Quant Phase 2: Alpha + Portfolio + Web

t-718·WorkTask·
·
·
Created1 month ago·Updated1 week ago·pipeline runs →

Description

Edit

Phase 2 plan:

1. Quant web dashboard at /fund/quant — show signals, eval, regime 2. Composite alpha: blend 252d momentum + 100d mean-reversion + macro overlay 3. Portfolio optimization: Kelly sizing with correlation matrix 4. Risk dashboard: drawdown monitor, regime alerts

Timeline (15)

🔄[human]Open → InProgress1 month ago
🔄[human]InProgress → Review1 month ago
💬[human]1 month ago

Quant web dashboard deployed to production at /fund/quant. Shows: macro regime banner, 4 signal tables, FRED summary, universe overview with prices, htmx eval button. All rendering with real market data.

💬[human]1 month ago

Phase 2 alpha + portfolio complete. Alpha.hs: IC-weighted composite alpha + shrinkage returns + Cholesky matrix inversion (10 tests). Portfolio.hs: N-asset correlated Kelly + constrained sizing + trade generation (9 tests). CLI: alpha + portfolio commands. Dashboard: alpha scores table + Kelly portfolio table. Top alpha: GLD 1.01, XLK 0.56. 13 positions at quarter-Kelly. Deployed.

💬[human]1 month ago

Phase 2 complete. New modules: Alpha.hs (505 lines, IC-weighted composite alpha + Cholesky matrix inversion, 10 tests), Portfolio.hs (378 lines, N-asset correlated Kelly + trade generation, 9 tests). Updated: Quant.hs (alpha + portfolio CLI commands), Web/Quant.hs (alpha table, Kelly portfolio table, backtest comparison). Dashboard shows: macro regime, 4 signal types, composite alpha, Kelly positions, portfolio backtest (momentum 35.4% vs equal-weight 33.4%), and walk-forward eval. Total: 10 quant modules, 3900+ lines, 89 tests. All deployed to production.

💬[human]1 month ago

Phase 3 initial modules built. Edgar.hs: EDGAR submissions API client with sector→CIK mapping for 55 companies across 11 sectors, cluster detection, JSONL cache (5 tests). Agent.hs: Op-based signal agents — insiderSignalAgent and narrativeSignalAgent using free monad for LLM inference, robust JSON parsing for signal extraction, prompt templates (9 tests). These are the first non-chat consumers of the Op free monad. Next: wire agents into the Quant CLI for live execution, then eval against deterministic baseline.

💬[human]1 month ago

Phase 3 complete. Built and tested: 1. Edgar.hs: EDGAR Form 4 client (5 tests) 2. Agent.hs: Op-based signal agents + lightweight interpreter (9 tests) 3. CLI commands: insider, narrative, agent-eval 4. End-to-end evaluation results:

  • Momentum-63d (deterministic): IC=+0.22 ✓
  • Narrative agent (LLM): IC=-0.03 ✗
  • Insider agent (mock data): IC=-0.95 (noise, N=4)
  • Composite 50/20/30 blend: IC=+0.19 (worse than deterministic alone)

5. VERDICT: Agent signals do not yet add alpha. Infrastructure is in place but agents need real-time EDGAR data and temporal evaluation to demonstrate IC > 0.02.

💬[human]1 week ago

Ava verified: implementation present in live tree (file/feature confirmed via inspection). Moving to Verified.

🔄[human]Review → Verified1 week ago