Research
We target markets where structural and operational constraints limit institutional participation, resulting in persistent fragmentation and mispricing.
Research Philosophy
Structural focus
We prioritize strategies that remain tradable under liquidity constraints, costs, and regime shifts. Core sources of mispricing include fragmentation, contract and market design, cross-market constraints, and resolution ambiguity.
End-to-end validation
Strategies are validated as executed: event-driven replay, explicit cost and slippage assumptions, and controlled stress scenarios across market conditions and failure modes.
Production feedback
Live performance is continuously measured from signal stability through execution quality to portfolio impact. We operate taker and maker strategies across multiple time horizons with always-closed execution. Sizing is adjusted and strategies are retired when degradation is observed.
Execution Infrastructure
Execution runs on infrastructure engineered per venue, ensuring consistently low-latency access to target venues.
Infrastructure highlights:
- Per-venue deployment across dedicated and cloud-based infrastructure
- Latency-optimized routing to venue APIs and real-time feeds
- Co-located ingestion for critical data paths
- Deterministic execution via hardware isolation and core pinning
- Continuous telemetry with automated circuit breakers
Core Research Pillars
Execution and Routing
- Multi-venue aggregation with consensus-based signals across prediction and derivatives venues, co-located data relays for low latency.
- Liquidity-aware order placement with funding rate and liquidation-level awareness embedded in routing decisions.
- Leverage-adjusted sizing and fill-risk modeling that accounts for margin requirements and forced liquidation thresholds on every venue.
Regime Modeling and Signal Conditioning
- Market state and transition modeling using price, liquidity, and volatility features across venue types.
- Funding regime detection: identifying zero/low/high funding environments and their effect on position carry.
- Liquidation pressure and leverage cycle modeling: tracking aggregate positioning from exchange data to anticipate cascade risk.
- Regime-conditioned behavior: regime-aware decay, thresholds, and sizing overlays applied consistently across strategy families.
Simulation, Replay, and Validation
- Deterministic tick-level replay with virtual clock: every strategy runs against historical sequences the same way it runs live.
- Funding event simulation: replay includes funding rate ticks, margin adjustments, and cross-margin reconciliation.
- CPCV fold validation: held-out periods, Pareto front selection across score, Sharpe, drawdown, fill rate, and margin utilization.
- Conservative fill modeling: explicit fees, FOK miss rates, partial-fill behavior, liquidation gap modeling, and staleness haircuts applied consistently.
- Attribution: shortfall decomposed into drift, latency, impact, fee, and margin cost for every outcome.
AI Agents and Research System
Research is supported by autonomous agents operating on code, configuration, and runtime state.
Outputs are validated, conflict-checked, and operator-reviewed.
The live trading path is strictly isolated.
Autonomy is gated by explicit permissions.
Systems are continuously monitored, reconfigured, and retired on degradation.
Interested in research collaboration?
Let's Talk