What it shows
Two lines. The blue line is US Manufacturing PMI, the headline number from ISM’s monthly survey of purchasing managers. The red-orange line is the share of the world’s central banks currently in a rate-cutting cycle, redrawn shifted forward in time — the buttons above the chart let you pick how far forward: 0 (no lead), 9, 11 or 12 months. Central-bank policy moves have historically shown up in US manufacturing conditions with a lag, and that lag has drifted from about 9 months before 2008 to 11-12 months since 2016.
The background colour is the regime the cutting signal was actually in that month: green where a majority of central banks were cutting, amber where a majority were hiking. Those two colours cover the whole chart, because in this framework every month is one or the other — there is no neutral state. The dotted grey line on the right-hand axis marks that 50% threshold itself.
How the lead was derived
Pick “0 (no lead)” and the two lines barely track each other — that comparison is the reason the other three buttons exist, not a decorative option. The lead itself is not a guess: for four overlapping windows of history, we z-scored both series and tested every lead from 0 to 24 months, keeping whichever one produced the strongest correlation:
| Window | Best lead | Correlation at best lead | Correlation with no lead |
|---|---|---|---|
| 1985 to present (full history) | 11 months | 0.61 | 0.09 |
| 2008 to present | 11 months | 0.69 | 0.11 |
| 2016 to present | 12 months | 0.77 | 0.22 |
| 2020 to present | 12 months | 0.78 | 0.28 |
Two things this shows. First, the lead matters: correlation with no lead applied at all is weak (0.09-0.28) in every window, and jumps to 0.61-0.78 once the right lead is applied — the relationship genuinely is a leading one, not a coincident one dressed up to look that way. Second, the best-fit lead itself has drifted, from 11 months over the full 1985+ history to 12 months in the two most recent windows, which is why the chart offers 9, 11 and 12 rather than committing to one number: whichever lead was “correct” pre-2008 is not necessarily the one that fits now, and the next few years could shift it again.
This is a different, older kind of evidence than the backtest below it — a same-window fit, not a held-out test — which is exactly why the regime-episode results below check the sturdier claim (out-of-sample forward returns) rather than resting on correlation alone.
How to use it
The shading is the actionable part, not the lines. A backtest across every historical episode since 1985 found that entering a majority-cutting regime (green) has been followed by positive returns in the S&P 500, Nasdaq 100 and Russell 2000 roughly 80-90% of the time at 12 months and beyond, with the relationship confirmed on data the backtest never saw when fitting it — not just a pattern that happened to fit the history it was built on.
Every cutting regime since 1985 that lasted 6 months or longer, and what happened to each index in the year and two years after it began (green background is a gain, red is a loss):
| Started | Ended | Length | S&P 500 | Nasdaq 100 | Russell 2000 | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| +3mo | +6mo | +12mo | +24mo | +3mo | +6mo | +12mo | +24mo | +3mo | +6mo | +12mo | +24mo | |||
| Mar 1985 | May 1988 | 3.3y | +6.2% | +0.8% | +32.2% | +61.5% | — | — | — | — | — | — | — | — |
| Jan 1991 | Nov 1991 | 0.9y | +9.1% | +12.8% | +18.9% | +27.6% | +13.4% | +17.1% | +45.6% | +59.4% | +18.3% | +19.8% | +42.3% | +58.2% |
| Sep 1992 | Sep 1994 | 2.1y | +4.3% | +8.1% | +9.8% | +10.7% | +15.0% | +14.8% | +22.2% | +25.8% | +14.6% | +18.8% | +31.1% | +32.8% |
| Apr 1995 | Dec 1997 | 2.8y | +9.2% | +13.0% | +27.1% | +55.7% | +21.2% | +27.5% | +42.0% | +86.3% | +12.6% | +11.3% | +30.8% | +28.9% |
| Jul 1998 | Dec 1999 | 1.5y | −2.0% | +14.2% | +18.6% | +27.7% | +1.7% | +54.5% | +64.9% | +162.1% | −9.9% | +1.8% | +6.0% | +19.3% |
| Jan 2001 | Nov 2004 | 3.9y | −8.5% | −11.3% | −17.3% | −37.4% | −28.5% | −35.1% | −40.2% | −62.1% | −4.5% | −4.6% | −5.0% | −26.8% |
| Nov 2008 | Jan 2011 | 2.3y | −18.0% | +2.6% | +22.2% | +31.7% | −5.8% | +21.1% | +49.1% | +78.6% | −17.8% | +6.0% | +22.5% | +53.7% |
| Oct 2011 | Sep 2018 | 7.0y | +4.7% | +11.5% | +12.7% | +40.2% | +4.6% | +15.4% | +12.2% | +43.1% | +7.0% | +10.2% | +10.5% | +48.5% |
| Jun 2019 | Feb 2022 | 2.8y | +1.2% | +9.8% | +5.4% | +46.1% | +1.0% | +13.8% | +32.4% | +89.7% | −2.8% | +6.5% | −8.0% | +47.5% |
| Aug 2024 | TBD | 2.1y so far | +6.8% | +5.4% | +14.4% | +36.1% | +6.9% | +6.7% | +19.6% | +50.5% | +9.8% | −2.5% | +6.7% | +33.3% |
Return profile, cutting regimes — % positive and median return at each horizon, across all 10 episodes above:
| Horizon | S&P 500 | Nasdaq 100 | Russell 2000 | |||
|---|---|---|---|---|---|---|
| % positive | Median | % positive | Median | % positive | Median | |
| +3mo | 70% | +4.5% | 78% | +4.6% | 56% | +7.0% |
| +6mo | 90% | +9.0% | 89% | +15.4% | 78% | +6.5% |
| +12mo | 90% | +16.5% | 89% | +32.4% | 78% | +10.5% |
| +24mo | 90% | +33.9% | 89% | +59.4% | 89% | +33.3% |
Cutting’s tailwind shows up almost immediately: already 70-78% positive at 3 months, before climbing further. Signal counts and downside risk:
| Index | Signals | Max drawdown (24mo) |
|---|---|---|
| S&P 500 | 10 | −40.3% |
| Nasdaq 100 | 9 | −67.9% |
| Russell 2000 | 9 | −28.7% |
Regime length, across the 10 episodes:
| Min | Median | Max |
|---|---|---|
| 0.9y (Jan 1991) | 2.5y | 7.0y (Oct 2011) |
Every index’s 12-month losses trace back to the same two episodes — January 2001 and, for the Russell 2000 only, June 2019 — and each index’s worst 24-month drawdown also happened during the January 2001 episode. The Nasdaq 100 and Russell 2000 have no price history before October 1985 and September 1987, hence the dashes on the episode table’s first row and their signal counts of 9 rather than 10. The one exception below is worth reading before treating any of this as a rule.
Every hiking regime since 1985 that lasted 6 months or longer:
| Started | Ended | Length | S&P 500 | Nasdaq 100 | Russell 2000 | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| +3mo | +6mo | +12mo | +24mo | +3mo | +6mo | +12mo | +24mo | +3mo | +6mo | +12mo | +24mo | |||
| Jun 1988 | Dec 1990 | 2.6y | −0.6% | +1.5% | +16.3% | +30.9% | −5.1% | −6.1% | +8.2% | +26.2% | −1.5% | −2.6% | +10.7% | +11.8% |
| Sep 1991 | Aug 1992 | 1.0y | +7.5% | +4.1% | +7.7% | +18.3% | +15.1% | +12.4% | +8.9% | +33.1% | +5.4% | +13.1% | +7.1% | +40.4% |
| Oct 1994 | Mar 1995 | 0.5y | −0.4% | +9.0% | +23.1% | +49.3% | −1.9% | +13.7% | +45.0% | +82.1% | −3.2% | +4.4% | +16.2% | +33.5% |
| Apr 2000 | Dec 2000 | 0.8y | −1.5% | −1.6% | −14.0% | −25.9% | −4.3% | −13.0% | −50.8% | −66.2% | −1.1% | −1.7% | −4.1% | +0.9% |
| Oct 2005 | Oct 2008 | 3.1y | +6.1% | +8.6% | +14.2% | +28.4% | +8.3% | +7.7% | +9.7% | +41.8% | +13.4% | +18.2% | +18.6% | +28.1% |
| Feb 2011 | Sep 2011 | 0.7y | +1.4% | −8.2% | +2.9% | +14.1% | +0.9% | −4.7% | +11.6% | +16.5% | +3.0% | −11.7% | −1.5% | +10.6% |
| Oct 2018 | May 2019 | 0.7y | −0.3% | +8.6% | +12.0% | +20.6% | −0.9% | +11.7% | +16.0% | +58.6% | −0.8% | +5.3% | +3.4% | +1.8% |
| Mar 2022 | Jul 2024 | 2.4y | −16.4% | −20.9% | −9.3% | +16.0% | −22.5% | −26.1% | −11.2% | +23.0% | −17.5% | −19.6% | −12.9% | +2.6% |
Return profile, hiking regimes — same measure, across all 8 episodes above:
| Horizon | S&P 500 | Nasdaq 100 | Russell 2000 | |||
|---|---|---|---|---|---|---|
| % positive | Median | % positive | Median | % positive | Median | |
| +3mo | 38% | −0.3% | 38% | −1.4% | 38% | −1.0% |
| +6mo | 62% | +2.8% | 50% | +1.5% | 50% | +1.3% |
| +12mo | 75% | +9.9% | 75% | +9.3% | 62% | +5.2% |
| +24mo | 88% | +19.5% | 88% | +29.6% | 100% | +11.2% |
This is the asymmetry that matters most on this chart: hiking starts as a near coin flip — 38% positive at 3 months, median slightly negative for every index — and only crosses 50/50 by 6 months, resolving to 75-100% positive by 24 months. Cutting never goes through that early dip. Signal counts and downside risk:
| Index | Signals | Max drawdown (24mo) |
|---|---|---|
| S&P 500 | 8 | −28.3% |
| Nasdaq 100 | 8 | −69.0% |
| Russell 2000 | 8 | −23.7% |
Regime length, across the 8 episodes:
| Min | Median | Max |
|---|---|---|
| 0.5y (Oct 1994) | 0.9y | 3.1y (Oct 2005) |
Hiking regimes have run shorter on the whole — a median under a year, against 2.5 years for cutting — which fits the same pattern as the near-term coin flip: this regime tends to resolve faster, for better or worse.
The Nasdaq 100’s worst drawdown here (−69.0%) is larger than any cutting episode produced, and both it and the S&P 500’s worst (−28.3%) trace to the same episode: April 2000, the start of the dot-com unwind, which was still underwater for both at 24 months (−66.2% and −25.9% respectively) even though the regime itself had long since ended.
That asymmetry between the two regimes makes some sense: hiking usually begins because the economy is already running hot, so near-term chop while policy bites is normal, whereas cutting usually begins in response to weakness, so the near-term relief tends to be faster.
Read the regime colour first, and the PMI/cutting lines second. PMI still tells you where manufacturing is right now, but the colour is what has actually pointed forward.
The one exception worth knowing
January 2001: the Fed started cutting into the dot-com bust, and it did not work — the S&P 500 was down 37% and the Nasdaq 100 down 62% two years later. One loss in ten green episodes is exactly the kind of thing to know going in, not to discover afterward. A green band is a historical tendency, not a guarantee.
Why Bitcoin isn’t in these figures
It was tested and dropped. Bitcoin has traded since 2010, but our free daily-price source only reaches back to September 2014 — a data-access limit, not a statement that Bitcoin didn’t exist before then. That window still covers just two cutting-regime starts (2019 and 2024) and two hiking-regime starts (2018 and 2022), and the two cutting episodes moved in opposite directions in the short term before eventually both turning positive. Two data points that disagree isn’t a signal, it’s a range of outcomes. The hiking pair was at least more consistent — both fell 45-57% in the first three months before recovering — but n=2 still isn’t enough to state as a finding. A source reaching back to 2011 would add one more episode of each kind, not enough to change this. We’d rather leave this out than publish a win rate built on two or three coin flips.
The fine print
The cutting share comes from the Bank for International Settlements’ monthly policy-rate data across up to 38 central banks. A bank counts as “cutting” in a given month if its rate fell that month, or if its rate held flat while the bank was still inside a cutting cycle it started earlier — a flat month right after a cut is read as a pause, not a reversal. The chart only starts in February 1985, the first month the panel reaches 20 comparable central banks; the raw data goes back to 1945, but with as few as 6 countries reporting, an early reading wouldn’t mean much. This is also why the lead-correlation table above says “1985 to present” rather than the “1945+” the old manual TradingView version of this chart used to quote — that earlier figure came from an unfiltered panel and isn’t reproducible from data we’d stand behind.
PMI is the real ISM Manufacturing PMI, not a proxy. History runs from January 1948; the latest month is added from ISM’s own public press release each month.
The backtest behind the regime shading measures the forward return from the start of each contiguous cutting or hiking episode, not from every month inside it — a 7-year episode and a 7-month one both count once, which is what makes the win rates above a fair read of history rather than one long episode counted dozens of times. Episodes lasting under 6 months (12 such blips since 1985 — 6 on the cutting side, 6 on the hiking side) are excluded from these figures as noise around the 50% line, though they’re still shown in the chart’s shading, since that’s the honest historical record.
“Max drawdown” in the summary rows is the worst monthly close reached relative to the entry price, at any point in the 24 months following the signal — not a running peak-to-trough measure. It answers “how far underwater could this have gotten before it worked out,” not “how much did it ever give back from a high.”
Even at the episode level, 8-10 events is a small sample by any standard — treat the win rates as a documented historical tendency, not a statistical proof, and read the regimes together with everything else on this site rather than as a signal on their own.