What this covers

The conversation pairs host Patrick with Leigh Drogen to compare systematic and discretionary investing. It starts with process. Drogen cites a study from Richard Thaler, described in Michael Lewis's The Undoing Project, in which oncologists failed to follow their own X-ray rubric. He also argues that research should start from an ex-ante hypothesis and that a strategy should be checked out-of-sample by fitting on one half of the data and testing on the other. Much of the middle covers the mechanics of earnings-based trading: why sell-side consensus misstates real expectations, how buy-side managers trade whisper numbers, and what post-earnings drift looks like.

The range is wide. Patrick presses on value investing as the original quant factor and on whether faster corporate turnover, with AutoZone-style names screening as cheap, makes the factor a disruption bet. They discuss why timing factor exposure tends to end badly. They also cover how a dataset's edge rises, crests with crowding and falls, and why data vendors often sell before crossing the trough. The IBES estimate dataset, which took about 30 years to be arbitraged, is the counterexample. Other threads include platform firms such as Millennium, Balyasny and WorldQuant building centralized data and risk infrastructure, and whether stat-arb funds supply liquidity that market makers withdraw.

The contested stretch is Drogen's case for a human-machine hybrid. He says a single z-score across all stocks cannot see that a signal fails in sectors like industrials. He also says sector specialists can size concentrated bets that diversified quants cannot. He cites a human-assisted AlphaGo result as support. The episode closes on trend-following rules for passive investors, which Drogen would like Betterment to offer, and on Numerai's crowdsourced modeling.

Sharpest takeaway

Leigh Drogen argues that alpha is migrating from informational edges that get arbitraged away toward behavioral and process-driven edges, and that discretionary managers can beat pure quant models by combining codified decision processes with sector expertise, while factors that once produced alpha become betas.

  • Informational edges decay through crowding, so the lasting edge is in decision-making discipline and behavioral inefficiencies
  • Pure systematic models fail to normalize across sectors, leaving room for human expertise to add value
  • Factors that persist become betas, and attempts to time beta exposure tend to blow up investors

The argument · threads3 threads · 80 claims
0.85

Factor performance should be judged by behavioral rationale and return distribution, not recent drawdowns or timing.

6 pointscentrality 5/5
  • Momentum strategies stopped working around late 2007 for about six months before recovering, and in 2009 momentum investors who relied on it, especially exclusively, were destroyed by a complete inversion whipsaw; only about five periods in the last 80 years were as bad for momentum investing.

    “Any quantitative manager that used momentum at all, especially those that used it exclusively got absolutely destroyed in 2009 with a complete inversion whipsaw.”

  • Patrick proposes thinking about signals in terms of half-lives, and says the test of whether a factor is irrevocably broken is whether its ex-ante behavioral reason still holds; value, grounded in behavioral overreaction, should persist even through seven or eight years of underperformance if the investor has a long horizon.

    “And my answer to that question for value is the ex-ante reason why it works, which is well, this is a behavioral phenomenon.”

  • Attempting to time factor exposures is extremely hard, and most investors who do it get caught on the wrong side and blow up; equal weighting robust factors and leaving them alone outperforms over-fitted timing tactics.

    “Whenever we've looked at it, the answer is you can't.”

  • Patrick argues that the distribution of volatility and factor returns within a long period matters as much as the average, so a factor's base rate of outperforming in three-year windows could fall from 80 to 60 percent even if average excess return is similar.

    “I think that the other variable here is lumpiness. So if there is on a 20-year period average amount of vol, how that's distributed within that period is interesting.”

  • Value investing has underperformed for an extended period, roughly seven to eight years, which tests the patience of even disciplined investors.

    “you go through periods we're in one right now where value investing stinks for a long period. It could be, I think in this case, seven, eight years of under-performance”

  • Value investing is the original quant factor, and it works because of a behavioral mechanism in which markets overreact to glamour and despondency, so it should persist as long as people price securities.

    “the basic idea is that at extremes, markets overreact. So they get too excited about glamor stocks. They get overly despondent about cheap stocks.”

0.71

Informational edges decay under competition, so investors must keep adapting or be displaced.

6 pointscentrality 4/5
  • Dmitry Balyasny has argued that a long-short manager a decade ago could succeed with a couple of good analysts and a chief operating officer, whereas competing today requires heavy investment in technology, infrastructure, data, recruiting, corporate access, portfolio finance, compliance, investor relations, and trading.

    “A long-short manager, 10 years ago, might have been okay with a couple of good analysts and a chief operating officer. Competing today requires a significant investment in technology, infrastructure, data”

  • The average lifespan of a company in the S&P 500 has fallen from roughly 50 to 60 years to 12 to 15 years, with cyclical swings but a clear secular downward trend.

    “the rolling lifespan of a company in the S&P 500 that shows this... There are some cyclical moves, but the secular trend is very clear where it was 50, 60 years now it's 12 or 15 years.”

  • AutoZone and other retail names score well on systematic value screens based on capital allocation and buybacks, so the value factor may be picking up exposure to disruption risk, which raises the question of whether accelerating technological change changes the value investing story.

    “Is how do you price disruption risk? And how does that affect value investing?”

  • Numerai crowdsources machine learning models on a secret, normalized dataset of about 240,000 observations with a binary outcome, and compensates contributors in a cryptocurrency called Numeraire whose value rests on the quality of their models, so compensation is tied to correlation with outcomes rather than cash.

    “So you get paid in what's called Numeraire.”

  • Informational edges keep appearing, and each one that once generated outsized alpha becomes table stakes once most players buy it, so the market is a continuous arms race and new data sets like the car, location, and distress-court datasets always emerge.

    “everybody has to buy it so it's like table stakes. But the alpha can go away very quickly in the informational game.”

  • Older discretionary firms fail to adapt to changing sources of edge, and progress in the industry occurs one retirement or death at a time, as in Kuhn's model of scientific change.

    “progress happens one retirement or death at a time. That some of these older firms are just not going to adapt.”

0.24

Repeatable process discipline matters, because automated investors cannot hold steady when markets crash.

2 pointscentrality 2/5
  • Patrick argues that the only interesting investment strategies are repeatable ones that follow a specific process, citing Atul Gawande's The Checklist Manifesto, and believes fundamental investors can gain advantage by strictly following a quant-like decision process with subjective fundamental inputs.

    “my belief that as an investor, the only interesting strategies are ones that are repeatable. And therefore follow a pretty specific process.”

  • Patrick argues that market bottoms happen because everyone sells, so in a severe downturn, passive and automated investors will drag everyone down, and no advisor or robo platform can manage client behavior through a real crash.

    “I think of the other way round bottoms happen because everyone's selling. So I think by definition, if we're in a really bad situation, people are going to drag the rest of us down with them.”