Robert Hanson had a post on Kelly betting, and I sort of trust hs proofs apriori (I shouldn't) Anyway, he summarized Kelly betting as betting according to the most probable equilibrium. In network theory, if we know what the structural distribution looks like, say in terms of the Shannon yield curve, then we should make a Kelly betting market to mimic that distribution. A Kelly gambler sets the i in -iLog(i), here i is the probability of the event. When the economy deals with it, aggregations are attended to, and we get the set i (let i be an integer set in our algebra that can be scaled to probabilities). So, the Kelly betters build a network such that the mutual entropy between the generalized yield curve and the Kelly curve match. Flash traders, who stuff a lot of small bets in short periods would get caught by a Kelly encoder, a run time Huffman encoder that went hunting for unattended aggregates.
For example, ever six months or so when Ben meets the traders, they agree on the Kelly network needed to push a probable debt increase event.
As an aside, let me say that the idea of Kelly betting is to make the gain from the next outcome to be a surprise to everyone; (there remains no public redundancies).
Kelly betting in the stock market, for example. If the Kelly better sees unattended aggregates, then he should act as if he is an intermediary fund manger, putting the aggregate into the selected -iLog(i) that matches the nearest economy wide -iLog(i). He inserts himself as a stage in a production system, and that unattended aggregate no longer shows up on other peoples Huffman encoder. The Huffman tree shuld give you a graph, whose which map to the real goods distribution graph, up to the specified precision.
No comments:
Post a Comment