Overall stress is elevated, but one credit gauge is flashing a warning.
A summary of the nine gauges below, not a probability of a crash or a verdict on AI valuations. What this score means
Stress band: Elevated · Formal alert: None · Shock flag: Off
Download the data (JSON) · Refreshed Nov 14, 5:09 AM UTC.
Elevated index. Stressed credit.
Critical on this index’s 0–100 scale, while overall stress is elevated. This signals pressure in riskier corporate borrowing, not a 93% probability of a credit event.
The index rose to 36 as Semiconductors stress rose 8 points, but CCC credit spreads remain at 93/100. The broad picture is calm; the weakest borrowers are not.
What would change our mind
Thesis to test: The AI build-out is breaking
Four signs would show the money behind AI data centres running short. Each is marked from KKReckons stories, and changes only when the evidence does.
- Funding on harder termsPartly met since 21 Oct
An AI-infrastructure borrower can't refinance, or pays sharply more to borrow.
Yields on CoreWeave's 2031 notes passed 16% and its five-year default insurance hit a record 1,194.5 basis points on 15 October. No one has forecast a default; it reports results in November.
CoreWeave's bond yield and CDS: an early warning for AI debt? · 21 OctCoreWeave's debt pile tests whether the AI build-out can be financed · 25 Oct
- A hyperscaler cuts capexNot met since 26 Oct
Alphabet, Microsoft, Amazon or Meta lowers its AI spending plans.
None has lowered its spending plans. Alphabet, Microsoft and Amazon report results this week.
- Chip and GPU rental prices fallNot met since 22 Oct
The price of renting AI chips, or of the memory they use, turns down.
No fall reported. CME plans futures on H100 and B200 rental prices, which would give this a public price to watch.
Want to short the AI boom? Futures on chip rental prices are coming · 22 Oct
- Banks can't offload AI loansNot met since 12 Oct
Lenders are left holding data-centre debt they meant to sell on.
Banks are still trying: a group including JPMorgan and MUFG has spent over six months placing $38bn of Oracle-leased project debt, and some lenders will sell at a discount.
Banks try to offload the AI data-centre debt they just underwrote · 12 Oct
Where is the pressure?
- Nasdaq-100 18%28+5.0
- Semiconductors 18%69+12.4
- AI & tech ETF 9%45+4.1
Bar = gauge score (0–100) · +pts = contribution to the index
- Data-center ETF 6%46+2.8
- BBB credit spreads 6%0+0.0
- CCC credit spreads 6%93+5.6
Bar = gauge score (0–100) · +pts = contribution to the index
- Listed BDCs 14%46+6.5
- Leveraged loans 13%0+0.0
- High-yield spreads 10%0+0.0
Bar = gauge score (0–100) · +pts = contribution to the index
Each component weight is the sum of its gauges’ weights, and its score is their weighted average. Contributions may not sum exactly because of rounding; all calculations use unrounded values. The index combines the components: 45% × 47.6 + 18% × 46.5 + 37% × 17.6 = 36.3, rounded to 36.
One set of bands labels the index and every gauge: Normal 0–24 (green) · Elevated 25–49 (amber) · Severe 50–74 (orange) · Critical 75–100 (red). Arrows show direction: ▲ stress rising · ▼ stress easing.
What changed since 12 Nov 2026
- ▲ Rose 8 Semiconductors: now 69/100
- ▲ Rose 6 Nasdaq-100: now 28/100
- ▲ Rose 4 AI & tech ETF: now 45/100
The nine gauges
Sorted by stress score (0–100). Tap a gauge for its latest value, contribution and why it matters, then open the details for thresholds, history and source.
CCC credit spreadsBroad-market proxyRisky corporate borrowing costs1,314 bp · contributes 5.6 pts93/100● Unchanged
Critical · AI financing
- Latest
- 1,314 bp
- Change from 52-week low
- +476 bp
- Contribution to index
- 5.6 points
- Data through
- 12 Nov 2026
Why it matters: Borrowing spreads on the riskiest rated corporate debt. It covers all CCC-rated borrowers rather than tech alone, but is a useful early-warning signal for vulnerable, highly leveraged companies.
Thresholds, history and source
- Calm → trigger
- 900 → 1500 bp
- Widening calm → trigger
- +150 → +500 bp above 52-week low
- Level test · widening test
- 69 · 93the higher one is the score
- Weight · maximum contribution
- 6% · 6 points
Contributes 5.6 today; at most 6 points, so it can add no more than 0.4.
Source: FRED BAMLH0A3HYC, observation 12 Nov 2026. Score = the higher of (spread − calm) ÷ (trigger − calm) and (rise above 52-week low − widening calm) ÷ (widening trigger − widening calm), limited to 0–100.
SemiconductorsAI-specificChip stocks10,146.32 · contributes 12.4 pts69/100▲ Stress rising
Severe · AI equities
- Latest
- 10,146.32
- From 2-year high
- -30.7%
- Contribution to index
- 12.4 points
- Data through
- 13 Nov 2026
Why it matters: Semiconductors are a major AI-infrastructure exposure and can be sensitive to changes in technology-sector risk appetite.
Thresholds, history and source
- Calm → trigger
- -10% → -40% from 2-year high
- Weight · maximum contribution
- 18% · 18 points
Contributes 12.4 today; at most 18 points, so it can add no more than 5.6.
Source: Yahoo Finance ^SOX, observation 13 Nov 2026. Score = (drawdown − calm) ÷ (trigger − calm), limited to 0–100.
Listed BDCsBroad-market proxyListed BDCs · private-credit proxy12.50 · contributes 6.5 pts46/100▼ Stress easing
Elevated · Credit markets
- Latest
- 12.50
- From 2-year high
- -14.3%
- Contribution to index
- 6.5 points
- Data through
- 13 Nov 2026
Why it matters: Listed business development companies (BDCs), a market proxy for private credit, a major lender to growth and tech firms. Their share prices reflect credit concerns and broader stock-market moves.
Thresholds, history and source
- Calm → trigger
- -5% → -25% from 2-year high
- Weight · maximum contribution
- 14% · 14 points
Contributes 6.5 today; at most 14 points, so it can add no more than 7.5.
Source: Yahoo Finance BIZD, observation 13 Nov 2026. Score = (drawdown − calm) ÷ (trigger − calm), limited to 0–100.
Data-center ETFAI-specificData-center infrastructure stocks25.37 · contributes 2.8 pts46/100▲ Stress rising
Elevated · AI financing
- Latest
- 25.37
- From 2-year high
- -21.6%
- Contribution to index
- 2.8 points
- Data through
- 13 Nov 2026
Why it matters: Data-center and digital-infrastructure companies: the physical AI build-out that much of the new debt is financing. Equity prices also reflect broader stock-market moves.
Thresholds, history and source
- Calm → trigger
- -10% → -35% from 2-year high
- Weight · maximum contribution
- 6% · 6 points
Contributes 2.8 today; at most 6 points, so it can add no more than 3.2.
Source: Yahoo Finance DTCR, observation 13 Nov 2026. Score = (drawdown − calm) ÷ (trigger − calm), limited to 0–100.
AI & tech ETFAI-specificAI & big-data stocks58.72 · contributes 4.1 pts45/100▲ Stress rising
Elevated · AI equities
- Latest
- 58.72
- From 2-year high
- -16.3%
- Contribution to index
- 4.1 points
- Data through
- 13 Nov 2026
Why it matters: A basket of AI and big-data companies. It moves closely with the Nasdaq-100, so it carries a smaller weight.
Thresholds, history and source
- Calm → trigger
- -5% → -30% from 2-year high
- Weight · maximum contribution
- 9% · 9 points
Contributes 4.1 today; at most 9 points, so it can add no more than 4.9.
Source: Yahoo Finance AIQ, observation 13 Nov 2026. Score = (drawdown − calm) ÷ (trigger − calm), limited to 0–100.
Nasdaq-100Broad-market proxyLarge-cap tech benchmark27,943.74 · contributes 5.0 pts28/100▲ Stress rising
Elevated · AI equities
- Latest
- 27,943.74
- From 2-year high
- -10.5%
- Contribution to index
- 5.0 points
- Data through
- 13 Nov 2026
Why it matters: The broad benchmark for large US tech and AI-exposed companies. A 25% fall from its 2-year high is a common bear-market yardstick.
Thresholds, history and source
- Calm → trigger
- -5% → -25% from 2-year high
- Weight · maximum contribution
- 18% · 18 points
This gauge would contribute 15.0 points at about -21.6% from its 2-year high (about 24,474.35), up from 5.0 today. With every other gauge unchanged, the index would rise by those same 10 points.
Source: Yahoo Finance ^NDX, observation 13 Nov 2026. Score = (drawdown − calm) ÷ (trigger − calm), limited to 0–100.
BBB credit spreadsBroad-market proxyInvestment-grade borrowing costs111 bp · contributes 0.0 pts0/100● Unchanged
Normal · AI financing
- Latest
- 111 bp
- Change from 52-week low
- +19 bp
- Contribution to index
- 0.0 points
- Data through
- 12 Nov 2026
Why it matters: Borrowing spreads on the lowest tier of investment-grade corporate debt, where much large-company and AI-related issuance prices.
Thresholds, history and source
- Calm → trigger
- 140 → 250 bp
- Widening calm → trigger
- +30 → +100 bp above 52-week low
- Level test · widening test
- 0 · 0the higher one is the score
- Weight · maximum contribution
- 6% · 6 points
Contributes 0.0 today; at most 6 points, so it can add no more than 6.0.
Source: FRED BAMLC0A4CBBB, observation 12 Nov 2026. Score = the higher of (spread − calm) ÷ (trigger − calm) and (rise above 52-week low − widening calm) ÷ (widening trigger − widening calm), limited to 0–100.
Leveraged loansBroad-market proxyLeveraged-loan prices20.52 · contributes 0.0 pts0/100● Unchanged
Normal · Credit markets
- Latest
- 20.52
- From 2-year high
- -0.2%
- Contribution to index
- 0.0 points
- Data through
- 13 Nov 2026
Why it matters: Leveraged loans: floating-rate funding for buyouts and below-investment-grade companies.
Thresholds, history and source
- Calm → trigger
- -2% → -8% from 2-year high
- Weight · maximum contribution
- 13% · 13 points
This gauge would contribute 10.0 points at about -6.6% from its 2-year high (about 19.20), up from 0.0 today. With every other gauge unchanged, the index would rise by those same 10 points.
Source: Yahoo Finance BKLN, observation 13 Nov 2026. Score = (drawdown − calm) ÷ (trigger − calm), limited to 0–100.
High-yield spreadsBroad-market proxyJunk-bond borrowing costs320 bp · contributes 0.0 pts0/100● Unchanged
Normal · Credit markets
- Latest
- 320 bp
- Change from 52-week low
- +60 bp
- Contribution to index
- 0.0 points
- Data through
- 12 Nov 2026
Why it matters: The broad junk-bond spread, the market-wide gauge of credit stress.
Thresholds, history and source
- Calm → trigger
- 400 → 600 bp
- Widening calm → trigger
- +75 → +250 bp above 52-week low
- Level test · widening test
- 0 · 0the higher one is the score
- Weight · maximum contribution
- 10% · 10 points
This gauge would contribute 10.0 points if the spread reached 510 bp, up from 0.0 today. With every other gauge unchanged, the index would rise by those same 10 points.
Source: FRED BAMLH0A0HYM2, observation 12 Nov 2026. Score = the higher of (spread − calm) ÷ (trigger − calm) and (rise above 52-week low − widening calm) ÷ (widening trigger − widening calm), limited to 0–100.
- Today
- 36
- 52-week high
- 475 Nov 2026
- 52-week low
- 35 Dec 2025
- 52-week average
- 13
Today’s score is higher than 244 of the 250 daily readings over the past 52 weeks (and equal to 2 others): about the 98th percentile. Each trading day counts once using that day’s latest data; ties are not counted as lower. Range over the year: 3 to 47.
Weekly highs and formal alerts
- L1 Stress WatchNot triggered
- L2 Financing StrainNot triggered
- L3 Broad CorrectionNot triggered
- L4 Systemic DislocationNot triggered
Each alert needs the index and specific gauges to stay high for 5 trading days in a row. The four alerts are independent tripwires, not a staircase: Systemic Dislocation can fire before Broad Correction, and Financing Strain can be skipped. Stress Watch and Financing Strain are the early warnings; the other two confirm damage already under way.
What this score means
An experimental monitoring dashboard: one 0–100 composite of nine AI-related stock, financing and credit-market gauges, under the published method below.
- Not a probability. 70/100 is not a 70% chance of a crash, and a low score doesn’t mean markets are safe.
- Not a valuation call. It reads market stress, not whether AI companies are fairly priced or will earn back their spending.
- Partly proxies. Six of the nine gauges are broad-market measures (marked on each gauge), so general credit or market stress can move the score without an AI-specific cause.
- Alerts can be late or wrong. They need 5 trading days to confirm, may come after a sell-off has started, and can fire without a later decline. The backtest below lists misses and false alarms.
Use it alongside established volatility and credit indicators and your own research. Not financial advice. Download today’s data (JSON)
Methodology
How the index works
Every weekday after the US close, the index reads 9 market gauges in three groups: AI equities (45%), AI financing (18%) and credit markets (37%).
Each gauge gets a stress score from 0 (calm) to 100 (its trigger level). For stocks and funds it is how far they have fallen from their 2-year high. For credit spreads (the extra interest riskier borrowers pay) it is how high they are, or how fast they have risen, whichever is greater.
The index is the weighted average of those scores, from 0 to 100. Formal alerts need the index and specific gauges to stay high for 5 trading days in a row, so a one-day move never counts.
Today’s example: CCC credit spreads stand at 1,314 bp, 476 bp above their 52-week low of 838. That fast rise gives a widening score of 93, above the level score of 69. With a 6% weight it adds 5.6 points to today’s index of 36.
What this index captures, and what it doesn’t
It captures stress in market prices tied to the AI cycle: AI-related stocks, proxies for how the AI build-out is financed, and broader credit conditions.
It does not measure AI business fundamentals, valuations, returns on AI spending or future profits. A low score is not proof that the AI boom is healthy.
The nine gauges are not nine independent signals. The Nasdaq-100, semiconductors, the AI ETF and the data-center ETF often move together, as do the credit spreads, so one market move can lift several gauges at once. Weights are set with this in mind, but the score is a combination of related proxies, not a count of separate risks.
Proxies: listed BDCs stand in for private credit and the data-center ETF for the AI build-out. Their share prices also reflect broader stock-market moves. The CCC spread covers all CCC-rated borrowers, not only technology companies.
Thresholds are fixed, judgement-based settings informed by past market episodes, not statistically optimised or universal market levels. Because they were shaped by the same history the backtest replays, the backtest shows fit, not forecasting power.
Why each gauge is weighted this way
- Nasdaq-100 · 18% · AI equities · Broad-market proxyThe broad benchmark for large US tech and AI-exposed companies. A 25% fall from its 2-year high is a common bear-market yardstick.This gauge would contribute 15.0 points at about -21.6% from its 2-year high (about 24,474.35), up from 5.0 today. With every other gauge unchanged, the index would rise by those same 10 points.
- Semiconductors · 18% · AI equities · AI-specificSemiconductors are a major AI-infrastructure exposure and can be sensitive to changes in technology-sector risk appetite.Contributes 12.4 today; at most 18 points, so it can add no more than 5.6.
- AI & tech ETF · 9% · AI equities · AI-specificA basket of AI and big-data companies. It moves closely with the Nasdaq-100, so it carries a smaller weight.Contributes 4.1 today; at most 9 points, so it can add no more than 4.9.
- Data-center ETF · 6% · AI financing · AI-specificData-center and digital-infrastructure companies: the physical AI build-out that much of the new debt is financing. Equity prices also reflect broader stock-market moves.Contributes 2.8 today; at most 6 points, so it can add no more than 3.2.
- BBB credit spreads · 6% · AI financing · Broad-market proxyBorrowing spreads on the lowest tier of investment-grade corporate debt, where much large-company and AI-related issuance prices.Contributes 0.0 today; at most 6 points, so it can add no more than 6.0.
- CCC credit spreads · 6% · AI financing · Broad-market proxyBorrowing spreads on the riskiest rated corporate debt. It covers all CCC-rated borrowers rather than tech alone, but is a useful early-warning signal for vulnerable, highly leveraged companies.Contributes 5.6 today; at most 6 points, so it can add no more than 0.4.
- Listed BDCs · 14% · Credit markets · Broad-market proxyListed business development companies (BDCs), a market proxy for private credit, a major lender to growth and tech firms. Their share prices reflect credit concerns and broader stock-market moves.Contributes 6.5 today; at most 14 points, so it can add no more than 7.5.
- Leveraged loans · 13% · Credit markets · Broad-market proxyLeveraged loans: floating-rate funding for buyouts and below-investment-grade companies.This gauge would contribute 10.0 points at about -6.6% from its 2-year high (about 19.20), up from 0.0 today. With every other gauge unchanged, the index would rise by those same 10 points.
- High-yield spreads · 10% · Credit markets · Broad-market proxyThe broad junk-bond spread, the market-wide gauge of credit stress.This gauge would contribute 10.0 points if the spread reached 510 bp, up from 0.0 today. With every other gauge unchanged, the index would rise by those same 10 points.
Exact rules, thresholds and data sources
Prices (Nasdaq-100, semiconductors, AI, data-center, BDC and loan ETFs): drawdown = dividend-adjusted close ÷ highest close of the last 2 years − 1; score = (drawdown − calm) ÷ (trigger − calm), limited to 0–100. The 2-year high reflects the AI-cycle peak without old pre-AI highs.
Credit spreads (BBB, CCC, high yield, in bp): score = the higher of the level test, (spread − calm) ÷ (trigger − calm), and the widening test, (spread − 52-week low − widening calm) ÷ (widening trigger − widening calm), each limited to 0–100. The 52-week window catches fast blow-outs.
Index = Σ(weight × score) ÷ Σ weights, rounded. A gauge’s contribution = its weight share × its score. Blank days carry the last value; a gauge not updated for more than 3 trading days is flagged stale and keeps its last value in the index. History needs 2 years of data before the first score.
Formal alerts (5 straight trading days): L1 Stress Watch: index ≥ 50 · L2 Financing Strain: index ≥ 60 and (BDCs ≥ 80 with loans ≥ 50, or CCC ≥ 80) · L3 Broad Correction: index ≥ 70, semiconductors ≥ 83, AI ETF ≥ 80 · L4 Systemic Dislocation: index ≥ 75, Nasdaq-100 at 100, and high-yield ≥ 500 bp or BDCs at 100. Each alert is an independent tripwire.
Exit rule (recovery mode): while the Nasdaq-100 is 20% or more above its lowest close since its 2-year high, alerts whose conditions still hold show as paused, so they don’t stay lit deep into a recovery. First-confirmed dates never move. Today the Nasdaq-100 is 4.9% above that low. Shock flag (not an alert): index ≥ 60 on a day when no alert has confirmed.
Threshold sensitivity: the range shown beside the index is today’s reading recomputed with every threshold moved 20% tighter or looser.
Sources (retrieved Nov 14, 5:09 AM UTC): Yahoo Finance dividend-adjusted daily closes, and ICE BofA option-adjusted spreads via FRED, which publish a day after the market date. Yahoo Finance ^NDX (13 Nov 2026) · Yahoo Finance ^SOX (13 Nov 2026) · Yahoo Finance AIQ (13 Nov 2026) · Yahoo Finance DTCR (13 Nov 2026) · FRED BAMLC0A4CBBB (12 Nov 2026) · FRED BAMLH0A3HYC (12 Nov 2026) · Yahoo Finance BIZD (13 Nov 2026) · Yahoo Finance BKLN (13 Nov 2026) · FRED BAMLH0A0HYM2 (12 Nov 2026) ·
Backtest & validation
How to read this backtest — honestly
- The thresholds were tuned on this same history. They were shaped by these very episodes, not chosen blind and then validated, so this replay shows the rules fit past crises, not that they would have predicted them. The real test is what comes next: the out-of-sample record starts 16 Oct 2026 and so far holds 0 alerts and 0 shock flags.
- It missed the Q4 2018 sell-off. The Nasdaq-100 fell 23.0% and the index hit 60, yet no alert confirmed: the index never held 50+ for 5 straight trading days. This system detects credit-driven stress, not fast equity drawdowns. The new shock flag (index ≥ 60 before any alert has confirmed; not itself an alert) would have fired on 24 Dec 2018.
- It missed the 2025 tariff shock. The Nasdaq-100 fell 22.9% and the index hit 72, yet no alert confirmed: the index never held 50+ for 5 straight trading days. This system detects credit-driven stress, not fast equity drawdowns. The new shock flag (index ≥ 60 before any alert has confirmed; not itself an alert) would have fired on 4 Apr 2025.
- The loudest alerts are late. The top two alerts confirm damage already under way; the quieter Stress Watch is the early signal. Timing against the Nasdaq-100 peak: Dot-com bust: Stress Watch 14 days after it, first Broad Correction or Systemic Dislocation 6 months after it; Global financial crisis: Stress Watch 19 days after it, first Broad Correction or Systemic Dislocation 11 months after it; 2022 rate-shock bear market: Stress Watch 6 months after it, first Broad Correction or Systemic Dislocation 10 months after it.
- The alerts don’t escalate. Each is an independent tripwire, so don’t read them as a staircase. 2022 rate-shock bear market: L3 Broad Correction fired without L2 Financing Strain.
- False alarms happen. Merging back-to-back alerts gives 12 alert episodes, each judged by the same rule: 8 gave early warning, 1 only confirmed a bear market already under way, and 3 led nowhere (20 May 2010, 8 Aug 2011, 13 Jan 2016). Several headline episodes run mostly on stand-in data (share of the weight: Dot-com bust 50%, Global financial crisis 64%, 2011 debt-ceiling & euro scare 51%, 2015–16 oil & credit scare 37%), so treat their precision as illustrative, not proven.
- Thresholds move the story. Tightening every threshold 20% turns the COVID crash into a Systemic Dislocation episode and moves today’s reading from 36 to 49; loosening them 20% takes today to 27. Small choices, big narrative swings.
- Before 14 Nov 2023, the credit-spread gauges are one signal. BBB credit spreads, CCC credit spreads, High-yield spreads all use the same stand-in, Moody’s Baa spread, times a fixed ratio (median ratio, 17 Oct 2023 – 16 Oct 2026). So 22% of the weight counts one investment-grade measure several times, and the CCC and high-yield conditions in the alerts run on it too. The ratios come from a calm credit period; in crises junk spreads widen proportionally more than Baa, so older spread stress is probably understated.
- Stand-in drift check. Re-running everything as if FRED’s spread window began 30 days later moves no alert dates, so the stand-in joins are stable.
- Alerts now stand down in recoveries. Since model v1.4, alerts pause in recovery mode: while the Nasdaq-100 is 20% or more above its lowest close since its 2-year high. Before this rule, the top alert stayed on for months after the 2002 and 2009 bottoms, because the 2-year high it measures from was set long before. First-confirmed dates are unaffected.
Read with care: historical results before late 2023 are indicative. Several gauges use substitute series because today’s instruments or comparable data did not exist, so historical alert dates and severity may differ from results with fully comparable data. This replay shows when formal alerts would have fired; it is not a test of predictive accuracy.
The replay starts on 2 Jan 1997 but scoring begins on 4 Jan 1999: each price gauge needs 2 years of history before its 2-year high exists. It runs day by day to 13 Nov 2026 (7,010 scored trading days) with no look-ahead.
| Selected sell-off | Nasdaq-100 fall | Peak index | First alert | Highest alert | Share of weight on stand-ins |
|---|---|---|---|---|---|
| Dot-com bust | -83% | 95 | 10 Apr 200014 days after the Nasdaq peak | L4 Systemic Dislocation5 Dec 2000 | 50%Mostly stand-in |
| Global financial crisis | -54% | 100 | 19 Nov 200719 days after the Nasdaq peak | L4 Systemic Dislocation26 Sep 2008 | 64%Mostly stand-in |
| 2011 debt-ceiling & euro scare | -16% | 73 | 8 Aug 201117 days after the Nasdaq peak | L2 Financing Strain18 Aug 2011 | 51%Mostly stand-in |
| 2015–16 oil & credit scare | -16% | 71 | 13 Jan 201671 days after the Nasdaq peak | L2 Financing Strain5 Feb 2016 | 37%Mostly stand-in |
| Q4 2018 sell-off | -23% | 60 | None | None | 28%Partly stand-in |
| COVID crash | -28% | 97 | 11 Mar 202021 days after the Nasdaq peak | L2 Financing Strain11 Mar 2020 | 28%Partly stand-in |
| 2022 rate-shock bear market | -36% | 79 | 9 May 2022171 days after the Nasdaq peak | L3 Broad Correction26 Sep 2022 | 22%Partly stand-in |
| 2025 tariff shock | -23% | 72 | None | None | 0%Real data |
Alerts outside the selected sell-off windows (since 1999): Stress Watch 3 · Financing Strain 1 · Broad Correction 0 · Systemic Dislocation 0. An alert outside these windows is not automatically a false alarm; some match real stress periods the list leaves out, such as the 2010 euro crisis and the March 2023 bank failures.
Every alert, judged the same way. Days with any alert active, merged when less than 20 trading days apart. Early warning: the Nasdaq-100 fell a further 10% or more within about 6 months. Late confirmation: no further fall, but it began with the Nasdaq-100 already 20% or more below its 2-year high. False alarm: neither. The same test applies inside and outside the selected sell-off windows.
| Alert period | Highest alert | Nasdaq-100 vs 2-year high at start | Worst over next 6 months | Verdict |
|---|---|---|---|---|
| 10 Apr 2000 → 1 Jun 2000Dot-com bust | L2 Financing Strain | -15% | -24% | Early warning |
| 18 Sep 2000 → 10 Apr 2001Dot-com bust | L4 Systemic Dislocation | -24% | -55% | Early warning |
| 10 Jul 2001 → 30 Oct 2001Dot-com bust | L4 Systemic Dislocation | -65% | -31% | Early warning |
| 21 Feb 2002 → 21 Feb 2002Dot-com bust | L4 Systemic Dislocation | -71% | -36% | Early warning |
| 9 Apr 2002 → 29 Oct 2002Dot-com bust | L4 Systemic Dislocation | -67% | -40% | Early warning |
| 7 Feb 2003 → 11 Mar 2003outside the selected windows | L4 Systemic Dislocation | -60% | -1% | Late confirmation |
| 19 Nov 2007 → 31 Mar 2009Global financial crisis | L4 Systemic Dislocation | -10% | -17% | Early warning |
| 20 May 2010 → 9 Jul 2010outside the selected windows | L2 Financing Strain | -12% | -4% | False alarm |
| 8 Aug 2011 → 11 Oct 20112011 debt-ceiling & euro scare | L2 Financing Strain | -15% | -1% | False alarm |
| 13 Jan 2016 → 29 Feb 20162015–16 oil & credit scare | L2 Financing Strain | -11% | -6% | False alarm |
| 11 Mar 2020 → 13 Apr 2020COVID crash | L2 Financing Strain | -18% | -13% | Early warning |
| 9 May 2022 → 10 Jan 20232022 rate-shock bear market | L3 Broad Correction | -26% | -12% | Early warning |
Out-of-sample record (alert rules frozen from 16 Oct 2026): no alerts or shock flags yet. Everything else on this page is in-sample.
Missed declines: of 7 periods when the Nasdaq-100 fell 20% or more below its 2-year high, 5 had a formal alert by the time they ended. No alert fired in: 21 Dec 2018 (-23%); 4 Apr 2025 (-23%).
Threshold sensitivity: the same replay with every calm and trigger threshold moved 20% tighter or looser.
| Thresholds | Index today | Systemic Dislocation fired in | Alerts outside windows (L1 · L2 · L3 · L4) |
|---|---|---|---|
| Thresholds 20% tighter | 49 | Dot-com bust, Global financial crisis, COVID crash | 6 · 4 · 0 · 0 |
| Current thresholds | 36 | Dot-com bust, Global financial crisis | 3 · 1 · 0 · 0 |
| Thresholds 20% looser | 27 | Dot-com bust, Global financial crisis | 1 · 0 · 0 · 0 |
Weighting check: a lower-redundancy weighting (less weight on gauges that move together) caught the same Systemic Dislocation episodes but produced 1 more alert outside the selected windows, so the current weights were kept.
Stand-ins: QQQ, then the Nasdaq-100 itself, for the AI ETF before 2018; the IYR real-estate ETF for the data-center ETF before 2020; Ares Capital for BDCs before 2013; a senior-loan fund (VVR) for leveraged loans before 2011. FRED publishes only about three years of the ICE credit spreads, so older spread history uses Moody’s Baa spread (BAA10Y) times a fixed ratio to each series, frozen so the replay doesn’t drift as FRED’s window rolls.
Model version and changes
Model v1.4 · methodology last changed 17 Oct 2026. Every run recalculates the full history under the current version. Thresholds were shaped on the same history the backtest replays, so the backtest is in-sample; only alerts after the out-of-sample date are a true test.
- v1.4Exit rule: alerts pause in recovery mode, when the Nasdaq-100 is 20% or more above its lowest close since its 2-year high (start dates, including the implosion date, never move). New shock flag for an index of 60+ before any alert has confirmed. Index shown with its range under thresholds 20% tighter/looser. Backtest now judges every alert by one rule, is labelled in-sample, and keeps an out-of-sample record from 16 Oct 2026. Alert start triggers, thresholds and weights unchanged.
- v1.3Wording and disclosure updates only; no change to scores.
- v1.2Financing Strain alert can also trigger on CCC spreads; stale-data flags.
- v1.1Oracle and CoreWeave replaced by the data-center ETF, BBB and CCC spreads; dividend-adjusted prices; spread widening test; 2-year highs.
- v1.0Launch: 8 gauges, 1-year highs.
Cite this reading: KKReckons AI Market Stress Index, 36/100 (Elevated), data through 13 Nov 2026. https://www.kkreckons.com/ai-stress-index · Method · Editorial standards
Embed this index
Free to use with the link back intact. The reading refreshes hourly after each US close.
Stress bands describe today’s reading. Formal alerts need extra conditions held for 5 straight trading days, so the index can be Elevated (25+) before the first alert, Stress Watch, triggers at 50. Gauge scores are normalised 0–100 readings, not probabilities. A Normal reading is not a prediction that markets are safe or will rise.
Model v1.4. Built and curated by KK Reddy. An indicator of market conditions, not a forecast. For information only; not financial advice. Market data as of 13 Nov 2026 · refreshed Nov 14, 5:09 AM UTC.