Stop Funding Backtest Illusions—Build Quantitative Strategies That Survive Live Markets
The most flattering thing a backtest can do is look inevitable. Across eight years of daily data, the Sharpe ratio climbs—yet charts never reveal whether a signal reflects an economic mechanism or one lucky winner from a 360-combination search. When live orders meet lagging timestamps, tight borrow, and vanishing midpoint fills, wishful arithmetic becomes misallocated capital.
Quantitative Trading Beyond Overfitting by Mila Bautista builds a disciplined path from hypothesis to evidence to live execution, treating statistical skepticism as an operational system rather than a pose.
Reconstruct the Search Before Capital Meets the Market
Separate repeatable signal from historical noise before risking capital:
— Search-Space & Refinement Audits: Inventory parameter grids, universe filters, and metric switches, separating ex ante reasoning from post-hoc stories while preserving dead ends and unselected configurations.
— Boundary-Locked Validation Matrix: Score holdouts, rolling walk-forward windows, and purged or embargoed designs across five criteria—balancing regime coverage, leakage control, and operational readiness.
— Data-Generating Path Checks: Trace timestamps, corporate actions, survivorship effects, and look-ahead bias through to executable orders.
Stress-Test the Spread Between Gross Edge and Live Execution
Replace flat average-cost shortcuts that hide spikes during drawdowns with structured execution evidence:
— Scenario and Trigger Modeling: Test base, conservative, and adverse cases, with a defined failure trigger across spread-aware fills, borrow frictions, capacity limits, and cost distributions by volatility bucket.
— Tiered Evidence & Calibration Loops: Anchor claims to observed fills and proximate spreads over generic benchmarks, running monthly bias checks and tracking missed trades as first-class outcomes.
— Operational Governance & Charters: Combine cash planning and pre-trade checklists with “unknowns” logs, “rejected for now” files, a research ledger, a cost laboratory, and bounded pilot charters.
Durable quantitative trading belongs to researchers who challenge their own argument before the market charges them to discover its flaws.