Frazzini, Israel, and Moskowitz (2015) use nearly $1 trillion in live institutional trading data from a large money manager (AQR) across 19 developed equity markets (1998-2011) to measure the actual transaction costs of implementing factor strategies.
Key findings
- Actual costs are ~1/10 of academic estimates: real-world trading costs for factor strategies are far lower than those assumed in academic studies, which typically rely on quoted spreads or Hasbrouck/Kyle measures
- Strategies are 10x more scalable than previous estimates suggest
- Value and momentum are the most scalable strategies, with large capacity before costs erode returns significantly
- Short-term reversal is the most constrained by trading costs, consistent with its microstructure origins
- Size strategies have intermediate scalability
Why academic cost estimates are too high
Academic studies typically estimate costs using:
- Quoted bid-ask spreads (overstate costs for patient traders)
- Market impact models calibrated to small/illiquid trades
- Failure to account for sophisticated execution algorithms
Real institutional traders use limit orders, algorithmic execution, spread across time and venues, and trade patiently, all of which dramatically reduce realized costs.
Implementation improvements
Strategies designed to reduce costs (trading more patiently, avoiding small/illiquid names, limiting turnover) can increase net returns and capacity without significant style drift. The gap between gross and net factor returns is manageable for well-constructed portfolios.
Significance
This paper is a critical rebuttal to the argument that factor premia are “paper profits” consumed by transaction costs (a concern raised in Arnott et al. 2019 and elsewhere). It provides the strongest evidence that the main anomalies (value, momentum, size) are robust, implementable, and sizeable in practice.