Tuesday, November 8, 2016

Watching Matt Levine 'get it' is fun

Queuing over finite networks using the informational model.  This is all implied Huffman coding:
Here's a fun paper by Marco Di Maggio, Francesco Franzoni, Amir Kermani and Carlo Sommavilla finding that investors who trade stocks through well-connected brokers do better than those who trade through less central brokers. Why? Maybe something like this:
We can imagine an informed trader submitting an order through a broker, who can infers the informational content of the trade and spreads it to other clients. The incentive for the broker is to build a reputation as a valuable source of information and attract more business. One may think that the informed clients would not like their information to be spread to other investors. However, if the informed investors have capacity constraints, they may actually solicit other investors to trade in the same direction, so that prices reach the new equilibrium faster
A big hedge fund buys a lot of XYZ, the broker gives market color to her other clients that there's a lot of activity in XYZ, and the other clients interpret that market color correctly and "back-run" the big hedge fund by buying more XYZ. That's good for the big hedge fund, since after all it owns XYZ and wants the price to go up. (It would be less good if the other clients front-ran the big hedge fund, buying before it built its entire position, and I suppose with a large position built up over time there is always the risk of sort of ... mid-running?) "The best clients of the broker executing the informed trade, and the asset managers affiliated with the broker, are among the first to benefit from the information about order flow."

Mr. Levine is very good.  What is he saying?

You are a big client, you are making a big portfolio shift.  That makes you a volatile price fixer, actually.  The big broker network has lotta sub networks to do the round off error.    That ensures you get your share of the round off error.

Let me go off into wonderland:

 This all  makes me think of the Gibbs effect, that sample noise we get on a sharp quantum shift. That guy was good, he knew, right away, that Gibbs effect was not the world obeying Newton's grammar, and he got to the Gibbs state, a quantum lock. He is becoming one of my favorite, for a long time I wandered in the wilderness thinking him as yet another goofy electrical engineer.

No comments: