Nov 20, 2025 · 55m · capital-allocators
Daniel Mahr – Glass Box Quant at MDT Advisers (EP.472)
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Ted Seides interviews Daniel Mahr, Head of MDT Advisers at Federated Hermes, exploring how their proprietary 'glass box' machine learning framework uses decision trees and multi-decade fundamental data to eliminate emotional bias and generate consistent equity alpha.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Ted holds 18.8% of the talking time here. How this is scored →
Ted as informed peer 3.5 Guest teaching 3.8 Guest disagreement 0.4 Ted pushing back 0.3
speaking balance: gold is Ted, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 36:57 Dismantling the quant blowup myth
Dan forcefully reframes the narrative around quant blowups like LTCM and the 2007 quant quake, arguing that leverage—not quantitative methodology—was the true culprit.
Hardest push from Ted ▶ 15:16 Pressing on machine learning explainabilityTed challenges Dan on whether machine learning models can truly be a 'glass box' given that complex algorithms naturally identify relationships humans cannot readily interpret.
Biggest teaching moment ▶ 46:27 LLM backtest contamination critiqueDan provides a clear technical critique of why off-the-shelf LLMs fail at backtesting equity returns due to unconstrained in-sample training and look-ahead bias regarding multi-trillion dollar winners.
Ted holds their own ▶ 39:30 Connecting leverage dynamics to hedge fund pod shopsTed demonstrates industry insight by linking the historical discussion of leverage and crowded factor risks directly to the modern growth of hedge fund multi-manager pod shops.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Ted as informed peer | Guest teaching | Guest disagreement | Ted pushing back | Why |
|---|---|---|---|---|---|---|
| Thanksgiving Reflections and Capital Allocators Team Tribute | 2 | 1 | 0 | 0 | The segment begins with Ted's Thanksgiving monologue and team appreciation before transitioning smoothly into Dan's college background flipping tech IPOs. The tone is entirely friendly, introductory, and collaborative. | |
| Early Career at MDT and the Evolution of Quant Investing | 3 | 4 | 0 | 0 | Ted asks foundational questions about the history and evolution of quant investing. Dan educates on how the industry transitioned from simple factor tilting to modern machine learning decision tree models. | |
| Investment Philosophy: Analytical Edge and Diversified Alpha | 4 | 4 | 1 | 1 | Ted probes into MDT's investment philosophy and differentiation against other quants. Dan emphasizes their 24-year head start using machine learning and their focus on diversified analytical edge rather than macro forecasting. | |
| The Glass Box Model and Discovering Price Reversals | 5 | 5 | 1 | 2 | Ted pushes on the tension between black-box machine learning and explainability. Dan explains MDT's 'glass box' model and shares an illustrative anecdote where the model identified counterintuitive 70-80% price reversal trades. | |
| Idea Generation and Long-Term Empirical Observations | 4 | 5 | 0 | 0 | Dan walks through how research ideas originate from 30+ years of empirical observations and introduces unconventional factors like 'company age' that provide contextual structure rather than standalone alpha. | |
| Balancing Human Hypothesis with Algorithmic Factor Selection | 4 | 4 | 0 | 0 | Ted asks how factors are systematically tested and removed. Dan details the removal of book-to-price due to the shift toward an intangible economy and the algorithmic verification process. | |
| From Single Trees to Forest: Solving Data Scarcity | 4 | 5 | 0 | 0 | Ted visualizes peering into the glass box, prompting Dan to explain the mathematical necessity of using a forest of shallow trees rather than one deep tree to avoid sample size degradation. | |
| Walkthrough of Tree Questions and Stopping Rules | 5 | 4 | 0 | 0 | Dan provides a concrete walkthrough of a tree branch starting with financing behavior and moving downstream into price momentum and volatility conditioning. | |
| Portfolio Construction, Optimization, and Trading Cost Modeling | 4 | 4 | 0 | 0 | Ted asks how thousands of individual tree outputs translate into an investable portfolio. Dan outlines the proprietary optimizer balancing alpha, tracking error, and explicit/implicit trading costs. | |
| Managing Market Impact Across Market Capitalizations | 4 | 4 | 0 | 1 | Ted asks about the role of human judgment in overriding quantitative models. Dan clarifies that overrides are handled through a strict data-centric lens rather than emotional discretion. | |
| Quant Reflexivity, Crowdedness, and the Risk of Leverage | 5 | 5 | 2 | 1 | Ted asks about quant reflexivity and crowdedness. Dan pushes back on the common narrative that quant models are inherently fragile, arguing that historical blowups (LTCM, 2007 quant quake, Archegos) were driven by leverage rather than quant methods per se. | |
| Changing Market Microstructure: Pod Shops, Passive, and Retail Flows | 4 | 4 | 1 | 1 | Ted brings up multi-manager pod shops. Dan observes that equity markets have experienced an unexpected resurgence in inefficiency over recent years, possibly driven by passive indexing, retail flows, and pod shops. | |
| Daily Operations at MDT: Overnight Data to In-House Research | 3 | 4 | 0 | 0 | Dan walks through the daily operating cadence at MDT, from overnight data downloads and automated re-optimization to trade review and fully in-house research tooling. | |
| Analytical Edge vs. Informational Edge in Data Strategy | 4 | 5 | 1 | 0 | Ted inquires about alternative data sets. Dan articulates MDT's deliberate strategic choice to focus on analytical edge over long-horizon, high-quality data (50 years) rather than participating in the expensive alternative data arms race. | |
| Evaluating Artificial Intelligence: LLM Limitations vs. Coding Co-Pilots | 3 | 5 | 1 | 0 | Ted asks about ChatGPT and modern LLMs. Dan warns against using commercial LLMs for stock backtests due to pervasive look-ahead bias and training leakage, while praising AI coding co-pilots. | |
| The Talent Landscape: Recruiting Engineers in an AI-Driven Market | 2 | 3 | 0 | 0 | Ted asks about current operational challenges. Dan describes the shifting talent dynamics between data scientists and software engineers in the current tech hiring cycle. | |
| Closing Questions: Mentors David Goldsmith and Sarah Stahl | 2 | 2 | 0 | 0 | Dan reflects warmly on his two foundational mentors at MDT, David Goldsmith ('the mad scientist') and Sarah Stahl ('the meticulous craftsman'). | |
| Closing Questions: Managing Competitiveness and Perspective on Setbacks | 1 | 1 | 0 | 0 | Dan shares reflections on managing personal competitiveness, learning resilience from setbacks, and concludes the interview followed by standard legal disclaimers. |
Statements from this episode (26)
Mahr: Dot-com brokerages distributed hot IPO shares first-come, first-served
“There were a lot of IPOs in that market environment, the dot-com bubble. They would go up a hundred percent, 200% or more on the day that they priced, and there were a small number of investment firms that would get allocations to these IPOs, and they would of…”
Mahr: MDT adopted decision trees after factor tilts struggled in 1998-1999
“In 2002, we had made a big transition at MDT. For the first decade, the strategies were traditional factor tilting strategies. There was a formula that used a small number of characteristics and the portfolios would tilt toward them. Those strategies generated…”
Mahr: MDT Advisers has used machine learning tools since 2001
“At MDT, we've been using these machine learning tools since 2001. So we have a 24 year head start on someone who is new to the game.”
Mahr: MDT runs 'glass box' models with fully traceable daily decisions
“At MDT, that is not the case. We like to position our investment strategies as being a glass box. There's a lot of machinery on the inside, but we can see into it. We can see how it's working and understand what's driving all of the decision making on a day-to…”
Mahr: Stocks down 70-80% yield strong returns when combined with value
“A number of years ago, we started adding price-based factors to our model, and the price-based factors found momentum effects, as was published in the academic literature and as we fully expected to see, but it also found some very powerful reversal effects wh…”
Mahr: Deep reversals work because investors emotionally avoid 'bad stories'
“This is precisely why this strategy works is 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.”
Mahr: Most published finance research fails to replicate or add value
“More often than not, we test an idea and either it's not replicable when we look at it with our data set or something else in our model essentially captures the same underlying effect.”
Mahr: MDT uses company age since IPO as a quantitative model factor
“One of the most unusual factors that we use, we call company age. We measure that simply as how long has the company been publicly traded and or filing financial statements.”
Mahr: Valuation matters far more for mature companies than new entrants
“Valuation is a lot more important for companies that have been around for a long time than a brand new entrant to the public markets.”
Mahr: Factor selection and interaction at MDT is 100% algorithmic
“The selection of factors is driven by the potential questions that can be asked, is driven by the investment team. That's a major area of focus for us on the research side. Once we present that list of factors to the algorithm, it's Completely mechanically det…”
Mahr: Book-to-price lost explanatory power as the intangible economy expanded
“We used book to price in our models. Going back to version one point O in 1991. But as markets evolved and more importantly, as the economy has changed, we saw less and less explanatory power to incorporating that in our model. And we had an intuitive sense of…”
Mahr: MDT originally audited trades using a paper tree taped to the wall
“Back in the OneTree day, we would print out the tree and tape it on the wall of our trading room. Every time we were reviewing a trade, we would simply walk through the sequence of questions On that paper tree on the wall to help inform what specifically was m…”
Mahr: MDT limits decision trees to two to five questions to avoid fragmentation
“Typically we ask between two and five questions in each tree. The reason we don't ask more questions is we found that as you ask questions deeper and deeper in the tree, you're working on smaller and smaller pools of data because the trees are customized to th…”
Mahr: Ensemble forests of shallow trees solve decision tree data scarcity
“You can quickly see that there's a sharp limit to how deep you want to make these trees. Fortunately, we have another approach to asking more questions about companies, which is rather than relying on a very deep tree, Relying on a forest of relatively shallow…”
Mahr: High-financing companies tend to underperform the market
“We find, as the academics have, that companies that are engaged in significant amounts of financing tend to underperform, and those that don't have better outcomes.”
Mahr: Strong Momentum Offsets Underperformance of High Financing
“And generally speaking, it's the strongest momentum companies that can generate good outcomes regardless of the financing.”
Mahr: Consistent Price Appreciation Yields Better Momentum Signals
“We tend to find momentum works better when it is consistent. When the stock price is rising in a consistent manner, it leads to better outcomes than companies that have one giant price move driving the momentum measurement.”
Mahr: Momentum Predicts Future Returns Better for Newer Companies
“We find that momentum typically is more meaningful when you're looking at newer companies than companies have been around for a long time.”
Mahr: LTCM Would Not Have Failed Without 50x Leverage
“If long-term capital management hadn't been running a fifty-x leveraged strategy, They wouldn't have ended up in the trouble that they ended up with.”
Mahr: 2007 Quant Quake was driven by crowding and added leverage
“The run on quant strategies that happened It was predicated by the fact that statistical arbitrage strategies had gotten more crowded over the years leading up to 2007. In response to that, certain managers began running those strategies with additional levera…”
Mahr: Decades-long trend toward market efficiency broke in recent years
“It does feel like the markets are different in the last couple of years than they were a decade ago. If you asked me five years ago, are markets on a never ending trend towards efficiency? And is your job as a systematic investor It's going to get harder and h…”
Mahr: MDT re-optimizes every portfolio nightly using decision tree forests
“Every night we download updated data from all of our vendors. We recalculate all of our characteristics. We run all the companies in the domestic equity market through our forests and have updated forecasts. Every portfolio that we run is re-optimized and gene…”
Mahr: MDT trains its machine learning models on 50 years of data
“We train our models on roughly 50 years worth of data, which I say that to some potential investors and they're surprised. We think that market data from the 19 seventies and eighties is still useful for forecasting mispricing.”
MDT Advisers does not use LLMs or ChatGPT in investment modeling
“Large language models and ChatGPT specifically are not anything that we're presently making use of in our modeling.”
Mahr: ChatGPT stock backtests are untrustworthy due to look-ahead bias
“When you're running a back test through the better part of the last decade, ChatGPT knows that Nvidia became a multi-trillion dollar company. ChatGPT knows what the mega trends were in the economy and the market over those timeframes. It's not realistic to tru…”
Mahr: Traditional software engineering demand softened while AI talent soars
“In the same way that folks with data science backgrounds and AI knowledge are super in demand, the software programming space has hit a little bit of a soft patch. There's a lot of opportunities to hire great engineers these days.”