Posts

Showing posts with the label prediction

On the poor man’s futachy

The main idea: conditional consols are a cheap and easy way to implement futarchy today. 0. Posts on this blog are ranked in decreasing order of likeability to myself. This entry was originally posted on 02.12.2022, and the current version may have been updated several times from its original form. 1.1 Robin Hanson’s futarchy proposes we give prediction markets some role in governing nations or other organisations, in the former case by having people vote on beliefs but bet on results. 1.2 A grave issue with this otherwise interesting proposal is the lack of a quantifiable goal (the “belief” part) that can be fed to the markets, this being AFAIK one of the vectors by which Moldbug attacked the idea. 1.3 But fear not, for whilst true futarchy may be a highly speculative proposal today, there’s a much simpler iteration available to us right now, and which delivers much of the promise but requires less of the suspension of disbelief. A poor man’s version of futarchy, if you will. I eve...

On a forecaster's prediction market

Main idea: use global Kelly allocations to translate probability estimates into an optimal betting portfolio. 0. Posts on this blog are ranked in decreasing order of likeability to myself. This entry was originally posted on 17.04.2026 and the current version may have been updated several times since its original form. 1.1 Consider a prediction market in which yes- and no-contracts on binary events trade via some AMM algorithm. So, like, all of them. 1.2 But users do not trade contracts manually. Instead, the market trades on behalf of all traders. Contracts are purely virtual, mimicked by the platform itself, which keeps track of everyone’s positions. 1.3 Users simply fund their wallet and input their probability estimate for any question they wish to bet on. 1.4 Once the platform has my estimate, it runs a Kelly calculation on the yes- and no-contracts, allocating to each a fraction of my wallet. The yes and no positions are netted, and the remaining weights are normalised across al...

On semi-useful retrodiction

The main idea: poll knowledgeable parties on the probability of past events within basic logical rules.    0. Posts on this blog are ranked in decreasing order of likeability to myself. This entry was originally posted on 05.07.2023, and the current version may have been updated several times from its original form.  1.1 Here’s a system that may be able to produce semi-useful probability estimates for past events. 1.2 Create a list of 10 - 20 past events, which include a combination of events that did and did not happen (ex. Germany to conquer France within 1940, Soviets to reach the moon first, etc.). 1.3 Ask a group of knowledgeable people to (independently of one-another) assign a probability to each event, assuming each event had a random chance of happening or not. Feel free to enforce logical rules if events are conditional upon one-another (ex. chance of Germans conquering France within 1940 must be lower than their chance of conquering France within September 1940...

[RETRACTED] On an alternative prediction scoring rule

0. This entry was originally posted on 09.02.2024 and retracted on 17.12.2025 when I finally bothered the check the expected outcome function. Fine, I get it. 1.1 You and a bunch of others have recorded your expectations that a given event will happen or not in the future, with the event being binary, and predictions in the form of 0 – 100%.  1.2 Further, each of you has put some money towards the pool. No odds were given to you when you did either of these things. 1.3 Now how you score the predictions and allocate the pool: the pool is allocated in weighted fashion, where each participant’s weight is the product of their contribution and the probability they assigned to the event (probability of occurrence if the event occurred, and one less that if the event didn’t occur). 1.4 With bets were made at different points in time, a bit more thought is required. Naively, one may just add a third term to the weight product there, this being the total time outstanding, but I fear there m...