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

Date: November 20, 2025
Host: Ted Seides
Guest: Daniel Mahr, Head of MDT, Federated Hermes

Episode Overview

This episode features Daniel Mahr, head of MDT, a $26 billion quantitative equity investment group within Federated Hermes. Mahr and host Ted Seides explore MDT’s “glass box” approach to quant investing—an investment process designed for transparency, interpretability, and analytical rigor. They discuss the evolution of quant investing, the development and implementation of decision tree models, the perennial challenges of model overfitting and underfitting, and the delicate balance between machine-driven decision making and human judgment in portfolio construction.

Key Discussion Points & Insights

1. Daniel’s Path from IPO Flipping to Quant Investing

• Background: Daniel describes an early interest in numbers, memories of poring through sports and business pages as a child ([03:51]).
• College Days: He reveals how, as a Harvard freshman during the late-90s dot-com bubble, he made money by flipping hot IPOs, leveraging fast internet and flexibility ([04:20]):

  “As a college student, I had a faster Internet connection than just about anyone else in the world...I managed to get a number of allocations to those hot IPOs.” ([04:20])

• Transition to Quant: Early wins gave way to losses as he overestimated his expertise. This taught him to value systematic, disciplined approaches, which led him into quant ([05:54]):

  “That experience was also very formative in my appreciation of a more disciplined, systematic, rigorous approach to investing, which I found in the quant space.” ([05:54])

2. MDT’s Early Quant History and Shift to Decision Trees

• MDT Timeline: Daniel joined MDT in 2002 as a junior analyst, working closely with the founder. The firm, an early quant pioneer since 1991, was later acquired by Federated ([06:31]).
• Factor Tilting to Decision Trees: Initially, MDT focused on factor tilting strategies, but their lumpiness and lack of adaptability led to a transition toward a “decision tree” approach in 2001 ([08:13]):

  “For the first decade, the strategies were traditional factor tilting strategies...That led us to the decision tree approach that we still use today, albeit in a much evolved way...” ([08:13])

3. Philosophy: The Value of Analytical Edge

• Quant Philosophy: Emphasis on analytical, not informational, edge for all-weather portfolio performance ([10:49]):

  “A disciplined quantitative approach to stock picking can lead to an analytical advantage that will help us generate superior portfolio outcomes...” ([10:49])

• Diversification: Portfolio construction via diversification, not prediction of macro trends—“leaning on diversification...to have a fighting chance at performing well no matter what market environment comes.” ([11:17])

4. Machine Learning: Experience, Transparency, and Pitfalls

• Long Head Start: MDT has deployed machine learning since 2001, providing a deep bench of experience ([12:41]).
• Overfitting vs. Underfitting: Balancing model simplicity and complexity; keeping the process “glass box,” not opaque ([13:51]):

  “Our view in the machine learning space is that transparency is exceedingly important to understand precisely how these models are working...We like to position our investment strategies as being a glass box.” ([13:51])

• Behavioral Edge: Sometimes, model outputs prompt emotional discomfort—especially buying “bad story” stocks, leading to behavioral alpha ([15:36],[00:00]):

  “This is precisely why this strategy works, is because even quantitative investors...find it hard to overcome the human emotions involved with buying a bad story.” ([16:55])

5. The R&D Process at MDT

• Idea Sources: 1) Academic & practitioner literature, and 2) Observations from running their own strategies ([17:55]).
• Unique Factors: Example—“Company age” as a contextual factor, not a direct return predictor ([19:13]):

  “We measure that simply as how long has the company been publicly traded and/or filing financial statements...What we find is that the important questions to ask about young companies...are a little different.” ([19:13])

• Model Updates & Removals: Model is revised as factors lose efficacy (e.g., “book to price” phased out as companies change, the intangible economy rises) ([22:11]).

6. Decision Trees: Mechanics and Interpretability

• From Single Tree to Forests: Initially, a single large decision tree (printed and taped to the wall!), now a “forest” for robust learning ([25:16]).

  “Back in the onetree day, we would print out the tree and tape it on the wall...As you move to a forest...we've built some tools...to synthesize and summarize what's happening across the thousand trees.” ([25:16])

• Tree Depth: Only 2–5 questions per tree to maintain statistical power ([26:37]).
• Example Path: Trees start with high-level questions (e.g., about financing activity), then drill into context (momentum, volatility, company age) ([27:55]).

7. Portfolio Construction and Risk Management

• Optimization Process: Portfolio construction folds in forecasts, hard risk constraints, and trading costs ([32:34]):

  “We have a portfolio optimizer...takes into account the alpha forecasts...risk management...statistical risk model...and trading costs.” ([32:34])

• Liquidity & Impact: Less liquid stocks get smaller positions; trading speed and size are scaled accordingly ([33:55]).

8. The Role of Human Judgment

• Intervention: Human overrides are rare and must be justified by new, “unseen” information. Glass box transparency is key ([35:21]):

  “We try to put all of the potential overrides...through the lens of data...That's the value of the glass box—being able to see how the decision making is being made allows us to be precise...” ([35:21])

• Model Reflexivity & Market Crowding: Quants have gotten more crowded, but leverage, not strategy itself, causes most blowups ([36:47], [37:57]).

9. Adapting to Market Structure & Efficiency

• Recent Shifts: In the past few years, markets have felt less efficient, possibly due to pod shops, passive investing, or retail trading. The causes are less important than maintaining adaptable, active strategies ([39:51]):

  “Something feels like it snapped in the last couple of years. I wish I knew what it was...The good news is we don't need to know what's driving it.” ([39:51])

10. A Day in the Life at MDT

• Daily Process: Overnight data ingest, forecast update, trade review (mainly checking data integrity, not second-guessing the model), and ongoing research ([41:31]):

  “The first thing every day is the trade review process...For most of the team, the focus is on research...idea generation and execution.” ([41:31])

• Tools: Reliance on proprietary in-house tools—no third-party backtesting or risk solutions ([43:00]).

11. Alternative Data & Analytical Edge

• Information vs. Analytics: MDT focuses on extracting analytical edges from long, high-quality datasets, not “arms race” alt data ([44:03]):

  “There are informational edges and there are analytical edges...for us, it's using these sophisticated machine learning tools to tease out differentiated insights.” ([44:03])

• Model Longevity: 50 years of stock data used in training, surprising to some investors ([44:50]).

12. AI & Large Language Models

• Practical Caveats: Not currently using AI LLMs (like ChatGPT) for portfolio construction due to in-sample bias concerns; more promise seen in coding (“copilot”) applications ([46:27]):

  “It's not realistic to trust a backtest that ChatGPT generated...One area in AI...is the idea of software development copilots.” ([46:27])

13. Research Frontiers and Talent Acquisition

• Factor Ownership, AI for Productivity: Looking at factors related to stock ownership anew; watching AI’s impact on workflow ([47:57]).
• Team Building Challenge: Hiring is now harder, as data scientists are in high demand, but opportunities exist for great engineers due to softness in tech hiring ([49:08]):

  “Twenty years ago, talented data science oriented programmers were not in demand by every single other firm in the entire world...We've also tried to adapt and be more flexible.” ([49:08])

14. Enduring Motivation & Life Lessons

• Never “Solved”: The excitement is the market’s unpredictability and the endless opportunity for technical improvements ([50:12]):

  “I've been at this for 23 years. We have learned a lot and improved our models...But you're never going to solve the financial markets.” ([50:12])

• Learning from Setbacks: Daniel reflects on competitiveness and resilience as a personal journey, a lesson he’s also sharing with his children ([53:47]):

  “I wish I knew earlier on that life is a journey and that no one wins everything...doors that seemed closed open in time.” ([53:47])

Memorable Quotes

• On Emotional Bias in Quant Models ([00:00]):

  “This is precisely why this strategy works: because even quantitative investors who are intentionally trying to buy these stocks find it hard to overcome the human emotions involved with buying a bad story.” — Daniel Mahr

• On Overfitting vs. Underfitting ([13:51]):

  “Figuring out what techniques can allow us to strike the right balance between having a model that's too complex versus having a model that's not complex enough is something that we have put a lot of thought into and evolved significantly over the decades.”

• On Analytical vs. Informational Edge ([44:03]):

  “We have intentionally focused not on that arms race, but on the analytical piece through the use of decision trees and machine learning...for us, it's using these sophisticated machine learning tools to tease out differentiated insights about companies to have differentiated alpha sources.”

• On the Limits of Large Language Models ([46:27]):

  “It's not realistic to trust a backtest that ChatGPT generated...One area in AI...is the idea of software development copilots.”

• On Team Building ([49:08]):

  “Twenty years ago, talented data science oriented programmers were not in demand by every single other firm in the entire world. We've also tried to adapt and be more flexible...”

Timestamps for Key Segments

• Daniel’s Quant Origin Story: [03:51] – [06:30]
• Evolution of MDT’s Investment Process: [08:13] – [10:33]
• MDT Philosophy & Diversification: [10:49] – [12:27]
• Machine Learning, Transparency, and Pitfalls: [12:41] – [16:55]
• Model R&D Process: [17:55] – [22:02]
• Decision Trees in Practice: [24:48] – [27:38]
• Portfolio Construction: [32:34] – [34:55]
• Human Judgment vs. Model Decisions: [35:21] – [36:47]
• Market Crowding and Reflexivity: [36:57] – [39:51]
• Daily Workflow at MDT: [41:31] – [44:03]
• Alternative Data & Analytical Edge: [44:03] – [46:27]
• Current Research and Team Building: [47:57] – [50:12]
• Personal Reflections & Life Lessons: [53:03] – [54:27]

Notable and Memorable Moments

• The tale of physically taping the entire decision tree to the trading room wall in the early days—a striking visual of model transparency ([25:16]).
• Daniel’s candid admission that even seasoned quants can feel deep discomfort when a model recommends “buying a bad story” stock—highlighting the persistent role of emotions, even in systematic investing ([00:00], [16:55]).
• His pragmatic stance on alternative data: “We don’t have to win the arms race,” focusing instead on long, clean, traditional datasets ([44:03]).
• Practical skepticism about current uses of AI/LLMs in investment models, preferring their application to workflow improvements ([46:27]).

Closing Notes

Daniel Mahr’s conversation is a rare, candid, and technical glimpse into the evolution of systematic investing at an institution with decades of track record. The episode demystifies the quant process, highlighting the power (and perils) of machine learning, the critical importance of interpretability, and the timelessness of human judgment in financial decision-making.