distbit - fut/acc @distbit0

underspecified futarchy proposals create adverse selection for prospective counter-traders. if a proposer hides counterparties or performance terms, selling PASS-asset is exposed to the chance that the hidden proposal details make it genuinely +ev, and will be revealed later. that lets -ev proposers imitate +ev proposers who also strategically delay info disclosure (to increase their trading profits on their own proposal's pass market), reducing corrective sell flow against -ev proposals, without proving the proposal is +ev.

@ryanchern related: chatgpt.com/s/t_6a43af947c…

Adverse selection is hard to measure robustly: the more control or info-asymmetry an insider has, the easier it is to make market activity look organic, via fake volume, depth, or price moves, because others fear calling their bluff and making it costly.

Keynesian stimulus treats recession as a spending gap for government to fill. Say’s Law says demand comes from exchange: you can buy what others produce only by first producing something they value. If sales collapse, the question is what broke that exchange: price or interest-rate controls, impaired banks, or malinvestment. By replacing private demand with public purchases or subsidies, state spending can keep labour and capital committed to production plans that buyers no longer value. Recovery requires prices, credit, labour, and capital to shift toward goods and services people will actually trade for.

@ryanchern would be interested in how this is operationalised :) similar to happenstance.ai ?

a minimal version of this can be implemented by just scheduling a daily skill which searches for surprising/useful implications/contradictions/combinations of ideas in one's zettelkasten then sends findings to inbox file/dms for manual review. already quite useful. ofc will be necessary to tune/adjust the skill over first week to ensure it actually returns useful findings gwern.net/ai-daydreaming

i am now LMSR-pilled LMSR USP (vs CPMM): - it solves the “pool value goes to zero if any outcome goes to zero” problem * specific to prediction markets: - it eliminates the tradeoff between capital efficiency and being forced to reprice outcomes unrelated to those you wanted to trade * i.e. kind of like independence of irrelevant alternatives - bonus: lp cost/trader-profit is proportional to KL divergence between pre-trade and post-trade prices * when # of outcomes > 2

manifold.markets/distbit/before… i would like to see more of these.

@lemonautzest 💛

decision selection bias should leave a fingerprint in the form of "conditional-market price decay" upon no information being revealed, which hence could be detected and used to disqualify that measurement window. traders will only rationally price in decay to the DSB-premium once the deadline for information revelation is near/past. unfortunately, this makes "dsb-decay" basically useless as a strategic-dsb heuristic, because the decay can be forced into an arbitrarily short window of time just before the market settles, hence making it indistinguishable from noise. this can only be addressed if the "info revelation deadline" can be enforced as being earlier than the market resolution date. the more you think about this requirement, the more impossible it appears.

distbit - fut/acc @distbit0

a week ago

to solve strategic decision selection bias without either randomisation or manual proposal vetting, a mechanism is required which resets the measurement window if any new information is revealed/priced-in during said window. BUT it is highly non-obvious how to detect such info

0 0 3 188 0

probably best intros are: overcomingbias.com/p/futarchys-mi… dynomight.net/futarchy-marke… tldr is that futarchy prices can sometimes be biased upwards, because you are asking traders to predict "how good will this proposal be if it is selected" not "how good will this proposal be". so the less you know about the proposal, the more plausible it is that "it being selected" means some v positive info about the proposal was revealed and caused it to be accepted. so you therefore should rationally predict it will be better than it really is.

Impact futures are markets on how much an event changes an asset’s price: the spread between the conditional price if YES happens and if NO happens. Example: if S&P futures trade at 5,600 conditional on a Fed rate cut and 5,500 conditional on no cut, the market-implied impact is +100 points. That lets traders price the effect of the event directly, rather than only betting on whether the event happens or exogenous factors impacting the asset's price. But the oracle should stop measuring before the event becomes one-sided. Once YES is at 98%, the NO branch is no longer a serious counterfactual market. It has little depth, so the measured spread becomes cheap to manipulate. Intuition: measure the counterfactual spread while people still disagree about the outcome, not after almost all trading has ceased. One design is to average only in-band periods where both outcomes have non-trivial probability, and refund if there is not enough in-band time.

to solve strategic decision selection bias without either randomisation or manual proposal vetting, a mechanism is required which resets the measurement window if any new information is revealed/priced-in during said window. BUT it is highly non-obvious how to detect such info revelation, given that it doesn't need to cause any observable impact to the market price, only a counterfactual impact.

a conditional-market hedging trilemma: a position which continuously rebalances to hold equal positions in LONG-conditional-on-PASS and SHORT-conditional-on-FAIL will be hedged against movements in the underlying (good), but incurs "impermanent loss" (not so good) from changes in the probability of each outcome, given their portfolio emulates that of a CPMM LP. decision randomisation addresses this, by making both outcomes equally likely if the market settles, but creates higher payout variance (not ideal), given that in most cases, the market will not settle.