Quantitative funds have long been the biggest users of artificial intelligence in the asset-management world. The advent of generative AI, though, could favor traditional, fundamentals-driven asset managers over the quants.
That’s the concern voiced by several quant fund managers and data providers in Asia to DigFin.
“AI applications in finance are still rare,” one quant manager said. “Data scientists aren’t applying it to capital markets. But if these tools are used to trade stocks, it will change the landscape. There will be new winners and losers.”
What’s a quant?
Quants buy and sell stocks based on huge computing power and customized software programs that model investment strategies. The rise of quants coincided with the decades-long decline in interest rates and the rise of passive investments – two trends that have made active stockpicking by humans an increasingly less competitive business.
The use of algorithmic or systematically programmed trades has given rise to an industry of ‘systematic investment’, with firms running platforms of single-strategy managers chasing a particular strategy or ‘factor’ (such as interest rates or a market’s volatility).
Such investors are not interested in being shareholders, only in quickly buying and selling stocks to drive strategies: long/short, market-neutral, statistical arbitrage, event-driven. There is an overlap with the high-frequency trading world, with the commonality being trades that are conceptualized and driven in purely numerical terms.
AI old-timers
These ideas aren’t new, but the availability of computing power and big data sets have fueled the rise of quants over the past two decades. Over the past ten years, quants have been early adopters of new AI techniques such as machine learning and the use of neural networks. They became voracious consumers of alternative data, such as sentiment analysis from social-media feeds.
The biggest problem with quant investors has been ‘explainability’, a more recent term for AI that goes back to quants’ ‘black box’. The 1998 collapse of Long-Term Capital Management epitomizes this risk, particularly as quants are typically leveraged.
But since then, quant shops such as Citadel, DE Shaw, Man AHL, Millennium Management, Renaissance Technologies and Two Sigma have become the biggest and most influential buy-side firms on Wall Street. Their success has spurred traditional fund houses such as BlackRock or Fidelity to launch their own quant strategies.
They also operate in non-US markets where they can find liquidity, low latency trading infrastructure, and hedging instruments (such as ETFs or futures contracts tracking local market indices). Japan has been the biggest market in Asia Pacific, but India is now a major playing ground. (One problem in Asia is regulatory caprice, as a recent South Korean ban on short selling and rising government interference in China attest.)
Quant funds are therefore not just influential apex predators: they are also at the forefront of adopting new digital technologies.
Enter GenAI
Which makes the new developments in AI a puzzle for quants.
These firms will of course use large-language models (LLMs), made possible by generative pre-trained transformers, to their full extent.
The holy grail for quants will be to turn LLMs into predictive tools. A human will interact with their computer buddies to detect patterns across time series and other data sets. In fact, quants do this already, it’s just that LLMs should make the process more intuitive, better integrate non-textual data, and let developers build models much faster.
Quant shops will also use genAI for more mundane purposes, such as learning how to write regulatory reports, interpret earnings reports, or sift through pitch decks. Customer onboarding and other back-office functions can be further automated.
But there is nothing mysterious about a quant shop doing these things, because it’s the same thing that everyone else will be using genAI for.
Everybody’s doing it
The difference is in developing predictive investment models and execution algorithms. That’s what makes quants special, but the early signs suggest genAI will enable traditional asset managers to do these things too. Ditto for managers of private-equity funds – a notoriously un-automated business, which could use LLMs to make investment decisions more systemic and data-driven.
Asset managers will all face questions with LLMs and their tendency to make up stuff. Products like OpenAI’s ChatGPT are the ultimate black box. Although quant funds rely on AI to divine strategies, these are still run by licensed professionals that understand the ramifications of a trade idea. That’s not the case with genAI tools.
Prompt engineering can add value by providing some of that transparency, by interrogating the LLMs to derive a sense of their processes and the factors and sources used to come to a decision. It’s theoretically possible that, one day, LLMs will be more transparent and accountable than a human.
Although the idea of handing investments over to the machine makes for a good headline, quants are likely to use LLMs in more specific ways.
For example, they will want tools to identify the true frictional cost of a trade, which involves a deep study of micro market structures. A typical metric to weigh a trader’s performance is called ‘implementation shortfall’, to figure out how closely they hew to a budget for a given trade. Such algos are already becoming more sophisticated, as firms search for moments during the day when liquidity is ripe or when they can trade without revealing their hand.
This is about finding market signals, which is the core of a quant’s mission. It’s likely that quant shops will use genAI to develop better ways to predict the best times and venues to execute a trade.
This is still very useful but it’s not like anyone’s handing the car keys to the Terminator. Nor does AI overcome the biggest hurdles in Asian markets, which is the lack of hedging instruments, followed by the high cost of hedging when a contract is available.
More importantly, this is not specific to quants. Big traditional buy sides also use these execution algos, whether designed in-house or by a sell-side broker.
The existential question for quants is how they maintain their edge when genAI tools can make a lot of what they do more readily available to fundamental asset managers. Quant shops avoid the limelight in part because they regard their AI models and execution algos as secret sauces. Could genAI turn these into commodities? Just how differentiated is your prompt engineering?
As one quant put it, “AI has been part of our tool set for years. GenAI isn’t getting rid of the barriers, but it will deliver more benefit to fundamental active managers, by making them more efficient at aggregating and analyzing data. Once those firms understand the drivers of return, they become our competitor.”
