The vote names only the dollar hedges, with 38% of its models
Which are markets in, how sure is the vote, how have assets moved since it began, and what does the macro say?
How this page works
What the page measures, how each part is made and how to read it, before the findings.Exhibit 2 · The page's parts, from the markets' correlations to what each regime meant
- The correlationsHow the core markets move together, day by dayLatest day: 13 Nov 2026
- The modelsEach groups days with similar correlations into regimes of its own8 models
- The voteThe models' regimes matched to shared ones; each model casts one vote a dayToday 3 of 8 name one regime; the vote is mixed
- The namesEach shared regime named by what cushioned stocks in itToday: Only the dollar hedges
- Since it beganHow markets moved through past spells, and how long spells lastedThis spell began between 11 Mar 2026 and 21 Apr 2026
- HedgesWhat cushioned the S&P 500's worst days in each regime
- The designsOther ways of finding regimes, held to the same yardsticks
- The macroThe economy in each regime's months, and today's nearest past months
About this exhibit
Exhibit 3 · Each measure's question, how it is made and how to read it
| Measure | Question | How it is made | How to read it |
|---|---|---|---|
| The correlations | How do the core markets move together? | Every pair of the core markets has a correlation each day, from a model that lets it drift as markets change. | The input every model reads. The map's two axes draw it as a picture. |
| The models | Which regime is each day in, model by model? | Each model groups days whose correlations look alike. Its settings were chosen on one stretch of years and scored on a later one. | The models often disagree, which is why the page counts their votes. |
| The vote and its status | Which regime do the models name, and how far do they agree? | Each model's regimes are matched to the shared regimes, and each model casts one vote a day. Short runs fold into the spell around them. | Clear when most models agree, mixed when two regimes are close, leaning otherwise. The note on the vote's chart gives the thresholds. |
| The names | What does a regime mean for a portfolio? | A regime is named by its legs: whether long Treasuries and the dollar rose when stocks fell in it, on average. | “Only the dollar hedges” means the dollar cushioned falls in stocks there and Treasuries did not. |
| The start window | When did the current spell begin? | The models are drawn again many times, and each draw dates the spell's start. | A wide window means the models disagree about when the change came. |
| The paths and the risk and return | How did markets move through past spells of a regime? | Each market's return counted from the week a spell began, across the regime's spells. The bands come from drawing the models and the spells again. | History, not a forecast. A band that holds zero leaves the sign open. |
| Survival | How long do spells last? | The share of a regime's spells still running after each week. | A steep fall means spells of that regime tend to be short. |
| Hedges on the worst days | What cushioned the S&P 500's worst days? | Each hedge's average return on the S&P 500's worst days, grouped by the regime in force that day. | A hedge is judged on the days it was needed. A regime with few such days is a thin record. |
| The designs | Do other ways of finding regimes agree? | Many designs scored on the same yardsticks: how often they switch, how stable they are, how they read at the time and what followed. | The live rule is one design among them. A design that matches hindsight only after the fact is of little use at the time. |
| The macro | Does the regime describe the economy? | Macro series averaged over each regime's months, and the past months whose macro and pricing sit nearest today's. | A description of what came with each regime, not a cause. |
The settings
| Setting | Options | What it changes |
|---|---|---|
| Model | The model vote by default, or one of the labellings the fund map publishes | The cards of Since it began and Hedges. The headline and the findings stay the vote's. |
| Regime | Any of the chosen model's regimes, the current one first | Which regime's spells the paths, the risk and return and the survival describe. |
| Units | Returns, or units of starting volatility | Returns as they came, or each path divided by the market's volatility before its spell began, so calm and wild starts compare. |
The headline, the findings and the key data describe the model vote. A setting changes only the cards that say so.
What the evidence says
Which regime the vote names, how sure it is, what followed it, and whether anything here forecasts next quarter.- The vote names only the dollar hedges, and resampling its models puts the spell's start between 11 Mar 2026 and 21 Apr 2026. When stocks fall, the dollar rises but Treasuries do not. Oil has returned +22.5% from the vote's start date and +37.5% from the earliest.→ Exhibit 16
- Mixed days have come before a change of regime more often: within a quarter after 35% of them, against 26% of all days. After a clear day the regime changed within a quarter on 20% of occasions, against a base rate of 26%.→ Exhibit 7
- The best next-quarter forecast here needs no regime: it scores +0.29 on co-movement. Half last quarter plus half a two-year average scores +0.29 on co-movement from 2012; the best model, time-series jump model, 126-day memory, scores +0.13.→ Exhibit 8
- No design forecasts next-quarter returns: the best scores ±0.00 out of sample. Out-of-sample R² for next-quarter returns is below zero for every design. So the paths since a regime began describe the past.→ Exhibit 22
- Today's macro is closest to Jun 2018, pricing to May 1998; the two align at the 16th percentile of their record. The regimes explain 61% of the variation across months in private credit growth but 2% of it in oil above its 3-year high.→ Exhibit 31
| What would change this view | Trigger | Today | Decided by |
|---|---|---|---|
| The vote turns | the top two regimes within 20 points, or another regime leads | 3 of 8 models (13 Nov 2026)crossed | the next session |
| A leg crosses the hedge line | stocks–Treasuries below −0.15, or stocks–dollar above −0.15, for 20 sessions | −0.05 · −0.11 (13 Nov 2026)not crossed | 20 sessions |
| The week goes off the 111-fund map | the distance from the nearest centre above 1.37 | 1.38 (week to 13 Nov 2026)crossed | the next week |
| The parts split | two parts outside the regime for four weeks | 5 of 5 in it (week to 13 Nov 2026)not crossed | four weeks |
Next: Risk sentiment — is risk being paid for?
Eight models vote on the regime each day since 1998
The nine core markets read daily from 2 January 1998. Each of eight models chooses its number of regimes and its penalty on 2004–2011, and is scored on 2012–2026.- Stocks' correlations with Treasuries and the dollar are what name the regimes. A leg hedges when the S&P 500's correlation with it in the regime is below −0.15. On 13 Nov 2026 stocks move with long Treasuries at −0.05 and with the dollar at −0.11. → Exhibit 5
- Tuned regimes still add nothing to the forecast. From 2012 the best tuned model, time-series jump model, 126-day memory, scores R² +0.13 on next-quarter co-movement. The regime-free blend scores +0.29. → Exhibit 8
- The 80% vote is a confidence flag, not a better forecast. At the time from 2012, 37% of days were clear, 15% mixed and 48% leaning. The leading regime matched hindsight on 89% of clear days and 24% of mixed days. → Exhibit 7
- Mixed days come before changes of regime more often than other days. After a mixed day the regime changed within a quarter on 35% of occasions. After a clear day it changed on 20%, against a base rate of 26%. → Exhibit 7
- Clear days separate next quarter's co-movement more sharply. The regimes explain 23% of the variance of next-quarter co-movement on clear days, against 27% on all days. → Exhibit 8
- Today 3 of the 8 models vote Only the dollar hedges, so the day is mixed. → Exhibit 1
Exhibit 1 · Each regime's share of the eight models' votes, weekly since 1998
About this exhibit
Exhibit 4 · Every week on the first two axes of the correlation map, by the vote's regime
About this exhibit
Exhibit 5 · The S&P 500's correlation with long Treasuries, the dollar, oil and gold, weekly
About this exhibit
Exhibit 6 · Validation score by model and number of regimes, 2004–2011
About this exhibit
Exhibit 7 · Share of days by status under each threshold, at the time from 2012
| Clear if top share ≥ | Mixed if top two within | Clear | Mixed | Leaning | Agreement when clear |
|---|---|---|---|---|---|
| 70% | 10 points | 62% | 7% | 31% | 85% |
| 70% | 20 points | 62% | 15% | 23% | 85% |
| 70% | 30 points | 62% | 27% | 12% | 85% |
| 80% | 10 points | 37% | 7% | 56% | 89% |
| 80% | 20 points | 37% | 15% | 48% | 89% |
| 80% | 30 points | 37% | 27% | 37% | 89% |
| 90% | 10 points | 8% | 7% | 85% | 99% |
| 90% | 20 points | 8% | 15% | 77% | 99% |
| 90% | 30 points | 8% | 27% | 66% | 99% |
About this exhibit
Exhibit 8 · Every model at its chosen settings, scored from 2012, daily
| Model | Regimes | Penalty | Switches a year | Bootstrap stability | At the time vs hindsight | Co-movement R² | Volatility R² | Co-movement blend vs no-regime | Volatility blend vs no-regime | Hedge gain, points |
|---|---|---|---|---|---|---|---|---|---|---|
| k-means + margin (correlation map) | 4 | — | 4.82 | 0.76 | 83% | +0.10 | −0.11 | −0.08 | −0.07 | 4.2 |
| Gaussian mixture | 3 | — | 8.94 | 0.30 | 78% | +0.03 | −0.07 | −0.09 | −0.04 | 4.4 |
| Gaussian HMM | 5 | — | 3.81 | 0.43 | 34% | −0.01 | −0.07 | −0.12 | −0.06 | 4.5 |
| Jump model (single day) | 6 | 8 | 1.73 | 0.50 | 52% | +0.08 | −0.08 | −0.11 | −0.06 | 5.1 |
| Time-series jump model, 63-day memory | 4 | 4 | 0.83 | 0.60 | 82% | +0.08 | −0.09 | −0.08 | −0.06 | 3.7 |
| Time-series jump model, 126-day memory | 6 | 1 | 0.62 | 0.64 | 74% | +0.13 | −0.15 | −0.06 | −0.07 | 4.0 |
| Time-series k-means + margin, 63-day memory | 6 | — | 1.70 | 0.51 | 69% | +0.11 | −0.12 | −0.09 | −0.06 | 3.7 |
| Wasserstein k-means + margin | 5 | — | 1.80 | 0.67 | 91% | −0.06 | +0.03 | −0.07 | −0.14 | 1.8 |
| The vote of the eight | 4 | — | 2.95 | — | 68% | +0.15 | +0.01 | −0.05 | −0.02 | 4.2 |
| No regimes: half last quarter + half the two-year average | — | — | — | — | — | +0.29 | +0.25 | ±0.00 | ±0.00 | — |
About this exhibit
The settings below change the cards of Since inception and Hedges. The headline and the findings stay the model vote’s.
Risk and return since each regime began
How the markets moved through each regime, counted from the week it began, with 90% bands that resample the models and the spells. This is history, not a forecast.- At week 35 the current spell runs outside the band for Gold. Gold: −22.9% against a band of −7.2% to +12.9%. → Exhibit 17
Exhibit 16 · Start date of the current spell, by source, with the resampled votes
Return to date from each start, to 13 Nov 2026
The spread across rows is how much a since-inception figure depends on where the spell is said to begin.
| Start | Date | S&P 500 | Nasdaq 100 | Semiconductors | Long Treasuries | 7–10-year Treasuries | TIPS | Gold | Oil | Dollar |
|---|---|---|---|---|---|---|---|---|---|---|
| Earliest plausible start | 11 Mar 2026 | +12.3% | +12.3% | +22.5% | −7.2% | −4.1% | −2.6% | −25.6% | +37.5% | +7.5% |
| The 111-fund map's published regime | 6 Mar 2026 | +11.6% | +12.0% | +23.1% | −7.8% | −4.1% | −2.6% | −23.7% | +51.2% | +7.4% |
| The model vote (the paths' start) | 18 Mar 2026 | +13.3% | +13.1% | +22.6% | −6.3% | −3.8% | −2.8% | −22.6% | +22.5% | +6.6% |
| Latest plausible start | 21 Apr 2026 | +7.0% | +5.3% | +4.9% | −6.3% | −3.8% | −2.8% | −19.6% | +20.0% | +8.0% |
About this exhibit
Exhibit 17 · Cumulative return since the regime began, by market
S&P 500
Nasdaq 100
Semiconductors
Long Treasuries
7–10-year Treasuries
TIPS
Gold
Oil
Dollar
About this exhibit
Exhibit 18 · Annualised risk and return in the regime, by market
| Market | Weeks | Return a year | 5% | 95% | Volatility | Return ÷ volatility | Worst drawdown | Corr. with S&P 500 | with long Treasuries | with the dollar |
|---|---|---|---|---|---|---|---|---|---|---|
| S&P 500 | 227 | +12.8% | +1.4% | +24.0% | 15.7% | 0.82 | −19.5% | 1.00 | +0.18 | −0.43 |
| Nasdaq 100 | 227 | +12.5% | −2.1% | +28.6% | 21.0% | 0.60 | −25.0% | +0.93 | +0.19 | −0.37 |
| Semiconductors | 227 | +21.8% | +0.6% | +48.5% | 32.8% | 0.67 | −29.9% | +0.77 | +0.13 | −0.32 |
| Long Treasuries | 227 | −4.0% | −14.5% | +6.6% | 14.2% | -0.29 | −24.7% | +0.18 | 1.00 | −0.22 |
| 7–10-year Treasuries | 227 | −0.1% | −5.5% | +5.4% | 6.9% | -0.01 | −11.0% | +0.26 | +0.90 | −0.40 |
| TIPS | 227 | +2.2% | −1.9% | +6.5% | 5.4% | 0.41 | −9.6% | +0.43 | +0.70 | −0.41 |
| Gold | 227 | +5.4% | −4.3% | +18.4% | 16.0% | 0.34 | −22.9% | +0.25 | +0.17 | −0.44 |
| Oil | 227 | +11.2% | −8.8% | +34.5% | 35.6% | 0.32 | −34.9% | +0.13 | −0.20 | −0.05 |
| Dollar | 227 | +5.2% | −0.8% | +10.3% | 6.5% | 0.80 | −9.1% | −0.43 | −0.22 | 1.00 |
About this exhibit
Exhibit 19 · Return over the 13 weeks after a week in the regime, by market
About this exhibit
Exhibit 20 · Share of spells still running, by weeks since they began
Hedges on the S&P 500's worst days, by regime
What each hedge returned on the S&P 500's worst tenth of days, by the chosen model's regime.- In this regime the dollar hedged. On the S&P 500's worst tenth of days the dollar returned +0.26% a day. Long Treasuries returned −0.02% and gold −0.58%. → Exhibit 21
Exhibit 21 · Mean daily return of each hedge on the S&P 500's worst days, by regime
Every design on the same yardsticks
Each way of cutting the weeks into four regimes, or as many as it finds, scored on the same weeks of the 111-fund map. How persistent it is, how stable under resampling, how often its label at the time matched hindsight, and how much it tells about next quarter.- No design matches its labels at the time better than the live rule. The live rule matches hindsight at the time on 80% of weeks; the next best, the TS k-means, 4-week window, on 78%. → Exhibit 22
- The regimes carry some information about next quarter's co-movement. Out of sample, the regime averages of the jump model score R² +0.19 against the running mean and +0.13 against last quarter. The live rule scores +0.17 and +0.11. → Exhibit 22
- Several designs read volatility better than last quarter. Against last quarter the live rule scores +0.04. Sixteen of the 25 designs are above zero; the best, the TS k-means, Euclidean + margin, scores +0.07. → Exhibit 22
- No design forecasts returns. Out-of-sample R² for next-quarter returns is below zero for every design. → Exhibit 22
- Bagging the jump model buys stability, not timeliness. Bagging the jump model moves its stability from 0.76 to 0.87, and its agreement at the time from 69% to 70%. → Exhibit 22
Exhibit 22 · Each design's persistence, stability, timeliness and forecast record
| k-means + margin (live design) | Cross-section | 1.10 | 20.5 | 0.66 | 80% | 28% | +0.18 | +0.11 | +0.19 | +0.04 | −0.03 | 4.6 | 1.00 |
| TS k-means, 4-week window | Time series window | 1.42 | 14.5 | 0.65 | 78% | 28% | +0.17 | +0.10 | +0.19 | +0.03 | −0.02 | 4.0 | 0.91 |
| Gaussian HMM | Markov | 0.89 | 32.0 | 0.61 | 77% | 30% | +0.16 | +0.09 | +0.21 | +0.05 | −0.02 | 4.6 | 0.81 |
| TS jump model, 4-week window | Time series window | 0.58 | 52.0 | 0.69 | 77% | 32% | +0.18 | +0.12 | +0.21 | +0.06 | −0.02 | 4.0 | 0.83 |
| TS k-means, Euclidean | Time series window | 0.79 | 33.0 | 0.69 | 73% | 27% | +0.18 | +0.13 | +0.20 | +0.04 | −0.04 | 3.4 | 0.77 |
| Consensus of 13 designs (weak included) | Blend | 0.68 | 43.5 | 0.82 | 72% | 29% | +0.18 | +0.13 | +0.20 | +0.04 | −0.04 | 3.4 | 0.76 |
| Consensus of 6 correlation designs | Blend | 0.47 | 57.5 | 0.72 | 72% | 32% | +0.19 | +0.13 | +0.21 | +0.06 | −0.02 | 3.6 | 0.73 |
| Consensus of 6 + margin | Blend | 0.47 | 57.5 | 0.72 | 72% | 32% | +0.19 | +0.13 | +0.21 | +0.06 | −0.02 | 3.6 | 0.73 |
| TS k-means, Euclidean + margin | Time series window | 0.47 | 54.0 | 0.70 | 71% | 29% | +0.19 | +0.14 | +0.22 | +0.07 | −0.04 | 3.2 | 0.65 |
| TS jump model, 26-week window | Time series window | 0.26 | 151.0 | 0.85 | 70% | 30% | +0.16 | +0.12 | +0.14 | −0.03 | −0.03 | 3.2 | 0.61 |
| Bagged TS jump model (25 fits) | Blend | 0.47 | 58.5 | 0.87 | 70% | 30% | +0.16 | +0.11 | +0.09 | −0.10 | −0.07 | 3.5 | 0.72 |
| Jump model | Cross-section | 0.63 | 48.0 | 0.69 | 69% | 31% | +0.19 | +0.13 | +0.21 | +0.06 | −0.02 | 4.2 | 0.82 |
| TS jump model, half penalty | Time series window | 0.68 | 43.5 | 0.79 | 69% | 27% | +0.17 | +0.12 | +0.19 | +0.03 | −0.06 | 3.3 | 0.77 |
| TS k-means, DTW | Time series window | 0.79 | 33.0 | 0.77 | 69% | 27% | +0.17 | +0.11 | +0.19 | +0.03 | −0.04 | 3.4 | 0.77 |
| TS jump model | Time series window | 0.47 | 57.5 | 0.76 | 69% | 30% | +0.17 | +0.12 | +0.20 | +0.04 | −0.06 | 3.4 | 0.73 |
| TS jump model, double penalty | Time series window | 0.26 | 148.5 | 0.76 | 67% | 32% | +0.17 | +0.11 | +0.20 | +0.05 | −0.06 | 3.5 | 0.68 |
| TS k-medoids, derivative DTW | Time series window | 26.96 | 2.0 | 0.47 | 63% | 1% | ±0.00 | −0.13 | −0.02 | −0.26 | −0.01 | 0.1 | -0.01 |
| Jump model, 4x penalty | Cross-section | 0.26 | 145.5 | 0.74 | 61% | 34% | +0.18 | +0.13 | +0.20 | +0.05 | −0.04 | 3.8 | 0.72 |
| Wasserstein k-means | Distribution | 2.10 | 18.0 | 0.51 | 59% | 6% | +0.02 | −0.11 | +0.19 | +0.01 | −0.01 | 0.4 | 0.13 |
| MiniRocket → k-means | Features | 6.04 | 7.5 | 0.39 | 50% | 3% | −0.02 | −0.14 | +0.06 | −0.15 | −0.05 | 0.8 | 0.03 |
| catch22 → k-means | Features | 3.26 | 11.0 | 0.41 | 49% | 15% | +0.06 | −0.03 | +0.08 | −0.12 | −0.06 | 1.8 | 0.32 |
| TS k-medoids, DTW | Time series window | 1.73 | 9.0 | 0.59 | 44% | 25% | +0.14 | +0.07 | +0.10 | −0.09 | −0.08 | 3.2 | 0.50 |
| k-Shape | Time series window | 1.89 | 11.0 | 0.51 | 28% | 26% | −0.09 | −0.20 | −0.19 | −0.50 | −0.22 | 3.6 | 0.39 |
| Kernel k-means (GAK) | Time series window | 39.83 | 1.0 | 0.00 | 25% | 1% | ±0.00 | −0.13 | +0.01 | −0.24 | ±0.00 | 0.3 | 0.00 |
| Greedy Gaussian segmentation | Segmentation | 0.05 | 497.0 | — | — | 15% | – | – | – | – | – | 0.7 | 0.28 |
About this exhibit
How each design cuts 2007–2026, and where the weeks sit on the map
One strip per design, each regime coloured by the live regime it overlaps most; and every week on the map's first two axes, coloured by the chosen design.Exhibit 23 · Each design's regime by month, 2007–2026
Exhibit 24 · Every week on the first two axes of the 111-fund map, by design and by the model vote
About this exhibit
How stable the regimes are
Whether a design gives the same labels when refitted on resampled data, and whether its labels drawn at the time match hindsight, year by year.Exhibit 25 · Bootstrap stability of each design's labels
About this exhibit
Exhibit 26 · Share of each year's weeks where the label at the time matched hindsight
The forecasts of next quarter, blended and stacked
These forecasts are : each is a rule fixed in advance and scored on next quarter. Blending labels is about a steadier regime name, and sits in the scorecard; stacking forecasts is about a better read of next quarter.- The best next-quarter forecast blends two numbers and needs no regime. Half last quarter plus half a two-year average beats last quarter for 8 of 8 markets on co-movement and 9 of 9 on volatility. Against last quarter it scores R² +0.27 on co-movement and +0.25 on volatility. → Exhibit 27
- It holds in both halves of the record. On co-movement it scores +0.30 in 2014-19 and +0.25 in 2020-26. → Exhibit 29
- A regime adds to it. The best variant with a regime (thirds: last quarter, the two-year average, the vote's regime mean) scores +0.034 against the regime-free blend. → Exhibit 29
- Learned weights do worse than equal ones. On volatility the fitted stack of last quarter and the two-year average scores +0.20 against last quarter; the equal blend scores +0.25. → Exhibit 28
Exhibit 27 · Next-quarter correlation with the S&P 500, R² by forecast
The half-and-half blend with a two-year average, in bold, is the bar to beat: +0.27.
About this exhibit
Exhibit 28 · Next-quarter volatility, R² by forecast
The half-and-half blend with a two-year average, in bold, is the bar to beat: +0.25.
About this exhibit
Exhibit 29 · R² of each forecast against last quarter and against the regime-free blend
| Last quarter | Correlation | no | ±0.00 | ±0.00 | ±0.00 | −0.380 | 0 of 8 |
| Regime mean: the vote | Correlation | yes | +0.16 | +0.23 | −0.01 | −0.176 | 7 of 8 |
| Regime mean: the 111-fund map | Correlation | yes | +0.19 | +0.33 | +0.05 | −0.116 | 8 of 8 |
| Regime mean: TS jump | Correlation | yes | +0.17 | +0.33 | +0.01 | −0.119 | 8 of 8 |
| Regime mean: return distribution | Correlation | yes | −0.16 | +0.17 | −0.42 | −0.576 | 4 of 8 |
| Half last quarter, half the vote | Correlation | yes | +0.24 | +0.28 | +0.19 | −0.054 | 8 of 8 |
| Half last quarter, half the 111-fund map | Correlation | yes | +0.26 | +0.31 | +0.22 | −0.020 | 8 of 8 |
| Half last quarter, half TS jump | Correlation | yes | +0.24 | +0.31 | +0.18 | −0.025 | 8 of 8 |
| Half last quarter, half return distribution | Correlation | yes | +0.19 | +0.27 | +0.12 | −0.115 | 7 of 8 |
| Half last quarter, half running mean | Correlation | no | +0.23 | +0.30 | +0.17 | −0.057 | 8 of 8 |
| Half last quarter, half 2-year average | Correlation | no | +0.27 | +0.30 | +0.25 | ±0.000 | 8 of 8 |
| Thirds, with the vote | Correlation | yes | +0.31 | +0.36 | +0.25 | +0.034 | 8 of 8 |
| Stack: last quarter, 2-year average | Correlation | no | +0.28 | +0.30 | +0.25 | +0.006 | 8 of 8 |
| Stack: plus every regime mean | Correlation | yes | +0.23 | +0.32 | +0.12 | −0.057 | 8 of 8 |
| Last quarter | Volatility | no | ±0.00 | ±0.00 | ±0.00 | −0.344 | 0 of 9 |
| Regime mean: the vote | Volatility | yes | −0.09 | −1.47 | +0.25 | −0.450 | 4 of 9 |
| Regime mean: the 111-fund map | Volatility | yes | +0.06 | −0.19 | +0.11 | −0.240 | 6 of 9 |
| Regime mean: TS jump | Volatility | yes | +0.06 | −0.19 | +0.12 | −0.235 | 7 of 9 |
| Regime mean: return distribution | Volatility | yes | +0.02 | −0.34 | +0.09 | −0.280 | 7 of 9 |
| Half last quarter, half the vote | Volatility | yes | +0.21 | −0.18 | +0.31 | −0.050 | 9 of 9 |
| Half last quarter, half the 111-fund map | Volatility | yes | +0.25 | +0.09 | +0.28 | −0.006 | 9 of 9 |
| Half last quarter, half TS jump | Volatility | yes | +0.24 | +0.09 | +0.27 | −0.017 | 9 of 9 |
| Half last quarter, half return distribution | Volatility | yes | +0.14 | +0.06 | +0.16 | −0.140 | 8 of 9 |
| Half last quarter, half running mean | Volatility | no | +0.25 | −0.05 | +0.32 | −0.002 | 9 of 9 |
| Half last quarter, half 2-year average | Volatility | no | +0.25 | +0.08 | +0.29 | ±0.000 | 9 of 9 |
| Thirds, with the vote | Volatility | yes | +0.24 | −0.13 | +0.32 | −0.015 | 8 of 9 |
| Stack: last quarter, 2-year average | Volatility | no | +0.20 | +0.09 | +0.23 | −0.068 | 9 of 9 |
| Stack: plus every regime mean | Volatility | yes | +0.19 | +0.04 | +0.22 | −0.088 | 9 of 9 |
About this exhibit
Does the regime describe the macro?
The macro and pricing in each regime and today, how much of each the regimes explain, the months most like this one, and whether pricing follows the macro.- Of the macro series, the regimes explain private credit growth best and oil above its 3-year high least. They explain 61% of the variation across months in private credit growth. They explain 2% of it in oil above its 3-year high. → Exhibit 33
- The macro looks like 2018; pricing looks like 1998. The months nearest today on the macro are Jun 2018 and Jul 2018. On pricing the nearest is May 1998. Their regimes: “Bonds hedge” on the macro, “Nothing hedges” on pricing. → Exhibit 31
- Pricing has drifted from what the macro alone would imply. The largest gap is in stocks with oil, −2.40 typical gaps from its macro analogues' pricing. → Exhibit 34
- Pricing follows the macro less closely than usual. How closely pricing similarity tracks macro similarity sits at the 16th percentile of its record since 1998. → Exhibit 35
Exhibit 30 · The macro today, in this spell and regime, and in the nearest months
| Productivity growth, 8 quarters annualised | Oil, 12-month change | Private credit growth | Credit impulse | Bank loans, year on year | Bank loans to non-bank lenders, year on year | Core PCE inflation | Industrial production, year on year | Unemployment rate | Real fed funds | Real fed funds minus r* | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Today (Oct 2026) | 2.4 | 51.7 | 4.4 | 0.9 | 7.6 | 17.3 | 3.1 | 2.2 | 4.1 | 0.6 | -0.4 |
| This spell (Apr 2026 –) | 2.4 | 36.1 | 4.4 | 0.9 | 7.1 | 17.4 | 3.1 | 1.7 | 4.2 | 0.6 | -0.4 |
| Only the dollar hedges (regime) | 1.9 | 18.7 | 4.6 | -0.9 | 4.5 | 12.2 | 3.7 | 1.4 | 4.2 | -0.5 | -1.6 |
| Nearest on the macro (Jun 2018) | 1.6 | 40.7 | 4.3 | -0.2 | 4.4 | 7.8 | 1.9 | 2.8 | 4.0 | -0.1 | -1.3 |
| Nearest on pricing (May 1998) | 2.2 | -33.4 | 9.7 | 3.9 | 8.4 | – | 1.3 | 6.9 | 4.4 | 4.2 | 1.0 |
Units: Productivity growth, 8 quarters annualised %; Oil, 12-month change log %; Private credit growth %; Credit impulse points of GDP; Bank loans, year on year %; Bank loans to non-bank lenders, year on year %; Core PCE inflation %; Industrial production, year on year %; Unemployment rate %; Real fed funds %; Real fed funds minus r* points.
About this exhibit
Exhibit 31 · The months nearest today on the macro, and on pricing
On the macro
| Month | Distance | Regime | Rank on pricing |
|---|---|---|---|
| Jun 2018 | 1.55 | Bonds hedge | 76 |
| Jul 2018 | 1.57 | Bonds hedge | 62 |
| Sep 2018 | 1.59 | Bonds hedge | 36 |
| Aug 2018 | 1.63 | Bonds hedge | 42 |
| Dec 2007 | 1.64 | Bonds hedge | 194 |
| Mar 2018 | 1.64 | Bonds hedge | 127 |
| Oct 2002 | 1.74 | Bonds hedge | 290 |
| Dec 2002 | 1.77 | Bonds hedge | 274 |
| Oct 2018 | 1.84 | Bonds hedge | 56 |
| Sep 2002 | 1.86 | Bonds hedge | 286 |
On pricing
| Month | Distance | Regime | Rank on the macro |
|---|---|---|---|
| May 1998 | 1.51 | Nothing hedges | 218 |
| Mar 2000 | 1.68 | Nothing hedges | 103 |
| May 1999 | 1.72 | Nothing hedges | 158 |
| Feb 2000 | 1.72 | Nothing hedges | 247 |
| Apr 1998 | 1.75 | Nothing hedges | 210 |
| Apr 1999 | 1.76 | Nothing hedges | 163 |
| Jun 1998 | 1.81 | Nothing hedges | 245 |
| Jan 2000 | 1.83 | Nothing hedges | 180 |
| Nov 1999 | 1.84 | Nothing hedges | 164 |
| Jun 1999 | 1.84 | Nothing hedges | 133 |
About this exhibit
Exhibit 32 · Every macro and pricing measure in each regime, spell and episode
| Months | Productivity growth, 8 quarters annualised | Oil, 12-month change | Oil above its 3-year high | Private credit growth | Credit impulse | Core PCE inflation | Industrial production, year on year | Real GDP, year on year | Unemployment rate | Unemployment, 12-month change | Real fed funds | Fed funds, 12-month change | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Today (Oct 2026) | 1 | 2.4 | 51.7 | 0.0 | 4.4 | 0.9 | 3.1 | 2.2 | 2.1 | 4.1 | -0.2 | 0.6 | -0.2 |
| This spell (Apr 2026 –) | 7 | 2.4 | 36.1 | 1.6 | 4.4 | 0.9 | 3.1 | 1.7 | 2.1 | 4.2 | -0.1 | 0.6 | -0.6 |
| Nothing hedges | 100 | 2.6 | 11.0 | 0.9 | 8.2 | 1.1 | 2.0 | 2.4 | 3.2 | 4.9 | -0.1 | 2.0 | 0.4 |
| Bonds hedge | 143 | 2.0 | -1.2 | 0.8 | 5.6 | 0.1 | 1.8 | 0.0 | 2.1 | 5.2 | 0.1 | -0.3 | -0.5 |
| Bonds and the dollar hedge | 53 | 1.8 | -4.3 | 0.0 | 0.5 | -1.9 | 1.5 | -0.5 | 0.9 | 8.7 | 0.6 | -1.3 | -0.6 |
| Only the dollar hedges | 50 | 1.9 | 18.7 | 0.6 | 4.6 | -0.9 | 3.7 | 1.4 | 3.3 | 4.2 | -0.8 | -0.5 | 0.9 |
| Nov 2020 – Dec 2020 | 2 | 4.4 | -28.7 | 0.0 | 6.4 | 3.7 | 1.5 | -4.1 | -0.9 | 6.7 | 3.1 | -1.4 | -1.5 |
| Mar 2021 – Nov 2021 | 9 | 3.7 | 72.7 | 1.5 | 4.9 | -2.7 | 3.7 | 6.6 | 7.4 | 5.3 | -4.0 | -3.6 | -0.1 |
| Apr 2022 – Apr 2023 | 13 | -0.4 | 12.6 | 0.4 | 6.6 | 1.1 | 5.2 | 0.8 | 2.1 | 3.6 | -1.1 | -2.2 | 2.8 |
| Jul 2023 – Oct 2024 | 16 | 1.9 | -3.6 | 0.0 | 2.9 | -3.3 | 3.2 | -0.4 | 3.1 | 3.9 | 0.3 | 2.0 | 1.1 |
| Jan 2025 – Mar 2025 | 3 | 2.8 | -7.7 | 0.0 | 3.5 | 1.1 | 2.7 | 0.7 | 2.3 | 4.1 | 0.3 | 1.6 | -1.0 |
| Late-1990s productivity boom | 36 | 3.2 | 12.8 | 0.9 | 9.6 | 1.3 | 1.5 | 4.5 | 4.5 | 4.2 | -0.3 | 4.1 | 0.3 |
| 2002–04 productivity surge | 30 | 3.6 | 10.7 | 0.5 | 6.9 | 0.5 | 1.7 | 1.1 | 2.7 | 5.8 | 0.4 | -0.4 | -1.2 |
| 2004–07 credit boom | 36 | 2.2 | 21.0 | 1.7 | 9.6 | 1.4 | 2.2 | 2.8 | 3.0 | 4.9 | -0.4 | 1.6 | 1.4 |
| 2007–08 oil spike | 11 | 1.5 | 50.5 | 5.3 | 8.1 | -1.8 | 2.2 | 0.8 | 1.6 | 5.1 | 0.6 | 1.1 | -2.0 |
| 2021–22 oil and inflation shock | 17 | 1.7 | 46.6 | 2.8 | 6.6 | 1.3 | 4.9 | 3.0 | 4.3 | 4.1 | -2.6 | -4.2 | 0.7 |
About this exhibit
Exhibit 33 · Share of each measure's monthly variation the four regimes explain
About this exhibit
Exhibit 34 · Today's pricing against the pricing of the months nearest on the macro
| Measure | Today | Macro analogues | Gap | Gap percentile | Record (share explained) |
|---|---|---|---|---|---|
| Stocks with long Treasuries (correlation) | 0.26 | -0.30 | +0.56 | 97th | -11% |
| Stocks with the dollar (correlation) | -0.23 | 0.01 | −0.24 | 62th | -28% |
| Stocks with oil (correlation) | -0.36 | 0.24 | −0.60 | 98th | -21% |
| Shiller CAPE | 40.71 | 28.34 | +12.37 | 83th | -63% |
| 10-year Treasury (%) | 5.13 | 3.38 | +1.75 | 84th | 11% |
| 10-year term premium (%) | 0.68 | 0.65 | +0.03 | 2th | -43% |
| 10-year minus 2-year (points) | 0.39 | 0.99 | −0.60 | 39th | -8% |
| VIX | 16.90 | 22.00 | −5.11 | 54th | -1% |
| Credit stress, OFR index | -0.94 | 0.55 | −1.49 | 64th | -33% |
| S&P 500, 12-month change (log %) | 11.33 | 1.79 | +9.54 | 48th | -20% |
| The dollar, 12-month change (log %) | 6.12 | -3.37 | +9.49 | 74th | -6% |
| 10-year TIPS yield (%) | 2.78 | 0.95 | +1.83 | 81th | -88% |
| Gold, 12-month change (log %) | -3.41 | 5.91 | −9.32 | 34th | -23% |
About this exhibit
Exhibit 35 · How closely pricing similarity has followed macro similarity, monthly
About this exhibit
Exhibit 36 · The macro and pricing measures since 1998, monthly, shaded by regime
Appendix
The rule that names a regime, then the data, the designs, the tests, the checks and the sources.99 checks, all pass
The vote that names a day
| Symbol | Meaning | From | Assumed or estimated |
|---|---|---|---|
| model m's regime on day t | computed | – | |
| model m's regimes mapped, many to one, onto the consensus regimes | computed | estimated | |
| the share of the votes for consensus regime k | computed | – | |
| the 8 models | the research record | – |
Runs of r under 21 sessions fold into the spell around them.
How the numbers are made
Yahoo Finance closes, nine core markets + H.15 yields + EIA oil futures + BLS CPI ──▶ daily log returns with the proxies before each fund; GARCH(1,1)-t per market; one DCC(1,1) by composite likelihood over the 36 pairs ──▶ the nine markets' correlations, daily since 1998 the nine markets' correlations, daily since 1998 ──▶ standardised; probabilistic PCA to 90%; the scores, their lags and each day's trailing return deciles ──▶ the features the features + the research record ──▶ eight models at their chosen settings; their regimes mapped onto four consensus regimes; each day's shares, status and leading regime, runs under 21 sessions folded; yearly refits from 2012 for the vote at the time ──▶ the vote the vote + the 111-fund map ──▶ names by the legs; spells, paths and bands, the start window, risk and return, survival, the hedges on the S&P 500's worst days ──▶ the page
Sources
- licensedYahoo Finance chart API, daily adjusted closes: the nine core funds (SPY QQQ SMH TLT IEF TIP GLD USO UUP) and the index and futures proxies before them (^GSPC, ^NDX, ^SOX, FSELX, DX-Y.NYB, GC=F)
Prices: Yahoo Finance. · https://query2.finance.yahoo.com/v8/finance/chart/
- publicBoard of Governors of the Federal Reserve System, H.15 Selected Interest Rates: the 7-, 10-, 20- and 30-year constant-maturity yields and the 5- and 10-year TIPS yields and their long-term average (the par bonds before TLT, IEF and TIP)
Source: Board of Governors of the Federal Reserve System. · https://www.federalreserve.gov/datadownload/Output.aspx?rel=H15&filetype=zip
- publicU.S. Energy Information Administration, NYMEX Cushing crude oil future contract 1, daily (before USO)
Source: U.S. Energy Information Administration. · https://www.eia.gov/dnav/pet/hist_xls/RCLC1d.xls
- publicU.S. Bureau of Labor Statistics, CPI-U, seasonally adjusted (the TIPS proxy's inflation accrual)
Source: U.S. Bureau of Labor Statistics. · https://api.bls.gov/publicAPI/v2/timeseries/data/
- licensedCboe Global Markets, VIX index history (read with the other daily inputs)
VIX: Cboe Global Markets. · https://cdn.cboe.com/api/global/us_indices/daily_prices/VIX_History.csv
- computed hereThe 111-fund map's regime, typicality and breadth, as this page's map publishes them
Model computed here on the sources listed.
- computed hereThe vote, its record, the profiles and the hedges computed here; the members' settings from the dated research record
Model computed here on the sources listed.
Pages this one copies from
| Page | Title | Updated | Data to | Arrays | Values | Source checks |
|---|---|---|---|---|---|---|
| Correlations & regime | Four regimes of co-movement | 2026-11-14 | 13 Nov 2026 | 0 | 11 | 27 pass |
Reproduction checks
| Check | Published | Computed | Tolerance | Result |
|---|---|---|---|---|
| The inputs: the nine markets' daily returns, H.15 and the VIX were fetched together (the day read) | 2026-11-14 | 2026-11-14 | 0 | pass |
| The correlations: a + b below one and the composite likelihood converged (a 0.0317, b 0.9569) | 0 | pass | ||
| The correlations: every value lies between −1 and 1 (largest absolute value ≤ 1) | 0 | pass | ||
| The members: k-means + margin (correlation map) runs at the research record's choice for its family (regimes, setting) | 4|None | 4|None | 0 | pass |
| The members: Gaussian mixture runs at the research record's choice for its family (regimes, setting) | 3|None | 3|None | 0 | pass |
| The members: Gaussian HMM runs at the research record's choice for its family (regimes, setting) | 5|None | 5|None | 0 | pass |
| The members: Jump model (single day) runs at the research record's choice for its family (regimes, setting) | 6|8.0 | 6|8.0 | 0 | pass |
| The members: Time-series jump model, 63-day memory runs at the research record's choice for its family (regimes, setting) | 4|4.0 | 4|4.0 | 0 | pass |
| The members: Time-series jump model, 126-day memory runs at the research record's choice for its family (regimes, setting) | 6|0.5 | 6|0.5 | 0 | pass |
| The members: Time-series k-means + margin, 63-day memory runs at the research record's choice for its family (regimes, setting) | 6|None | 6|None | 0 | pass |
| The members: Wasserstein k-means + margin runs at the research record's choice for its family (regimes, setting) | 5|None | 5|None | 0 | pass |
| The vote: each day's shares of the members' votes sum to one (largest gap) | 0 | 0 | 0.002 | pass |
| The vote: each day's leading regime holds the largest share (days where it does not) | 0 | 0 | 0 | pass |
| The vote: each day's status is the thresholds applied to its shares (clear at 80%, mixed within 20 points; days differing) | 0 | 0 | 0 | pass |
| The regime: every day's regime is the vote's leading regime with runs under 21 sessions folded into the spell around them (days differing) | 0 | 0 | 0 | pass |
| The regime: every regime holds days | 4 | 4 | 0 | pass |
| The regime: every week takes the label of its last session (weeks differing) | 0 | 0 | 0 | pass |
| The names: every regime's name is the leg rule applied to its published correlations and volatility (a leg hedges below −0.15; regimes differing) | 0 | 0 | 0 | pass |
| The names: the regimes' names are distinct | 4 | 4 | 0 | pass |
| The record: the agreement on clear days drawn at the time reproduces from the published labels (difference) | 0.8946 | 0.8946 | 0 | pass |
| The record: the chance of a change of regime within a quarter after a mixed day reproduces from the published regime (difference) | 0.346 | 0.346 | 0 | pass |
| The hedges: the regime's worst-day returns reproduce from the daily returns and the labels (max abs difference) | 0 | 0 | 0.001 | pass |
| The hedges: in every regime the S&P 500's mean on its worst days is below zero | 0 | pass | ||
| The map: the coordinates equal the scores projected on the stored loadings (max abs difference) | 0 | 0 | 0 | pass |
| Returns to date: the S&P 500 from the regime's start reproduces from the daily returns (difference, points) | 13.34 | 13.34 | 0.01 | pass |
| Dates: the vote's last day is the nine markets' last session (13 Nov 2026, fetched 2026-11-14) | 2026-11-13 | 2026-11-13 | 0 | pass |
| The hero: each week's shares are its days' mean shares and sum to one (largest gap) | 0 | 0 | 0.002 | pass |
| The hero: one row a week (Saturday to Friday) from the first day to the last (weeks) | 1507 | 1507 | 0 | pass |
| The hero: the first week's shares reproduce from its days' daily shares (largest difference) | 0 | 0 | 0 | pass |
| The start card: every source's start lies on or before the last day and in the current regime's record | 0 | pass | ||
| The Correlations & regime page: the source page is published with all its checks passing (dated 14 Nov 2026, 27 checks) | 1 | 1 | 0 | pass |
| The Correlations & regime page: the site still serves the page this one copied (dated 14 Nov 2026) | 1 | 1 | 0 | pass |
| The Correlations & regime page: every copied single value equals the source page (11 values; mismatches) | 0 | 0 | 0 | pass |
| Decisions: every desk decision this page's specification names is in the decisions file (D-2026-10-24-RL2, D-2026-10-24-RN1, D-2026-10-24-RP1; ids missing) | 0 | 0 | 0 | pass |
- Yahoo Finance chart API, daily adjusted closes: the nine core funds (SPY QQQ SMH TLT IEF TIP GLD USO UUP) and the index and futures proxies before them (^GSPC, ^NDX, ^SOX, FSELX, DX-Y.NYB, GC=F) — Prices: Yahoo Finance.
- Board of Governors of the Federal Reserve System, H.15 Selected Interest Rates: the 7-, 10-, 20- and 30-year constant-maturity yields and the 5- and 10-year TIPS yields and their long-term average (the par bonds before TLT, IEF and TIP) — Source: Board of Governors of the Federal Reserve System.
- U.S. Energy Information Administration, NYMEX Cushing crude oil future contract 1, daily (before USO) — Source: U.S. Energy Information Administration.
- U.S. Bureau of Labor Statistics, CPI-U, seasonally adjusted (the TIPS proxy's inflation accrual) — Source: U.S. Bureau of Labor Statistics.
- Cboe Global Markets, VIX index history (read with the other daily inputs) — VIX: Cboe Global Markets.
- The 111-fund map's regime, typicality and breadth, as this page's map publishes them — Model computed here on the sources listed.
- The vote, its record, the profiles and the hedges computed here; the members' settings from the dated research record — Model computed here on the sources listed.