Mechanisms for Honest Values

Harberger insurance and prediction markets

Two scenes, apparently unrelated. In the first, a defamation suit: the plaintiff swears under oath that the reputation destroyed was worth millions; the defendant swears it was worth next to nothing; both numbers were manufactured for the occasion, both sides know it, and the court spends years and fortunes splitting the difference. In the second, a television panel: a pundit predicts with total confidence, is wrong, and is back the following week predicting with total confidence, because nothing in his world attaches any cost to having been wrong.

These look like different failures — one about valuing, one about forecasting. They are the same failure. In both cases an institution asks people to report something private — what a thing is worth to them, how likely they believe an event to be — inside an incentive structure where the honest report costs money and the distorted one pays. Then the institution acts surprised when it harvests distortion. You cannot exhort your way out of this. Preaching honesty to people who are paid to lie is not a mechanism; it is a mood. The serious response is to redesign the game so that directional distortion carries a cost. This chapter examines two mechanisms designed to do that: one for valuation reports, one for beliefs.

The Self-Assessment Trick

Start with the older idea, because it contains the trick in its purest form.

A Harberger tax works like this: you declare, publicly, what your own property is worth, and you pay tax on the value you declared. The catch that disciplines the report: you must stand ready to sell at that declared price to any buyer who shows up. Declare high and the tax rises; declare too low and you risk a forced sale below your reservation price. The opposed pressures make strategic distortion costly. Under specified tax rates, sale rules, liquidity, transaction costs, and risk tolerance, the cheapest report can move toward an actionable reservation price. It does not reveal a context-free “true valuation,” and a constrained owner may still report a number shaped by the mechanism.

Notice what has not happened here. No assessor has divined the “objective value” of your property, because there is no such thing to divine — value is always value to some agent. The mechanism does not extract a cosmic truth or the owner’s complete preference ordering. It elicits a self-assessed sale price under stated rules by making reports on either side costly in different ways.

Insuring the Intangible

Now point the same trick at harm.

When harm occurs — property damage, reputation loss, health impairment — society faces a perennial problem: quantifying the damages. Current practice estimates harm after the fact, which is precisely when every incentive is at its worst. The victim, post hoc, has every reason to exaggerate; the party who caused the harm has every reason to minimize; there is no pre-agreed benchmark to arbitrate between them, so the gap gets closed by litigation, which is slow, expensive, adversarial, and arbitrary. Everyone leaves poorer and angrier, except the lawyers.

Harberger insurance moves a coverage figure to before the harm. You assign coverage to assets — including the intangible ones that courts handle worst: reputation, privacy, intellectual property, digital identity, even personal health — and pay premiums against that declaration. Inflate the figure and you pay more for a loss that may never occur. Lowball it and any covered payout is correspondingly low. The tradeoff pressures the report toward a level the owner is prepared to finance, but income, liquidity, risk tolerance, beliefs about loss, the premium schedule, and coverage limits still shape it. The mechanism elicits an insurable valuation under those conditions, not everything the asset means to its owner.

Three properties can follow from the design. First, pre-committed coverage: the number exists before the dispute and narrows later argument about quantum, though causation, scope, fraud, and whether a covered loss occurred may remain contested. Second, incentive-compatible reporting: the premium schedule makes exaggeration costly without requiring an assessor to infer subjective worth. Third, faster claims: where the trigger and extent of damage can be established, the pre-agreed figure can reduce litigation cost, delay, and strategic reconstruction.

Imagine reputational harm from online defamation, a privacy violation, the distress of sustained harassment — each with a pre-established coverage benchmark rather than a damages figure invented only after the dispute. The deep change is not that society learns the true worth of intangible wealth. It gains owner-declared, condition-bound evidence that may resolve some claims faster while leaving nonfinancial meaning and uncovered harms outside the number.

The Pundit’s Dilemma

So much for values. The same disease infects beliefs, and there the stakes are larger, because beliefs steer policy.

Expert opinion is cheap to produce and difficult to falsify. A pundit can make confident forecasts that shape public discourse, move policy, and influence billions of dollars, and face no structured accountability whatsoever; the misfires dissolve into the churn of the news cycle. The asymmetry is brutally simple: the cost of being wrong is borne by others, while the rewards of being theatrically confident are kept by the pundit. Behavior follows incentives, and this incentive landscape selects systematically for performance over accuracy, coalition signaling over clarity, rhetorical dominance over model-building. Punditry is not a truth-seeking institution that happens to fail. It is an attention-seeking institution working exactly as designed.

Prediction markets are the countermeasure — the Harberger trick applied to belief. To assert a probability in a prediction market, you must buy or sell at the prices on offer, which means your confidence is no longer a costume; it is a position, and the position can lose. A prediction market forces belief to bear the cost of its own implications. It replaces performative epistemology with epistemology grounded in consequence.

Markets as Coherence Filters

What a prediction market price expresses is the balance of conditional judgments made under incentive pressure — not a prophecy, not a verdict, and not any claim to certainty. The price is indifferent to narrative flair, institutional prestige, and rhetorical force. It reflects one thing only: the relative performance of the competing models whose holders were willing to stake something. In this sense a prediction market is a coherence filter — a mechanism that extracts a structured signal from distributed noise by charging admission to incoherence.

The mechanics align naturally with everything the conditionalist volume established about probability. A market price is a conditional claim: if the offered odds are wrong, traders with better models profit by moving them, so a standing price is the frontier of collective inference under specified constraints — liquidity, information flow, the incentive structure, the set of models actually in circulation. Change the conditions and the meaning of the number changes with them. Nor is the price any single agent’s Credence; it is what happens when many Credences are forced to bid against one another with real stakes attached, each trader’s confidence disciplined by the possibility of paying for it. A price is never unconditioned truth. It is something more useful: the current output of a decentralized engine that continuously rewards whoever discovers a better pattern, and continuously taxes whoever clings to a worse one.

Calgary, 1994

I did not come to this argument as a spectator.

Prediction markets did not begin as speculative entertainment or a financial novelty. They began as an attempt to repair the epistemic failures of academia and public discourse. Through the late 1980s and early 1990s, Robin Hanson had been developing a proposal he called Idea Futures: a market mechanism for evaluating scientific claims. The premise was disarmingly simple — if you want to defend a scientific assertion, you should be willing to stake something on its accuracy. Peer review asks what the referees think; Idea Futures asks what you will bet.

I brought Hanson’s article to a discussion group I belonged to in Calgary, and it caught. In 1994 — when the web itself was barely out of its cradle — a small group of us, engineers, programmers, and researchers, decided to stop discussing the idea and build it. I served as coordinator and one of the main software developers: I helped shape the early design discussions, maintained the mailing lists that held the collaboration together, and worked on the implementation itself, including the early interface and infrastructure. What we shipped was the first public web-based prediction market — play-money, improvised, running on the enthusiasm of a handful of engineers and idealists — a working prototype of the systems that would later evolve into today’s crypto-based markets. In 1995 the project won the Golden Nica at Ars Electronica,¹ one of the earliest major awards given to an experiment in decentralized collective intelligence.

Two lessons from that project have held up across three decades, and they are the reason I trust this chapter’s argument at a level no literature review could supply.

First: prediction markets are technically simple and socially disruptive, in exactly that combination. Almost none of our real obstacles were engineering or economics. They were regulation and institutional discomfort — the quiet, persistent resistance of people and organizations whose authority rested on pronouncing likelihoods without ever being scored on them. The technology was the easy part in 1994. It is trivially easy now. The barrier was, and remains, that the mechanism embarrasses the wrong people.

Second: the value of a prediction market is the epistemic discipline it imposes on belief, not the forecast it emits. Ours ran on play money, and even play-money stakes — mere reputation among peers, a scoreboard anyone could check — changed how people asserted things. Claims got sharper. Confidence got quieter. Attaching consequence to a claim, any consequence at all, forces clarity in a way that no norm of intellectual honesty ever has.

Competitive Model Selection

When well-designed prediction markets outperform selected polls or experts, the advantage need not come from a mystical wisdom of crowds. Prices can aggregate heterogeneous information while rewarding correction. Performance depends on question resolution, liquidity, participant selection, subsidies, fees, manipulation, time horizon, and the comparison forecast. A trader can also profit from noise, hedge for non-epistemic reasons, or face limits to arbitrage; the incentive is useful without guaranteeing truth.

Models must compress reality, because hand-waving has no cash value. Confidence must scale with evidence, because overconfidence is expensive — the market enforces, automatically and impersonally, the calibration discipline that a lone reasoner must impose on himself by force of will. Belief updating becomes mandatory, because stubbornness is a losing strategy with a visible running total. And distributed information becomes usable: a market integrates heterogeneous data from thousands of sources without any central authority deciding whose input counts.

Trading, done seriously, is deliberate epistemic labor. The trader generates hypotheses, tests them against the market’s current state, discovers the mismatches between her model and the world, and refines the model through feedback — active inference, scaled across a population and paid for its accuracy. The pundit produces confident noise without consequence; the trader produces calibrated signal because the alternative comes out of her account. The market does not make anyone smarter. It makes incoherence costly and coherence valuable, and then lets selection do the rest.

The Thinness Problem

Honesty requires admitting where the mechanism fails. Prediction markets fail when they are thin — when liquidity is low or participation has been truncated by regulation or risk aversion. A thin market is fragile: it can be manipulated by a single motivated actor, and its prices carry weak signal dressed in the same authoritative-looking numbers as strong signal. Anyone who quotes a market price without asking how deep the market is has missed the point of the whole apparatus, which is that the number means something because of the conditions behind it.

Automated market makers, subsidies, and low-friction participation can reduce thinness while introducing model, subsidy, oracle, and platform risks. Legal pressure also has multiple stated grounds: gambling law, consumer protection, manipulation, insider information, sanctions, election integrity, and jurisdiction. Some restrictions may protect incumbent forecasters; that motive requires evidence rather than inference from the restriction itself. The institutional question is which rules preserve useful aggregation without pretending a market price is authoritative merely because it is numerical.

Honesty as Infrastructure

Set the two mechanisms side by side and the shared design principle is plain: attach consequence to the report. The Harberger policyholder who inflates pays premiums; the one who lowballs accepts a lower covered payout. The trader who overclaims can lose capital; the one who underclaims can leave profit for someone better calibrated. Neither mechanism adjudicates what a reputation is really worth or what the true probability is. Each makes some directional misreports costly, with evidential strength limited by the rules, participation, resources, and alternatives already described.

Understood this way, these are not financial curiosities. They are epistemic infrastructure. Prediction markets can serve as distributed inference engines for policy analysis, accountability mechanisms for experts and institutions, self-correcting scaffolds for scientific inquiry, and voluntary governance tools — the load-bearing instrument, in fact, of Axiocracy, governance by value discovery. Harberger valuation gives the asset side of the ledger a related treatment: declared coverage made explicit and potentially more liquid before disputes arise rather than after.

None of this comes with guarantees. The strength of these mechanisms is the strength of their participation, resolution rules, incentives, and safeguards. The Calgary experiment was an early demonstration that a web interface could support distributed forecasting and that even play-money scoring changed discussion. It did not establish comparative accuracy or a universal governance instrument without a published benchmark, sample, and evaluation protocol. The tools make claims more scoreable; whether they produce clarity is itself a claim that should be scored.