Post-Trade Wrap — Sentiment + kNN Catalyst Basket (Dec 8–12, 2025)
This report closes the loop on the Monday open entry note (equal-weight across 10 liquid names) and marks the exit at the Friday close (Dec 12, 2025). The objective here is not storytelling — it is a clean attribution: what we earned from market beta, what we earned from selection, and what the model got wrong (and why).
Line-Item Performance
Entry is the 2025-12-08 open. Exit is the 2025-12-12 close. MAE/MFE are measured using the week’s daily low/high relative to entry (a conservative proxy for stop/target tagging).
| Ticker | Entry | Exit | Return | Contrib. | MAE | MFE | β (SPY) | RVOL | Sent | kNN | kNN hit | −8% tagged |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SPOT | 584.98 | 617.84 | +5.62% | +56.2 bp | -4.39% | +5.62% | 1.71 | 1.18× | -0.75σ | -0.17% | 44% | No |
| ABNB | 132.76 | 137.22 | +3.36% | +33.6 bp | -1.81% | +5.16% | 1.35 | 0.82× | +0.19σ | +0.77% | 60% | No |
| QCOM | 151.91 | 155.00 | +2.03% | +20.3 bp | -2.55% | +4.32% | 1.34 | 0.96× | -0.08σ | +0.68% | 60% | No |
| PLTR | 71.70 | 72.47 | +1.07% | +10.7 bp | -2.40% | +8.51% | 2.74 | 0.89× | +0.75σ | +0.33% | 56% | No |
| SHOP | 118.66 | 119.31 | +0.55% | +5.5 bp | -5.04% | +3.25% | 1.78 | 0.68× | +0.24σ | +2.26% | 72% | No |
| COIN | 339.40 | 332.50 | -2.03% | -20.3 bp | -4.18% | +4.20% | 4.08 | 0.78× | +0.28σ | +1.46% | 60% | No |
| DBX | 29.68 | 28.60 | -3.64% | -36.4 bp | -8.71% | +3.10% | 0.69 | 1.07× | -0.18σ | -0.57% | 44% | Yes |
| AMD | 127.08 | 122.26 | -3.79% | -37.9 bp | -6.03% | +2.00% | 2.51 | 0.90× | +0.09σ | +0.21% | 52% | No |
| UBER | 71.03 | 66.60 | -6.24% | -62.4 bp | -8.87% | +1.75% | 1.28 | 0.73× | +0.29σ | +0.02% | 44% | Yes |
| HOOD | 41.39 | 36.88 | -10.89% | -108.9 bp | -12.10% | +2.85% | 4.61 | 0.62× | +0.75σ | +1.13% | 64% | Yes |
Stop diagnostics: 3 names tagged −8% intraweek lows — UBER, DBX, HOOD. In this specific tape, two of those (UBER, DBX) mean-reverted enough that a hard intraday stop would likely have reduced final P&L.
Signal Scoreboard: Sentiment vs kNN
For this week, the sentiment layer was directionally correct on 60% of names (6/10), while the kNN layer was correct on 40% of names (4/10). The net P&L was dominated by factor exposure (high beta / growth tilt), not by a single catastrophic idiosyncratic event.
| Ticker | Realized | Sent (z) | kNN pred | Sent ok | kNN ok |
|---|---|---|---|---|---|
| SPOT | +5.62% | -0.75σ | -0.17% | × | × |
| ABNB | +3.36% | +0.19σ | +0.77% | ✓ | ✓ |
| QCOM | +2.03% | -0.08σ | +0.68% | × | ✓ |
| PLTR | +1.07% | +0.75σ | +0.33% | ✓ | ✓ |
| SHOP | +0.55% | +0.24σ | +2.26% | ✓ | ✓ |
| COIN | -2.03% | +0.28σ | +1.46% | × | × |
| DBX | -3.64% | -0.18σ | -0.57% | ✓ | ✓ |
| AMD | -3.79% | +0.09σ | +0.21% | × | × |
| UBER | -6.24% | +0.29σ | +0.02% | × | × |
| HOOD | -10.89% | +0.75σ | +1.13% | × | × |
Tape Context: What likely drove the week
The basket leaned heavily into high-beta, growth and “attention” categories (AI, software, and crypto sensitivity). In a week where large-cap tech/AI sentiment was mixed and risk appetite rotated in and out, the basket’s beta profile did most of the explaining. News flow around AI infrastructure valuations and crypto’s correlation with risk assets remained key narrative drivers into the week’s close.
- AI / semis narrative volatility: pockets of concern around “AI trade” positioning and valuation punctured momentum into Friday. (Reuters market coverage)
- Crypto beta behaved as “risk proxy”: crypto-sensitive equities continued to trade with the broader risk complex rather than as isolated idiosyncratic stories. (Reuters crypto correlation coverage)
- Index divergence mattered: SPY down modestly while QQQ was meaningfully weaker, mechanically penalising our high-beta tilt.
Note: macro headlines are used only as context. The performance math is computed from daily OHLCV.
Algorithm Patch Notes
The goal of this patch is simple: reduce unpriced beta, enforce attention confirmation, and align the kNN target to the actual strategy horizon.
Changes in v6.2
- Hold-period alignment: kNN labels updated to match Mon open → Fri close returns (not generic fwd-N close-to-close).
- Regime gate: neighbour set is conditioned on index state (SPY trend + QQQ impulse). If QQQ is below its 20-day mean and breadth is weak, the model shifts toward lower-beta historical analogues.
- Beta budgeting: weights are scaled to target a portfolio beta (default 1.0–1.2). High-beta names remain eligible, but they no longer dominate the risk budget by default.
- Volume-surge enforcement: mandatory RVOL filter is raised. Names with RVOL below threshold are either excluded or down-weighted materially.
- Week diagnostic: only 2/10 names printed RVOL ≥1.0× on the entry day (median volume vs 30-day median). That is structurally inconsistent with an “attention week” and is now explicitly penalised.
- Stop logic refit: replace a universal −8% intraday stop with ATR-aware + time-aware protection (reduce “stop-and-reverse” damage in mean-reverting tapes).