AI, Data and Models

In 2020, a year when many quantitative funds struggled, Renaissance Technologies' Medallion fund reportedly gained about 76%. The funds Renaissance sold to outside institutions lost money; its Renaissance Institutional Equities Fund was reported down about 19–23%, and investors later withdrew billions from the outside funds (reported). The same firm, the same mathematicians and the same offices on Long Island had produced the best and some of the most ordinary results in the industry in the same year.

The contrast was not an accident. Jim Simons, a former code-breaker and chairman of Stony Brook University's mathematics department, launched Medallion in 1988, and Gregory Zuckerman's history reports average annual returns of about 66% before fees and about 39% after them from 1988 to 2018. Renaissance closed Medallion to outside investors in 2005 and has kept it at about $10 billion, paying out its profits (reported). Its best signals are short-horizon, small and secret, and they would stop working if more money traded on them. The outside funds were built for capacity, holding positions for longer in larger size, and they earned roughly what large, long-horizon strategies earn. This chapter explains where models make money in finance, why their edge decays, and how the same lessons apply when a model decides who gets a loan.