Capital Allocators• Nov 20, 2025• 55:47Interview

Daniel Mahr – Glass Box Quant at MDT Advisers (EP.472)

From Capital Allocators

Daniel Mahr•Head of MDT

Executive Summary

Daniel Mahr, head of MDT at Federated Hermes, details the evolution of their $26 billion quantitative equity strategy, which has used machine learning since 2001.
MDT employs a "glass box" philosophy using a "forest" of decision trees, emphasizing transparency and understanding the context-dependent nature of investment factors, unlike opaque "black box" models.
The strategy focuses on gaining an "analytical edge" by applying sophisticated models to public data, rather than pursuing an "informational edge" with alternative datasets.
Key model findings include a powerful reversal effect in heavily sold-off stocks and that the predictive power of factors like momentum and valuation differs significantly based on a company's age and financing activities.

12 quotes

Concerns Raised

The declining efficacy of traditional factors like 'book-to-price' requires continuous model evolution.
A potential structural shift in markets over the last few years may be reducing overall market efficiency.
Recruiting top-tier data science and programming talent has become significantly more competitive.
Inability to properly backtest commercial LLMs prevents their use in current stock selection models.

Opportunities Identified

Leveraging a 20+ year head start in machine learning to maintain a durable analytical edge.
Exploiting behavioral biases, such as the difficulty investors have buying stocks with 'bad stories' that are down 70-80%.
Using a proprietary, fully-integrated technology stack provides flexibility and a competitive advantage.
Applying AI software development co-pilots to enhance the productivity and efficiency of the in-house technology team.

Key Themes

Evolution of Quantitative Investing

The discussion traces the progression of MDT's strategy from traditional factor-tilting in the 1990s to a sophisticated machine learning approach using decision trees starting in 2001. This evolution was driven by the need to adapt to changing market dynamics, like the dot-com bubble, and enabled by exponential growth in computing power and data availability.

This highlights the dynamic nature of quant investing, where firms must continuously innovate their models and technology to maintain an edge as markets and economies evolve.

The 'Glass Box' Philosophy

MDT deliberately positions its models as a "glass box," ensuring transparency into how and why every investment decision is made. This contrasts with "black box" approaches, allowing the team to understand factor interactions, identify factor decay (like the declining power of book-to-price), and maintain conviction during periods of underperformance.

This philosophy is crucial for risk management and model improvement, as it allows for human oversight and intuition to complement algorithmic precision, preventing blind reliance on a model that may no longer be effective.

Context-Dependent Factor Efficacy

A core tenet of MDT's approach is that investment factors are not universally predictive; their effectiveness is highly dependent on context. For example, the models find that momentum is a stronger predictor for younger companies, while valuation is more critical for mature ones, and high-financing companies can still outperform if they exhibit strong, consistent momentum.

This sophisticated, non-linear view challenges simplistic factor investing and demonstrates the power of machine learning to uncover nuanced relationships that traditional linear models would miss.

Analytical vs. Informational Edge

MDT's strategy is explicitly focused on creating an "analytical edge"—using their proprietary machine learning models and long-term data to generate superior insights from widely available information. They consciously avoid the race for an "informational edge," which relies on acquiring unique or alternative datasets that may have fleeting value.

This strategic choice emphasizes building a durable, process-driven competitive advantage through superior analysis rather than relying on a constant, expensive search for proprietary data.

Cautious Adoption of New AI

While a pioneer in using machine learning, MDT is cautious about integrating the latest AI, such as Large Language Models (LLMs), directly into its stock-picking models. The primary barrier is the inability to control for look-ahead bias in historical data, making robust backtesting impossible. They are, however, exploring AI co-pilots to improve developer productivity.

This illustrates a disciplined approach to innovation, prioritizing analytical rigor and verifiable results over adopting hyped technology, a key consideration for any firm integrating AI into its core processes.

Get started free

Topics

Quantitative Investing Machine Learning Decision Trees Factor Investing Glass Box Model Analytical Edge Proprietary Technology Backtesting Risk Management Stock Selection Momentum Factor Value Investing Factor Decay Book-to-Price Federated Hermes MDT Advisors

Processed Apr 6, 2026 yt-dlp + mlx-whisper + Gemini