This interview says making money in investing is getting harder because too many quant traders turn profitable strategies (Alpha) into common knowledge (Beta). Guest Leigh Drogen argues that human stock-pickers can still beat pure machines by combining quant tools with industry expertise, especially in sectors like industrials. He highlights three holdings: AutoZone is bearish (electric cars and self-driving will kill DIY repairs), Zynga is risky (fast innovation may disrupt its business in 5 years), and Apple is neutral (market underestimates iPhone sales resilience).
At a Glance This episode of Invest Like the Best features Leigh Drogen, founder of Estimize, who discusses the shift from discretionary stock picking to quantitative strategies in investing. The core argument is that alpha is becoming increasingly difficult to capture, as the influx of quantitative
Leigh Drogen, founder of Estimize and former statistical arbitrage portfolio manager, discusses the transition from discretionary stock picking to quantitative strategies in the investment field. Core judgment: Alpha returns are becoming increasingly difficult to capture, as the influx of quantitative investors rapidly converts Alpha into Beta; any potential Alpha edge must be separated across three dimensions—data, execution, and model—while discretionary managers can still outperform systematic strategies through human judgment.
Drogen argues that once-effective statistical arbitrage strategies are being "arbitraged away" by the influx of quantitative investors, rapidly converting Alpha into Beta.
Drogen begins with his own experience: at Geller Capital, he used the IBIS (sell-side analyst earnings forecast) dataset to construct a market-neutral strategy—standardizing the entire market via Z-scores, going long the top decile and short the bottom decile. This strategy exploited inherent biases in analyst forecasts: institutional prejudice, investment bank pressure, and compliance hurdles that prevent analysts from updating forecasts weeks before earnings, while also encouraging companies to "beat" lowered expectations.
"We were essentially arbitraging inefficiencies in the sell-side analyst dataset." The strategy performed exceptionally well in 2006–2007, but suffered a devastating reversal in 2009—one of the worst five periods for momentum strategies in 80 years. Drogen’s mentor had moved entirely to cash by mid-2008, manually trading index futures to precisely capture subsequent volatility, but ultimately closed the fund.
Key data point: The 2009 shock to momentum strategies was an extreme event occurring only five times in 80 years.
Drogen’s framework: Any Alpha edge should be decomposed into three dimensions—information edge (being rapidly arbitraged away), analytical edge (scarcity of quantitative talent), and behavioral edge (most critical for discretionary managers).
Drogen explains that Estimize, by aggregating crowdsourced earnings estimates (averaging hundreds of forecasts per company), provides hedge funds with consensus data that is more accurate and less biased than sell-side estimates.
Estimize collects EPS and revenue forecasts for U.S.-listed companies, covering approximately 2,100 companies (with at least 3 estimates), of which 1,400 have more than 10 estimates. For a giant like Apple, Estimize can gather around 1,000 estimates, whereas sell-side analysts typically provide only 30–40.
Key data point: The Estimize consensus is more accurate than the sell-side consensus 70% of the time. For small- and mid-cap stocks (where sell-side coverage is only 7–10 estimates), the accuracy rate rises to 75–80%.
Incentive mechanism: Contributors gain free access to others' data (via anonymous accounts). Drogen describes this as "cannibalizing the industry" — the industry's annual revenue is roughly $400 million, and Estimize is compressing it toward $100 million.
Drogen's warning: If everyone buys Estimize data, the alpha within it will disappear. However, the alpha lifecycle of a dataset follows a bell curve — Estimize is still in its upward phase ("there is more alpha today than yesterday"), with an estimated 4–5 years remaining. If it can navigate the "trough" and become a "must-have arbitrage dataset" (like IBIS or short-sale data), it could grow into a $100 million annual revenue company.
Drogen argues that the most cutting-edge quantitative firms are pushing the concept of "crowdsourcing" to its extreme, shifting from centralized research to distributed model generation.
Drogen describes three models:
1. Traditional Platform Model (Millennium): Provides capital and risk control to "pods," but research is independent
2. Centralized Research Model (AQR): A single research team manages multiple portfolios
3. Crowdsourcing Model (WorldQuant): 500 global contract analysts independently generate alphas, and PMs select from the "alpha pool" to construct portfolios — PMs are not even aware of the specific alpha content
Numerai goes a step further: It provides anonymized datasets (240,000 data points, with both variables and outcomes processed), and global participants submit predictive models. Top performers receive cryptocurrency (Numeraire) rewards. Drogen notes: "This corresponds to how we do Force Rank — we don't care about magnitude, only whether the ranking is correct."
Key Analogy: The AlphaGo team handed AlphaGo to ordinary chess players, and human + machine defeated pure machine. Drogen believes that discretionary managers can outperform systematic strategies in specific areas (e.g., industrials, utilities), because systematic strategies "hit singles," while discretionary managers can "hit doubles and triples."
Drogen argues that fundamental managers who combine quantitative tools with industry expertise can outperform pure quantitative strategies in specific areas.
Core case: the post-earnings announcement drift model. Drogen finds that industrial stocks react more weakly to earnings than other sectors (because industrials trade on cyclical earnings rather than quarterly expectations), but a pure quantitative model cannot distinguish this — it only outputs a Z-score. A fundamental analyst can "reduce weight in the industrial sector" to avoid false signals.
Key data: Quantitative strategies typically hold hundreds of positions, with maximum allocations below a low threshold; fundamental managers can hold a 10% single-stock position and significantly increase exposure when signals are "all green."
Drogen's framework: Fundamental managers should:
Falsification condition: The hybrid model fails if fundamental managers cannot strictly adhere to the decision-making process (e.g., as in Michael Lewis's The Undoing Project, where oncologists did not use their own scoring criteria).
Drogen believes the market is undergoing a structural shift: more data, fewer "dumb money" participants, and lower volatility — but the next crisis may originate from the collective behavior of passive investors.
Recalling the 2010 Flash Crash: At the time, Drogen was at StockTwits and observed the market "leaking" for several consecutive days, eventually resulting in "no bids." His mentor once taught him: "After a crash, the bottom will always be retested." The index did indeed retest its lows weeks later, as long-term funds like Fidelity had placed buy limit orders at that level.
Key Judgment: Algorithms and automation do not necessarily increase the risk of unforeseen dislocations — market makers shut down systems during volatility, while statistical arbitrageurs (StatArb) profit from volatility, thereby adding liquidity. The real question is: "If 75% of capital is passive, who will shout 'buy' during a crash?"
Drogen's Hypothesis: Lower volatility may stem from:
Falsification Condition: If users on platforms like Betterment massively drag their risk sliders to zero during the next crash, passive capital could become a volatility amplifier.
| Position | Guest Stance | Key Data |
|---|---|---|
| AutoZone | Bearish (long-term zero) | Vehicle electrification + autonomous driving will eliminate DIY repairs; Drogen believes "you won't drive a car in 5-10 years" |
| Zynga | Risk Warning | Innovation cycle accelerating, business model could be disrupted within 5 years |
| IBM | Risk Warning | "Classic value stock," but Drogen believes it struggles to innovate out of its current cycle |
| Apple | Neutral (mentions market misjudgment) | Market consistently underestimates iPhone sales resilience |
| Chipotle | Not explicitly stated | Used as a credit card data case study: different funds draw opposite conclusions from the same data |
1. “Alpha is turning into Beta, and the pace is accelerating.” (Drogen) — The IBIS dataset took 30 years to be arbitraged away; Estimize may only take 10-15 years. The alpha lifecycle of a dataset follows a bell curve, and only by crossing the “valley of despair” can it become a “must-have arbitrage dataset.”
2. “Human plus machine can beat pure machine.” (Drogen) — After the AlphaGo team handed the AI to ordinary chess players, they defeated the pure AI. Fundamental managers can outperform systematic strategies in specific areas (e.g., industrials, utilities) because systematic strategies “hit singles,” while fundamental managers can “hit doubles and triples.”
3. “A priori assumptions are the starting point of quantitative research — if you don’t know why it works, you won’t know why it fails.” (Drogen) — Drogen criticizes many products for skipping this step, leaving them unable to exit in time when a strategy fails.
4. “Oncologists don’t use their own scoring criteria, and neither do fundamental managers.” (Drogen, citing The Undoing Project) — Experts deviate from the rules they themselves set in decision-making. Drogen believes the greatest advantage of fundamental managers is their “behavioral edge” — strictly adhering to the decision-making process.
5. “After a crash, the bottom is always retested.” (Drogen’s mentor) — After the 2010 Flash Crash, indices retested the lows within weeks because long-term funds like Fidelity had buy limit orders set at that level.
6. “WorldQuant’s PMs don’t know what’s in the alpha pool — it’s infinitely scalable.” (Drogen) — 500 global contract analysts independently generate alphas, and PMs select from the pool to build portfolios. Drogen considers this the most cutting-edge quantitative organizational model.
7. “The next crisis for passive capital: Will Betterment users drag the risk slider to zero?” (Drogen) — If passive investors panic collectively during a crash, the market may lose its stabilizer. Drogen argues that “bottoms are formed because people sell.”
8. “AutoZone is zero — in 5-10 years, you won’t drive, and Tesla will fix it for you.” (Drogen) — Drogen believes AutoZone is a classic value trap, as technological disruption (electric vehicles + autonomous driving) will destroy its business model.