The Axiomatic Trader: Defining the Hard Limits of Quantitative Investment Systems

Most retail systems treat alpha as something to extract. A new arXiv paper published this week treats it as a bounded resource — and the constraint list is brutally honest. The framework, titled "The Axiomatic Trader: Latent Regularity, Information Budgets, and the Canonical Form of a Quantitative Investment System," argues that latent market states, recurrence bounds, representation stability, signal ceilings, and regime dependence are not optimization knobs but architectural ceilings. For systematic traders, that reorders where to spend engineering hours and where to stop.

What the framework actually claims

The authors do not sell a strategy. They sketch the hard limits any valid quantitative system must satisfy before a line of signal code gets written. Latent states carry a finite information budget — only so much predictive content bleeds through into observable price action. Recurrence bounds cap how stable that leakage is over time. Representation stability constrains how your feature set behaves across distribution shift. Signal ceilings mark the point at which stacked features stop adding independent edge and start encoding the same noisy factor twice. Regime dependence is then the multiplier on top of all of it.

Translated into trader language: there is a hard ceiling on how much edge a given instrument class can leak, and your model's representation must remain stable across regimes or you are not measuring alpha. You are measuring noise that happened to align in-sample, within a single regime, at one point in the volatility surface.

Why this matters for working setups

I have backtested enough curve-fit systems to know the smell. Strip the regime tag and the numbers collapse. The Axiomatic Trader formalism is essentially an attempt to codify that experience into axioms. If recurrence bounds are tight, no amount of feature engineering rescues you past the signal ceiling. If representation stability fails across regimes, every "confluence" indicator you stack is a redundant encoding of the same factor, and your Sharpe is just compounded overfitting.

The practical checklist collapses to the paper's axioms, reduced to engineer-level tasks:

• Measure latent state leakage before adding features. If the state explains little variance in walk-forward, stop adding indicators and stop calling it confluence.
• Audit representation stability. The feature set must produce similar conditional outputs across regimes, or the backtest is a mirage dressed up as a report.
• Treat signal ceiling as a hard cap. Anything beyond it is curvature the model is misreading as signal.
• Stress-test every component under regime-shifted backtests. The drawdown profile there, not the headline equity curve, is the real report card.

What to watch

The paper sits inside an arXiv cluster that has been quietly pushing formal constraints over predictive theater for the better part of a year. If the recurrence bound theorem replicates on instruments outside US equities — crypto perps and short-dated options are the cleanest probes for regime dependence — the next wave of vendor tools will rebrand themselves "axiomatic" with little of the underlying rigor. Pressure-test the claims yourself before the marketing cycle catches up. Treat the framework as a methodological lens for filtering signal ideas, not as a magic bullet that produces them.