Quantifying DeepSeek’s Impact on 1/27/25

TL;DR

• The shockwave that hit US Tech on Monday and sent ripple effects throughout global markets underscores the need for sophisticated portfolio analytics in order to track thematic and factor movement in the broader market and within your portfolio.
• The correlation between AI themes had increased materially since the US election, which provided a warning sign that the trade was too concentrated and subject to a selloff on any bearish news. The themes of AI Power Generation, AI Infrastructure, and AI Datacenter saw an avg correlation increase of 14% since 8/1/24, and took a significant beating on Monday (ranging from -10% to -18%) , while AI Adopters were less impacted (-2.36%).
• Monday saw several extreme factor movements, as the market rotated away from Momentum (-5.23 standard deviations), Beta (-3.27 STD), and Exchange Rate Sensitivity (-7.08 STD) towards the safety of Earnings Yield (+4.14 STD) and Value (+2.31 STD).
• Communications Equipment , Construction & Engineering, Electrical Equipment, and Electronic Equipment were the hardest hit sectors on Monday, suggesting the unraveling of an Industrials AI theme.

Monday’s market rout was chiefly precipitated by investor revelations around DeepSeek’s R1, a new Chinese open source large language model. R1’s performance ostensibly rivals OpenAI’s o1 model at a fraction of the cost to DeepSeek and its end users (although the true cost and resources used to develop and train the model are hotly contested).

As the AI market was already seen as priced to perfection, both directly and peripherally in the AI CAPEX ecosystem, the end result was a substantial sell-off for nearly all associated names. The S&P 500 closed down 1.5%, the Nasdaq 100 fell by 3%, and Nvidia itself lost $600 billion in market value. Energy firms benefiting from the AI power generation theme were also punished, led by Constellation Energy Corp (-21%).

The rally in US stocks over the past two years has been largely predicated on US AI dominance and the moat afforded by hundreds of billions in AI investments. Now, the DeepSeek news has led investors to feel less confident in how they have been valuing AI-linked companies.

Fundamentals aside, in the following we will leverage the Omega Point platform to provide an empirical perspective on Monday’s movement using factors and themes, to help you better understand how the structure of the market changed before and after Monday’s selloff. We will follow up with another post detailing how investors can utilize Omega Point to be prescriptive with their portfolio’s positioning to follow-on DeepSeek/AI-related thematic movements on a go forward basis.

‍The Aftermath: Extreme Factor Movements

‍If you recall from our previous analyses during the Regional Banking Crisis and Ukraine Invasion, it’s important to understand which factors saw significant movement during major market events such as Monday’s.

To effectively analyze many factors across different models, we normalized Monday’s 1-day factor moves by the expected volatility of each factor. This approach allows us to isolate the factors most sensitive to Monday’s events, which we can also refer to as the “Surprise” metric. The table below summarizes the style factors that exceeded a 2-standard deviation performance move on Jan 27, 2025.

‍Extreme Style Factor Movements: 1/27/2025‍

‍‍Markets rotated away from Momentum, Beta, and Exchange Rate Sensitivity‍

• Exchange Rate Sensitivity was the biggest loser, falling by an average of -7.08 standard deviations in a single day. As this factor essentially tracks net exporters vs net importers, it saw weakness as US tech companies that export semiconductors and hardware sold off.
• The Momentum category fell -5.23 standard deviations on average, while Beta declined by -3.27 standard deviations.
• Conversely, we saw a flight to safety via outperformance in Earnings Yield, Value, and Barra’s Long-Term Reversal factor.
• Size factor experienced notable weakness within the Wolfe US2 Tech, Media, & Telecom model, highlighting the underperformance of megacap tech.

This analysis can help you understand the significant factor moves that occur around major events and predict how your portfolio might react on a go-forward basis. If other news comes out that threatens US AI dominance and investment, we might expect these factors to behave in a similar way again.

‍Digging Into the Beta Rotation

‍One factor that caught our eye was Barra’s Beta factor, which fell by 4.48 standard deviations on Monday. In order to get a better view under the hood of this extreme movement, we applied a simple yet powerful portfolio construction approach called High-Minus-Low (“HML”). HMLs can help analyze residual characteristics of a factor and “project” that factor onto additional risk models and datasets.

We built a Beta HML using the following method:

• Universe: Russell 3000 Index
• Long Stocks: Top Quintile Based on Exposure to the Barra US Beta factor
• Short Stocks: Bottom Quintile Based on Exposure to the Barra US Beta factor
• All stocks are equally weighted on the long and short sides of the portfolio

Beta HML Portfolio Performance: 1/27/25‍

While the high Beta names on the long leg of this portfolio declined by 3.99%, weakness in the portfolio was also driven by the outperformance of low beta stocks (up 1.29%), representing a flattening of the beta curve as dispersion between high and low beta stocks narrowed.

If we isolate just the the short side of the book, we can see the sectors that drove the outperformance of low Beta - namely Real Estate, Consumer Staples, Financials, Healthcare, and Utilities. Note that negative absolute performance is due to the short side of the book being weighted negatively (so stocks going up denote negative performance).

As the AI CapEx theme continues to play out in the market, we will continue to track what is driving performance within Beta since it has proven to be quite sensitive to this dynamic.

‍Extreme Sector Movements: 1/27/2025

‍We ran the same analysis to determine which sectors had a greater than 3-standard deviation move on Monday.

• The Communications Equipment, Construction & Engineering, Electrical Equipment, and Electronic Equipment plummeted on Monday as investors wrestled with the implications of reduced CapEx.
• Again, we saw a flight to safety to Insurance, Tobacco, Water Utilities, and Consumer Staples.

Forewarning: How AI Themes Were Sounding the Alarm

‍It isn’t every day that a new market entrant disrupts the global economy after releasing a new product to little fanfare, but DeepSeek managed to do just that. The notion that no one could have prepared for this, however, is not accurate. In the following, we will discuss how Omega Point’s proprietary thematic package was sounding a warning klaxon for the AI space.When we think about AI, there are three predominant themes:

• AI Producers
• Power Generation
• AI Adopters

Thanks to our partners at Goldman Sachs, JP Morgan, and Morgan Stanley, we have several baskets representing each of these themes (along with hundreds of others) on Omega Point.

Check out our Resources page for more information on Omega Point’s thematic work

Of course, we would expect most of the AI-focused baskets to have been decimated on Monday, but the magnitude of these single day moves was still staggering.

‍Performance of AI Thematic Baskets 1/27/25

‍While the AI Power Generation, Infrastructure, and Datacenter themes were pummeled, it’s noteworthy that the AI Adopters basket saw relatively little carnage. This basket includes names that the market thinks will benefit from the lower costs of AI implementation if DeepSeek’s efficiency is truly replicable.

If we take a look at these baskets relative to NVDA, the AI bellwether, we can see that the Returns Correlation for eac basket to NVDA was relatively high on an arbitrary date last summer, as we might expect.

If we look at the same metric post-election, we see that correlation to NVDA increased materially for most of these themes, although the AI Adopters basket remained less correlated than the others.

Between 8/1/24 and 1/28/25:

• AI Power Generation - went from 56% to 74% correlated
• AI Semi and Tech Infrastructure - correlation remained high above 80% (NVDA is a constituent)
• AI Data Centers - went from 72% to 80% correlated
• AI Adopters - went from 44% to 59% correlated
• On the flip side, the Long Term AI Beneficiaries basket became less correlated, falling from 62% to 38%

In essence, ex-the AI Adopters, all of the AI themes became one trade since the election. If an investor were tracking these themes, they might have become more wary of their exposure to these themes given the excessive investor concentration. While we didn’t necessarily know when the next shoe was going to drop, it became increasingly evident that the market was at risk for a drawdown like the one we experienced on Monday.

The purpose of our thematic analysis is to capture transient systematic risks in the market, and there are a lot of ways that we can leverage this analysis. For instance, if we know each theme’s correlation to your portfolio, we can quantify your exposure and predict how these themes may impact your positions going forward.‍

Conclusion

‍This remains a very fluid situation as even today (Wed, 1/29) AliBaba released its Qwen 2.5 Max model (a lesser model that didn’t spark a market reaction but indicates the AI arms race is heating up). Market participants will also be tuning in to earnings calls from the AI heavyweights of the world, from which we can expect questions around the DeepSeek bombshell as well as forward CapEx guidance.

There is a lot to parse as the market continues to validate and digest this latest disruption, and we will be back with a second installment discussing how asset managers and hedge funds can use Omega Point to inoculate their portfolios from market shocks similar to the one caused by DeepSeek this week.