Better utlity

[akolonin]

Current "Utility" and "Weighted Utility" seem to have very little significance and variance and seem not informative. Likely, because they are including all the metrics made by scammers.

Consider having better Utility metrics - at least three of them: 1) "Organic Utility" - computed only based on organic ratings made by honest non-gamin buyers 2) "Dynamic Significant Utility" ("Dynamic Liquid Utility") - weighted not with financial weights, but with reputation ranks of the buyers (measured at the time of the ratings being made) 3) "Eventual Significant Utility" ("Eventual Liquid Utility") - weighted not with financial weights, but with reputation ranks of the buyers (measured at the time of the end of simulations)

Need to explore how the metrics above perform compared to existing "Utility" and "Weighted Utility" in different market conditions with different reputation systems and select the best one.

[deborahduong]

In my opinion, its important to keep the definition of Utility simple and straightforward. Utility is simply the average rating of all honest transactions. It does not include ratings purchased in scams. My proposed metric was weighted utility, that is, weighted by the price one paid. For example, you dont want a consumers satisfaction with a car it buys to be worth exactly the same as the consumers satisfaction with a matchbook it buys. You recognized this because your BSL metric is really 1 - my weighted utility suggestion. I didnt insist on keeping the weighted utility because it made sense to have a bunch of metrics that you minimize in context, however, in the future, it may make more sense to talk about a weighted utility when something to maximize is better.

[deborahduong]

I will look into why utility didnt move in our most recent case, but note that it did in other cases. One place I will look is the fact that there are other goods on the market, but the requirements stated that only one was open to scams. Another place I will look is the agents good memory of good agents, and to see if increasing enterers and leavers would make this more realistic. And I will implement your metrics as well, but I have some questions. Why would you want to weight the rating of a transaction by an agents reputation, which could be inaccurate, when the simulation already knows whether a particular transaction is honest or not? Also, in some scenarios the reputation is based on products, not agents, however, we always know the honesty of a transaction. As for the organic utility, I think it is better to look at transactions and not gaming agents because gaming consumers also make honest buys, and the simulation knows which ones.

[deborahduong]

PLEASE CLOSE (I can not). The utility separated adequately in the runs for the technical report, and does not need new definition. https://docs.google.com/document/d/1FEr4ir0jBZ5PZtOBas7naeHef8AnPo_E5IArSoxN0kE/edit