TL;DR:
On a simple Mon 17 → Fri 21 close-to-close basis, the
equal-weight 10-name sentiment + kNN basket returned approximately
−6.4%, against the S&P 500 at about
−1.0% over the same window. :contentReference[oaicite:0]{index=0}
Losses clustered in MSTR, COIN, PLTR and
HOOD. Crypto-beta and high-beta AI skew hurt; RIVN was essentially flat.
No intraday stop logic or slippage is included in these figures.
Measurement convention:
Mark entry at the regular-session close on Mon 17 Nov and exit at the
close on Fri 21 Nov for each name, using exchange-reported daily history
(Yahoo Finance / StockAnalysis / Investing.com / FT). :contentReference[oaicite:1]{index=1}
This is an approximation to the intended “post-open settle” entry and ignores intraweek
stops, gaps and execution details.
Headline result:
Basket: −6.4% (equal-weight, 10 × 1 unit notionals).
S&P 500: −1.0% over the same dates.
Realised dispersion was negative; there was no compensating right tail.
Context:
Into 21 Nov, “AI bubble” fatigue weighed on semis (ARM, NVDA, SMCI), while
crypto-beta (COIN, MSTR) and single-name tech momentum bled alongside a soft tape. :contentReference[oaicite:2]{index=2}
Realised performance vs SPX
Using the close on 17 Nov 2025 as the entry mark and the close on 21 Nov 2025 as the exit, the basket and benchmark behaved as follows:
| Series | Mon 17 Close | Fri 21 Close | Return (Mon→Fri) | Notes |
|---|---|---|---|---|
| Basket (EW) | — | — | −6.4% | Simple average of the 10 name-level returns (1/10 each). |
| S&P 500 | 6,672.41 | 6,602.99 | −1.0% | Cash index, close-to-close via FT historical series. |
In other words, relative performance was roughly −5.4 percentage points versus the broad market over this particular week, with the basket picking the wrong side of high-beta dispersion.
Per-name realisation (Mon 17 → Fri 21 close)
The table below shows the closing prices and close-to-close returns used in the calculation. All prices are in USD.
| Ticker | Company | Mon 17 Close | Fri 21 Close | Return | Comment |
|---|---|---|---|---|---|
| NVDA | NVIDIA | 186.60 | 178.88 | −4.1% | Moderate drawdown as AI enthusiasm cooled; beta worked against the basket. :contentReference[oaicite:4]{index=4} |
| ARM | Arm Holdings | 140.26 | 131.57 | −6.2% | Underperformed as renewed “AI bubble” nerves hit the post-IPO ARM complex. |
| SMCI | Super Micro Computer | 34.10 | 32.19 | −5.6% | High-beta AI hardware lagged; no upside dispersion to offset downside. :contentReference[oaicite:6]{index=6} |
| PLTR | Palantir | 171.25 | 154.85 | −9.6% | Sharp give-back after prior strength; sentiment momentum reversed intraweek. :contentReference[oaicite:7]{index=7} |
| TSLA | Tesla | 408.92 | 391.09 | −4.4% | Classic high-beta move: market down, TSLA down more, but not catastrophically. :contentReference[oaicite:8]{index=8} |
| RIVN | Rivian | 14.87 | 14.86 | −0.1% | Essentially flat; idiosyncratic EV headlines dominated over the week’s beta. |
| COIN | Coinbase | 263.95 | 240.41 | −8.9% | Crypto-beta worked against the basket; right tail never materialised this week. :contentReference[oaicite:10]{index=10} |
| MSTR | MicroStrategy | 195.42 | 172.74 | −11.6% | Largest detractor; leveraged BTC proxy amplified crypto wobble. :contentReference[oaicite:11]{index=11} |
| HOOD | Robinhood | 115.97 | 107.30 | −7.5% | Retail-beta faded; flows pulled back after earlier strength. :contentReference[oaicite:12]{index=12} |
| SNAP | Snap | 8.18 | 7.69 | −6.0% | Comm-services dispersion skewed negative; no catalyst to re-ignite the story. :contentReference[oaicite:13]{index=13} |
Where the model went wrong
The design of the basket was explicitly to lean into attention-driven dispersion via the Sentiment Momentum Score and kNN state guards, assuming that a cluster of high-attention names would, on average, throw off enough positive outliers to pay for the inevitable left tails. This particular week delivered the opposite:
- No meaningful positive tail: RIVN was roughly flat; none of the ten names printed a +10–15% three-to-five day burst. The best outcome was essentially “no damage”.
- Concentrated downside in correlated themes: ARM, NVDA, SMCI and PLTR all suffered as the market questioned AI valuations again, while COIN and MSTR absorbed crypto-beta weakness in tandem. :contentReference[oaicite:14]{index=14}
- Market backdrop: SPX was down only about 1% on the week, so there was no full-blown risk event to justify the basket’s drawdown. The portfolio effectively bought a local maximum in crowding.
Execution and measurement caveats
The original playbook specified entries shortly after the open on Monday, using a first-10m VWAP anchor and hard −8% stops, with optional +12% OCO orders and a time stop at Friday’s close. In this wrap:
- We approximate entry and exit strictly by official daily closes on 17 Nov and 21 Nov 2025. :contentReference[oaicite:15]{index=15}
- We assume no stopouts and no intraday re-entries. Any strategy that mechanically stopped some of the worst movers (for example MSTR or COIN on sharp down-days) would have realised a different, likely less negative, P&L path.
- Transaction costs, slippage and borrow fees are ignored. For liquid names in this size, they should be second-order relative to the −6.4% mark, but they are not literally zero.
What this says about the framework
One week does not invalidate a framework, but it does provide useful stress points:
-
Sentiment is not direction: The inputs (tone, social velocity, breadth,
options intent) can be extreme even as positioning is already crowded. The
SMSgate may need explicit crowding penalties or a cap on simultaneous exposure to the same narrative complex. - kNN state guard needs regime awareness: Conditioning on recent analogues of “benign volatility” around AI and crypto may fail once macro or valuation narratives shift. A regime label (for example, “bubble-fatigue” vs “acceleration”) should be part of the state vector.
-
Portfolio construction: The basket respected sector caps in design,
but practically carried a heavy AI/crypto tilt. A more explicit
max theme exposureconstraint would have pushed some risk into less correlated names.
Do Your Own Research.
These figures are mechanical close-to-close calculations based on public historical data
for 17–21 Nov 2025 and do not incorporate actual fills, stops, or trade
management. Use them as a diagnostic of the framework, not as audited track record.