Abstract
A printed trade reports that a transaction happened, not that an order could have traded there: odd lots, off-exchange prints, and condition-coded trades are unreachable by a marketable order or excluded from the official high, low, and last. A bar built from all prints therefore sits above the executable high and below the executable low, so every stop, target, and breakout is scored against prices that never existed. From one trades feed we build naive and executable bars by print-time condition codes, across the US equity top-1000 (2007-2026, 1.46 billion regular-hours bars); three results follow. First, 42.3% of bars carry a phantom-affected extreme, rising to 90% by 2026, led by odd-lot prints (76% of the range inflation). Second, the headline is selection fidelity, not profit: across 7,232 strategies in an edge-free universe, in-sample selection on executable bars predicts the executable out-of-sample ranking better than on naive bars (rank-IC 0.9987 versus 0.9948; gap +0.0039, p < 0.001 under an effective-N cluster bootstrap). Third, a CME futures negative control near zero and the 2013 odd-lot-tape-onset natural experiment isolate the effect to equity microstructure. A formal companion traces the mechanism partly to persistence, so a strategy can pass naive validation in and out of sample yet die on execution (the deployment trap); a known-answer planted-drift experiment on a 60-name panel finds the clean-selected book higher at every tested drift strength, consistent with a deployment flip it suggests rather than directly observes. The claim is one-sided: cleaning restores selection fidelity rather than manufacturing an edge.
Keywords: market microstructure, odd lots, off-exchange trading, sale condition codes, OHLC bars, backtest overfitting, selection fidelity, walk-forward validation
The views, analysis and figures are the author’s own and do not necessarily reflect those of algoseek. This content is for educational purposes only and is not investment advice.