← All projects

QRS research · Trading agents · Financial NLP

AlphaAllocator

Bounded news overlays for risk-aware asset allocation

Don’t ask one language model for the trade. Ask small models narrow questions, and let a risk model allocate.

A common way to build a news-reading trading agent is to hand the news to a large language model and let it produce the trade, an overlay or an event signal directly. AlphaAllocator splits that job into the two questions news can answer: which asset is this about, and is it good or bad, each with its own small model. A classical risk model decides the allocation, and the news may only nudge it within a fixed budget.

Details

Published

Participants: Yuhe Sui

NLPCC 2026: 1st of 10 in Track 1 · ICAIF 2026: coming soon

How AlphaAllocator reads the newsThree news items published before the cutoff are linked to the assets they are about, marked good or bad, and pooled into one score per asset. A late item is ignored. The centred scores tilt an illustrative risk-model portfolio, and the whole tilt is capped at one percent of the portfolio. Illustrative example.News visible todayAsset · news scoreWeightPolicy easingpublished 09:58+Earnings warningpublished 11:20−Safe-haven demandpublished 14:05+Late itempublished 15:40too lateCSI 300Govt bondsGoldMetalsNew energy1 · Relevance: which asset?2 · Tone: good or bad?3 · Risk model: how much?News tilt budget‖Δw‖₁ ≤ 1% of the portfoliotilt drawn 10× larger
One day’s news, step by step: link each item to the assets it is about, mark it good or bad, pool it into one score per asset, then let the risk model allocate while the news nudges within a capped budget. Each news item published before the cutoff is linked to the assets it is about and marked good or bad; the pooled scores nudge a risk-model portfolio within a capped budget. Illustrative example; the agent reads up to 20 items a day.

AlphaAllocator at a glance

The idea

For news to help a portfolio, an agent has to know two things: which asset an item is about, and whether it is good or bad news. We give each question its own small model and multiply the answers:

news scorej=relevancej×tonej

The scores, weighted for recency and certainty, give one score zj per asset. They tilt a risk-parity portfolio by at most η = 1.0% of it in total:

d=η(z−z¯),Δw=dmin(1, η‖d‖1)
The budget the news may move the portfolio withinEvery tilt the news may apply to the portfolio, for three assets. Centring makes the tilt sum to zero and the cap holds its total size at or below 1.0% of the portfolio, so the set the news can reach is exactly the teal hexagon around today’s risk-model portfolio; each corner moves 0.5% out of one asset and 0.5% into another. A quiet day stays inside it. A loud day is stopped at the edge: the arrow cannot leave the hexagon however large the news score grows, where an uncapped overlay (dashed) would carry on. The inset places the same budget against every portfolio of the three assets, at true scale: the whole of it is smaller than the dot that marks today’s portfolio. Illustrative assets and scores.Every portfolio of three assetsEnlarged: what the news may domore equitiesmore bondsmore gold0.5% out of gold,0.5% into equitiestoday’s portfolio,set by pricesa quiet day: inside the budgeta loud day: stopped at the cap, 1.0% of the portfoliowithout the cap: an overlay with no bound
The budget the news may move the portfolio withinEvery tilt the news may apply to the portfolio, for three assets. Centring makes the tilt sum to zero and the cap holds its total size at or below 1.0% of the portfolio, so the set the news can reach is exactly the teal hexagon around today’s risk-model portfolio; each corner moves 0.5% out of one asset and 0.5% into another. A quiet day stays inside it. A loud day is stopped at the edge: the arrow cannot leave the hexagon however large the news score grows, where an uncapped overlay (dashed) would carry on. The inset places the same budget against every portfolio of the three assets, at true scale: the whole of it is smaller than the dot that marks today’s portfolio. Illustrative assets and scores.All portfoliosEnlargedequitiesbondsgold0.5% out of gold,0.5% into equitiestoday’s portfolioset by pricesa quiet day: inside the budgeta loud day: at the 1.0% capwithout the cap: no bound
Every allocation the news can reach. Centred, and scaled down whenever it would exceed the cap, so three assets let the whole reachable region be drawn in the plane: a hexagon, each corner moving 0.5% out of one asset and 0.5% into another. The inset sets it against every portfolio of the three, at true scale: all of it is smaller than the dot marking today’s portfolio. Clipping and rescaling afterwards cannot enlarge it. Drag the news, turn it up, or raise every score equally. Illustrative assets and scores: the geometry the equation fixes, not a measurement.

Experiments

NLPCC 2026 · Chinese funds tested

Track 1SharpeAnn. returnMax DD
1st1.114413.52%−6.23%
2025 replay Sharpe, with news2.270same agent, news zeroed2.198

Track 1: 11 macro assets, 10 teams, organizer-run on a hidden 2026 period (next-best Sharpe 0.7792). Track 2 (sectors): 6th of 13.

ICAIF 2026 · US equities Coming soon

The same design, adapted to 30 US stocks and 7 decisions a day, is in the ACM ICAIF 2026 Trading Agent Competition.

Results coming soon.

Every team’s official SharpeOfficial Sharpe ratio of every team on the hidden 2026 evaluation, in rank order, AlphaAllocator highlighted.Track 1 · 10 teams · ours 1st0−112Rank 1: Sharpe 1.1144 (AlphaAllocator)11.1144Rank 2: Sharpe 0.77922Rank 3: Sharpe 0.73223Rank 4: Sharpe 0.49074Rank 5: Sharpe 0.39985Rank 6: Sharpe 0.39686Rank 7: Sharpe 0.29967Rank 8: Sharpe 0.20378Rank 9: Sharpe -0.37669Rank 10: Sharpe -0.508710official rankTrack 2 · 13 teams · ours 6th0−112Rank 1: Sharpe 1.97351Rank 2: Sharpe 1.70772Rank 3: Sharpe 1.27313Rank 4: Sharpe 0.84954Rank 5: Sharpe 0.79465Rank 6: Sharpe 0.7566 (AlphaAllocator)60.7566Rank 7: Sharpe 0.71447Rank 8: Sharpe 0.70358Rank 9: Sharpe 0.66289Rank 10: Sharpe 0.615510Rank 11: Sharpe 0.210311Rank 12: Sharpe -0.160512Rank 13: Sharpe -1.296513official rank
Every team’s official SharpeOfficial Sharpe ratio of every team on the hidden 2026 evaluation, in rank order, AlphaAllocator highlighted.Track 1 · 10 teams · ours 1st0−112Rank 1: Sharpe 1.1144 (AlphaAllocator)11.1144Rank 2: Sharpe 0.77922Rank 3: Sharpe 0.73223Rank 4: Sharpe 0.49074Rank 5: Sharpe 0.39985Rank 6: Sharpe 0.39686Rank 7: Sharpe 0.29967Rank 8: Sharpe 0.20378Rank 9: Sharpe -0.37669Rank 10: Sharpe -0.508710official rankTrack 2 · 13 teams · ours 6th0−112Rank 1: Sharpe 1.97351Rank 2: Sharpe 1.70772Rank 3: Sharpe 1.27313Rank 4: Sharpe 0.84954Rank 5: Sharpe 0.79465Rank 6: Sharpe 0.7566 (AlphaAllocator)60.7566Rank 7: Sharpe 0.71447Rank 8: Sharpe 0.70358Rank 9: Sharpe 0.66289Rank 10: Sharpe 0.615510Rank 11: Sharpe 0.210311Rank 12: Sharpe -0.160512Rank 13: Sharpe -1.296513official rank
Every team’s official Sharpe, NLPCC 2026. The organizers’ hidden 2026 evaluation, in rank order, ours in teal. Track 1: 1st of 10. Track 2: 6th of 13. More results →

Limitations and what didn’t work →

Papers

  • NTU-QRS AlphaAllocator: Bounded Transformer News Overlays for Chinese Macro-Asset Allocation

    NLPCC 2026 Shared Task 4 · poster and oral · oral at Evaluation Workshop I, 5 November 2026 [OpenReview] [Talk slides, PDF]

  • ACM ICAIF 2026 Trading Agent Competition

    Write-up coming soon

Recognition

  • NLPCC 2026 Shared Task 4 · Track 1 · 1st
  • NLPCC 2026 Shared Task 4 · Track 2 · 6th
QR code for qrsntu.org/projects/alphaallocator/
Scan to open this page