A market that closes in an autoregressive manner. ie, the chance of this market closing is a function of the previous markets' results. It is not deterministic, as a large amount of the variance is still random, but correctly estimating by way amount each lagged day is related to the current market, you can get a statistical edge predicting this market. so the end goal really is to correctly statistically estimate the coefficients I used for each lagged term. If you're familiar with time series stuff, this is just an AR(4) model of sorts

Each market lasts one day. This is day 4 It will run the ar(4) version as the three previous days (yes, no, yes) plus the assumed yes vote to begin the first round. So full thing is: [yes, no, no, yes] with first being most recent market / lagged term and last being last relevant one. previous market:

https://manifold.markets/inerati/autoregressive-market-day-3?r=aW5lcmF0aQ

Procedure:

Every day I draw a random decimal fraction (0 to 1) from random.org. Let's call that R. Then according to the following equation:

F(R, X) = (1-𝛼)R + Xβ,

where 𝛼 is the sum of all the autoregressive coefficients β, and X is a vector of the previous 4 games, encoded as 1 for yes and 0 for no. β is a vector of the corresponding coefficients of those autoregressive terms.

If F(R,X) is greater than or equal to .5, the market resolves as yes. If not, it resolves no.

If there have been less than 4 games, I will artificially not include the later terms and use the appropriate 𝛼 for the number of terms actually in use. The exception of this is the first one, which I will assume ended as yes for the first AR term. This is just to give a reason for people to participate at all.

I am capping the number of potential terms at AR(4), but some could be zero. All coefficients are less than 1 and in discrete values of .05. ie, 0, .05,.1, .15 all the way up to .9. They are decided arbitrarily by me. They must also sum to less than .9 all combined. Keep this in mind (!). If this pops off I will make more markets trying to guess the values. I will not be participating in this market.

I would also argue that this is not "non predictive", as it is trying to predict something (the coefficients)

example w fake coefficients:

If previous 3 markets ended in [yes, no, yes] then beta would be [1, 0, 1]. Take random guess (say it was .35), multiply it by the sum of beta 1,2,3 and add it to Beta 1 and beta 3 (no beta 2 since that's zero). Let's say beta is [.2, .3. .1]. Then 𝛼 is .6. So...

F(.35, [1,0,1]) = (1-.6)*.35 + 1*.2 + 0*.3 + 1*.1

= .14 + .2 + .1 = .44

.44 is less than .5, so resolves no. This example is not using real betas and the full market will be 4 long not 3, so just an example. This particular market is assuming AR(2) with the first (non existent, used last time market as YES and the previous one as NO (so x is [0 1])