← Back to list
Colossus (Invest Like the Best / Business Breakdowns)Podcast4 Sep 2018Source: traffic.libsyn.comHost: Patrick O'Shaughnessy

Richard Craib - Crowdsourcing Predictive Algorithms - [Invest Like the Best, EP.102]

In plain words

This interview covers Numerai, a company using blockchain and crowdsourcing for quantitative investing. Founder Richard Craib argues that letting data scientists model on blind data (stock info hidden) and staking NMR tokens (like a deposit) to filter good models beats relying on data or algorithms alone. He says staked models boosted the Sharpe ratio (risk-adjusted return) from 1.5 to 2.1. Key holdings: Numerai (NMR) – token up 10x this year; Renaissance Technologies – questioned if its edge is data, not algorithms; Quantopian – mentioned as contrast, requiring model code.

AI SummaryAI-generated · may contain errors · verify against the original

Numerai founder Richard Craib discussed on the program how to leverage crowdsourcing and cryptocurrency incentive mechanisms to bring together global data scientists to build better quantitative investment models. The core argument is that data scientists stake cryptocurrency to back the quality of

~12 min full read · 7 sections
Deep Analysis

At a Glance

Richard Craib is the founder and CEO of Numerai, a company that integrates quantitative investing, cryptocurrency, crowdsourcing, and machine learning. The central thesis of this issue is: by leveraging a staking mechanism on the blockchain, global data scientists are incentivized to "bet" on their own predictive models, thereby filtering out superior quantitative signals—Craib argues that the improvement in the Sharpe ratio (from 1.5 to 2.1) driven by this gamified design is more significant than the data or algorithms themselves.


Theme 1: The Core Mechanism of Numerai—Blind Data + Crowdsourced Modeling + Staking Screening

Richard Craib argues that making financial data "blind," opening it to global data scientists, and using a cryptocurrency staking mechanism to filter models is a system design unprecedented in quantitative investing.

  • Blind Data Design: Numerai normalizes stock market data (containing 50 feature dimensions) so that data scientists cannot identify specific stocks or feature meanings ("feature one isn't value or anything like that"). This solves the pain point of traditional quant funds being "afraid to share data"—data can be made public but cannot be used to build competing funds.
  • Crowdsourced Modeling: After downloading the blind data, data scientists use machine learning algorithms (neural networks, random forests, support vector machines, etc.) to fit the target variable (0 = decline, 1 = rise) and submit predictions. Numerai does not require submission of model code, only prediction values.
  • Staking Screening: Data scientists must stake Numerai's native token, NMR, to "support" their predictions. If predictions are accurate, they earn more NMR; if predictions are wrong, the stake is destroyed. Craib emphasizes: "If your predictions do well, then you earn more money. And if they do badly, then we destroy your stake."

Key Data: The Sharpe ratio for unstaked models is approximately 1.5, while for staked models it jumps to 2.1—Craib considers this "the biggest thing we've done by far." Approximately 600 stakes are placed each week, with cumulative staked amounts exceeding $1 million.


Theme 2: Data vs. Algorithms vs. Staking — Ranking Their Contribution to Alpha

Craib argues that data quality is foundational, but the "game theory" improvement introduced by the staking mechanism outweighs the importance of both data and algorithms.

  • Data: Numerai does not use "alternative data"; instead, it employs "very clean, very long, structured" traditional financial data (such as value factors, momentum factors, and other underlying features). Craib cites the view of Two Sigma's founder: "it's the data that's the most important thing."
  • Algorithms: Global data scientists use various machine learning algorithms, but Craib notes that the commonly used algorithms are actually only "a handful" (neural networks, decision trees, random forests, support vector machines, etc.). The real differentiation lies in "much more creative approaches"—some users employ highly unique methods that Numerai itself cannot fully understand.
  • Staking: Craib believes that staking is "the critical thing not really in the list of three things." It addresses the core problem of internet crowdsourcing—the extremely low cost of spam. By requiring data scientists to have "skin in the game," the system automatically filters for high-quality models.

Reader's Note: As the founder of Numerai, Craib naturally emphasizes the uniqueness of its staking mechanism. However, it should be noted that Numerai has not publicly disclosed its track record, so whether this model is truly effective remains unknown.


Theme 3: From "Market for Lemons" to "Staking Market" — A New Paradigm for Data Signal Trading

Craib argues that the traditional financial data signal market suffers from the "Market for Lemons" problem (Akerlof's "Market for Lemons"), and that the staking mechanism can resolve this market failure.

  • The Lemon Market Dilemma: If someone emails you saying, "I have a perfect model for predicting the stock market," you would not even reply — because "the probability that he's actually got something is so, so low." Good signals cannot be traded in the market because buyers cannot distinguish between genuine and fake ones.
  • Staking as a Signal: If that person says, "I've actually staked 10,000 NMR on the fact that it's good," the situation changes. Staking transforms "empty promises" into "costly commitments," allowing good signals to stand out.
  • Future Direction: Numerai is exploring "crowdsource the data part" — enabling external data providers to also stake NMR to vouch for the quality of their data. Craib believes this could unlock a massive market: "there's huge markets do not exist for I have signals for stocks."

Falsification Condition: If Numerai fails to attract enough external data providers to participate in staking, or if the staking mechanism cannot effectively differentiate good data from bad data, this direction may fail.


Theme 4: A Critical Perspective on the Quantitative Investment Industry

Craib argues that the traditional quantitative fund industry is stuck at a "Schelling point"—a locally optimal but globally suboptimal equilibrium.

  • Industry Status: "All these smart people from the top universities joining hedge funds that are identical to other hedge funds in every single way... buying the same data and none of them are sharing any knowledge."
  • Insights from Renaissance: Craib points out that Renaissance Technologies may not owe its success to algorithmic advantages, but rather to data advantages—"they just have the best data." Citing insiders, he claims that Renaissance does not use machine learning.
  • Numerai's Alternative: By combining blind data, crowdsourcing, and staking, Numerai aims to build a "distributed hedge fund"—a system never seen before. Craib describes himself as seeking to become "the Vitalik of hedge funds," thinking from the perspective of system design rather than profit generation.

Reader's Note: Craib's claim that Renaissance does not use machine learning comes from "some of the people who work there"—this is second-hand information, and given Renaissance's extreme secrecy, it cannot be verified.


Mentioned Positions

Position Guest Sentiment Key Data
Numerai (NMR) Bullish (founder's perspective) Staked model Sharpe ratio 2.1 vs unstaked 1.5; 600 stakings per week; cumulative payouts exceed $8 million; NMR up 10x year-to-date
Renaissance Technologies Neutral (mentioned for commentary) Reportedly does not use machine learning; data advantage may be its core
Quantopian Neutral (mentioned for comparison) Requires users to submit model code, differing from Numerai's "predictions only" model

Judgments Worth Remembering

1. “The power of the NMR staking, it's very hard to beat taking a simple average of the staked models.” — Craib found that the simplest equal-weighted average staking model outperforms any complex meta-model. The staking mechanism itself acts as a powerful self-filter.

2. “If you have a 52% edge, to turn that into 53% has a really, really big impact on your Sharpe.” — Craib explains why finance is particularly suited for crowdsourcing: a tiny improvement in win rate can lead to a massive improvement in the Sharpe ratio (the difference between 1 and 2), while other fields (e.g., healthcare) lack this leverage effect.

3. “We are not trusting our users with our data. We're totally obfuscating it. Why should they trust us with their models?” — Craib explains why Numerai does not require users to submit model code, only predictions. This “mutual distrust” design actually creates a healthier incentive structure.

4. “The easier they made it to withdraw the money that you had in PayPal, the less people would withdraw it.” — Craib cites the PayPal case to illustrate that “low exit costs” actually increase user stickiness. Numerai allows users to leave at any time, which paradoxically makes them more willing to stay.

5. “Renaissance... maybe they just have the best data.” — Craib challenges the market narrative that “Renaissance relies on the smartest people,” arguing that its core advantage may be data (long-term collection and storage of stock data) rather than algorithms. This is a counterintuitive judgment.

6. “If you're an alien coming to earth... capital allocation is a super important problem... the way that the humans have set this up is... a huge, weird, zero-sum-esque game.” — Craib critiques the traditional hedge fund industry from an alien perspective: all funds buy the same data, do not share knowledge, hire similar talent, and are stuck in a “Schelling point” (locally optimal but globally suboptimal).

7. “It only took a couple of weeks for the average person to stake in 4.25 of the tournaments.” — After Numerai launched five parallel tournaments (with different target variables), data scientists adapted quickly. Craib believes this demonstrates the flexibility of the crowdsourcing model: traditional funds take years to enter new asset classes, while Numerai can simply “stick it on and see what people can do with it.”

8. “Maybe blockchain isn't about consumer things like CryptoKitties, but maybe it's about enabling weirder companies that couldn't exist like Numerai.” — Craib argues that the true value of blockchain may not lie in consumer applications, but in enabling organizational forms that cannot exist under traditional legal/business frameworks (e.g., a distributed hedge fund).