Bridgewater CEO: AI Fund Generating "Unique Alpha"

Hey, it's Matt. Here’s what’s up in AI + Wall Street.
Was this email forwarded to you? Sign up below:

IN THIS ISSUE

• Bridgewater’s AI Fund

• AI Outperforming Factor Models

• New Podcast! w/ SigTech’s Bin Ren 

• AI Boosting Productivity: Fed

HEDGE FUNDS

Bridgewater CEO: AI Fund Generating "Unique Alpha”

Bridgewater Associates' $2 billion AI fund is generating "unique alpha that is uncorrelated to what our humans do," CEO Nir Bar Dea said at a Bloomberg conference this week.

Though it may not sound like it, “unique alpha” (market-beating returns) that’s “uncorrelated” is a big deal on Wall Street, especially from the head of one of the world’s largest hedge funds. The fund, which was launched last summer, is delivering returns "comparable" to the firm’s human-led strategies, though specific figures weren't disclosed, according to Bloomberg.

Run by Co-CIO Greg Jensen, the fund combines Bridgewater's proprietary technology with plans to include models from OpenAI, Anthropic, and Perplexity, Bloomberg reported last summer.

Bridgewater formed its Artificial Investment Associate (AIA) Labs division in 2023, combining large language models, machine learning, and reasoning tools. Designed to understand causal relationships in markets, the tech serves as the primary decision-maker in the fund, while human professionals oversee risk management, data acquisition, and trade execution.

AI adoption in finance will really take off once some major funds start reporting outperformance. (Bloomberg)

RESEARCH

AI Outperforms Factor Models

New research shows how AI is flipping investment analysis upside down.

The reason AI is so powerful/exciting/scary is because it fundamentally changes how we solve problems. Before, engineers came up with ideas, wrote rules, and deployed code—a top-down exercise. With AI, particularly transformer models like those behind ChatGPT, we can now work bottom-up.

This turns traditional factor models (like Fama-French) upside down. Instead of pre-selecting what matters, AI “discovers” what’s driving returns. Give a transformer model tons of data and it draws connections, understands patterns, and makes predictions by learning from examples, similar to how children learn language.

That’s the core idea behind ‘Artificial Intelligence Asset Pricing Models.’

I had a fun chat with one of the authors, Semyon Malamud, who broke it down well:

“In the old days, people would just count words... but then people realized you can't treat words out of context. You have to understand the contextual environment of every given word. We extended this idea to financial markets - you can't understand Tesla or Google or Microsoft alone. You have to treat it in context, determined by other stocks.”

A New Approach to Asset Pricing

The results from their AI-based approach are pretty striking:

• Their transformer-based model achieved a Sharpe ratio of 4.57, significantly outperforming traditional models, which typically range between 1.05 and 1.80.

• It performed especially well with large-cap stocks, where making accurate predictions is typically more difficult.

• The AI model delivered the most precise predictions, with the lowest pricing error among all machine learning models.

• The benefits of “cross-asset information sharing” were most pronounced in mega-cap stocks, where the transformer model reached a Sharpe ratio of 1.84 compared to 1.18 for competing methods.

• Each time the researchers increased the model’s complexity by adding more transformer “blocks,” performance consistently improved.

These systems learn patterns directly from data, even though we don't fully understand how it all works. This is the unsettling part of AI in finance, trusting a system we don’t fully understand over our theories.

These sorts of breakthroughs are happening across disciplines. Research published in Nature showed that an AI-driven weather forecasting model outperformed sophisticated models built by expert meteorologists—despite having zero knowledge of atmospheric physics. It simply learned patterns from historical data.

As Malamud put it: "Physicists were saying, 'Oh, this is so stupid. You don't understand what [the model] is doing.' But a machine learning model that knows nothing about physics is better at predicting weather than the most advanced model used by physicists."

What's striking (and weird frankly) is that this success doesn't require a theoretical understanding of market dynamics. If pattern recognition consistently outperforms theory-based models, understanding 'why' may matter less than predicting ‘what' happens next.

Paper: Artificial Intelligence Asset Pricing Models
Authors: Bryan Kelly, Boris Kuznetsov, Semyon Malamud, and Teng Andrea Xu.

PODCAST

SigTech's Bin Ren on AI Agents for Investors

For the second episode of the Alpha Intelligence Podcast, my co-host, Francesco Fabozzi, and I speak with Bin Ren, the founder of SigTech, on using AI agents in financial analysis.

Ren founded SigTech in 2019 after leading quantitative investment funds as Chief Investment Officer at Brevan Howard’s Systematic Investment Group, from which SigTech was spun out. He previously worked as an equity exotics trader at Barclays and holds a PhD in computer science from Cambridge.

We dive into MAGIC, SigTech’s GenAI product, the rise of AI agents in hedge funds, banking, and equity research, and what the shift towards automation means for the future of investment decision-making. We also explore the economics of compute costs, the role of knowledge vs. experience in AI-driven markets, and how LLMs are redefining financial workflows.

This episode is useful for engineers looking to understand how to build GenAI products for finance, as well as portfolio managers and analysts exploring ways to integrate LLMs into their workflows. Let us know what you think!

Listen on Spotify 

ADOPTION

AI Boosts Productivity: Fed Survey

Workers reported saving a substantial number of work hours by using AI, according to research from the Federal Reserve Bank of St. Louis.

This is the first independent study I’ve seen on the impact of generative AI on productivity. The stats:

• 33% Productivity Boost – Workers are 33% more productive in each hour they use generative AI.

• More Than 1 in 5 Users Save 4+ Hours Weekly – 21% of AI users saved 4 or more hours per week.

• Frequent Users Benefit More – 34% of daily AI users saved 4+ hours per week, compared to only 12% of weekly users.

Amazon Cloud on Wall Street

• JPMorgan runs 1,000+ AI-powered applications on AWS, with 5,000 employees using SageMaker monthly to streamline model development.

• Bridgewater’s AIA Labs uses AI agents to break down complex investment strategies, speeding up research while keeping analysts in the loop.

• MUFG’s AI sales tool, which analyzes client data and market signals, has achieved a 30% conversion rate, transforming corporate sales efficiency.

ICYMI

In Case You Missed It: Markets Edition

Extracting Reliable Data From Earnings Calls With Hudson Labs

For Last Sunday’s Markets Edition, I spoke with Kris Bennatti, CEO of Hudson Labs, about “AI nonsense,” when AI mixes up accurate numbers, like up reported results with forward guidance.

In our conversation, Bennatti explains how Hudson Labs fine-tunes AI to separate actual earnings from guidance, ensuring more reliable data extraction for investors.

We walk through Nvidia’s latest earnings release as a case study, highlighting why revenue jumped 78% but the stock still sold off.

If you’re not receiving the Markets edition and you’d like to, change your preferences here.

WHAT ELSE I’M READING