SI405: Why Most Trend Following Improvements Should Fail ft. Rob Carver

— Perpetual futures approval: The CFTC approved the first perpetual futures contract (crypto-linked), which settles daily rather than on fixed expiration dates. The CME has sued the regulator, arguing these should be classified as swaps rather than futures, threatening traditional exchange business models reliant on rolling volume.

— Quantica research on trend attribution by asset class: Over three decades, trend returns have shifted dramatically: 2000–2009 was diversified (equities, currencies, commodities equally); 2010–2019 was dominated by fixed income; 2020–2026 is almost entirely commodities-driven. Chasing the strongest trend does not add value beyond what continuous signal systems already capture.

— Overfitting and AI-generated strategies: AI tools can generate trading ideas but should not be trusted to backtest or validate them. Robust process—separating idea generation from rigorous, out-of-sample testing—is critical. Public AI models introduce additional risk because training data provenance is opaque.

— Factor concentration and diversification risk: Academic research shows the "factor zoo" collapses into a handful of true return drivers. When one factor dominates, diversification provides little benefit; this has major implications for portfolio construction.

— Drawdown patterns and crisis alpha: Different asset classes have distinctive drawdown profiles (frequency, depth, duration). The key insight is identifying which assets draw down *together* versus independently, rather than simply measuring drawdown size in isolation.

— Economic data degradation and Fed communications: New Fed Chair Walsh signaled plans to overhaul central bank data collection and abandon forward guidance. Declining statistical quality (from budget cuts and shutdowns) risks undermining both systematic macro strategies and basic economic visibility.

— Resist chasing recent performance: The Quantica paper demonstrates that different asset classes dominate in different multi-year regimes. A continuous trend-following system already allocates capital to stronger trends; manually rotating into "hot" sectors adds complexity and execution risk without statistical edge.

— Maintain genuine commodity exposure: Over 30 years, commodities have been the most reliable crisis diversifier and have driven trend returns in multiple regimes. Reducing commodity allocation for liquidity or AUM reasons sacrifices true portfolio resilience. A 40%+ allocation to convergent (non-trend) strategies may require balancing with deeper commodity participation.

— Evaluate AI strategies on *process*, not output: When assessing AI-driven trading strategies (yours or others'), focus on the research workflow—how ideas are separated from backtesting, whether out-of-sample validation is genuine, and whether data leakage is ruled out. Do not trust an AI model to validate itself; do not rely on public AI models without understanding their training data.