Timing Arrays — Dashboard

What is the model saying today?

The system watches 14 financial markets around the world — American stocks, European stocks, Asian stocks, gold, Bitcoin, bonds — and tries to guess whether each one might fall a lot (more than 8%) over the next 4 months. For each market it gives a score from 1 (very calm) to 10 (warning).

Bottom line: 3 markets flashing a WARNING right now, and 2 of the 14 are calm. Most of the world is fine, but there are spots to be careful about.

By region:

USA: WARNING
Europe: neutral
Asia (Japan, Korea, HK, Taiwan): Japanese stocks a bit nervous, South Korean stocks neutral, Hong Kong stocks WARNING, Taiwanese stocks WARNING
Chinese tech (Hong Kong): Chinese internet stocks a bit nervous, Chinese tech stocks neutral, Hong Kong tech stocks neutral
Australia: neutral
Gold and the US Dollar: all calm
Bitcoin: neutral
US government bonds: a bit nervous

What's changed in the past week:

Getting riskier: South Korean stocks (up 2 steps to dec 6), European stocks (up 2 steps to dec 6), Australian stocks (up 2 steps to dec 5). The model has been raising its concern about these over the past week.
Calming down: gold (down 3 steps to dec 3), Hong Kong tech stocks (down 3 steps to dec 4), the US dollar (down 2 steps to dec 3), US government bonds (down 2 steps to dec 8). The model has been relaxing its worry about these over the past week.

What might happen next:

Even though the current decile reading isn't a full warning, US government bonds, Japanese stocks, South Korean stocks have been persistently elevated for weeks. The model's persistence signal tells us that markets which have been chronically stressed often fail eventually — worth keeping an eye on.

Things to be careful about right now:

US stocks are in the riskiest zone right now. Historically when the model has said this, prices fell meaningfully over the next 4 months about 11% of the time. The honest uncertainty range is 5–24% — the model isn't precise about exactly how risky. This has been the case for weeks already — the stress is chronic, not a fresh spike.
Hong Kong stocks are in the riskiest zone right now. Historically when the model has said this, prices fell meaningfully over the next 4 months about 28% of the time. The honest uncertainty range is 31–59% — the model isn't precise about exactly how risky. This has been the case for weeks already — the stress is chronic, not a fresh spike.
Taiwanese stocks are in the riskiest zone right now. Historically when the model has said this, prices fell meaningfully over the next 4 months about 19% of the time. The honest uncertainty range is 14–57% — the model isn't precise about exactly how risky. This has been the case for weeks already — the stress is chronic, not a fresh spike.

Important caveat:

The model has been doing a bad job of predicting Gold lately. Don't fully trust what it says about it until things change.

Colour key: green = calm · yellow = neutral · orange = a bit nervous · red = warning

Today's read — October 01, 2026· auto-generated from current model state · focus: ^GSPC

Where we are

The Equity Risk Dial reads WATCH (71/100, 29th-pctile trailing 1y convergence). ECM phase: late decline (wave 7/8 of the current 8.6-yr cycle; next main pivot 2028-08-28, +1.9y out). Current ^GSPC: 7,684 (Sep 28, 2026); last 30 trading days -0.6%.

What the model is pointing at

Calibrated drawdown probabilities (P(120d return < −8%) | current decile; base rate 7%): ⚠ Composite 14% (CI 11–17%) · ⚠ Dial 11% (CI 9–14%) · Yield curve 3% (CI 2–5%) · VIX term structure 6% (CI 4–9%) · ⚠ MOVE 17% (CI 14–19%) · ⚠ SOX leadership 15% (CI 12–17%) · ⚠ Breadth 11% (CI 9–15%) · ⚠ JPY carry stress 13% (CI 11–16%) · ⚠ CRE / real estate 13% (CI 10–15%) · ⚠ Credit spread 11% (CI 8–13%) · ⚠ Mortgage spread 10% (CI 8–12%) · ⚠ IPO market 19% (CI 16–22%) · ⚠ CFTC positioning extreme 10% (CI 8–13%) · EPU 5% (CI 4–8%) · ⚠ TIC foreign-official UST flow 15% (CI 14–17%) · ⚠ GPR 12% (CI 10–14%) · L-Wave 5% (CI 3–7%) · ⚠ Long Term 11% (CI 10–11%) · ⚠ Direction Change 12% (CI 11–13%) · ⚠ Consumer stress 10% (CI 8–12%) ⚠ = CI lower bound above base rate.
Nearest resistance above 7,684: 7,791 (+1.4%, Bollinger upper (20,2σ)) · 7,834 (+2.0%, pivot R1 (monthly))
Nearest support below 7,684: 7,669 (−0.2%, pivot P (monthly)) · 7,551 (−1.8%, Bollinger lower (20,2σ))
Convergence score: latest +0.21, 20-day trend flat (-0.43).

Honest caveats

The risk dial has 6/6 historical crash recall but only ~29% precision (walk-forward OOS) — it warns early and often. Cost-aware strategy backtest showed mechanical de-risking underperforms buy-and-hold. Treat the dial as situational awareness, not an execution rule.
Validated signal scorecard (Phase 4–7): Dial = 6/6 named-crash recall, ~29% precision (early & often). Yield curve T10Y2Y = 4/6 named-crash recall; flagged COVID 149d ahead and 2022 bear 41d ahead; +18.7pp drawdown saved at 2022 bear with Aggressive sizing. L-Wave = 4/6 recall, +9.8pp DD at GFC. Long Term = 2/6 recall, +24.6pp DD at COVID. Direction Change = lowest Brier (0.0897) on probability scale but only 1/6 recall via decile-≤2 alert rule.
All reversal/MA/Fibonacci/trend lines are deterministic price levels derived from past pivots — they are not forecasts.
Forward-projected cycle dates assume cycle-spacing persistence, which may not hold. Bare ECM cycle did not beat random in our validation (info-ratio ~1.0).

Regime probabilities — P(120d ^GSPC return < −8%) calibrated from walk-forward backtest, Wilson 95% CI

Base rate (no signal): 7.3% · ⚠ = CI lower bound above base rate (calibrated warning). Bar scale 0–60%, white tick = base rate.

Signal	Decile	P(drawdown)	95% CI	sample
Composite (7-signal mean) ⚠✓	9	13.6%	[10.6%, 17.3%]	n=404
BEST — equal-weight mean of all 7 calibrated signals; Brier 0.0661 vs base 0.0680 (beats every individual); decile 10 → 27.7% drawdown rate (CI 23.5-32.3%) vs 7.3% base = +20pp lift
Dial (cross-asset convergence) ⚠	5	11.3%	[9.0%, 14.2%]	n=582
weak — Brier barely beats base rate; high crash recall but high false-alarm rate
Yield curve (T10Y2Y)✓	2	2.8%	[1.8%, 4.6%]	n=564
strong — flagged COVID 149d ahead + 2022 bear 41d ahead; +18.7pp drawdown saved at 2022; non-monotonic (danger zone at deciles 6-8)
VIX term structure	2	6.0%	[4.1%, 8.8%]	n=400
strongest OOS — only signal with statistically significant OOS IC (+0.227, CI +0.094/+0.374); 4/4 named-crash recall; flagged 2024-25 decline 149d ahead; modest per-prediction spread (+4pp lift at peak decile)
MOVE (bond vol) ⚠✓	10	16.6%	[14.2%, 19.4%]	n=777
weak — 3/5 named-crash recall (GFC, 2011, 2018), flagged 2024-25 OOS; era IC flipped 2009-15 (QE distortion); decile 10 hits 16.6% (+8pp lift)
SOX leadership (60d vs SPX) ⚠○	1	14.6%	[12.4%, 17.1%]	n=844
strong — 6/6 named-crash recall (matches dial); 3/3 OOS including 2024-25 +149d ahead; precision 32% (+22pp lift); when semis lag broad market, drawdowns historically follow
Breadth (RSP vs SPX 60d) ⚠	2	11.3%	[8.7%, 14.6%]	n=450
strong — 5/5 testable named-crash recall (perfect within data window); 36% OOS precision (+28pp lift); flagged COVID/2022/2024-25 OOS; captures the narrow-leadership / mega-cap-concentration risk that the cross-asset composite is structurally blind to
JPY carry stress (60d USD/JPY) ⚠○	1	13.2%	[11.0%, 15.7%]	n=803
strong — 5/6 named-crash recall, statistically significant OOS IC at -0.286 (CI strictly < 0, only 2nd signal to clear this); dual warning: decile 1 (active carry unwind = Aug 2024 episode) AND decile 9 (peak carry euphoria, 15.7% drawdown rate); captures the global leverage risk channel
CRE / real estate (VNQ vs SPY 60d) ⚠	1	12.6%	[10.4%, 15.2%]	n=715
moderate — 3/5 named-crash recall including GFC +148d ahead (CRE was its epicenter); 22% OOS precision (+14pp lift); decile 10 (RE strong) = strongest calm signal (4.7%, -5.2pp); captures the credit-channel risk that no other signal sees
Credit spread (HYG-LQD 60d) ⚠	10	10.5%	[8.3%, 13.2%]	n=609
strong — 4/4 testable named-crash recall (perfect within data window); flagged COVID +138d, 2022 bear +150d, 2024-25 +147d; decile 9 = peak warning 13.0%; captures broad corporate credit-risk pricing
Mortgage spread (30y - 10y T) ⚠	2	9.8%	[7.7%, 12.4%]	n=620
strong — 5/6 named-crash recall (only missed COVID exogenous shock); all flagged crashes signaled 140-149 days ahead; decile 1 (tight spread) = STRONGEST single-signal calm at -5.4pp; decile 7-8 = peak warning at 14.5%; the housing/MBS channel
IPO market (FPX vs SPY 60d) ⚠✓	1	19.0%	[16.1%, 22.1%]	n=654
strong — 5/5 testable named-crash recall (FPX-era only, dot-com excluded); asymmetric U-shape: decile 1 (IPO bid leaving) hits 19.0% (+8.3pp, STRONGEST single-decile warning in dashboard); decile 10 (IPO euphoria) hits 13.8% (+3.0pp); captures issuance-market temperature and the rotation-from-speculative pattern that historically front-runs broader risk-off
CFTC positioning extreme ⚠	2	10.3%	[8.1%, 12.9%]	n=604
moderate — 4/6 named-crash recall; +12.3pp drawdown saved at COVID; non-monotonic peak at decile 9 (20.5%); IC sign flipped 2009-2015 (momentum regime)
EPU (econ policy uncertainty)○	5	5.4%	[3.8%, 7.7%]	n=552
strong — 5/6 named-crash recall (only missed COVID, an exogenous shock); IS/OOS IC consistent at +0.15; decile 3 hits 20% drawdown rate (CI 17-23%); non-monotonic peak in mid-range
TIC foreign-official UST flow ⚠✓	10	15.3%	[13.7%, 16.9%]	n=1940
strong — 5/6 named-crash recall, 3/3 OOS (COVID -12d, 2022 +150d, 2024-25 +79d); 30% OOS precision (+22pp); bimodal warning at decile 1 (foreign UST dumping) AND decile 10 (safe-haven accumulation)
GPR (geopolitical risk) ⚠	10	11.9%	[10.1%, 14.0%]	n=1080
narrow-peak — IS/OOS IC near zero; but decile 6 alone hits 27.5% drawdown rate (+17pp, CI 24-31%) — biggest single-decile spike of any signal. Contributes strongly when GPR is in mid-range, dilutes elsewhere
L-Wave (FFT cycle)○	6	5.0%	[3.4%, 7.3%]	n=480
moderate — Brier 0.0904 vs 0.0940 base
Long Term (ECM cycle) ⚠	10	10.6%	[9.7%, 11.5%]	n=4744
weak — narrow probability range across deciles
Direction Change (density) ⚠○	10	12.2%	[11.1%, 13.5%]	n=3039
moderate — lowest Brier (0.0897); decile 3 hits 47% drawdown rate
Consumer stress (K-shaped economy) ⚠	6	10.0%	[8.4%, 11.8%]	n=1231
strong — Bonferroni-clean on both tails at all 4 horizons (30/60/120/252d); composite of 8 FRED series (UMich sentiment + inflation expectations, card delinquencies, saving rate, debt service ratio, retail sales, real income, revolving credit); h=252 decile 8 hits 44.9% (+31.7pp) vs 13.2% base, decile 10 = 54.5% (+41.4pp); decile 2-3 protective at -10 to -12pp; CAVEAT: high-decile readings cluster into only 3 independent episodes (2007-08 GFC, 2015-16 China slowdown, 2022 inflation shock) — directionally consistent across horizons but small effective-N

How to read: each row is a separate signal's calibrated estimate of the chance ^GSPC drops >8% in the next 120 trading days, given that signal's current decile. The signals are not multiplied: naive combination tested worse than the base rate (see Phase 5c). Treat multiple warnings as confirmatory context, not arithmetic risk. Significance badges next to signal names (glyph = strictness, colour = direction): ✓/✓ = Bonferroni pooled (Phase 48); ✓/✓ = per-target Bonferroni only (Phase 52); ○/○ = BH-FDR-clean only (Phase 51). Red/amber = warning (more drawdowns than baseline); green/blue = calm.

Technical Analysis · focus: ^GSPC

Price + convergence oscillator + direction-changes
Zoom timeframe: 3m / 6m / 1y / 5y / 10y / All buttons just above chart | Toggle overlays: MA 50 / 100 / 200, Bollinger (20,±2σ), Donchian (55d), Pivots (monthly) — via the upper button row or legend entries (top-right) | Draw your own trend lines: click the line icon (top-right modebar), then click-drag from anchor #1 (e.g. a prior low) through anchor #2 (e.g. a later low). On release, the line auto-extends both directions across the chart. Use erase shape to delete. | All overlays are deterministic price levels — no probability claim

Readout drag ☰

loading…

Forecast term structure · focus: ^GSPC · calibrated drawdown probabilities at multiple forward horizons

Signal	30d base 3%	60d base 6%	120d base 7%	252d base 6%
Composite BEST — equal-weight mean of all 7 calibrated signals; Brier 0.0661 vs base 0.0680 (beats every individual); decile 10 → 27.7% drawdown rate (CI 23.5-32.3%) vs 7.3% base = +20pp lift	1%○ CI 0–3% decile 4 · n=405	3%○ CI 2–5% decile 6 · n=405	14% ▲✓ CI 11–17% decile 9 · n=404	2%○ CI 1–4% decile 8 · n=387
Dial weak — Brier barely beats base rate; high crash recall but high false-alarm rate	7% CI 5–9% decile 5 · n=590	6% CI 5–9% decile 5 · n=590	11% CI 9–14% decile 5 · n=582	11% CI 9–14% decile 5 · n=582
Yield curve strong — flagged COVID 149d ahead + 2022 bear 41d ahead; +18.7pp drawdown saved at 2022; non-monotonic (danger zone at deciles 6-8)	3%○ CI 2–4% decile 1 · n=1644	5%○ CI 4–6% decile 1 · n=1643	3%✓ CI 2–5% decile 2 · n=564	6%✓ CI 5–7% decile 1 · n=1642
VIX term structure strongest OOS — only signal with statistically significant OOS IC (+0.227, CI +0.094/+0.374); 4/4 named-crash recall; flagged 2024-25 decline 149d ahead; modest per-prediction spread (+4pp lift at peak decile)	3% CI 2–5% decile 3 · n=401	8% CI 5–11% decile 3 · n=401	6% CI 4–9% decile 2 · n=400	5% CI 3–8% decile 3 · n=385
MOVE weak — 3/5 named-crash recall (GFC, 2011, 2018), flagged 2024-25 OOS; era IC flipped 2009-15 (QE distortion); decile 10 hits 16.6% (+8pp lift)	2% CI 1–4% decile 3 · n=509	10%○ CI 7–13% decile 3 · n=505	17% ▲✓ CI 14–19% decile 10 · n=777	3%✓ CI 2–5% decile 3 · n=497
SOX leadership strong — 6/6 named-crash recall (matches dial); 3/3 OOS including 2024-25 +149d ahead; precision 32% (+22pp lift); when semis lag broad market, drawdowns historically follow	5% CI 4–7% decile 10 · n=786	8% CI 6–10% decile 10 · n=786	15%○ CI 12–17% decile 1 · n=844	15% CI 12–17% decile 10 · n=771
Breadth strong — 5/5 testable named-crash recall (perfect within data window); 36% OOS precision (+28pp lift); flagged COVID/2022/2024-25 OOS; captures the narrow-leadership / mega-cap-concentration risk that the cross-asset composite is structurally blind to	7%○ CI 5–9% decile 1 · n=846	12%✓ CI 10–14% decile 1 · n=846	11% CI 9–15% decile 2 · n=450	6%✓ CI 4–7% decile 1 · n=841
JPY carry stress strong — 5/6 named-crash recall, statistically significant OOS IC at -0.286 (CI strictly < 0, only 2nd signal to clear this); dual warning: decile 1 (active carry unwind = Aug 2024 episode) AND decile 9 (peak carry euphoria, 15.7% drawdown rate); captures the global leverage risk channel	7% CI 5–9% decile 4 · n=614	7% CI 5–9% decile 4 · n=607	13%○ CI 11–16% decile 1 · n=803	10% CI 8–13% decile 4 · n=598
CRE / real estate moderate — 3/5 named-crash recall including GFC +148d ahead (CRE was its epicenter); 22% OOS precision (+14pp lift); decile 10 (RE strong) = strongest calm signal (4.7%, -5.2pp); captures the credit-channel risk that no other signal sees	4% CI 3–7% decile 2 · n=403	7% CI 5–10% decile 2 · n=403	13% CI 10–15% decile 1 · n=715	10% CI 7–13% decile 2 · n=402
Credit spread strong — 4/4 testable named-crash recall (perfect within data window); flagged COVID +138d, 2022 bear +150d, 2024-25 +147d; decile 9 = peak warning 13.0%; captures broad corporate credit-risk pricing	2%○ CI 1–3% decile 10 · n=610	5%○ CI 3–7% decile 10 · n=610	11% CI 8–13% decile 10 · n=609	11%○ CI 9–14% decile 10 · n=604
Mortgage spread strong — 5/6 named-crash recall (only missed COVID exogenous shock); all flagged crashes signaled 140-149 days ahead; decile 1 (tight spread) = STRONGEST single-signal calm at -5.4pp; decile 7-8 = peak warning at 14.5%; the housing/MBS channel	4% CI 3–7% decile 3 · n=522	7% CI 5–10% decile 3 · n=517	10% CI 8–12% decile 2 · n=620	12% CI 10–15% decile 3 · n=505
IPO market strong — 5/5 testable named-crash recall (FPX-era only, dot-com excluded); asymmetric U-shape: decile 1 (IPO bid leaving) hits 19.0% (+8.3pp, STRONGEST single-decile warning in dashboard); decile 10 (IPO euphoria) hits 13.8% (+3.0pp); captures issuance-market temperature and the rotation-from-speculative pattern that historically front-runs broader risk-off	7% CI 5–10% decile 8 · n=401	6% CI 4–8% decile 8 · n=401	19% ▲✓ CI 16–22% decile 1 · n=654	11% CI 8–14% decile 8 · n=401
CFTC positioning extreme moderate — 4/6 named-crash recall; +12.3pp drawdown saved at COVID; non-monotonic peak at decile 9 (20.5%); IC sign flipped 2009-2015 (momentum regime)	4% CI 2–6% decile 3 · n=551	4%○ CI 3–6% decile 3 · n=551	10% CI 8–13% decile 2 · n=604	12% CI 9–15% decile 3 · n=518
EPU strong — 5/6 named-crash recall (only missed COVID, an exogenous shock); IS/OOS IC consistent at +0.15; decile 3 hits 20% drawdown rate (CI 17-23%); non-monotonic peak in mid-range	2%○ CI 1–3% decile 2 · n=593	4%○ CI 2–6% decile 2 · n=593	5%○ CI 4–8% decile 5 · n=552	8%○ CI 6–11% decile 2 · n=572
TIC foreign-official UST flow strong — 5/6 named-crash recall, 3/3 OOS (COVID -12d, 2022 +150d, 2024-25 +79d); 30% OOS precision (+22pp); bimodal warning at decile 1 (foreign UST dumping) AND decile 10 (safe-haven accumulation)	7%✓ CI 6–9% decile 10 · n=1957	12%✓ CI 11–14% decile 10 · n=1940	15%✓ CI 14–17% decile 10 · n=1940	20% ▲✓ CI 19–22% decile 10 · n=1940
GPR narrow-peak — IS/OOS IC near zero; but decile 6 alone hits 27.5% drawdown rate (+17pp, CI 24-31%) — biggest single-decile spike of any signal. Contributes strongly when GPR is in mid-range, dilutes elsewhere	4% CI 3–5% decile 10 · n=1117	9% CI 7–10% decile 10 · n=1096	12% CI 10–14% decile 10 · n=1080	12% CI 10–14% decile 10 · n=1080
L-Wave moderate — Brier 0.0904 vs 0.0940 base	4% CI 4–5% decile 10 · n=4480	10%○ CI 9–11% decile 10 · n=2782	5%○ CI 3–7% decile 6 · n=480	12% CI 9–17% decile 8 · n=297
Long Term weak — narrow probability range across deciles	5% CI 4–5% decile 10 · n=4971	7% CI 7–8% decile 10 · n=4907	11% CI 10–11% decile 10 · n=4744	15%○ CI 14–16% decile 10 · n=4381
Direction Change moderate — lowest Brier (0.0897); decile 3 hits 47% drawdown rate	5% CI 4–6% decile 10 · n=3101	9% CI 8–10% decile 10 · n=3080	12%○ CI 11–13% decile 10 · n=3039	18% ▲✓ CI 17–20% decile 10 · n=2949
Consumer stress strong — Bonferroni-clean on both tails at all 4 horizons (30/60/120/252d); composite of 8 FRED series (UMich sentiment + inflation expectations, card delinquencies, saving rate, debt service ratio, retail sales, real income, revolving credit); h=252 decile 8 hits 44.9% (+31.7pp) vs 13.2% base, decile 10 = 54.5% (+41.4pp); decile 2-3 protective at -10 to -12pp; CAVEAT: high-decile readings cluster into only 3 independent episodes (2007-08 GFC, 2015-16 China slowdown, 2022 inflation shock) — directionally consistent across horizons but small effective-N	1%○ CI 1–3% decile 8 · n=547	7% CI 5–9% decile 8 · n=525	10% CI 8–12% decile 6 · n=1231	45% ⚠✓ CI 40–49% decile 8 · n=477

Cell colour: CI overlaps base = no warning · +0–5pp above base · +5–15pp ▲ · +15pp+ ⚠
How to read: each cell is today's calibrated probability that ^GSPC drops ≥8% over the column's horizon, given the row signal's current decile in walk-forward history. Cells are independent estimates — see the regime-probabilities panel above for the unified 120d view.
Significance badges — glyph = correction strictness, colour = direction. ✓/✓ = Bonferroni pooled (7,119 tests, Phase 48); ✓/✓ = per-target Bonferroni only (~720 tests, Phase 52); ○/○ = BH-FDR-clean only (Phase 51, ~5% expected FDR). Red/amber = warning direction (more drawdowns than baseline); green/blue = calm direction (fewer).

Daily commentary · what's moved across the cross-asset panel · 1d / 7d / 30d decile + probability deltas ✉ Email this view

What's changing

Arrow direction: ▲ = decile rose (more risk), ▼ = decile fell (less risk). Number = absolute change in decile. Probability shift in pp shown in parentheses. Sorted by 7-day movement magnitude — biggest movers first.

Threshold crossings (past 7 days)

Gold (GC=F) dropped to calm zone (dec 6→3)
HSTECH (3033.HK) dropped out of elevated zone (dec 7→4)
US Dollar (DX-Y.NYB) dropped to calm zone (dec 5→3)
10y T-Notes (ZN=F) left warning regime (dec 10→8)
China Tech (CQQQ) dropped out of elevated zone (dec 8→6)
China Internet (KWEB) left warning regime (dec 9→8)

Market	Now	p̂	1d Δ	7d Δ	30d Δ
Gold (GC=F)	dec 3	6.7%	▼ -1 (-0.6pp)	▼ -3 (-1.4pp)	flat (-0.3pp)
HSTECH (3033.HK)	dec 4	40.6%	flat	▼ -3 (-3.5pp)	flat (+0.3pp)
US Dollar (DX-Y.NYB)	dec 3	3.3%	▼ -1 (-0.2pp)	▼ -2 (-0.4pp)	▼ -2 (-0.4pp)
10y T-Notes (ZN=F)	dec 8	0.8%	▼ -2 (-0.4pp)	▼ -2 (-0.3pp)	▼ -1 (-0.2pp)
KOSPI (^KS11)	dec 6	16.3%	▲ +2 (+1.1pp)	▲ +2 (+1.0pp)	▼ -2 (-1.1pp)
China Tech (CQQQ)	dec 6	30.9%	flat	▼ -2 (-1.7pp)	▼ -4 (-4.1pp)
Stoxx Europe 600	dec 6	11.0%	flat	▲ +2 (+1.5pp)	▲ +2 (+1.4pp)
ASX 200 (^AXJO)	dec 5	9.9%	flat	▲ +2 (+1.1pp)	▲ +2 (+0.8pp)
Bitcoin (BTC-USD)	dec 4	29.0%	▼ -1 (-0.4pp)	▼ -1 (-0.4pp)	▼ -1 (-0.3pp)
China Internet (KWEB)	dec 8	31.2%	flat	▼ -1 (-2.5pp)	▲ +3 (+3.1pp)
S&P 500 (^GSPC)	dec 9	10.8%	flat	flat (-0.3pp)	▲ +1 (+0.2pp)
Nikkei 225 (^N225)	dec 7	16.2%	flat	flat (-0.5pp)	▲ +2 (+0.7pp)
Hang Seng (^HSI)	dec 10	27.5%	flat	flat (+1.6pp)	▲ +7 (+7.5pp)
Taiwan (^TWII)	dec 10	19.3%	flat (+0.6pp)	flat	▲ +3 (+4.4pp)

Model drift monitor · rolling 90-day Brier percentile per market · ALERT = recent reliability in worst 10% historically

⚠ DRIFT ALERT — 1 market: Gold
Current 90-day Brier is in the WORST 10% of historical 90-day windows. Recent reliability is materially worse than usual — treat readings with extra caution.

Model drift monitor

For each market we compute a 90-day rolling Brier score (sum of squared prediction errors) and rank TODAY's window against the market's own historical distribution.

OK (percentile ≤ 70): performance in normal range. WATCH (70–90): mildly worse than usual. ALERT (> 90): performance in WORST 10% historically — the model is failing to predict accurately right now, treat readings with caution.

Bias = recent 90-day mean predicted probability minus mean realised drawdown rate. Large negative bias = model is UNDER-warning (missing real drawdowns). Large positive bias = model is OVER-warning (false alarms in calm markets).

Market	Status	Percentile	Current 90d Brier	Historical median	Bias (pred − real)
Gold	ALERT	98%	0.5357	0.0081	-54.2pp
China Internet	WATCH	88%	0.4077	0.1638	-45.8pp
HSTECH	WATCH	84%	0.2753	0.2210	-13.0pp
KOSPI	OK	69%	0.1275	0.0286	-0.6pp
Bitcoin	OK	66%	0.2222	0.1332	-0.0pp
10y T-Notes	OK	65%	0.0001	0.0000	+0.7pp
Hang Seng	OK	63%	0.1609	0.0836	-1.2pp
S&P 500	OK	55%	0.0098	0.0096	+9.8pp
China Tech	OK	49%	0.1727	0.1772	+10.9pp
US Dollar	OK	35%	0.0014	0.0017	+3.7pp
Stoxx 600	OK	26%	0.0095	0.0145	+9.7pp
Taiwan	OK	16%	0.0182	0.0228	+13.4pp
Nikkei 225	OK	14%	0.0218	0.0340	+14.7pp
ASX 200	OK	14%	0.0086	0.0112	+9.2pp

Market dials · one gauge per market · needle = current decile (1-10), color zones encode risk regime · sorted by current decile, highest risk first

How to read the dials

Each dial shows one market's current decile on its own 120-day drawdown-risk composite, ranked against the last 252 trading days of its own signal history. The composite mixes the universal signals (JPY carry, HYG/LQD credit, MOVE bond vol) with whichever extras transferred to that market in the calibration audit — so each market is judged on the signals that actually work for it.

dec 1–3 (green) = calm regime, historical drawdown rate well below base. dec 4–6 (yellow) = neutral. dec 7–8 (orange) = elevated; tail risk rising. dec 9–10 (red) = warning regime, historical drawdown rate sharply above base.

Subtitle reads dec X · YY.Y% · base Z% — the calibrated drawdown probability at the current decile vs that market's overall base rate. A WARNING tag means the decile's 95% confidence-interval lower bound is above the base rate. Dials are sorted highest-risk first, so the most actionable markets cluster at the top.

dec 1-3 calmdec 4-6 neutraldec 7-8 elevateddec 9-10 warningsorted by current decile, highest risk first · horizon = 120 trading days

Composite trajectory · current vs persistent stress · momentum + acceleration of composite per market

Reading the trajectory columns

Each column captures a different aspect of where the composite is RIGHT NOW vs WHERE IT'S BEEN:

Composite — current decile + calibrated drawdown probability. The existing model reading.
Persistence — how many of the last 30 days the composite has been at decile ≥7. Captures chronic stress. Strongest single feature in the system (D10 lift +15-41pp on 13 of 14 markets).
Momentum — 30-day change in composite decile. High = stress rising; low = stress falling.
Acceleration — rate of change of momentum. High = acceleration of regime shift.
Trajectory score — simple average of composite + persistence deciles. Quick aggregate of "how bad is it AND how long has it been bad".

A market with low composite but high persistence is chronically elevated. A market with high composite but low persistence is a fresh spike. The combination tells you whether you're early or late in the cycle.

Market	Composite	Persistence	Momentum	Acceleration	Trajectory
S&P 500	dec 9 14% / base 7%	dec 10 26% / base 7%	dec 6	dec 5	9.5
Gold	dec 3 1% / base 10%	dec 6 4% / base 10%	dec 7	dec 10	4.5
US Dollar	dec 3 0% / base 4%	dec 3 3% / base 4%	dec 5	dec 4	3.0
Bitcoin	dec 4 16% / base 25%	dec 3 36% / base 26%	dec 9	dec 8	3.5
10y T-Notes	dec 8 0% / base 1%	dec 10 1% / base 1%	dec 8	dec 9	9.0
Nikkei 225	dec 7 18% / base 14%	dec 9 15% / base 13%	dec 7	dec 5	8.0
KOSPI	dec 6 14% / base 15%	dec 9 14% / base 14%	dec 1	dec 2	7.5
Hang Seng	dec 10 48% / base 23%	dec 10 38% / base 22%	dec 10	dec 8	10.0
Taiwan	dec 10 37% / base 12%	dec 10 24% / base 12%	dec 7	dec 4	10.0
China Internet	dec 8 40% / base 29%	dec 10 49% / base 29%	dec 5	dec 4	9.0
China Tech	dec 6 46% / base 38%	dec 7 46% / base 36%	dec 3	dec 3	6.5
HSTECH	dec 4 33% / base 42%	dec 1 20% / base 44%	dec 6	dec 6	2.5
Stoxx 600	dec 6 6% / base 11%	dec 8 9% / base 10%	dec 10	dec 7	7.0
ASX 200	dec 5 5% / base 10%	dec 4 2% / base 9%	dec 8	dec 5	4.5

Cross-market contagion · stress-propagation matrix · when leader (row) hits dec ≥9, conditional P(follower entering stress within 14 trading days)

How to read the contagion matrix

Each cell answers: when LEADER (row) enters a stress regime (decile ≥9), what fraction of the time does FOLLOWER (column) enter stress within 14 trading days? The number shown is lift — the conditional probability minus the follower's baseline rate.

Red = strong contagion (stress in leader reliably precedes stress in follower). Yellow = mild positive. Grey = neutral / no relationship. Blue = strong decoupling (leader stress is associated with REDUCED follower stress; the two regimes are anti-correlated, often safe-haven flows).

Hover any cell for the conditional probability, sample size, and median lag in trading days. Pairs marked with * have fewer than 10 leader-onsets in joint history and should be treated as unreliable.

Strongest contagion (top 8, reliable sample)

Leader		Follower	Lift	P(cond/base)	Median lag	Sample
Gold	→	Bitcoin	+13.2pp	50% / 37%	7 trading days	n=12
HSTECH	→	10y T-Notes	+8.7pp	26% / 17%	5 trading days	n=81
Stoxx 600	→	US Dollar	+8.5pp	30% / 22%	8 trading days	n=56
ASX 200	→	US Dollar	+6.8pp	28% / 22%	8 trading days	n=193
Bitcoin	→	10y T-Notes	+6.8pp	19% / 12%	2 trading days	n=27
Gold	→	Stoxx 600	+5.5pp	23% / 17%	5 trading days	n=111
Gold	→	US Dollar	+5.0pp	27% / 22%	8 trading days	n=111
Bitcoin	→	Stoxx 600	+4.3pp	11% / 7%	5 trading days	n=27

Vol-Bond divergence dials · Phase 28 · capitulation-buy + bond-leading-warning binary flags · regime-dependent, standalone (not in composite)

Capitulation Buy (vol-only panic)

off

VIX_TS dec 2 (need ≥9) · MOVE dec 10 (need ≤5)

Track record (273 firings, h=120d):
Mean fwd return: +8.13% (base +5.44%)
Hit-rate (positive): 85.7% (base 74.9%)

Bond-Leading Warning (bond-only panic)

FIRING

MOVE dec 10 (need ≥9) · VIX_TS dec 2 (need ≤5)

Track record (301 firings, h=120d):
Mean fwd return: +0.84% (base +5.44%)
Hit-rate (positive): 61.5% (base 74.9%)

How to read. Both signals exploit MOVE-vs-VIX_TS divergence. Capitulation Buy fires when equity vol panics while bond vol stays calm — historically a buyable equity-specific shock (best in QE-era regimes; missed 2011 Eurocrisis). Bond-Leading Warning fires when bond vol panics while equity vol stays calm — the bond market historically led (caught the 2021-22 Fed pivot cleanly; false alarms cluster when bond vol moves for non-equity reasons). Both are regime-dependent and shown for eyeball reference; not integrated into the leading drawdown composite.

As of 2026-09-28

Food inflation watch · Phase 29 · CPI food · PPI farm/processed · IMF global food · per-commodity prices · CPI Food at home YoY now integrated as leading drawdown signal (+7.5pp D10 lift, 1990-2026)

Signal	YoY	5y pctile	Zone	As of
US CPI Food at homeIN COMPOSITE	+2.13%	33.3	neutral	2026-08
US CPI Food	+2.63%	26.7	low	2026-08
PPI Processed foods	-3.47%	45.0	neutral	2026-08
PPI Farm products	+17.68%	80.0	elevated	2026-08
IMF Global Food Index	+4.78%	68.3	neutral	2026-07
Global Beef	+8.80%	40.0	neutral	2026-07
Global Cocoa	-23.78%	15.0	low	2026-07
Global Coffee	+13.40%	46.7	neutral	2026-07
Global Wheat	+38.41%	85.0	elevated	2026-07
Global Corn	+10.78%	68.3	neutral	2026-07
Global Rice	+26.24%	88.3	elevated	2026-07
PPI Nitrogen Fertilizer	+19.92%	61.7	neutral	2026-08

How to read. YoY = annual change in the price series. 5y percentile is the rank vs the trailing 5 years of YoY readings — this normalises across regimes. US CPI Food at home YoY is the only signal currently in the leading drawdown composite: when it hits 5y decile 10, ^GSPC 120-day forward drawdown probability is +7.5pp above base rate (Wilson CI [12.9%, 16.8%] vs base 7.3%) over 1990-2026. The other rows are watch indicators — will become integration candidates if food inflation regime-shifts higher.

Cross-asset composite · equal-weight calibrated drawdown probability per target instrument

Multi-asset meta-composite

avg decile 6.4/10

elevated across targets · avg P(drawdown) 19.9%

State distribution

3 in warning decile ≥ 9 · 5 in decile ≥ 8 · 2 in calm decile ≤ 3

out of 14 targets covered

Target	State	P(drawdown)	95% CI	vs base
S&P 500 (^GSPC) ⚠✓	20 sigs · dec 9	13.6%	[11–17%]	base 7.3%
Gold (GC=F)✓	7 sigs · dec 3	0.7%	[0–2%]	base 9.6%
US Dollar (DX-Y.NYB)✓	7 sigs · dec 3	0.2%	[0–1%]	base 4.0%
Bitcoin (BTC-USD)	6 sigs · dec 4	16.3%	[11–24%]	base 25.1%
10y T-Notes (ZN=F)	6 sigs · dec 8	0.0%	[0–1%]	base 0.8%
Nikkei 225 (^N225) ⚠○	5 sigs · dec 7	18.3%	[15–22%]	base 13.7%
KOSPI (^KS11)	4 sigs · dec 6	14.4%	[11–18%]	base 15.1%
Hang Seng (^HSI) ⚠✓	5 sigs · dec 10	48.4%	[44–53%]	base 22.9%
Taiwan (^TWII) ⚠✓	6 sigs · dec 10	36.7%	[32–42%]	base 12.3%
China Internet (KWEB) ⚠✓	5 sigs · dec 8	39.5%	[34–45%]	base 28.6%
China Tech (CQQQ) ⚠	6 sigs · dec 6	46.2%	[39–53%]	base 38.0%
HSTECH (3033.HK)	5 sigs · dec 4	33.3%	[26–42%]	base 42.3%
Stoxx Europe 600○	6 sigs · dec 6	6.4%	[4–9%]	base 11.1%
ASX 200 (^AXJO)○	6 sigs · dec 5	5.1%	[3–8%]	base 10.2%

How to read: each row is the equal-weight composite of that target's validated signals. The meta-composite header summarises across all 5 targets: average decile, count of targets in warning vs calm zones. Multi-asset agreement (many targets simultaneously warning) is qualitatively different from any single target firing alone -- it distinguishes localised equity events from broad macro stress.

Enhanced composite · base composite + persistence + momentum · lower Brier on every market, sharper D10 on trajectory-sensitive markets

Enhanced composite

Each market's enhanced composite is the equal-weight mean of:

Base composite — the existing per-market signal stack
Persistence — days at decile ≥7 over last 30 days (added everywhere except treasuries)
Momentum — 30-day decile change (added only where standalone D10 lift was ≥+3pp: BTC, HSTECH, KWEB, CQQQ, Nikkei, AXJO)

Brier score improves on every market. D10 actual rate jumps materially on the markets where persistence and momentum show standalone lift (HSTECH 94%, BTC 68%, CQQQ 69%, KWEB 69%). On markets where trajectory features are weak (KOSPI, HSI, Treasuries), enhanced and base read similarly. Δdec shows how the enhanced reading differs from the base reading today.

Market	Base composite	Enhanced composite	Δdec	95% CI (block-bootstrap)
S&P 500	dec 9 (10.8%)	dec 9 (19.9%)	=	[5–24%] ±9pp
Gold	dec 3 (6.7%)	dec 3 (2.4%)	=	[1–9%] ±4pp
US Dollar	dec 3 (3.3%)	dec 5 (1.4%)	+2	[0–1%] ±0pp
Bitcoin	dec 5 (29.4%)	dec 7 (28.9%)	+2	[2–29%] ±14pp
Nikkei 225	dec 7 (16.2%)	dec 8 (15.0%)	+1	[10–28%] ±9pp
KOSPI	dec 6 (16.3%)	dec 7 (14.4%)	+1	[5–17%] ±6pp
Hang Seng	dec 10 (27.5%)	dec 10 (43.1%)	=	[31–59%] ±14pp
Taiwan	dec 10 (19.3%)	dec 10 (30.2%)	=	[14–57%] ±21pp
China Internet	dec 8 (31.2%)	dec 9 (41.2%)	+1	[26–54%] ±14pp
China Tech	dec 6 (30.9%)	dec 9 (46.5%)	+3	[28–61%] ±17pp
HSTECH	dec 4 (40.6%)	dec 2 (30.8%)	-2	[19–48%] ±15pp
Stoxx 600	dec 6 (11.0%)	dec 7 (7.7%)	+1	[1–12%] ±6pp
ASX 200	dec 5 (9.9%)	dec 2 (4.7%)	-3	[1–9%] ±4pp

95% CI shows the block-bootstrap (60-day blocks, 2000 resamples) confidence interval on the true P(drawdown) at this market's CURRENT base-composite decile. Width colour: green <10pp tight, yellow 10-20pp moderate, orange 20-30pp wide, red >30pp very wide. Wide CIs reflect limited sample size and time-series autocorrelation -- be cautious about taking the point estimate at face value when CI ± is large.

Regime-conditional reading · today's deciles re-calibrated through the current VIX regime's table

Regime-conditional calibration

Current global VIX regime: LOW_VOL (trailing-252 VIX percentile: LOW_VOL ≤ 33, MID_VOL 33–67, HIGH_VOL > 67).

The standard composite calibration averages across all regimes. But signal effectiveness CHANGES with macro regime — warnings during LOW_VOL (calm market) tend to predict drawdowns more strongly than warnings during HIGH_VOL (already-stressed market, where the 120-day clock has often already started running). This panel re-reads today's composite deciles through the lens of the current VIX regime's own calibration table.

Compare Unconditional vs Conditional columns: when they differ materially the standard reading is biased for the current environment. A green "lower" delta means the regime view is more bullish; red "higher" means more bearish.

Market	Today	Unconditional p̂
China Tech	dec 6	46.2%
Bitcoin	dec 9	45.5%
Gold	dec 10	32.9%
KOSPI	dec 9	26.6%
Hang Seng	dec 7	25.0%
China Internet	dec 5	22.1%
HSTECH	dec 3	20.0%
Stoxx 600	dec 6	6.4%
Taiwan	dec 6	10.0%
Nikkei 225	dec 3	8.1%
10y T-Notes	dec 10	7.2%
ASX 200	dec 1	0.2%
US Dollar	dec 3	0.2%
S&P 500	dec 5	2.0%

Signal contribution · what's driving each market's reading today · per-signal lift vs base rate

Why is the composite reading what it is today?

For each market we decompose today's composite into per-signal contributions. Bearish signals are pushing the probability UP vs base rate (they're flagging risk). Bullish signals are pushing it DOWN (suggesting calm). Each pill shows signal (dec X, ±Y.Ypp) = signal name, its current decile, and its contribution in percentage points relative to base. The avg column shows the mean contribution across all signals (positive = composite is leaning bearish today, negative = leaning bullish).

Use this to sanity-check: if a market's composite is in dec 9 driven by a single signal, that's fragile. Multiple signals pulling the same direction is corroborated.

Market	Top bearish (raising prob)	Top bullish (lowering prob)	Net
S&P 500	ipo (dec 1, +8.3pp)move (dec 10, +8.0pp)tic_flows (dec 10, +4.7pp)	t10y2y (dec 2, -7.7pp)lwave (dec 6, -5.5pp)epu (dec 5, -5.1pp)	+0.65pp avg
Nikkei 225	credit (dec 10, +4.3pp)jpy_carry (dec 1, +3.7pp)usdjpy (dec 10, +3.7pp)	consumer_stress_jp (dec 5, -7.3pp)china_stress (dec 1, -5.1pp)	+0.36pp avg
KOSPI	move (dec 10, +5.2pp)credit (dec 10, +0.4pp)mortgage (dec 2, +0.2pp)	china_stress (dec 1, -3.7pp)jpy_carry (dec 1, -0.4pp)consumer_stress_kr (dec 7, -0.3pp)	+0.23pp avg
Hang Seng	dc (dec 7, +19.6pp)move (dec 10, +2.6pp)jpy_carry (dec 1, +0.9pp)	mortgage (dec 2, -1.2pp)	+4.56pp avg
Taiwan	credit (dec 10, +9.1pp)move (dec 10, +8.1pp)dc (dec 10, +4.0pp)	--	+4.68pp avg
China Internet	credit (dec 10, +10.9pp)gpr (dec 10, +7.1pp)sox_lead (dec 1, +0.2pp)	jpy_carry (dec 1, -5.0pp)move (dec 10, -0.6pp)	+2.52pp avg
China Tech	gpr (dec 10, +5.4pp)credit (dec 10, +5.3pp)lwave (dec 10, +3.7pp)	jpy_carry (dec 1, -9.1pp)move (dec 10, -3.1pp)	+0.75pp avg
HSTECH	gpr (dec 10, +13.9pp)move (dec 10, +6.4pp)mortgage (dec 2, +0.8pp)	jpy_carry (dec 1, -15.8pp)credit (dec 10, -8.2pp)	-0.58pp avg
Stoxx 600	move (dec 10, +7.5pp)vstoxx_proxy (dec 3, +3.7pp)jpy_carry (dec 1, +0.7pp)	btp_bund (dec 7, -11.6pp)mortgage (dec 2, -7.1pp)consumer_stress_eu (dec 6, -5.1pp)	-1.82pp avg
ASX 200	credit (dec 10, +3.9pp)move (dec 10, +3.6pp)	t10y2y (dec 2, -6.5pp)mortgage (dec 2, -1.6pp)jpy_carry (dec 1, -1.3pp)	-0.47pp avg

Year-by-year reliability · when the model warned at D10 in a given calendar year, did a drawdown actually follow?

Year-by-year reliability

Each cell: when the model put this market at decile 10 (top warning) during this calendar year, how often did a drawdown actually follow within 120 days?

Red (high %) = the warning was right; drawdown followed. Green (low %) = false alarm year; D10 days didn't lead to drawdowns. Blue (blank) = no D10 days that year (composite never warned strongly). Hover any cell to see the D10 sample size and the year's actual base rate.

What this exposes: the model is calibrated to AVERAGE behavior across all years. Looking at year-by-year, recall is variable: very high in real-crisis years (2008, 2011, 2015, 2020, 2022) but lower in mixed years. Calm years often show 0% (no drawdown after D10) which means the model issued benign false alarms. This is honest information about model behaviour that the headline Brier metric hides.

Cycle phase & projected next inflection · deterministic math projections from validated cycle signals · focus: ^GSPC

Cycle signal	Last extreme · phase	Position within cycle	Projected next inflection
L-Wave (monthly FFT) period: 40 months status: weak signal	Last peak Aug 2026 phase: early decline from peak	extrememid-cyclenext inflection	next inflection: May 2028 in ~578 days
Trading Cycle (weekly FFT) period: 16 weeks status: not built (cycle math only)	Last peak Aug 2026 phase: approaching trough	extrememid-cyclenext inflection	next inflection: Oct 2026 in ~9 days
Long Term (ECM 8.6yr cycle) period: 8.6 years (π×1000d main; 4.3y half-cycles) status: weak signal	Last ECM pivot May 2024 phase: mid phase	extrememid-cyclenext inflection	next inflection: Aug 2028 in ~697 days

How to read: these are deterministic math projections, not probability forecasts. The L-Wave and Long Term signals have measured but weak edge in walk-forward backtests (see Track record); the Trading Cycle signal hasn't been validated. The position bar shows where we are between the last extreme and the next projected inflection. Phase names describe the section of the cycle (e.g. 'mid decline' = halfway from peak to next trough).

Sector breadth & rotation · capital concentration / late-cycle leadership context

3-month leaders: XLE +15.9% · XLV +6.5% · XLK +4.9%

3-month laggards: XLI -7.6% · XLRE -7.9% · XLU -14.7%

Equal-weight underperforming cap-weight by 4.8pp over 3 months — narrow leadership / mega-cap concentration.

Sector 3m return dispersion (1σ): 8.1pp — high rotation regime

SPDR sector ETFs

Sector	1m	3m	6m
XLE Energy	-0.9%	+15.9%	+0.2%
XLV Health Care	+0.1%	+6.5%	+19.1%
XLK Technology	+4.8%	+4.9%	+52.6%
XLC Communication	-1.6%	+3.1%	+3.0%
XLF Financials	-6.7%	+0.9%	+12.1%
XLB Materials	-7.0%	-2.3%	+0.8%
XLP Cons. Staples	-3.7%	-2.5%	+0.5%
XLY Cons. Discretionary	-7.0%	-6.9%	+3.2%
XLI Industrials	-4.7%	-7.6%	+7.8%
XLRE Real Estate	-7.0%	-7.9%	+2.8%
XLU Utilities	-8.1%	-14.7%	-14.5%

Sub-sector & factor proxies

ETF	1m	3m	6m
IGV Software / SaaS	-3.7%	+17.3%	+35.8%
RSP S&P equal-weight	-5.0%	-1.6%	+11.5%
SMH Semiconductors	+8.5%	-5.1%	+65.5%
IWM Russell 2000	-5.3%	-6.3%	+16.9%

How to read: sector returns are price returns (no dividends added back) for SPDR ETFs. Left table sorts by 3m return descending; right column shows sub-sector and factor proxies (semis vs SaaS captures the AI-infrastructure cycle vs application-software divergence). SMH leading and IGV lagging by a wide margin is the late-1990s-style capex-cycle pattern: capital concentrates in physical infrastructure while application-layer software gets disrupted. Wide dispersion + narrow leadership (RSP < ^GSPC) historically precedes broader market turbulence.

Track record · composite probability over time vs realised 8%+ drawdowns · focus: ^GSPC

shaded bands = days where the realised 120d forward return was below −8% (i.e. days that led into a real 8%+ drawdown over the next 120 trading days). │ vertical dotted lines = peak dates of named crashes. Reading the chart: probability spike before a shaded band = true warning; spike with no shaded band that follows = false alarm; flat-and-low going INTO a shaded band = a miss.

Track record (Jan 2010 — Jan 2026): composite hit decile 9-10 (high warning) on 806 days; of those, 201 (25%) preceded a real 8%+ drawdown over the next 120 days · base rate 7.2%

Strategy backtest · signal-driven equity/cash rotation vs Buy & Hold · 0bp = signal quality only; 5bp = realistic ETF friction

Signal-quality vs cost-blame decomposition (Aggressive 100/50/25 sizing scenario):
  Gap vs B&H at 0bp (signal quality only): −3.77pp/yr
  Gap vs B&H at 5bp (with realistic costs): −4.88pp/yr
  Of which costs explain: −1.12pp/yr (the rest is foregone upside when the model is mistakenly elevated)

At 0bp per rebalance (signal quality only)

Scenario	CAGR	MaxDD	Sharpe	Time in	Rebals
Aggressive 100/50/25	+8.83%	-33.3%	0.64	75%	833
Moderate 100/75/50	+10.28%	-33.6%	0.70	85%	833
Cut-on-alert 100/100/50	+10.75%	-33.9%	0.71	90%	419
Buy & Hold	+12.59%	-33.9%	0.78	100%	0

At 5bp per rebalance (realistic ETF execution)

Scenario	CAGR	MaxDD	Sharpe	Time in	Rebals
Aggressive 100/50/25	+7.71%	-33.3%	0.57	75%	833
Moderate 100/75/50	+9.53%	-33.6%	0.66	85%	833
Cut-on-alert 100/100/50	+10.04%	-33.9%	0.67	90%	419
Buy & Hold	+12.59%	-33.9%	0.78	100%	0

Reading the table: green CAGR / Sharpe = beats Buy & Hold; red = worse. The 0bp column isolates signal quality from execution friction. Even at 0bp, the signal-driven scenarios underperform B&H because the composite cuts equity during sustained bull markets where the eventual drawdown didn't happen — i.e. it warns more often than it should. Realistic 5bp execution adds friction but isn't the primary constraint. The honest answer to 'does the signal pay?': not as a mechanical equity/cash rebalance rule, but the calibrated probabilities may still be useful for discretionary sizing or hedge timing — strategies this backtest does not model.

Reference · cross-market state + recent events

Recent significant events ( |score| ≥ 2.0 ) · ↑ positive score = risk-on convergence (directional marker) ↓ negative score = panic classifier output

Date	Score	Class filled = panic classifier · outline = directional	Contributing instruments
2026-05-29	+3.29 UP	risk-on	3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CBOE_VIX9D,C
2026-05-28	+3.29 UP	risk-on	3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CBOE_VIX9D,C
2026-05-26	+3.30 UP	risk-on	3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CBOE_VIX9D,C
2026-05-25	+3.66 UP	risk-on	3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CBOE_VIX9D,C
2026-05-22	+3.66 UP	risk-on	3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CBOE_VIX9D,C
2026-05-20	+3.66 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-19	+3.85 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-18	+3.85 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-15	+4.04 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-14	+4.04 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-12	+3.90 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-11	+3.90 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-08	+3.92 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-07	+3.92 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-05-01	+3.76 UP	risk-on	000001.SS,3033.HK,BITO,BTC-USD,BZ=F,CBOE_VIX3M,CBOE_VIX6M,CB
2026-02-20	+2.36 UP	risk-on	BDRY,BITO,BTC-USD,BZ=F,CC=F,CCJ,CEG,CL=F,ETH-USD,GC=F,GEV,GF
2026-02-17	+3.34 UP	risk-on	BDRY,BITO,BTC-USD,BZ=F,CC=F,CCJ,CEG,CL=F,ETH-USD,GC=F,GEV,GF
2026-02-13	+3.69 UP	risk-on	3033.HK,BDRY,BITO,BTC-USD,BZ=F,CC=F,CCJ,CEG,CL=F,ETH-USD,GC=
2026-02-12	+3.69 UP	risk-on	3033.HK,BDRY,BITO,BTC-USD,BZ=F,CC=F,CCJ,CEG,CL=F,ETH-USD,GC=
2026-02-10	+3.49 UP	risk-on	3033.HK,BDRY,BITO,BTC-USD,BZ=F,CC=F,CCJ,CEG,ETH-USD,GC=F,GEV

Timing Arrays — Dashboard ▸ HOME ▸ MARKET INTELLIGENCE

Equity risk dial

In plain English · what the model is saying today, written so anyone can follow it