The Best Investment Writing Volume 3: Corey Hoffstein – Factor Fimbulwinter

Episode Summary

Executive Summary: This episode presents Corey Hofstein’s essay on factor investing, arguing that well-known anomalies like value, size, and momentum are hard to “kill” statistically because decades of evidence and high return variance make disconfirming them extremely slow. The core message: factor failure is more often a prolonged “Fimbulwinter” of disappointing performance than a sudden Ragnarok, and investors need both statistical humility and a coherent rationale for why an anomaly should persist.

Main Topics: Factor investing as 'Fimbulwinter' rather than Ragnarok (Priority: 5/5): The essay frames factor underperformance as a long, painful winter of weak returns rather than a dramatic collapse, especially for value investing. Price-to-book and the value factor's recent underperformance (Priority: 5/5): Price-to-book has been in a large drawdown and has prompted debate about whether value is broken or merely suffering from implementation issues. Bayesian test of how long it takes to 'disprove' a factor (Priority: 5/5): A simulation-based experiment estimates how many years of new zero-alpha evidence would be required before factors are no longer statistically significant. The difficulty of distinguishing randomness from structural decay (Priority: 4/5): The speaker argues that a decade of weak performance may be insufficient to conclude a factor no longer works, because history could have unfolded differently. Momentum and drawdown-driven distortions in factor testing (Priority: 4/5): Momentum’s return path includes rare crashes that complicate simple resampling tests, making the experiment potentially skewed. Statistical significance versus the need for a 'why' (Priority: 5/5): The episode closes by questioning whether statistical evidence alone is enough and emphasizing that investors also need a plausible causal explanation.

Key Arguments: A factor’s failure is more likely to appear as prolonged underperformance than as an obvious catastrophic break. Price-to-book’s weak decade may reflect randomness, an implementation flaw, crowding, or a genuine false positive; performance alone cannot distinguish these cleanly. Because many factors have decades of supporting evidence and substantial return variance, it may take several careers to gather enough contrary evidence to reject them. The experiment suggests that the statistical tools used to validate factors also make them slow to invalidate. Momentum’s distribution is unusual because a small number of severe crashes dominate losses, affecting the resampling logic and making many sampled 12-month periods look positive. Quant investors should not rely only on historical significance; they should also ask why a factor should continue to exist going forward.

Data Points: Price-to-book drawdown from prior high: about 25% - The academic price-to-book factor remains well below its December 2006 peak. Time since price-to-book last made a new high: since December 2006 - Used to illustrate the long period of weak performance. Years since the size premium was first published: 35+ years - The size anomaly has only recently faced widespread skepticism. Median time to declare Price-to-Book/HML dead: 67 years - Result from the Bayesian simulation experiment. Simulation repetitions per factor: 10,000 - Each factor was tested repeatedly to create a distribution of future disconfirmation times. Confidence threshold: 99% - The test stopped when annualized return was no longer statistically non-zero at this level. Rolling 12-month positive periods for value, size, quality, low-volatility: just under 50% - Random 12-month draws from demeaned histories were often still positive. Rolling 12-month positive periods for momentum: about 55% - Momentum had a higher share of positive sampled periods due to its return profile. Momentum average positive vs. negative period size: 1.4x larger - Excluding the worst 5% of rolling 12-month periods, positive momentum periods were larger on average.

Pivotal Quotes: "the death of a factor will be more fimblewinter than Ragnarok" — Corey Hofstein: Summarizing the core metaphor: factor failure is gradual, prolonged, and emotionally painful rather than sudden. "the same tools we use to establish and defend factors may actually prevent us from tearing them down" — Corey Hofstein: The closing thesis explaining why established factors are so hard to abandon. "there is a certain faith required to use them going forward" — Corey Hofstein: Conclusion about the limits of statistical evidence and the need for investor conviction.

Implications: Investors should expect factor decay, if it happens, to be slow and ambiguous. Historical significance can outlast practical usefulness, so decisions should combine data, economic logic, and humility rather than relying on backtests alone.