Welcome back to AI Street. I'm Matt. Here’s the latest on AI + Wall Street.

HEDGE FUNDS
AQR’s Asness Says AI Has Taken Over Parts of His Job

AQR Capital Management’s Cliff Asness recently spoke to Bloomberg TV on how the quant firm is using AI. The billionaire outlined three ways the asset manager, which runs about $116 billion, is using the technology.

He discusses the impact of AI starting around 17:00 to 22:00 in the clip below, which is worth watching. (I find him to be pretty candid relative to a typical money manager). See the highlights of his interview below:

Signal Generation:

• Upgraded natural language processing (NLP) to analyze written and verbal information

• Previous quant models used simple word counts (e.g., +1 for "increasing")

• New AI models understand context ("massive embezzlement is increasing" is negative)

Portfolio Construction:

• AI helps determine factor weights in portfolios

• Traditionally, Asness made final calls on weights to avoid overfitting

• AI is now taking over some of this decision-making

Productivity Enhancement:

• Coding assistance

• Speeds up innovation cycles

• Streamlines routine tasks

Some quotes that stood out to me:

On AI vs. Traditional Investing

• “I do think the AI world will be far more dynamic than what you might call the quant factor investing world. You do a very good version of what Warren Buffett’s doing, you’ll have bad times, but if you stick with it, I think, I think the whole world can know about it, and you can still make money from it.”

• “AI, Big Data alternative data sets will be, I think, a more constant arms race, where you have to keep reinventing what you do. I’m actually optimistic we can do that, but it will have more of that element.”

On AI Asness

• “AI is coming for me now, because I’ve always been one of the AQR that makes kind of final call” on weighing certain investment factors. “Turns out it’s annoyingly better than me.”

On Overfitting and Complexity

• “Forever we’ve talked about not overfitting. That a danger of quantitative processes is you see things in the data that were random. You think you can trade them, but they weren’t real… AI pushes us a little on the spectrum away from some of the traditional things we’ve talked about, and that was uncomfortable for me.”

On AI’s Impact on Jobs

• “Not to get philosophical, but on significantly long enough time horizon, [AI] is coming after all our jobs, right? Which is not necessarily a bad thing depending on the transition, you know, we get to a world of Star Trek where we have a replicator that can make whatever we want. That’s not necessarily terrible news, but it could be a lot of upheaval on the way to that.”

From June:
Cliff Asness Says the Machines Are Making More Decisions at AQR

COMING SOON
AI Street Sunday Editions

In the new year, I’ll be writing a new AI Street edition on how to use the latest AI tools for market insights. Look for an opt-in invitation in the coming weeks.

Please take a moment to answer the poll below:

INTERVIEW
Using ChatGPT to Measure Profitability: Five Minutes with USC’s Matthew Shaffer PhD

Regular readers know that I wrote about white collar crime for a number of years at Bloomberg News in New York. And I learned that one of the most challenging types of cases to prove was accounting fraud — in part because there are so many judgement calls required.

Tax fraud cases are (generally) more straightforward: ‘You reported earning X dollars last year, but you actually earned Y dollars.’

Prior to ChatGPT, accounting judgement calls had to be done by human analysts. It’s hard to create deterministic rules for that task.

That’s what makes a recent academic paper about using GPT-4o, the current state of the art model provided by ChatGPT, to measure “core earnings” so interesting. The research, written by Matthew Shaffer at the University of Southern California and Charles C.Y. Wang at Harvard Business School, showed that ChatGPT could be used to estimate a company’s recurring profitability better than standard measures.

I spoke with Matt for this week’s Five Minutes With Interview.

The paper, Scaling Core Earnings Measurement with Large Language Models, shows how AI could bridge the gap between quantitative metrics, like price-to-earnings ratios, and qualitative factors which have been historically been hard to analyze at scale.

This interview has been edited for clarity and length.

What did you seek to study in your paper?

Our paper examined whether large language models could effectively quantify companies’ “core earnings”—the recurring, core profitability from their main business activities—at scale. We chose this question because we thought that this is the kind of task that is most amenable to the distinctive new capabilities opened up by large language models—as opposed to the prior “rule-based” symbolic artificial intelligences. This is a task that requires reasoning over a lot of unstructured text, applying common sense judgments contextualized in general background knowledge, about accounting and industries, etc. 

With that research question in mind, we approached in a quite straightforward way, developing two strategies and judging the models’ outputs.  

Then we evaluated the quality of the measures we got from these approaches. Now, by its nature, “core earnings” is not directly reported anywhere, there’s no one external benchmark for the “correct” number. So our approach to testing the quality of the measures is to say, “if this were indeed a good core earnings measure, what properties would it have empirically?” For example, a good core earnings measure should be informative for predicting future Net Income. We have many such tests, and we benchmark the LLM-based measures to widely-used proxies from Compustat, a standardized financial data provider.

Overall, our LLM-based measure from our gold standard approach outperforms the Compustat alternatives in most, though not all, standard tests. For example, Compustat’s OPEPS does slightly better in regressions predicting future Net Income – but, interestingly, our LLM-based measure does better predicting average Net Income over the next two years. 

But, I should emphasize that our goal wasn't really about earnings prediction per se – these are just standard ways of empirically testing whether it’s a good core earnings measure. That’s what we’re most interested in - this concept of core earnings, that matters to investors but typically requires qualitative, firm-specific analysis. We wanted to see if LLMs could be used to quantify this at scale.

Read the full interview here

IN CASE YOU MISSED IT
Recent Five Minutes with Interviews

• Moody’s Sergio Gago on scaling AI at the enterprise level.

• Ravenpack | Bigdata.com’s Aakarsh Ramchandani on AI and NLPs

• PhD candidate Alex Kim on executive tone in earnings calls

• MDOTM’s Peter Zangari, PhD, on AI For Portfolio Management

• Arta’s Chirag Yagnik on AI-powered wealth management

MARKET MOVERS
Top AI-Driven Stock Movers

Salesforce (CRM):
📈 +12.0%
Shares soared to a record as CEO Marc Benioff touted Agentforce’s 200 deals and its role in driving digital labor and AI collaboration.

SoftBank (9984):
📈 +5.1%
Shares gained on plans to invest $1.5bn in OpenAI, reinforcing its AI leadership ambitions under Masayoshi Son.

Amazon (AMZN):
📈 +6.0%
Shares rose as AWS announced an “Ultracluster” AI supercomputer powered by Trainium chips, advancing its AI infrastructure efforts.

Marvell Technology (MRVL):
📈 +31%
Shares jumped on strong results driven by AI chip sales to Amazon and others, beating consensus. FY4Q25 guidance also topped expectations at $1.8bn revenue vs. $1.65bn forecast.

Data provided by Linq Alpha, an AI copilot for hedge funds.

DEEPFAKES
FBI Warns of AI-Boosted Financial Fraud

The FBI is warning that criminals are leveraging generative AI to make financial fraud more convincing and efficient, according to a December 3 advisory. Tactics include:

Voice Cloning: The FBI reports fraudsters are using AI-generated audio to impersonate bank clients and gain account access. In some cases, scammers creating "short audio clips containing a loved one's voice" to request urgent financial transfers.

Market Manipulation: AI-generated images and videos are being deployed in market manipulation schemes, though the FBI didn't provide specific examples.

The advisory notes that AI helps criminals overcome traditional red flags by:

• Eliminating language errors that previously helped identify foreign scammers

• Creating convincing fake IDs and credentials

• Generating realistic social media profiles for investment fraud

"Since it can be difficult to identify when content is AI-generated," the FBI is advising financial institutions to verify caller identity through established channels and watch for subtle imperfections in synthetic images or videos.

Unfortunately, tech advancements usually come with criminal exploitation.

REGULATION
New Bill Asks Regulators to Map AI’s Financial Impact

A new bipartisan bill introduced by Rep. Maxine Waters and Rep. Patrick McHenry would require financial regulators to conduct comprehensive studies on AI's impact across banking, securities, and housing within 180 days.

Federal Reserve, FDIC, OCC, CFPB and NCUA must examine:

• AI use in loan underwriting, fraud detection, and customer service

• How smaller banks and credit unions can leverage AI

• Challenges in hiring and retaining AI talent

• Current regulatory barriers

SEC to analyze:

• AI in market research and portfolio management

• Exchange use of AI for surveillance and fraud detection

• Agency's own AI capabilities and limitations

Housing regulators to study:

• AI in mortgage lending and property management

• Impact on homebuyers in underserved communities

• Use by real estate agents and online platforms

The bill follows a study released in July from congressional staffers that raised concerns over potential bias and discrimination that may be harder to detect. 

MORE TOP NEWS
Brevan Howard Spinoff Leverages LLMs To Transform Analysts Into Quants

Bin Ren, the CEO of SigTech, recently spoke at Fintech Connect 2024, saying LLMs will cause a "quantum leap in productivity.”

The firm was spun out of Brevan Howard's systematic investments group in 2019 and implements multi-agent generative investment copilots (MAGIC).

Finance has historically been split into "two modalities" according to Ren: numbers and text. Quants worked in the former and analysts worked in the latter, and infrequently the two shall mix. Through LLMs, however, Ren says we can "converge those modalities into one."

AI of the past has helped automate processes, but Ren says what held it back from allowing employees to branch into new fields was its lack of awareness of "tacit knowledge," which generative AI excels at. (eFinancial Careers)