Subject: A one-year backtest!
AQR’s ‘Hard to Believe’ Study Spurs Clash Over AI Use for Quants
https://finance.yahoo.com/news...
Quant traders, who use rules-based strategies derived from data analysis, have long believed their models get less effective when they become too complicated. That’s because they suck in too much of the distortive noise that makes predicting markets such a challenge in the first place.

But a researcher at AQR Capital Management has sparked a backlash with a study claiming the opposite — that rather than being a liability, bigger and more complex models might offer advantages in finance. The paper, titled , showed that a US stock market trading strategy trained on more than 10,000 parameters and just a year of data beat a simple buy-and-hold benchmark.

“This idea of preferring small, parsimonious models is a learned bias,” said Bryan Kelly, head of machine learning at AQR and one of the paper’s three authors...

After digging into the details of the study, Nagel concluded that because the model was dissecting just 12 months of data, it was simply copying signals that had worked more recently. In other words, it was following a momentum strategy — a well-established trading approach.

DB2