Prioritizing Cluster Enhancements
Slipping from a strong start to the year of roughly +6.5% to down -2.04% (-1.84% relative to NAV) by month-end is frustrating, given the slide came in a matter of a few days. We acknowledge this is a short-term view, as EHLS prioritizes long-term upside participation, which could bring volatility in certain periods. However, EHLS aims to be the premier core equity fund, and as such, we seek to reduce its volatility from current levels. We are actively refining our clustering techniques to help accomplish this goal, which we may discuss in more depth in future updates. It’s an incredibly exciting development, which we believe will better align the fund’s exposures.
Despite the recent decline, the fund remains ahead of the S&P 500 as of the end of February since the August lows. This month provided us with vital data and motivation to continue improving our clustering models. We view the clustering process as ever-evolving in our ultimate investment selection criteria to alleviate overconcentration in various exposures that can be quite difficult to detect. This will be our primary focus over the next month.
In This Update
Momentum Factors: Momentum investments broadly experienced an unusually severe five-day decline starting February 19th, revealing difficulties in our portfolio's exposure clusters.
Short Misalignment: Our short positions failed to provide adequate protection during this correction, while key holdings like Carvana, Sprouts Farmers Market, and Robinhood peaked and pulled back sharply on earnings.
Improving Clusters: February's market disruption highlighted necessary improvements in our clustering methodology, prompting a comprehensive reassessment of how we group investments based on their risk characteristics.
Exposing Exposures: We're ecstatic about implementing enhanced clustering protocols that will better identify hidden correlations and reduce portfolio volatility while maintaining a focus on upside potential.
Looking Back
At Even Herd, our investment approach combines momentum principles with diverse portfolio construction and data-driven clustering. While EHLS adapts across sectors and monitors emerging trends like AI, momentum strategies typically follow steady progression with occasional pullbacks. However, from February 19th through month-end, momentum investments experienced an extreme convergence of short-term correlations, adversely skewing the fund’s performance given our proprietary system is most comparable or exposed to momentum based factors.
We observed a sharp decline across momentum investments in multiple sectors. Speculation around Nvidia's upcoming earnings call likely triggered much of the initial market rotation, as the AI-focused companies have seen extreme volatility this year. Despite S&P 500 declines, market conditions revealed targeted severe downturns in specific momentum factors rather than widespread deterioration. AI infrastructure investments faced sharper declines, with companies like Sterling Infrastructure (STRL) and IES Holdings (IESC) experiencing price pressure. Our system has reduced these exposures while identifying opportunities that continue to lean more international like banking firms HSBC Holdings (HSBC, UK) and Mizuho Financial (MFG, Japan). The fund has had exposure to Chinese technology firms, particularly data center providers Kingsoft Cloud (KC) and GDS Holdings (GDS), but has now included larger names like Alibaba (BABA).
Unrelated companies including Carvana (CVNA), Sprouts Farmers Market (SFM), Redwire (RDW), IonQ (IONQ), Robinhood Markets (HOOD), and Palantir Technologies (PLTR) all peaked and corrected substantially during this period. Earnings releases provided catalysts of some extreme declines in names, but as a whole, momentum factors seemed explicitly targeted. Crypto-related companies showed high volatility, though our exposure remains minimal through holdings like Strategy (MSTR) and Cango (CANG). Argentina, a top-performing region in 2024, has also underperformed this year, which the fund maintains exposure to in diverse sectors, although almost all reside within the same cluster.
It’s not shocking to see speculative investments, such as space-related or quantum computing-focused companies previously discussed, correlate in such a way during sharp market movements, which is an additional reason we maintain modest exposure in these areas. But the breadth of the correlation extremes in such a small period was remarkable while also being quite selective. The widespread declines challenged our portfolio, with our short positions failing to provide adequate protection as underperforming, lower-momentum stocks remained rather stable in the period.
Though five trading days is too short for meaningful conclusions, we find the magnitude of February's portfolio-wide fluctuations something to be addressed and a current priority. Utilities and Real Estate showed unexpected surges relative to other sectors, while Communications and Technology reversed sharply starting mid-month, requiring adjustments to our positioning. One of the only constants observed was that Financials remained the leading sector, with Basic Materials consistently underperforming since early November. We previously expressed the closeness of most sectors, however, most previous trends seem to abruptly snap in a few days, which is somewhat unique without a glaring catalyst and something we must closely monitor.
Looking Forward
Our diversification strategy continues to push beyond sector or industry allocation. We use advanced clustering methods to analyze how securities relate to each other based on our proprietary data. February's disruption revealed exposure concentrations that were once more difficult to visualize, prompting a reassessment of our clustering approach, which we are now focused on improving. But, we should clarify, as computing power and AI advance, we'll always be incorporating new technologies to improve our clustering systems—while our proprietary system remains our primary investment tool, the clustering process serves as a vital secondary component that enhances our selection process and contributes to long-term performance stability.
For those unfamiliar, clustering is a technique that groups investments based on their exposure characteristics or correlation patterns. Even Herd's proprietary models allow us to dynamically adjust these clusters based on our own data, giving us unique analytical advantages over traditional sector-focused approaches. We are striving to balance portfolio weights across these custom-defined clusters while still strategically overweighting leading sectors and underweighting laggards through our dynamic allocation process. This dual approach enables us to identify and capitalize on market movements. We're continuously enhancing our proprietary clustering algorithms to reduce volatility while maximizing gains and minimizing drawdowns like those experienced in February. We will always have room for improvement within this aspect of our strategies. But again, most of this move came in a matter of days, which is less concerning than witnessing this kind of disconnect occurring over several months.
Extreme market events provide valuable insights that normal conditions don't reveal. While January's AI correction following DeepSeek concerns was informative how the market views a variety of AI-specific investments, February's broader momentum disruption highlighted important gaps in our cluster methodology. We're optimistic about implementing improvements based on these findings. This brief market dislocation showed exposure patterns we consider unacceptable, but we also won't overreact. Our approach emphasizes capturing upside while avoiding unnecessary volatility, knowing it will come with periods of short-term disconnects. But, ensuring any unexpected spikes in volatility like witnessed this month is a short-term blip is critical and our primary focus right now.
Our systems are designed for intermediate-term trends rather than daily or even weekly movements—a principle tested during periods of extreme price swings. Improved clustering will help prevent concentration in areas whose true risk characteristics only become apparent during market extremes. We'll share more about these methodological improvements as they develop and how we utilize them, as they'll enhance our ability to identify correlations that would otherwise remain masked. Improving this process will also help align our long and short positions.
Comments