The death and life of prediction markets at Google
Prediction markets at Google — internal platforms where employees bet (mostly with play money and prizes) on future events — are used as tools to aggregate information and inform management decisions, but their value and limitations are hotly debated. Commenters argue over whether any prediction market can remain a neutral forecast once participants have incentives that might influence outcomes, drawing parallels to stock, insurance, and futures markets, as well as moral hazards and manipulation risks. Others highlight practical issues such as poorly chosen questions, operationalization problems, shallow liquidity, and the tension between using these markets for genuine forecasting versus treating them as gambling or reputation games.
Limits and Risks of Prediction Markets
- Many argue truly “pure” prediction markets can’t exist: once money or reputation is at stake, participants are incentivized to influence outcomes, not just forecast them.
- Self-referential effects are highlighted: markets can become about predicting their own impact or enabling manipulation (e.g., insider trading, match-fixing, “assassination markets”).
- Others counter that endogeneity is manageable in many domains (natural disasters, competitors’ behavior) and that markets can still be useful even if partly self-fulfilling.
Hedging, Insurance, and What Markets Measure
- Examples: election bets as portfolio hedges; weather futures and crop risk; insurance as a de facto disaster prediction market.
- Disagreement over whether such trades reflect beliefs, risk preferences, or simply hedging.
- Some say markets “determine” rather than “predict,” especially when design and regulation shape outcomes.
Corporate Prediction Markets at Google
- Internal markets used play money plus prizes (e.g., devices), not real downside risk; some see this as reducing legal and ethical issues.
- Markets were used both for internal milestones (hiring, project success) and competitor forecasts.
- A key tension: using market signals to change decisions can undermine their value as neutral predictions but increase their value as management tools.
- Reputation-based systems and “clout” are also seen as distorting, yet still motivating.
Forecasting Quality and Question Design
- Participants stress that many corporate/online questions are poorly chosen: they track the wrong entities, depend on fragile benchmarks, or hinge on resolution technicalities.
- Calibration charts from Google’s markets suggest prices can align reasonably with probabilities, but others note theoretical reasons why prices ≠ average beliefs.
Market Design, Incentives, and Volatility
- Discussion of using superforecasters or “sharps” and selling their aggregated signals separately from public odds.
- Platforms face a trade-off: stable, information-rich markets vs. volatile, gambling-friendly ones that attract more users and profit.
- Some believe intellectual, “academic” prediction markets struggle to compete with gambling-focused platforms.
Google Culture and “Pioneering” Practices
- Debate over claims that Google pioneered cafés, dogfooding, and A/B testing; several point to earlier corporate examples.
- Some suggest Google more “popularized” certain perks and practices, while others consider even that overstated.