Momentum Trading Meets AI

Key Takeaways

• Volatility targeting raises Sharpe from 0.53 to 0.97.
• Max drawdown cut from 96.7% to 45.2% with targeting.
• LLaMA2 outperforms GPT‑3.5 in signal generation.
• Momentum risk 77% strategy‑specific, predictable.
• AI‑driven execution improves risk‑adjusted returns.

Pulse Analysis

Momentum investing has long attracted academic and practitioner attention, yet its practical implementation often suffers from severe tail risk and inconsistent performance. Recent advances in natural language processing allow models to ingest vast streams of news, earnings releases, and macro commentary, converting sentiment into actionable signals. By feeding this enriched information into a momentum framework, researchers can move beyond simple price‑based rankings and capture nuanced drivers of price continuation, creating a more adaptive and data‑rich investment process.

A core insight from Garmash’s work is the power of volatility targeting to tame momentum’s inherent risk profile. The analysis shows that only about a quarter of the strategy’s variance stems from market movements, while the remaining 75% is strategy‑specific and highly predictable. Scaling positions inversely to forecasted volatility therefore stabilizes returns, nearly doubling the Sharpe ratio and halving the worst‑case drawdown. This risk‑adjusted improvement aligns momentum with institutional risk mandates, making it a viable component of diversified portfolios rather than a niche, high‑risk play.

The comparative test between LLaMA2 and GPT‑3.5 underscores the competitive edge that model architecture and training data can provide in financial applications. LLaMA2’s superior signal generation translated into higher annual returns and sharper risk‑adjusted metrics, suggesting that model selection matters as much as the underlying strategy. For quantitative firms, the implication is clear: integrating state‑of‑the‑art LLMs with disciplined volatility targeting can unlock more reliable alpha, reduce capital erosion during market stress, and justify the computational investment in AI infrastructure.