Phantom Gains in Backtests
Phantom gains are backtested profits that exist only because the simulation deleted a structural cost of real trading. The term comes from Lalor Swishchuk 2025, who show that a Market Making framework assuming the price process and the order-arrival process are independent generates “large phantom gains” — the agent gets filled without ever paying the Adverse Selection penalty real liquidity providers face. Closely related inflations: simulating fills at the mid-price rather than the actual bid/ask, assuming posted orders are automatically at the front of the queue, and ignoring the fixed price-tick grid.
This is the reason a market-making or HFT backtest result must be graded, not believed. Lalor & Swishchuk warn that “many models built using much of the standard mathematical finance theory in algorithmic and HFT have often been shown to over-inflate results,” and recommend reading any backtest “with a grain of salt.” Phantom gains sit alongside Overfitting in Quantitative Trading and Data-Snooping Bias as a reason the MDP/RL trading literature’s profitability evidence is simulation-based and weak.
Connections
- Market Making — relates (the strategy most affected), source: https://arxiv.org/html/2410.14504v2
- Lalor Swishchuk 2025 — relates (names and quantifies the artefact), source: https://arxiv.org/html/2410.14504v2
- Adverse Selection — causes (omitting adverse fills is the primary source), source: https://arxiv.org/html/2410.14504v2
- Out-of-Sample Backtesting — opposes (phantom gains survive even nominal out-of-sample tests if costs are missing), source: https://arxiv.org/html/2410.14504v2
- Overfitting in Quantitative Trading — relates, source: https://arxiv.org/html/2410.14504v2