MacroBessent

 Account
S&P 5007,650.50+0.17%Nasdaq26,522.54+0.39%Dow51,682.64-0.18%Nikkei 22565,018.95+1.38%Hang Seng24,750.78+0.60%EuroStoxx 506,236.20-1.37%Dollar (DXY)100.22+0.00%EUR/USD1.15+0.18%USD/JPY156.85+0.46%GBP/USD1.34+0.26%USD/CNY6.70-0.15%US 10Y yield5.00+1.03%US 5Y yield4.86+1.15%Gold4,424.90+0.57%Silver66.56+1.66%Copper6.62+0.43%Platinum1,803.50+0.68%Palladium1,306.20+1.24%WTI100.30-1.58%Brent103.87-0.91%Henry Hub2.91+0.38%Gasoline3.53+0.58%Heating oil5.06-1.10%Wheat714.25-1.75%Corn527.50-0.57%Soybeans1,303.50-1.23%Coffee294.55+1.38%Sugar17.36-0.34%Cocoa5,327.00-7.71%Cotton77.38-1.31%Lithium ETF70.50-0.56%Uranium ETF41.65-2.41%Rare Earths ETF69.06-0.06%Copper Miners ETF87.28+0.77%Bitcoin81,244.98+0.42%Ethereum2,637.29+0.99%Solana111.48-0.99%S&P 5007,650.50+0.17%Nasdaq26,522.54+0.39%Dow51,682.64-0.18%Nikkei 22565,018.95+1.38%Hang Seng24,750.78+0.60%EuroStoxx 506,236.20-1.37%Dollar (DXY)100.22+0.00%EUR/USD1.15+0.18%USD/JPY156.85+0.46%GBP/USD1.34+0.26%USD/CNY6.70-0.15%US 10Y yield5.00+1.03%US 5Y yield4.86+1.15%Gold4,424.90+0.57%Silver66.56+1.66%Copper6.62+0.43%Platinum1,803.50+0.68%Palladium1,306.20+1.24%WTI100.30-1.58%Brent103.87-0.91%Henry Hub2.91+0.38%Gasoline3.53+0.58%Heating oil5.06-1.10%Wheat714.25-1.75%Corn527.50-0.57%Soybeans1,303.50-1.23%Coffee294.55+1.38%Sugar17.36-0.34%Cocoa5,327.00-7.71%Cotton77.38-1.31%Lithium ETF70.50-0.56%Uranium ETF41.65-2.41%Rare Earths ETF69.06-0.06%Copper Miners ETF87.28+0.77%Bitcoin81,244.98+0.42%Ethereum2,637.29+0.99%Solana111.48-0.99%
ESC clears

Through time — level, distribution, seasonality and range Cas of 2026-09-19 13:06 UTC

36 headline series across 10 desks, each read four ways: level against its own trailing mean, position in its full history, calendar-month seasonality, and an empirical 63-day range. The most stretched is Copper at the 100th percentile of 415 observations back to 1992-01-01. Only 3 of 36 show a calendar pattern larger than their own noise (Gasoline (Gulf), 2s10s slope, 5y breakeven) — the rest are shown with that stated, because a seasonal chart drawn on noise is the most convincing bad chart in finance. Behind the tabs: 18 distinct series decomposed with statsmodels 0.14.6, 18 with a two-state Markov filter and 18 with an analog episode table scored against a linked price.

level
7,637.76
idx · as of 2026-09-17
percentile
99
of 2,513 obs since 2016
percentile, 5y
98
of 1,254 obs
vs 13w mean
+44.31
level minus its own quarter
vs 52w mean
+545.04
level minus its own year
year on year
+14.6%
same week last year
3,3315,6848,0382021-10-012022-12-302024-03-292025-06-272026-09-18level 7,638MSTL … 7,706
MSTL with periods 5 and 52 on 522 weekly observations from 2016-09-23. The trend is drawn over the level and the remainder under it; nothing here is extrapolated. The trend explains 99% of the variance the remainder does not (Hyndman's trend strength); the seasonal component explains 25%. · FRED SP500 · statsmodels 0.14.6
-562.258-57.296447.6662021-10-012022-12-302024-03-292025-06-272026-09-18re… -103.031
What the trend and the seasonal do not account for, on the same dates. A remainder that trends is a decomposition that has missed something, which is why it is drawn rather than described. · FRED SP500
2,0857,799now
The whole history as a distribution, with today marked. Median 3,963.94 over 2,513 observations. · FRED SP500
0%50%100%last 5 yearsnow · 99th2,0857,799
The same record read cumulatively: today sits at the 99 percentile of the full history (n=2,513) and the 98 of the last five years (n=1,254). A level extreme against forty years and ordinary against five is a different fact from one that is extreme against both. · FRED SP500
+2.0Jan-0.7Feb-1.1Mar+1.9Apr+1.7May+1.6Jun+2.9Jul+1.1Aug-1.5Sep+0.6Oct+4.0Nov-0.0Decno significant pattern~10 yrs/bar
The spread of monthly means is no larger than the noise — this series has no seasonal worth trading. Kruskal-Wallis across calendar months; the corner figure is the average number of years behind each bar. · FRED SP500
-488.363-140.379207.6052021-10-012022-12-302024-03-292025-06-272026-09-18sea… -13.162
The MSTL seasonal component itself, at periods 5 and 52. Drawn rather than summarised because a seasonal that changes shape year to year is not one a reader can plan against, and the chart shows that where a single average would hide it. · FRED SP500 · statsmodels 0.14.6
8,6587,1634,3577,638
Where this has historically gone over 63 trading days: 10th to 90th percentile of 2,425 realised moves, applied to today. A base rate for sizing, not a forecast. · FRED SP500
downside (10th)
7,163.31
-6.2%
median path
8,088.62
50th percentile
upside (90th)
8,658.44
+13.4%
Markov-filtered mean
+0.397
per week, state-weighted
The band above is unconditional: it does not know what state we are in. The filtered mean beside it does — it is the two-state Markov model's mean weighted by today's filtered probability (3 of the high-volatility state). The two disagree when the current state is unusual, which is the point of printing both.
state
low volatility
since 2026-05-01
P(high vol) now
3
filtered, this week
persisted
21
weeks (147 days)
P(exit in 4w)
0.0%
0 of 7 spells this old
spells on record
46
23 completed in this state
-4%51%106%2021-10-012022-12-302024-03-292025-06-272026-09-18P(high v… 3%
Two-state Markov-switching model with switching variance, fitted on 521 weekly % log change. The line is the FILTERED probability of the high-volatility state — a statement about weeks that have happened, not a forecast. Exit counts are empirical spell counts, not the model's expected durations. · FRED SP500 · statsmodels 0.14.6
3,3315,6848,0382021-10-012022-12-302024-03-292025-06-272026-09-18level 7,638
The same weeks on the level: shaded stretches are the ones the filter puts in the high-volatility state. Mean +0.413 per week at 1.400 standard deviation in the calm state against -0.237 at 3.980 in the volatile one. · FRED SP500 · MarkovRegression(k_regimes=2, trend='c', switching_variance=True)
level
80,844.00
usd · as of 2026-09-18
percentile
90
of 4,275 obs since 2014
percentile, 5y
76
of 1,826 obs
vs 13w mean
+11684.30
level minus its own quarter
vs 52w mean
+730.86
level minus its own year
year on year
-30.2%
same week last year
10,15169,484128,8162021-10-012022-12-302024-03-292025-06-272026-09-18level 80,844MSTL… 59,913
MSTL with periods 5 and 52 on 614 weekly observations from 2014-12-05. The trend is drawn over the level and the remainder under it; nothing here is extrapolated. The trend explains 96% of the variance the remainder does not (Hyndman's trend strength); the seasonal component explains 15%. · FRED CBBTCUSD · statsmodels 0.14.6
-21,803.6111,539.49124,882.5932021-10-012022-12-302024-03-292025-06-272026-09-1815,726.844
What the trend and the seasonal do not account for, on the same dates. A remainder that trends is a decomposition that has missed something, which is why it is drawn rather than described. · FRED CBBTCUSD
120124,720now
The whole history as a distribution, with today marked. Median 16,452.39 over 4,275 observations. · FRED CBBTCUSD
0%50%100%last 5 yearsnow · 89th120124,720
The same record read cumulatively: today sits at the 90 percentile of the full history (n=4,275) and the 76 of the last five years (n=1,826). A level extreme against forty years and ordinary against five is a different fact from one that is extreme against both. · FRED CBBTCUSD
-4.1Jan+8.7Feb-0.9Mar+8.6Apr+4.0May-2.0Jun+8.0Jul+0.3Aug-1.7Sep+16.4Oct+2.9Nov+7.6Decno significant pattern~12 yrs/bar
The spread of monthly means is no larger than the noise — this series has no seasonal worth trading. Kruskal-Wallis across calendar months; the corner figure is the average number of years behind each bar. · FRED CBBTCUSD
-7,788.2831,626.44311,041.1692021-10-012022-12-302024-03-292025-06-272026-09-1810,032.448
The MSTL seasonal component itself, at periods 5 and 52. Drawn rather than summarised because a seasonal that changes shape year to year is not one a reader can plan against, and the chart shows that where a single average would hide it. · FRED CBBTCUSD · statsmodels 0.14.6
146,89859,23348,14580,844
Where this has historically gone over 63 trading days: 10th to 90th percentile of 4,187 realised moves, applied to today. A base rate for sizing, not a forecast. · FRED CBBTCUSD
downside (10th)
59,232.51
-26.7%
median path
87,998.52
50th percentile
upside (90th)
146,897.92
+81.7%
Markov-filtered mean
+0.573
per week, state-weighted
The band above is unconditional: it does not know what state we are in. The filtered mean beside it does — it is the two-state Markov model's mean weighted by today's filtered probability (33 of the high-volatility state). The two disagree when the current state is unusual, which is the point of printing both.
state
low volatility
since 2026-09-04
P(high vol) now
33
filtered, this week
persisted
3
weeks (21 days)
P(exit in 4w)
92.1%
35 of 38 spells this old
spells on record
131
65 completed in this state
7%56%105%2021-10-012022-12-302024-03-292025-06-272026-09-18P(high … 33%
Two-state Markov-switching model with switching variance, fitted on 613 weekly % log change. The line is the FILTERED probability of the high-volatility state — a statement about weeks that have happened, not a forecast. Exit counts are empirical spell counts, not the model's expected durations. · FRED CBBTCUSD · statsmodels 0.14.6
10,15169,484128,8162021-10-012022-12-302024-03-292025-06-272026-09-18level 80,844
The same weeks on the level: shaded stretches are the ones the filter puts in the high-volatility state. Mean +0.206 per week at 2.956 standard deviation in the calm state against +1.302 at 11.411 in the volatile one. · FRED CBBTCUSD · MarkovRegression(k_regimes=2, trend='c', switching_variance=True)
level
4.94
pct · as of 2026-09-17
percentile
46
of 16,163 obs since 1962
percentile, 5y
99
of 1,248 obs
vs 13w mean
+0.26
level minus its own quarter
vs 52w mean
+0.59
level minus its own year
year on year
+0.80
same week last year
1.133.165.182021-10-012022-12-302024-03-292025-06-272026-09-18level 4.94MSTL t… 4.64
MSTL with periods 5 and 52 on 3377 weekly observations from 1962-01-05. The trend is drawn over the level and the remainder under it; nothing here is extrapolated. The trend explains 99% of the variance the remainder does not (Hyndman's trend strength); the seasonal component explains 13%. · FRED DGS10 · statsmodels 0.14.6
-0.6170.0320.6822021-10-012022-12-302024-03-292025-06-272026-09-18remai… 0.240
What the trend and the seasonal do not account for, on the same dates. A remainder that trends is a decomposition that has missed something, which is why it is drawn rather than described. · FRED DGS10
0.5215.84now
The whole history as a distribution, with today marked. Median 5.40 over 16,163 observations. · FRED DGS10
0%50%100%last 5 yearsnow · 46th0.5215.84
The same record read cumulatively: today sits at the 46 percentile of the full history (n=16,163) and the 99 of the last five years (n=1,248). A level extreme against forty years and ordinary against five is a different fact from one that is extreme against both. · FRED DGS10
+0.0Jan+0.0Feb+0.0Mar+0.1Apr-0.0May-0.0Jun+0.0Jul-0.0Aug-0.0Sep-0.0Oct-0.1Nov+0.0Decno significant pattern~65 yrs/bar
The spread of monthly means is no larger than the noise — this series has no seasonal worth trading. Kruskal-Wallis across calendar months; the corner figure is the average number of years behind each bar. · FRED DGS10
-0.2030.0010.2042021-10-012022-12-302024-03-292025-06-272026-09-18seaso… 0.056
The MSTL seasonal component itself, at periods 5 and 52. Drawn rather than summarised because a seasonal that changes shape year to year is not one a reader can plan against, and the chart shows that where a single average would hide it. · FRED DGS10 · statsmodels 0.14.6
5.684.141.484.94
Where this has historically gone over 63 trading days: 10th to 90th percentile of 16,075 realised moves, applied to today. A base rate for sizing, not a forecast. · FRED DGS10
downside (10th)
4.14
-0.80
median path
4.97
50th percentile
upside (90th)
5.68
+0.74
Markov-filtered mean
+0.001
per week, state-weighted
The band above is unconditional: it does not know what state we are in. The filtered mean beside it does — it is the two-state Markov model's mean weighted by today's filtered probability (6 of the high-volatility state). The two disagree when the current state is unusual, which is the point of printing both.
state
low volatility
since 2025-07-04
P(high vol) now
6
filtered, this week
persisted
64
weeks (448 days)
P(exit in 4w)
0.0%
0 of 3 spells this old
spells on record
48
24 completed in this state
-4%51%106%2021-10-012022-12-302024-03-292025-06-272026-09-18P(high v… 6%
Two-state Markov-switching model with switching variance, fitted on 1300 weekly change in level. The line is the FILTERED probability of the high-volatility state — a statement about weeks that have happened, not a forecast. Exit counts are empirical spell counts, not the model's expected durations. · FRED DGS10 · statsmodels 0.14.6
1.133.165.182021-10-012022-12-302024-03-292025-06-272026-09-18level 4.94
The same weeks on the level: shaded stretches are the ones the filter puts in the high-volatility state. Mean +0.001 per week at 0.086 standard deviation in the calm state against -0.001 at 0.148 in the volatile one. · FRED DGS10 · MarkovRegression(k_regimes=2, trend='c', switching_variance=True)
level
15.44
pct · as of 2026-09-17
percentile
35
of 9,276 obs since 1990
percentile, 5y
26
of 1,287 obs
vs 13w mean
-0.49
level minus its own quarter
vs 52w mean
-2.71
level minus its own year
year on year
-0.01
same week last year
9.9328.6247.312021-10-012022-12-302024-03-292025-06-272026-09-18level 15.44MSTL … 17.65
MSTL with periods 5 and 52 on 1916 weekly observations from 1990-01-05. The trend is drawn over the level and the remainder under it; nothing here is extrapolated. The trend explains 67% of the variance the remainder does not (Hyndman's trend strength); the seasonal component explains 16%. · FRED VIXCLS · statsmodels 0.14.6
-9.5684.02817.6252021-10-012022-12-302024-03-292025-06-272026-09-18rema… -1.450
What the trend and the seasonal do not account for, on the same dates. A remainder that trends is a decomposition that has missed something, which is why it is drawn rather than described. · FRED VIXCLS
9.1482.69now
The whole history as a distribution, with today marked. Median 17.58 over 9,276 observations. · FRED VIXCLS
0%50%100%last 5 yearsnow · 36th9.1482.69
The same record read cumulatively: today sits at the 35 percentile of the full history (n=9,276) and the 26 of the last five years (n=1,287). A level extreme against forty years and ordinary against five is a different fact from one that is extreme against both. · FRED VIXCLS
+0.3Jan+0.8Feb-0.4Mar-1.2Apr-0.7May-0.2Jun+0.4Jul+1.4Aug+1.5Sep+0.3Oct-1.9Nov-0.4Decno significant pattern~37 yrs/bar
The spread of monthly means is no larger than the noise — this series has no seasonal worth trading. Kruskal-Wallis across calendar months; the corner figure is the average number of years behind each bar. · FRED VIXCLS
-5.2792.0539.3852021-10-012022-12-302024-03-292025-06-272026-09-18seas… -0.184
The MSTL seasonal component itself, at periods 5 and 52. Drawn rather than summarised because a seasonal that changes shape year to year is not one a reader can plan against, and the chart shows that where a single average would hide it. · FRED VIXCLS · statsmodels 0.14.6
23.097.9621.1515.44
Where this has historically gone over 63 trading days: 10th to 90th percentile of 9,188 realised moves, applied to today. A base rate for sizing, not a forecast. · FRED VIXCLS
downside (10th)
7.96
-7.48
median path
14.85
50th percentile
upside (90th)
23.09
+7.65
Markov-filtered mean
-0.144
per week, state-weighted
The band above is unconditional: it does not know what state we are in. The filtered mean beside it does — it is the two-state Markov model's mean weighted by today's filtered probability (3 of the high-volatility state). The two disagree when the current state is unusual, which is the point of printing both.
state
low volatility
since 2026-06-19
P(high vol) now
3
filtered, this week
persisted
14
weeks (98 days)
P(exit in 4w)
21.7%
5 of 23 spells this old
spells on record
173
86 completed in this state
-3%51%106%2021-10-012022-12-302024-03-292025-06-272026-09-18P(high v… 3%
Two-state Markov-switching model with switching variance, fitted on 1300 weekly change in level. The line is the FILTERED probability of the high-volatility state — a statement about weeks that have happened, not a forecast. Exit counts are empirical spell counts, not the model's expected durations. · FRED VIXCLS · statsmodels 0.14.6
9.9328.6247.312021-10-012022-12-302024-03-292025-06-272026-09-18level 15.44
The same weeks on the level: shaded stretches are the ones the filter puts in the high-volatility state. Mean -0.161 per week at 1.687 standard deviation in the calm state against +0.388 at 5.723 in the volatile one. · FRED VIXCLS · MarkovRegression(k_regimes=2, trend='c', switching_variance=True)
level
107.02
usd · as of 2026-09-15
percentile
98
of 9,502 obs since 1986
percentile, 5y
95
of 1,247 obs
vs 13w mean
+21.61
level minus its own quarter
vs 52w mean
+28.85
level minus its own year
year on year
+69.8%
same week last year
52.7588.66124.582021-10-012022-12-302024-03-292025-06-272026-09-18level 107.02MSTL … 92.89
MSTL with periods 5 and 52 on 2122 weekly observations from 1986-01-03. The trend is drawn over the level and the remainder under it; nothing here is extrapolated. The trend explains 96% of the variance the remainder does not (Hyndman's trend strength); the seasonal component explains 16%. · FRED DCOILWTICO · statsmodels 0.14.6
-17.4984.48726.4732021-10-012022-12-302024-03-292025-06-272026-09-18remai… 7.869
What the trend and the seasonal do not account for, on the same dates. A remainder that trends is a decomposition that has missed something, which is why it is drawn rather than described. · FRED DCOILWTICO
-36.98145.31now
The whole history as a distribution, with today marked. Median 47.02 over 9,502 observations. · FRED DCOILWTICO
0%50%100%last 5 yearsnow · 97th-36.98145.31
The same record read cumulatively: today sits at the 98 percentile of the full history (n=9,502) and the 95 of the last five years (n=1,247). A level extreme against forty years and ordinary against five is a different fact from one that is extreme against both. · FRED DCOILWTICO
+0.8Jan-0.2Feb+1.4Mar+3.0Apr+1.1May+0.7Jun+1.1Jul+0.6Aug+1.3Sep-2.4Oct-3.6Nov+0.3Decno significant pattern~41 yrs/bar
The spread of monthly means is no larger than the noise — this series has no seasonal worth trading. Kruskal-Wallis across calendar months; the corner figure is the average number of years behind each bar. · FRED DCOILWTICO
-10.7671.16413.0952021-10-012022-12-302024-03-292025-06-272026-09-18seaso… 2.204
The MSTL seasonal component itself, at periods 5 and 52. Drawn rather than summarised because a seasonal that changes shape year to year is not one a reader can plan against, and the chart shows that where a single average would hide it. · FRED DCOILWTICO · statsmodels 0.14.6
139.7984.6976.01107.02
Where this has historically gone over 63 trading days: 10th to 90th percentile of 9,414 realised moves, applied to today. A base rate for sizing, not a forecast. · FRED DCOILWTICO
downside (10th)
84.69
-20.9%
median path
110.02
50th percentile
upside (90th)
139.79
+30.6%
Markov-filtered mean
+0.059
per week, state-weighted
The band above is unconditional: it does not know what state we are in. The filtered mean beside it does — it is the two-state Markov model's mean weighted by today's filtered probability (10 of the high-volatility state). The two disagree when the current state is unusual, which is the point of printing both.
state
low volatility
since 2026-08-14
P(high vol) now
10
filtered, this week
persisted
6
weeks (42 days)
P(exit in 4w)
0.0%
0 of 11 spells this old
spells on record
36
18 completed in this state
-6%50%106%2021-10-012022-12-302024-03-292025-06-272026-09-18P(high … 10%
Two-state Markov-switching model with switching variance, fitted on 1300 weekly % log change. The line is the FILTERED probability of the high-volatility state — a statement about weeks that have happened, not a forecast. Exit counts are empirical spell counts, not the model's expected durations. · FRED DCOILWTICO · statsmodels 0.14.6
52.7588.66124.582021-10-012022-12-302024-03-292025-06-272026-09-18level 107.02
The same weeks on the level: shaded stretches are the ones the filter puts in the high-volatility state. Mean +0.289 per week at 4.313 standard deviation in the calm state against -1.995 at 15.569 in the volatile one. · FRED DCOILWTICO · MarkovRegression(k_regimes=2, trend='c', switching_variance=True)
level
118.21
idx · as of 2026-09-11
percentile
78
of 5,188 obs since 2006
percentile, 5y
16
of 1,250 obs
vs 13w mean
-1.36
level minus its own quarter
vs 52w mean
-1.48
level minus its own year
year on year
-1.5%
same week last year
112.70121.77130.842021-09-242022-12-232024-03-222025-06-202026-09-11level 118.21MSTL… 119.45
MSTL with periods 5 and 52 on 1080 weekly observations from 2006-01-06. The trend is drawn over the level and the remainder under it; nothing here is extrapolated. The trend explains 98% of the variance the remainder does not (Hyndman's trend strength); the seasonal component explains 11%. · FRED DTWEXBGS · statsmodels 0.14.6
-4.1171.0496.2152021-09-242022-12-232024-03-222025-06-202026-09-11rema… -0.600
What the trend and the seasonal do not account for, on the same dates. A remainder that trends is a decomposition that has missed something, which is why it is drawn rather than described. · FRED DTWEXBGS
85.47130.04now
The whole history as a distribution, with today marked. Median 110.21 over 5,188 observations. · FRED DTWEXBGS
0%50%100%last 5 yearsnow · 78th85.47130.04
The same record read cumulatively: today sits at the 78 percentile of the full history (n=5,188) and the 16 of the last five years (n=1,250). A level extreme against forty years and ordinary against five is a different fact from one that is extreme against both. · FRED DTWEXBGS
+0.0Jan+0.2Feb-0.1Mar-0.6Apr+0.4May+0.0Jun-0.4Jul+0.6Aug+0.3Sep+0.4Oct+0.3Nov-0.3Decno significant pattern~21 yrs/bar
The spread of monthly means is no larger than the noise — this series has no seasonal worth trading. Kruskal-Wallis across calendar months; the corner figure is the average number of years behind each bar. · FRED DTWEXBGS
-1.2460.3131.8722021-09-242022-12-232024-03-222025-06-202026-09-11seas… -0.630
The MSTL seasonal component itself, at periods 5 and 52. Drawn rather than summarised because a seasonal that changes shape year to year is not one a reader can plan against, and the chart shows that where a single average would hide it. · FRED DTWEXBGS · statsmodels 0.14.6
124.12113.45113.76118.21
Where this has historically gone over 63 trading days: 10th to 90th percentile of 5,100 realised moves, applied to today. A base rate for sizing, not a forecast. · FRED DTWEXBGS
downside (10th)
113.45
-4.0%
median path
118.34
50th percentile
upside (90th)
124.12
+5.0%
Markov-filtered mean
-0.005
per week, state-weighted
The band above is unconditional: it does not know what state we are in. The filtered mean beside it does — it is the two-state Markov model's mean weighted by today's filtered probability (0.6 of the high-volatility state). The two disagree when the current state is unusual, which is the point of printing both.
state
low volatility
since 2020-11-20
P(high vol) now
0.6
filtered, this week
persisted
304
weeks (2128 days)
P(exit in 4w)
not countable
0 of 0 spells this old
spells on record
38
19 completed in this state
-2%21%44%2021-09-242022-12-232024-03-222025-06-202026-09-11P(high v… 1%
Two-state Markov-switching model with switching variance, fitted on 1079 weekly % log change. The line is the FILTERED probability of the high-volatility state — a statement about weeks that have happened, not a forecast. Exit counts are empirical spell counts, not the model's expected durations. · FRED DTWEXBGS · statsmodels 0.14.6
112.70121.77130.842021-09-242022-12-232024-03-222025-06-202026-09-11level 118.21
The same weeks on the level: shaded stretches are the ones the filter puts in the high-volatility state. Mean -0.007 per week at 0.641 standard deviation in the calm state against +0.266 at 1.554 in the volatile one. · FRED DTWEXBGS · MarkovRegression(k_regimes=2, trend='c', switching_variance=True)

S&P 500 · last 7,637.76 as of 2026-09-17 · 2,513 observations from 2016-09-19

3,3315,6848,0382021-10-012022-12-302024-03-292025-06-272026-09-18S&P 5… 7,638
Five years of S&P 500 with the high-volatility weeks shaded, from the same two-state filter the Regime tab draws. One state, one shading, every tab. · FRED SP500

The last 8 times this looked like today — and what S&P 500 did next

episodezlevelS&P 500 +5dS&P 500 +21d
2026-08-071.867,757.64+0.36%-1.08%
2026-07-171.537,457.69-0.61%+3.85%
2026-06-261.517,354.02+2.49%+0.80%
2026-06-051.727,383.74+0.65%+2.08%
2026-04-291.657,135.95+3.21%+5.99%
2026-04-081.066,782.81+3.54%+8.17%
2026-02-231.336,837.75+0.64%-4.12%
2026-02-021.756,976.44-0.17%-2.29%
z in [1.0, 2.0) · n=99 episodes · 21-session hit rate 73% · median +5d 0.54, +21d 1.83 · grade C
Episodes when this series last sat in the same trailing-365-observation z bucket, with the forward 5- and 21-session move of S&P 500. 99 episodes at least 21 days apart; the table shows the most recent 8. A conditional distribution with its count, not a signal.

Bitcoin · last 80,844.00 as of 2026-09-18 · 4,275 observations from 2014-12-01

10,15169,484128,8162021-10-012022-12-302024-03-292025-06-272026-09-18Bitc… 80,844
Five years of Bitcoin with the high-volatility weeks shaded, from the same two-state filter the Regime tab draws. One state, one shading, every tab. · FRED CBBTCUSD

The last 8 times this looked like today — and what Bitcoin did next

episodezlevelBitcoin +5dBitcoin +21d
2026-01-14-0.3996,852.91-7.14%-31.85%
2025-11-06-0.07101,447.93-2.51%-11.93%
2025-10-160.48106,463.30+4.30%-10.93%
2025-03-290.4282,379.98-4.87%+15.37%
2025-03-080.3780,699.17+2.31%-1.93%
2024-10-010.4960,631.37-0.09%+15.78%
2024-09-010.4057,347.49-0.51%+6.00%
2024-08-050.3154,025.90+9.83%+6.38%
z in [-0.5, 0.5) · n=47 episodes · 21-session hit rate 53% · median +5d 1.18, +21d 0.87 · grade C
Episodes when this series last sat in the same trailing-365-observation z bucket, with the forward 5- and 21-session move of Bitcoin. 47 episodes at least 21 days apart; the table shows the most recent 8. A conditional distribution with its count, not a signal.

10y Treasury · last 4.94 as of 2026-09-17 · 16,163 observations from 1962-01-02

1.133.165.182021-10-012022-12-302024-03-292025-06-272026-09-1810y Tr… 4.94
Five years of 10y Treasury with the high-volatility weeks shaded, from the same two-state filter the Regime tab draws. One state, one shading, every tab. · FRED DGS10

The last 8 times this looked like today — and what S&P 500 did next

episodezlevelS&P 500 +5dS&P 500 +21d
2026-08-032.434.70+2.01%+0.41%
2026-07-132.094.62-0.96%+2.83%
2026-05-192.234.67+2.25%+0.90%
2023-10-122.304.70-1.65%+1.51%
2023-09-212.114.49-0.70%-2.44%
2022-11-082.164.14+4.27%+2.76%
2022-10-182.364.01+3.74%+6.42%
2022-09-272.653.97+3.94%+5.03%
z in [2.0, 99.0) · n=139 episodes · 21-session hit rate 9% · median +5d 0.00, +21d 0.00 · grade C
Episodes when this series last sat in the same trailing-365-observation z bucket, with the forward 5- and 21-session move of S&P 500. 139 episodes at least 21 days apart; the table shows the most recent 8. A conditional distribution with its count, not a signal.

VIX · last 15.44 as of 2026-09-17 · 9,276 observations from 1990-01-02

9.9328.6247.312021-10-012022-12-302024-03-292025-06-272026-09-18VIX 15.44
Five years of VIX with the high-volatility weeks shaded, from the same two-state filter the Regime tab draws. One state, one shading, every tab. · FRED VIXCLS

The last 8 times this looked like today — and what S&P 500 did next

episodezlevelS&P 500 +5dS&P 500 +21d
2026-07-31-0.6615.99+3.58%+2.62%
2026-07-06-0.7415.57-0.29%+2.64%
2026-06-15-0.6016.20-1.08%-0.14%
2026-05-25-0.5216.59+1.08%-2.04%
2026-01-22-0.5815.64+0.81%-0.06%
2025-12-29-0.9014.20-0.05%+1.06%
2025-12-03-0.5116.08+0.54%+0.13%
2025-09-05-0.5215.18+1.59%+3.99%
z in [-1.0, -0.5) · n=293 episodes · 21-session hit rate 18% · median +5d 0.00, +21d 0.00 · grade C
Episodes when this series last sat in the same trailing-365-observation z bucket, with the forward 5- and 21-session move of S&P 500. 293 episodes at least 21 days apart; the table shows the most recent 8. A conditional distribution with its count, not a signal.

WTI crude · last 107.02 as of 2026-09-15 · 9,502 observations from 1986-01-02

52.7588.66124.582021-10-012022-12-302024-03-292025-06-272026-09-18WTI … 107.02
Five years of WTI crude with the high-volatility weeks shaded, from the same two-state filter the Regime tab draws. One state, one shading, every tab. · FRED DCOILWTICO

The last 8 times this looked like today — and what WTI crude did next

episodezlevelWTI crude +5dWTI crude +21d
2026-06-032.0699.76-6.09%-30.10%
2026-05-072.3998.38+6.38%-4.13%
2026-04-162.7796.46+2.91%+12.99%
2026-03-263.8396.18+17.73%+2.33%
2026-03-052.7680.88+18.21%+40.96%
2022-06-162.14117.56-10.05%-15.29%
2022-05-262.32116.19+0.59%-6.13%
2022-05-052.07108.17-1.87%+9.98%
z in [2.0, 99.0) · n=106 episodes · 21-session hit rate 56% · median +5d 0.01, +21d 1.25 · grade C
Episodes when this series last sat in the same trailing-365-observation z bucket, with the forward 5- and 21-session move of WTI crude. 106 episodes at least 21 days apart; the table shows the most recent 8. A conditional distribution with its count, not a signal.

Broad dollar · last 118.21 as of 2026-09-11 · 5,188 observations from 2006-01-02

112.70121.77130.842021-09-242022-12-232024-03-222025-06-202026-09-11Broa… 118.21
Five years of Broad dollar with the high-volatility weeks shaded, from the same two-state filter the Regime tab draws. One state, one shading, every tab. · FRED DTWEXBGS

The last 8 times this looked like today — and what Broad dollar did next

episodezlevelBroad dollar +5dBroad dollar +21d
2026-04-30-1.05118.67-0.56%+0.17%
2026-04-08-1.04119.06-0.59%-0.88%
2026-03-10-1.19118.73+0.93%+0.28%
2026-02-13-1.66117.53+0.40%+2.19%
2026-01-23-1.36118.90-0.84%-0.81%
2026-01-02-1.20119.61+0.51%-1.04%
2025-12-11-1.15119.98+0.06%+0.20%
2025-10-20-1.01120.62+0.15%+0.49%
z in [-2.0, -1.0) · n=74 episodes · 21-session hit rate 46% · median +5d -0.03, +21d -0.21 · grade C
Episodes when this series last sat in the same trailing-365-observation z bucket, with the forward 5- and 21-session move of Broad dollar. 74 episodes at least 21 days apart; the table shows the most recent 8. A conditional distribution with its count, not a signal.

What changed since last week

leafthennowchangecompared with
horizon days63.0063.00unchanged2026-09-12
n series33.0036.00+32026-09-12
n panels18.0018.00unchanged2026-09-18
1 of 3 headline leaves moved. A repeated value can mean the upstream had not published, not that nothing happened.
ACF of the remainder · 6 rows
seriesnlags outside 95 bandfirst lag outsideconf95reading
S&P 5002601510.12Autocorrelation of the MSTL remainder over 24 lags, with the 95% band at ±0.122,
Bitcoin2601910.12Autocorrelation of the MSTL remainder over 24 lags, with the 95% band at ±0.122,
10y Treasury2602110.12Autocorrelation of the MSTL remainder over 24 lags, with the 95% band at ±0.122,
VIX260510.12Autocorrelation of the MSTL remainder over 24 lags, with the 95% band at ±0.122,
WTI crude2601810.12Autocorrelation of the MSTL remainder over 24 lags, with the 95% band at ±0.122,
Broad dollar2602110.12Autocorrelation of the MSTL remainder over 24 lags, with the 95% band at ±0.122,
Harvey-Jaeger style structural check: the autocorrelation of the decomposition's remainder against a 95% band. Lags outside the band mean the trend and seasonal have not taken everything systematic out of the series. This is Method content. It is not a forecast and nothing is extrapolated from it.

cumulative abnormal return vs the sample's own mean drift, aligned by DATE to the nearest session; only the first day of each episode counts as an event; 95% CIs from 2,000 bootstrap resamples. Events are rebuilt from the underlying official series, so these are historical relationships rather than a record of calls we published. Window 10 sessions before, 20 after. As of 2026-09-19 13:12 UTC. Floor for a conditional claim: 20 episodes (grade_t_thresholds.json).

2s10s crosses into inversion — S&P 500 · n=44 · earliest 1978-08-18
horizonnmeanmedian95% CIthit rateCI
+1d44+0.03%+0.03%[-0.2, +0.3]0.355%straddles zero
+5d44-0.13%-0.25%[-0.6, +0.4]-0.4846%straddles zero
+10d44+0.03%+0.41%[-0.8, +0.8]0.0759%straddles zero
+20d44+0.37%+1.04%[-0.8, +1.4]0.6564%straddles zero
series · 6 rows
labelseries idunitkindlastas offirst obsn obsma 13wvs 13w
S&P 500SP500idxprice7,637.762026-09-172016-09-192,5137,593.4544.31
BitcoinCBBTCUSDusdprice80,844.002026-09-182014-12-014,27569,159.7011,684.30
10y TreasuryDGS10pctlevel4.942026-09-171962-01-0216,1634.680.26
VIXVIXCLSpctlevel15.442026-09-171990-01-029,27615.93-0.49
WTI crudeDCOILWTICOusdprice107.022026-09-151986-01-029,50285.4121.61
Broad dollarDTWEXBGSidxprice118.212026-09-112006-01-025,188119.58-1.36
scalars
fieldvalue
horizon days63
n series36
statsmodels version0.14.6
statsmodels pinned0.14.6
statsmodels errornot reported
4 rows and 1 field withheld at this tier: absent from the page, not hidden in it.
MacroBessent is a research publication and is not investment advice; nothing here is a recommendation to buy or sell anything. Every figure carries an evidence grade (D descriptive, C conditional with counts, N tested-null, T tested-positive — none exists). Nulls are public in the registry. Sources and as-of dates are on every chart.