New to Curatia?

See why finance counts on us for key market intelligence and industry trends.

Curatia Analysis for Tue, Aug 13

After Slow Start, AI Use in Trading Is Beasting

Although forward-thinking finance firms have been leveraging AI to automate routine back-office tasks for years, many balked initially at the prospect of deploying the black-box technology in the high-stakes, unforgiving realm of trading.

That hesitancy has dissipated of late as firms increasingly use AI to analyze reports, calculate risk, develop hedging strategies, evaluate alt-data sets, and trade — with tremors heralding seismic industry shifts already reverberating.

Research & Analysis

In the analytical domain, the fusion of man and machine is yielding a next generation of tools and techniques.

As more traders look to AI for an edge in unearthing alpha, FactSet has partnered with fintech startup DataRobot to allow its workstation users to model equity volatility, bond performance, and macroeconomic event predictions.

The solution gives users lacking data science expertise “‘guardrails…to build and deploy advanced machine-learning models.’” When paired with FactSet’s alt-data marketplace, the offering conjures visions of a future in which wonky active fund managers arrive at unique investing strategies by cobbling together numerous data-driven AI engines on the fly.

Morgan Stanley has meanwhile devised a trading strategy using natural-language processing to extract sentiment from its own research reports.

Using a two-month holding period following a report’s release, the strategy performed solidly in back-testing reaching back to 2013 and handily outperformed broader markets in 2018. Morgan Stanley’s algorithm assigns a confidence rating to each sentiment score, mitigating long-standing concerns about AI’s opacity.

Interestingly, the much-discussed automation of research authoring could introduce a scenario in which machines parse the words of other machines. In that case, stock moves based on prior report analysis could impact future research recommendations, creating a feedback loop.

Fintech startup Prattle is using a neural network to interpret central bank meeting minutes and policy statements more rapidly than humans can while sidestepping human confirmation bias. As such techniques grow more commonplace, central banks are starting to use AI to vet policy statements before their release.

Modeling, Trade Automation

Closer to the metal, AI is operating largely without human collaboration — with much pioneering work coming out of JPMorgan, which sports a mammoth $11.4B tech budget.

The bank is looking to increase adoption of the machine learning-based FX algo trading tool it launched in April. It has also developed “deep-hedging” strategies, which use a neural network and reinforcement learning in place of pricing models like Black-Scholes to develop hedging strategies in index options and, more recently, stocks.

The technology has outperformed classical models, and BAML and Société Générale are already working on similar projects. And, in a similar vein, Pimco has also employed neural networks to forecast prepayment risk in mortgage-backed securities more accurately.

While those projects showcase AI’s impacts on discrete segments of the trading ecosystem, others are approaching AI-based trading holistically.

China’s first AI fund, for instance, is opening this month. Trained by analyzing the strategies of more than 80 of China's best-performing fund managers, the fund’s AI engine returned 26.4% from Sep-Jun, while CSI 300 Index benchmark returned roughly 12%.

Other hedge funds including Bridgewater and Man Group have also deployed AI-only funds. But since 2018, such funds have lagged a broader index of hedge-fund performance.

With highly differentiated performance among AI-based funds in mind, JPMorgan filed with regulators in June to create a fund-of-funds to invest in AI-driven hedge funds.

Market & Industry Implications

Broadly speaking, the diversity of AI projects in the thick of the trading jungle spotlights an accelerating automation movement — with significant industry implications.

For starters, AI project teams are managing to solve — or at least blunt — oft-cited AI shortcomings such as black-box decision-making. Simultaneously, they are introducing others like short back-testing periods that leave AI engines vulnerable to faulty assumptions.

Heightened use of AI in trading is also touching off an arms race among hedge funds striving to “stand out in a crowded field.” In just ten months from August 2017 to May 2018, the proportion of hedge funds using AI leaped from 20% to 56%.

As a result, AI positions continue to be in high demand. AI job postings grew 29% in the 12-month period ended in May, with machine-learning engineers boasting an average salary of around $143k annually.

Those dynamics are great for coders but could also presage the demise of the human trader — a topic we’ll explore in greater detail tomorrow.