Why Eliminating Discretion from Quantitative Models is Harder Than You Think

Episode Summary

Executive Summary: The episode argues that quantitative investing is never fully “robotic”: human judgment is required in strategy design, evolution, and interpretation of changing market conditions. The hosts emphasize that emotions can distort model construction, that strategies should ideally be based on long-term evidence, and that adapting models—especially around valuation and intangible assets—must be done carefully to avoid short-term bias or data mining.

Main Topics: The myth of fully emotionless quantitative investing (Priority: 5/5): The hosts explain that quant strategies are designed to reduce emotional decision-making, but humans still make critical choices in construction and maintenance. Initial strategy construction requires discretionary judgment (Priority: 5/5): Decisions about factors, portfolio size, rebalancing frequency, and universe selection are all made by humans and shape strategy outcomes. Recency bias and data mining in model design (Priority: 5/5): They warn that investors may overweight recent winners like growth and underweight long-term factors like value, creating fragile models. Why some strategies evolve over time (Priority: 4/5): Using O'Shaughnessy Asset Management as an example, they discuss moving from a single valuation metric to composite measures to reduce dependence on one potentially weak signal. When to modify a factor like price-to-book (Priority: 5/5): They debate whether factors can be declared “dead” using data alone and argue that judgment is needed when structural changes, such as intangible assets, undermine older metrics. A disciplined approach to quant research (Priority: 4/5): The hosts highlight their own practice of implementing strategies exactly as published while also continuing to research new ideas, balancing discipline with innovation.

Key Arguments: Quantitative investing reduces emotion, but it cannot eliminate human judgment; people still decide what enters a model and how it evolves. Initial portfolio design choices—factors, weighting, universe, rebalancing, and portfolio size—can materially change outcomes and are influenced by emotion. Recency bias can push investors toward recent winners like growth, even if long-term evidence favors a different approach such as value. Altering a strategy to fit current market conditions can lead to data mining and overfitting, producing models that look good in backtests but fail live. Composite valuation measures can be preferable to single-factor measures because they reduce dependence on one metric being “best” at any given time. The rise of intangible assets challenges older valuation metrics like price-to-book, especially for companies where most value is not on the balance sheet. There is insufficient data, over a normal investing lifetime, to definitively prove some factors are dead or alive; judgment must fill the gap. The goal is not to reject quant investing, but to recognize that every quant process has a human decision-maker behind it.

Data Points: Portfolio size: 10 or 20 stocks - The hosts describe the model portfolios they run at Validia Capital. Rebalancing frequency: monthly, quarterly, annually, and tax efficiently - Examples of human decisions embedded in their portfolio construction process. Research time horizon for factor conclusions: 100 years vs. 10 years - They contrast long-term evidence with recent performance when evaluating factors like value and price-to-book. Intangible assets share: 50% or more - They suggest intangible assets may represent half or more of corporate assets, making book-value-based metrics less reliable.

Pivotal Quotes: "there is no pure quantitative model" — Jack: Summarizing the core thesis that human judgment always remains in quant investing. "And so the point was to say a lot of people sell quantitative models as, all right, these are 100% quantitative and we're taking all your emotions out of the process. And in the real world, that just doesn't exist." — Jack: Explaining that quant strategies still depend on human decisions behind the scenes. "the less discretion, the better" — Jack: Contrasting quant investing with discretionary investing and explaining why the hosts prefer rule-based approaches.

Implications: Investors should not assume quant strategies are fully automatic or free from bias. Even rule-based systems depend on thoughtful human design, periodic review, and discipline to avoid overfitting and chasing short-term trends.