Why Wall Street should have had its ChatGPT moment (but hasn’t)?
Artificial Intelligence & Machine Learning
The asset management industry is due for a huge shake up.
As LLMs are revolutionizing the tech world, asset management has been a massive laggard in this regard. Over the last few years, Silicon Valley has given us LLMs, liquid nets, symbolic AI, Fintech, robotics and quantum computing startups.
On the other hand, Wall Street is still wallowing in factor models, smart betas, technical analysis and (at best) XG Boost or random forest models. Its traders are still executing trades by phone call or messaging, with very little automation, let alone AI. Having been at several AI panels, it appears that most asset managers are getting limited use of AI models- primarily, workflow automation and report generation. Most are using the same hedging and position sizing frameworks invented before the 70s, with minor adjustments. Why is Wall Street a casual bystander in the AI revolution, left way behind by Silicon Valley? More pertinently, why didn’t it have one of its own?
Truly speaking, the next revolution in data science and AI should have come from Wall Street.
We have some of the most brilliant and brightest technical minds grappling with forecasting and optimisation challenges with far fewer and noisier data that any other industry has to deal with, along with a plethora of unstructured datasets. Yet why do Wall Street quant funds and banks transform pioneering physicists and engineers into passive consumers of models, using plug-and-chug libraries and running lots of backtests? It is really very tempting to go down this route and I speak from experience.
Rather than spend months building conceptually sound models and frameworks, it’s much easier to plug some existing model into a backtester, fiddle with it and get the dopamine high of a beautiful backtest. And when your neck is not on the line, there’s no downside in presenting a nice overfitted or hidden-beta-riding backtest (or several) to your manager and getting an allocation. If it works, awesome. If it doesn’t then you go to the next quant fund and play the same song and dance there.
While Silicon Valley is producing startup after startup, each with a new application of AI models or new foundation models themselves (see Symbolica and Liquid AI), Wall Street seems stuck in the dinosaur age.
There is no fundamental innovation in the mainstream, and most large funds and asset managers are following variants of old factor rotation and portfolio optimisation models. At most there are tired attempts at copying LLMs’ success, but there isn’t enough relevant training data for a copy-paste to work. You might claim that there is exciting work going on in secret behind closed doors at quant funds, but having interacted with and interviewed scores of quants, I am yet to find any evidence of novel approaches beyond regurgitations of Bayesian, trend following, stat arb or some overfitted AI approaches.
In general across industries, innovation does not come from big companies which are built for predictability and reliability- it usually does from startups, which have much more freedom and flexibility to learn, pivot, grow and explore.
The truth is Wall Street does not have a startup culture. Most of the recognisable names in the quant hedge fund industry were created in the 1985-2000 era. Our industry does not have an ‘Open AI’ or ‘Anthropic’ equivalent, nor do we have anyone remotely close to building its likeness. What we have is buy-side institutions chasing overfitted backtests and sell-side institutions publishing mostly trite or abstract research. The ‘startup culture’ becomes dominated by spin-offs of big funds that regurgitate variants of the same old strategies from fund to fund.
What are the reasons for this lack of startup culture? There are a few:
- Lack of funding opportunities: Simply put, VCs do not fund startup hedge funds, systematic or discretionary. A quant hedge fund is much more similar to a tech company than a discretionary fund in terms of IP and capital requirements. Yet there is no Angel, VC or YCombinator equivalent for quant funds.
- Age of founders: Most hedge fund founders are well into their late forties, fifties and sixties, and usually well out of date with the latest innovations. Allocators believe gray hair equates better risk management but there is no data to back that.
- Prime Brokers: Possibly the largest hindrance to startup innovation in quantitative finance are prime brokers. It is getting harder and harder to custody and execute with prime brokers with AUM below 50 mm, and retail brokers are not suitable for systematic trading. Apart from AUM, the only way to get a prime brokerage relationship is by having a sell side background; i.e, more factor model and smart beta strategies.
- Regulation: Regulation is getting harder and more expensive every year for startup hedge funds.
- Short term thinking : A good quant fund should operate as a deep tech company. It should take years of research and development, and if successful long lasting alpha. Very few funds have followed this model. It needs long term thinking, which is missing from seeding platforms. We at FMI Tech see ourselves as a developer of sound, long lasting IP that happen to be monetised by asset management, and not a hedge fund that just happens to create IP.
- Propensity to backtest: I’ve seen talented mathematician quants convince themselves that they have found the holy grail using convolutional neural networks, Bayesian filters and the like. It is very very easy to fall into the trap and chase good backtests with mediocre overfitted techniques. At FMI, any model with more than a handful of free parameters becomes automatically disregarded.
- Lack of transparency: Due to a high amount of secrecy, we have very few big funds posting any research; if they do it is outdated or plain wrong. Remember that Google produced the paper on transformer models that kickstarted the current revolution. At my firm, FMI, we have made a conscious effort to publicize and publish our non-alpha research as much as possible. We have research articles on microstructure, options pricing and crash prediction for instance.
- High failure rates and lack of appetite for risk: We have all heard that 90 percent of funds fail in five years, which is roughly equal to the same number for startups. However while VCs are happy to take the risk and get multiples on their returns. Nevertheless, the limited seeders and family offices that exist still appear spooked by these numbers. I met a well known family office CIO who after one failed GP investment in a quant fund, vowed never to touch the space again.
- Ties to academia: While Silicon Valley has ties to Stanford, Berkeley and the like, very few hedge funds do so. We at FMI are an exception, working with Princeton, Stony Brook, UCLA and the like.
We on Wall Street missed the AI Revolution!
However, there is another, arguably bigger one coming in the quantum space. This is more suited to financial use cases because it is able to deal with noisy datasets with very few number of points. From the work of Jack Sarkissian, myself and Halperin, we know that markets are quantum in nature, and quantum methods are much more data efficient and noise-resilient. Quantum could be Wall Street’s ChatGPT moment, and we do not want to miss it.
Written by Revant Nayar
