Quantarded Weekly Signals #004 — Week 3, 2026
Market note: U.S. markets are closed tomorrow for MLK Jr. Day (January 19).
Welcome to the third issue of Quantarded for 2026.
Week 3 continues the pattern that emerged earlier in the year. Participation remains elevated, but consensus is selective. A small number of names attract sustained agreement, while the rest of the universe shows the familiar mix of attention, disagreement, and decay.
For context, this week Quantarded processed 440 House trade disclosures filed during the week, 232,658 Reddit comments analyzed and 28,554 stock ticker mentions detected and classified. As always, what matters is not volume alone, but how agreement evolves as volume grows.
How to read this newsletter
This issue looks at three layers:
1. Reddit sentiment, capturing where retail attention and conviction are clustering.
2. House trade disclosures, to detect position building rather than one-off trades.
3. Performance tracking, showing what actually happened after publication.
None of these is a trading system on its own. The value is in where they reinforce each other, and where they do not.
Reddit picks — strong activity, selective consensus
To assign a BUY or SELL, Quantarded requires a minimum sentiment imbalance. Volume alone is not enough. That said, unusually large discussion is still worth calling out, even when conviction is not absolute.
New this week: instead of labeling picks as high/medium/low confidence, each pick gets a share percentage. This mimics a simple portfolio allocation where weights are proportional to the absolute score (higher absolute score, higher share).
Here are the most relevant Reddit signals for Week 3.
$ASTS: BUY, 30% share
$ASTS is the top-ranked signal this week. It is the largest allocation in the Week 3 basket, reflecting the strongest combined evidence of sustained attention and directional agreement in the data.
$META: SELL, 20% share
$META is the strongest SELL signal in this week’s list. It earns the second-largest share because the underlying signal is both strong and directionally consistent, but in the opposite direction to the rest of the basket.
$MU: BUY, 20% share
$MU ranks in the core of the basket. The signal is clearly positive, with enough consistency to justify a meaningful allocation, though it is not as dominant as the top two names.
$RKLB: BUY, 15% share
$RKLB remains a solid BUY candidate this week, very close to $MU in allocation. It reflects sustained interest with a directionally consistent tone, placing it among the higher-weight picks.
$NFLX: BUY, 15% share
$NFLX rounds out the Week 3 set. It qualifies as a BUY, but at a smaller share due to a lower absolute score relative to the names above.
This Week 3 basket is top-heavy: one dominant BUY ($ASTS), a meaningful SELL ($META), and two mid-weight BUYs ($MU, $RKLB), with a smaller fifth name ($NFLX).
House trades — activity without conviction
House trade disclosures this week tell a quieter story than the headline numbers suggest.
Yes, there are a few larger trades, but most trades are small position changes. The typical disclosed trade size is around $1,001, and only a handful of entries reach the level that Quantarded treats as more meaningful (around $100k).
The other important nuance is concentration: the only $100k+ cluster this week comes from a single filer on a single trade date, which makes it notable, but not a clean multi-person consolidation signal. A few examples help illustrate the shape of the week.
- Markwayne Mullin (R, Senate) disclosed a basket of BUYs just above the $100k mark on 2025-12-29, filed 2026-01-16, including $AAPL, $NVDA, $AMZN, $GOOGL, and *$MSFT\\* (each $100,001). This is the clearest "large trade" footprint in the dataset, but it is still one person, one day, one style of exposure.
- Tommy Tuberville (R, Senate) disclosed a SELL in $AAPL for $50,001 (traded 2025-12-17, filed 2026-01-15). This is a meaningful individual disclosure, but it is not repeated enough to read as a broader signal.
- Gilbert Cisneros (D, House) disclosed a BUY in $CPB for $50,001 (traded 2025-12-30, filed 2026-01-12). Again, noteworthy in isolation, but not reinforced by repeated accumulation elsewhere.
If you zoom out to recurrence across the full extended set, a few names show up frequently by count, including $MSFT, $AAPL, $META, and $AMZN. But in most cases, those repeats are driven by small-dollar trades rather than scaling position size.
The important takeaway is not that "nothing happened", but that nothing consolidated across multiple people at meaningful size.
This is exactly the kind of week where House data is useful by what it does not show. There is activity, and even a visible large-trade cluster, but no clear multi-person build-up of exposure that would justify a strong signal.
Performance review — what actually happened
With the full week completed, we can evaluate how last week’s picks behaved after publication.
Last week’s results
At the portfolio level, gains again arrived in bursts, with strong contributors offset by short-term reversals in lower-confidence names.
End of 26W3 return: +8.16%
YTD (2026) return: +13.89%
Cumulative return since inception (weighted): +25.46%
From the initial $10,000 baseline, the portfolio now stands at $12,278. The overall pattern remains unchanged: uneven paths, clustered gains, and results driven by periods of concentrated consensus rather than continuous exposure.
Knowing the algorithm - more stable week-long picks
This week we shipped a small but meaningful change to the WSB scoring logic. The first version worked, but the list could feel jumpy when you checked it repeatedly throughout the day. In the current, improved revision we kept the core model intact (weekly structure, direction, imbalance, volume hygiene), and changed only two stability levers: recency weighting and confidence determinism.
Mixed recency decay (fast + slow)
In the first version, each event used a single exponential decay with a half-life. That is responsive, but it can overweight short bursts and cause leaders to reshuffle too easily mid-week.
In the new version, we still use exponential decay, but we mix two time-scales:
- fast half-life:
12h(reacts to fresh discussion) - slow half-life:
72h(adds inertia so a weekly leader does not get replaced by a brief spike) - mixing factor:
0.75(bias toward fast, but not exclusively)
Concretely, for an event with age in hours age_h, recency weight becomes:
recency_w = α * exp(-ln(2) * age_h / 12) + (1 - α) * exp(-ln(2) * age_h / 72)where α = 0.75 by default.
This is a standard idea in time-series smoothing: you retain short-term responsiveness while reducing sensitivity to transient noise by adding a slower component. Good references:
The practical effect for the newsletter is simple: the Sunday snapshot reads more like "how the week went" and less like "what happened in the last few hours".
What did not change
To be explicit, this new adjustment does not change the fundamentals of the model:
- still ISO-week scoped
- still sentiment-derived direction (BUY vs SELL)
- still imbalance-gated picks
- still volume hygiene and attention-share logic
The goal of these changes is not to "improve performance". It is to make the signal more honest and easier to interpret week-long, without rewarding brief bursts that do not persist.
Disclaimer
This newsletter is NOT FINANCIAL ADVICE.
All content is provided for informational and educational purposes only. Markets involve risk, including loss of principal. Past performance does not guarantee future results. Always do your own research and consider consulting a licensed financial professional.
Links
Website: https://quantarded.com
Subscribe: https://quantarded.kit.com/products/monthly-subscription?step=checkout