Pull-quote: “Three uncorrelated edges with deterministic risk are worth more than one elegant edge with stochastic risk.”
Why this matters
A quantitative research stack with a single edge is a strategy — and strategies decay. A stack with three uncorrelated edges is closer to a portfolio. When one edge fades, the others continue to produce. The combined Sharpe and the combined drawdown profile are both better than any single edge — if the edges are actually uncorrelated.
The phrase “uncorrelated” is the load-bearing word. A typical research shop runs three strategies that all bet on momentum-after-volatility-spikes and calls them three edges. They are not. They will all suffer in the same regime.
Three actually-uncorrelated edges
Sigma Axion’s research framework is structured around three:
- Maker-rebate market making — capturing the spread on venues that pay rebates for providing liquidity. The edge is small per-trade, large in aggregate, and correlated with venue microstructure rather than market direction.
- Cross-venue arbitrage — exploiting the fact that the same instrument trades simultaneously on multiple venues, sometimes including retail-accessible venues, with intermittent dislocations. The edge is correlated with infrastructure quality (latency, fill ratios) rather than market regime.
- Structural mispricing between adjacent markets — for example, prediction markets vs. the underlying spot market for a related event. The edge depends on the slow flow of arbitrage capital between asset classes, and is correlated with regulatory and structural friction rather than direction.
These three are uncorrelated because they depend on different kinds of friction — venue, infrastructure, and structural. A regime that destroys one does not destroy the others.
The deterministic risk engine
The other half of the design is the risk engine. Deterministic means: given the same portfolio state and the same instrument data, the engine returns the same gates and the same sizes. No stochastic resampling. No model jitter. No surprise.
Risk engines that are stochastic at the gate level cannot be reasoned about by the trader. Deterministic risk engines can. The difference shows up in the worst-case scenario, when it matters most.
Evidence-based LLM routing
Where the LLMs come in: research synthesis, earnings note summarization, news sentiment, and methodology critique. Different models are better at different tasks. Sigma Axion routes based on measured benchmark performance, not vibes. We benchmark the models on each task class periodically and update the routing table.
What this is not
This is not a black-box trading system. It is a research framework. The edges are documented, the backtests are walk-forward, the risk gates are deterministic, the LLM routing is benchmarked. Everything that should be auditable is auditable.
Closing
Three uncorrelated edges with a deterministic risk engine is worth more than one elegant edge with a stochastic risk engine. Edges decay; uncorrelated portfolios of edges decay slower. Sigma Axion is built on that thesis.
