

Separate checks with a single lump sum and 50 random windows, rather than every start month, told the same story on the long windows. Over 100 months the signal beat holding in 84% of windows on small caps, 70% on the Nasdaq-100 and 70% on financials, and the worst drawdown fell from -83% to -60% on the Nasdaq-100 and from -83% to -43% on financials.

Does the recession signal still win if you invest every month?

The tests above are one payment at the start. Most people add money every month, and a monthly plan already buys some of the dips that a lump sum has to sit through. So I ran the same R2 rule as a monthly plan: $500 deposited every month, nothing up front, on 50 random windows drawn from January 2000 to October 2026. While the signal was out, that month’s deposit sat in cash. The comparison is plain DCA into the same fund, the same $500 on the same dates, left invested the whole time. Exit when the leading composite crosses below zero, buy back when the S&P 500 closes a month above its 10-month average. No tax, 0.10% per switch. I ran each fund twice, once with no interest on the cash and once at 2%.

The signal version beat plain DCA on average in every one of these runs, and the worst fall was smaller in every one, including the runs where cash earned nothing.

Across the ten runs, plain DCA averaged 5.3% a year and the signal averaged 6.2%, so the average CAGR improvement was 0.8 points. The average drawdown inside a window went from -36% to -25%, an improvement of 11 points. The worst fall in each run was 23 points shallower on average. The gap is smaller than in the lump-sum tests, which is what you would expect: each new deposit buys after the crash as well as before it, so stepping aside protects a smaller share of the money. The columns above are the average annual return across the 50 windows of that run.

Developed markets outside the US are the honest exception on the win rate. The average was ahead, by 1.1 points a year with no interest on cash and by 0.3 points at 2%, but the signal won only 38% and 40% of those 50-month windows, and the median window was a fraction behind. A few windows that contained a recession carried the average, which is the same shape as the lump-sum results. Over 100 months, where the monthly plan was run on small caps and financials, the signal won 84% and 64% of windows.

A single year is too short for this rule, and that is true of monthly deposits too. Two draws of $500 a month into the S&P 500 for 12 months, with no interest on cash, went different ways: 20 windows came out 0.8 points a year behind plain DCA, and 50 windows came out 0.7 points ahead. In most of those years the signal never fired, so the two versions held the same fund.

Recession signal exit vs the ATH staircase momentum strategy

The momentum figures are the ones published in the original post, not a re-run, and the drawdown column uses the same measure that post used (the worst drop inside any single calendar year). On that measure my buy-and-hold matched the post’s own benchmark almost exactly (-47.0% vs -47.1% in 2008), which tells me the comparison is lined up. One difference is worth stating: my benchmark includes dividends and ends at $789,844, while the momentum post’s benchmark ended at $603,958 after tax. Against my benchmark, after tax on both sides, the staircase leads by about 2.4 points a year, and by about 1.3 points against the best recession exit.

Even so, the staircase beat every recession-exit version on return and matched the best of them on pain. It also does something a macro switch cannot: in 2022 it lost about 9% while the index lost 19%, because the stocks still making staircase highs were not the ones being crushed by rate hikes. But it is a different job. It owns five stocks at a time, it needs a monthly scan and discipline, and it can have years like 2018 or 2023 where it trails the index by 10 points or more. For someone who wants to own the index and nothing else, it is not an option, and this comparison should not be read as “switch to stock picking.”

What I would actually do with a recession signal

• Use the leading composite for the exit. Check it at each month-end. When it crosses below zero, move the index money to T-bills. That is the part that kept 2001-02 and 2008 from doing their full damage.
• Let price alone decide the way back. Buy back at the first month-end close above the 10-month average. Do not wait for the economy to look healthy: that was the worst version in every test, and adding the dial as a second way back in made it worse too.
• Use one signal for every equity fund you hold. Exit everything on the composite, and bring everything back on the S&P 500 trend. That worked on all six funds I tested, including the Nasdaq-100, small caps and developed markets outside the US, and it means you only watch two numbers.
• Keep adding every month. The rule does not need a lump sum. In the monthly tests the same $500 went in either way, and stepping aside on the signal still finished ahead of plain DCA on average, with a smaller worst fall, even when the cash earned nothing.
• Expect to pay the premium for years. From 2010 to 2025 the rule trailed holding by about two and a half points a year, with several false alarms. That is the cost of having a much smaller hole when a real recession arrives. If you drop the rule after three bad years, you keep the cost and lose the payout.
• Do not expect it to catch every bear market. In 2022 the market fell without any help from the economy, and the signal stayed quiet. It protects against recession bear markets, not against every sell-off.
• Do not use the full Bull/Bear index as the switch. It reacts to everything and flips too often. The leading composite is the part that worked.

Limitations

The composite history is point-in-time: every figure enters on the day it was first published, with its first-release value, taken from the St. Louis Fed’s ALFRED archive of data vintages. That rebuild changed this article. My first draft used the history as it was stored then, with monthly data dated to the first day of the month it describes and with revised values. That version sold in August 2000, just before the dot-com bear market, and again in October 2022, right at the bottom, and both of those well-timed exits disappeared once only published data was allowed. Exact first-release dates are available from 1997 for inventories to sales, 2001 for real retail sales, 2011 for the CFNAI and 2018 for the OECD indicator. Before that, I date each figure using the typical publication delay and use today’s revised values, so the early part of the long sample is still an approximation. The CFNAI only exists from March 2001 and the Weekly Economic Index from April 2020, so the composite has fewer inputs before those dates. The composite was also designed with the past in view, which is a quiet form of hindsight. The 25-year window holds two real recession bear markets, and the edge rests mostly on them. The API does not expose the individual components as separate histories, so I could only test the blend, not each part on its own. The six-fund check was added after the main results, uses a flat 2% on cash and no tax, and covers only 2000 onwards, so it confirms the S&P 500 result rather than adding new recessions. I left out bonds and gold on purpose: they often rise in recessions, and a rule that sells on a recession warning was never meant for them. The monthly-deposit checks are 50 random windows per run, not every possible start month, and I ran them after the lump-sum results were already in.

Conclusion

Acting on the recession signal beat sitting still. Selling the S&P 500 when the leading composite crossed below zero and buying back when price turned up made 9.76% a year after tax against 8.71% from 2001 to 2025, and 11.74% against 10.71% over 69 years, with a worst drawdown of -34% instead of -55% in both. The same signal beat holding on all six ETFs I tested from 2000, by 2.1 to 3.7 points a year, and cut the worst fall on every one. It also beat plain monthly investing: with $500 added every month, the signal version had a higher average return than DCA in every 50-month and 100-month test. Across those ten runs the average CAGR rose from 5.3% to 6.2% (0.8 points) and the average drawdown in a window improved by 11 points, from -36% to -25%. The edge came in the recession bear markets and was partly given back in the long bull market after 2010, which is what insurance looks like over time.

The condition is the way back in. Waiting for the economy to recover before buying back lost to holding (7.22%), because the data turns months after the market. My read: for an index investor who knows a 50% fall would shake them out, a leading-indicator exit with a price re-entry is a better plan than holding through everything, on return and even more on comfort. If you want to beat the index outright rather than hold it with a smaller hole, the ATH staircase did more on this window, but that is stock picking with its own bad years.

FAQ: selling at a recession signal