Rodrigo Gordillo - “This Craftsmanship Perspective Is About Identifying The Difference Between Complex Versus Robust” | #180

Episode Summary

Executive Summary: Rodrigo Gordillo of Resolve Asset Management discusses how personal experience with inflation, currency collapse, and market crashes shaped his quant-first, diversification-focused investing philosophy. The conversation centers on using ensembles, breadth, and risk parity to reduce luck and fragility, while machine learning is framed as a useful but highly filtered tool rather than a silver bullet.

Main Topics: Formative macro experiences in Peru and Canada (Priority: 5/5): Gordillo explains how childhood exposure to bombs, blackouts, hyperinflation in Peru, and later housing and market losses in Toronto taught him the importance of nominal vs. real returns and concentration risk. Momentum as an ensemble problem (Priority: 5/5): The paper 'Global Equity Momentum: A Craftsman’s Perspective' argues that using multiple lookbacks, moving averages, and asset classes produces a more robust trend signal than a single fixed rule such as the 12-month or 200-day moving average. Why single-parameter strategies can be fragile (Priority: 5/5): He critiques the industry tendency to market one clean rule because it is easy to explain and sell, even when an ensemble approach is statistically more resilient and less exposed to luck or regime dependence. Machine learning as signal generation plus strict filtering (Priority: 4/5): Machine learning is presented as a way to generate many candidate strategies, but the real edge comes from human judgment, validation, and pruning via a 'sentinel' process that excludes non-viable or overfit ideas. Risk parity and adaptive asset allocation (Priority: 5/5): Gordillo defends risk parity as an asset-class-agnostic framework that equalizes risk contributions across inflation, growth, and deflation assets; adaptive asset allocation adds tactical exclusion and momentum overlay. Breadth, weighting, and portfolio construction (Priority: 4/5): A major theme is that portfolio results often improve more from better weighting and broader exposure than from better stock picking; he argues that thoughtful weighting can materially raise Sharpe ratio. Productization, naming, and investor behavior (Priority: 3/5): The discussion notes that investors prefer binary, memorable rules and often resist nuanced, blended approaches, which creates a marketing advantage for simplistic strategies despite their limitations.

Key Arguments: A single lookback or valuation metric is often a lucky parameter choice; ensembles reduce the chance of being specifically wrong and produce more stable outcomes over time. The difference between complex and robust matters more than simplicity; complexity that improves resilience is preferable to a simple rule that only looks good in one regime. Momentum, value, and other factors should be viewed as recipes with multiple ingredients; the choice of asset classes, lookbacks, and weighting schemes can matter as much as the signal itself. Machine learning is useful for discovering candidate patterns, but without a strong validation/filtering layer it mostly finds noise and overfit relationships. Risk parity is not simply a levered bond portfolio; it is a framework for balancing risk across asset classes tied to different macro regimes. A well-constructed portfolio should be asset-class agnostic and humble about certainty, avoiding 100% binary bets whenever possible. Breadth and weighting can generate substantial performance improvements without changing the underlying signal, making portfolio construction a key source of edge.

Data Points: Peru inflation (1988): 7,200% - Gordillo describes the hyperinflation that wiped out family wealth and shaped his views on real returns and debtors vs. savers. Peru inflation before crisis: 23% - He notes inflation had previously seemed 'normal' before the jump to hyperinflation. Family grandfather savings: Equivalent of a million dollars reduced to zero USD value - Illustrates how currency collapse can erase nominal wealth. Toronto housing drawdown: 50% to 55% - He recalls Toronto home prices falling sharply after his family bought with zero money down. Assets gathered as early advisor: $2.5 billion - He worked at a family office managing money and evaluating managers before going independent. Timing of independent start: 2006 - He left to build a quantitative practice during the commodity boom. 2008 performance: Single-digit positive returns - He says the strategy made money for clients during the crisis year, drawing attention afterward. 2020? no—paper universe size: About 1,200 different strategies - The ensemble paper tested many parameter specifications for momentum. Ensemble ranking: Around the 90th percentile Sharpe ratio - He says the ensemble was not the best single backtest but sat near the top in robustness. Value weighting example: Sharpe ratio improved from 0.94 to 1.23 - He cites a portfolio construction example showing risk parity weighting materially improved results versus 50/50. Weighting impact: About 30% Sharpe ratio increase - He argues thoughtful weighting alone can significantly improve a portfolio's risk-adjusted return. Risk parity leverage example: Targeting 15% volatility - He explains that levering the risk-parity portfolio to equity-like risk improved outcomes over equities in the 1940-1981 example. Real return of bonds, 1940-1981: -68% - Used to counter the claim that risk parity is just a levered bond portfolio. Momentum paper backtest period: 1970 to 2012 - Original book/paper period that produced strong results for a single 12-month specification. Extended backtest period: 1950 to 1970 and 2012 to present - Adding more history moved the single-specification result back toward the median.

Pivotal Quotes: "I'd rather be broadly correct about capturing that specific signal than specifically wrong." — Rodrigo Gordillo: Explaining why Resolve prefers ensembles and robustness over a single elegant rule. "There is no magic in machine learning. The magic is in what opportunity SETA gives you, and then how do you filter those out?" — Rodrigo Gordillo: Describing machine learning as a candidate-generation tool that still requires human judgment and validation. "You tell me the risk that you're willing to take, and I will lever that portfolio to the point where we hit that risk target or that return expectation." — Rodrigo Gordillo: Summarizing the core logic of risk parity as a risk-targeting framework rather than a bond bet.

Implications: For investors, the episode argues for breadth, humility, and robust process over single-rule certainty. For managers, the edge may come from portfolio construction, validation, and regime awareness more than from one clever signal.