AQR's former machine-learning head says quant funds should start 'nowcasting' to react to real-time data instead of trying to predict the future

By Bradley Saacks

For many big quant funds, March was rockier than 2007's infamous "Quant Quake" that decimated Goldman's systematic trading unit. A former AQR Capital executive is recommending that to avoid further shocks quants should adopt the practices of the people that track real earthquakes.

Marcos Lopez de Prado, who ran machine-learning at $143 billion hedge fund AQR before joining Cornell as a professor, said on a webinar Friday hosted by alt-data company Thinknum that quants are too reliant on forecasts and models that attempt to predict the future. Instead, he said, they need to "nowcast" — react to the latest news that has actually happened. 

"Forecasting made a lot of sense decades ago when we had very limited data, that was delayed and slow," Lopez de Prado said. "Today we have an abundance."

He laid out several examples of finance and non-finance already doing this, including those that track real earthquakes. In finance, he said, inflation and liquidity are often tracked using near-real-time data, like The Billion Prices Project, an initiative to find the real inflation by listing the prices of millions of items every day. 

"There are very few assumptions involved in direct measurements," he said.  "It's time for quants to pay less attention to crystal balls and add nowcasting to their arsenal."

Big-name quants like Schonfeld Strategic Advisors, Renaissance Technologies, Bridgewater, Point72's Cubist, and Quantedge were crushed in March as the novel coronavirus pandemic introduced never-before-seen volatility into the markets. Bridgewater's billionaire founder Ray Dalio told investors last month that "of course we are questioning if we should have done things differently." 

Lopez de Prado did an informal poll of the attendees on the webinar, and found that 80% use forecasting, not nowcasting, for price discovery and risk management — which is dangerous, he said, because of quants' reliance on backtesting to find trading rules. He said backtests should be used to refute hypotheses, not create them, but many funds and academics have done the opposite. 

"It is almost assured there will be a false positive" using backtests to create trading rules, he said. As Richard Michaud, the CEO of New Frontier Advisors in Boston, said on Thursday in a CFA Society of Boston webinar, "backtests are notorious for misleading investors, and ruining careers."

The quant of the future, Lopez de Prado believes, will take advantage of the reams of data flowing into the asset management space, and focus on adapting their strategies to the environment as opposed to building a strategy that can endure any environment. The of creating an all-regime or all-weather fund, he says, is slim as markets continually adapt and discretionary investors learn from their mistakes.

"It is a futile exercise."