Can LLMs Fix Momentum Investing? Defiance Bets on Real-Time AI

MIAMI, FL – October 19, 2026 – Momentum investing has a glaring structural flaw: it drives by looking in the rearview mirror. Traditional quantitative models excel at identifying which stocks have accelerated over the past twelve months, but they are chronically late to market reversals. By the time a price-based algorithm registers a shifting narrative, the alpha has often already been extracted by faster, qualitative actors.

Today, the asset management industry is witnessing a live attempt to engineer a cure for this lag. Defiance ETFs has officially rebranded the QRAFT AI-Enhanced U.S. Large Cap Momentum ETF—a five-year-old quantitative fund formerly known by the ticker AMOM—into the Defiance LLM Catalyst Momentum ETF. Now trading on the NYSE Arca under the ticker AIED, the newly positioned fund pairs an established price-momentum screen with a real-time news sentiment engine powered by large language models.

The transition, which follows the sponsor's takeover of the fund on July 1, 2026, marks a critical pivot in how artificial intelligence is deployed in retail investment products. For years, AI has been packaged primarily as a thematic destination—a sector to invest in. With this relaunch, generative AI is being operationalized as an active portfolio management tool, shifting from a marketing narrative to an execution engine.

From Theme to Tool: Curing the Momentum Lag

To understand the mechanical shift under the hood of AIED, one must look at the division of labor between its quantitative and qualitative signals. The fund’s underlying AI engine, developed by South Korean fintech firm Qraft Technologies, filters a universe of U.S. large-cap stocks down to a pool of 100 to 200 candidates. It does this using a proprietary formula based on 12-month momentum and risk-adjusted momentum factors.

Historically, the system then utilized deep learning models to estimate relative price appreciation over a four-week investment period, relying on Bayesian neural networks to gauge the uncertainty of those forecasts before selecting a final 50-stock portfolio.

The updated strategy introduces a critical new layer: an LLM-driven scoring framework that reads company-specific news in real time. This system is designed to flag developments that are unusually positive or negative, effectively acting as a qualitative catalyst detector.

"Momentum has a well-known limitation: it only knows what the price has already done," said Sylvia Jablonski, Chief Investment Officer of Defiance ETFs. "The strategy behind AIED adds a second input, a large language model that reads company-specific news in real time and scores what it finds, so the portfolio reflects both the trend and what is driving it. We think that is a more complete read on a stock than either signal alone."

By marrying a quantitative momentum factor with a real-time news signal, the strategy attempts to capture persistent trends while remaining highly responsive to abrupt narrative changes that traditional price screens would miss until the next rebalance.

The Commercial Reality Behind the Rebrand

While the technological integration is compelling, the repositioning of the fund is equally a story of commercial ETF survival. Despite boasting a solid track record since its inception in May 2019, the original vehicle struggled to accumulate significant assets.

Historical performance data reveals a strategy that frequently outpaced broader benchmarks. In 2024, the fund returned 35.79%, comfortably beating the S&P 500's 24.89%. Year-to-date in 2026, it sits at 21.68%, again outperforming the broader market. Yet, as of this month, the fund's assets under management hovered between $21 million and $27 million—a modest footprint for an ETF with a five-year live history and a competitive 0.75% expense ratio.

The decision by the new sponsor to step in, replacing Qraft in that specific capacity, reflects a strategic marriage of convenience. Qraft, fundamentally a technology provider, retains its role as the licensor of the AI engine. Meanwhile, the new sponsor brings its formidable distribution network and marketing machinery, specializing in first-mover thematic products.

"First-generation AI ETFs often suffered from a translation problem," noted one independent quantitative research analyst familiar with the fund's history. "They had complex, black-box methodologies that were difficult to explain to retail advisors. Slapping 'LLM' on the label isn't just a trend-chasing move; it gives advisors a tangible, understandable narrative. They know what ChatGPT is. Now they can explain how that same underlying technology is reading the news to pick stocks."

The Execution Challenge: Hallucinations and High Turnover

For leaders who value execution over hype, the integration of LLMs into active portfolio management introduces a fresh set of operational hurdles. Chief among them is the well-documented phenomenon of AI "hallucinations"—instances where large language models confidently generate false or deeply flawed interpretations of text.

If an LLM misreads a sarcastic earnings call comment or misinterprets a complex regulatory filing as a negative catalyst, the resulting algorithmic signal could force an unwarranted trade. This is where the fund’s governance structure becomes paramount.

The strategy is not fully autonomous. Exchange Traded Concepts, LLC remains the investment adviser and retains full discretion over every investment decision. The adviser can, and presumably will, depart from the AI recommendations to address corporate actions, mergers, spin-offs, or instances where the LLM's sentiment analysis appears fundamentally disconnected from market reality. This human-in-the-loop architecture provides a necessary safeguard against rogue algorithmic behavior, satisfying regulatory compliance while mitigating model risk.

Furthermore, the practical application of this high-frequency sentiment analysis carries structural costs. The fund refreshes its 50-stock database every four weeks. Historically, this strategy has resulted in substantial portfolio turnover—with past metrics indicating turnover rates highly variable and occasionally exceeding 400% annually. High turnover inherently increases transaction costs, which can quietly erode the alpha generated by the AI signals.

To manage concentration risk during these frequent rebalances, the portfolio adheres to strict weighting methodologies. It implements a 10% cap on any single company at the time of rebalance and ensures that no more than 40% of total assets are held in positions weighted above 5%.

A New Benchmark for Active Machine Management

As the ETF landscape becomes increasingly saturated with products targeting the artificial intelligence boom, a clear dividing line is emerging. On one side are funds that offer exposure to the companies building the hardware and software infrastructure of the AI revolution.

On the other side are funds attempting to use that exact infrastructure to outsmart the market. By explicitly integrating natural language processing to solve a known quantitative deficiency, the newly minted AIED represents a maturation of this second category.

The coming months will serve as a live stress test for this hybrid model. The market will be watching to see if the qualitative overlay of an LLM can successfully navigate sudden macroeconomic shocks or complex corporate crises faster than a standard price-momentum screen. If the real-time news signal can consistently identify catalysts before they are fully priced in, while human oversight successfully filters out the noise and hallucinations, this rebranded ETF may finally attract the assets its historical performance suggests it deserves.