The everyday blog of Richard Bartle.

RSS feeds: v0.91; v1.0 (RDF); v2.0.

Previous entry. Next entry.

8:17am on Wednesday, 15th June, 2005:



Yesterday, I spoke at the WEHIA conference at Essex University. WEHIA is an acronym for Workshop on Economic Heterogeneous Interacting Agents; lest that sound obscure, I should point out that this was the tenth such workshop, it had 300 attendees, and Benoit Mandelbrot was guest of honour (he looks like my wife's brother's wife's father did before he died, if that helps you picture him).

I was speaking at the conference because earlier this year I spoke at the Digital Money Forum and one of the attendees, Sheri Markose, came over to talk because she's a lecturer in Economics also at Essex. As organiser of WEHIA 2005, she asked me (and Dave Birch, organiser of the Digital Money Forum) to do a double act in front of 300 easily-unimpressed economists. Stirred by the thought of addressing a conference with a word in its title I don't even know how to pronounce ("Heterogeneous"), I accepted.

This is just background.

What I'm going to write about today concerns the two sessions I attended of the conference. With 6 parallel sessions at a time, I had plenty of papers to choose from, but none of them had titles that made the remotest sense to me. Every academic subject has its own vocabulary, and although I understood the meanings of individual worlds in the titles, I didn't know what they meant when strung together into a 14-word extravaganza with a colon in the middle. Anyway, feeling that I should attend at least some sessions, I picked the one at which Sheri was speaking.

Bizarrely, I actually understood what she was saying. Even more bizarrely, I was probably one of only two people in the room who did (the other being Sheri herself). She was talking about modelling economic activity using theories of computation — fixed points, universal Turing machines, Gödel's Theory, the works! Those years spent in Tony Brooker's CC204 lectures and daily dining with Ray Turner and Martin Henson finally paid off. I knew what she was talking about. I've no idea what it had to do with economics, but I put that down to the fact she had a mathematically dense, 40-slide, 2-hour talk prepared and only 25 minutes to deliver it...

The title of Sheri's talk, by the way, was "On Hostile Agents and the Red Queen Principle: Formalizing the Arms Race in Innovation and Complex Dynamics". See? You understand every word, but have no idea what it's about.

The other talk I attended was not called "Heterogeneity-Driven Asymmetry in the Stock Market: Agent-Based Modelling with Frequency Decomposition and an Evolutionary Artificial Neural Network". It would have been called that, but we got a different talk instead from someone from Bristol University. This one was about auctions.

In an English auction, the bidders say everything and the sellers say nothing. In a Dutch (flower) auction, the sellers say everything and the bidders say nothing. In a double auction, both bidders and sellers speak. Auctions can thus be characterised on a scale of 0 to 1 depending on how often seller or bidder speaks, with double auction being 0.5. The Big Idea of the talk was that in digital trading environments we could have, say, a 0.8 system, where the bidder speaks a fifth of the time and the seller four fifths of the time. We don't get this in "natural" markets.

My first reaction was that actually we might get this. If stock exchange bidding patterns were analysed, we may indeed discover that in some cases sellers are more likely to change their prices than buyers and vice versa. Still, no matter, on with the show.

What the guy did was take 4 different supply and demand curves and put 11 computerised bidders against 11 computerised buyers, then run a genetic algorithm (basically a fancy hill-climbing algorithm) to find what the optimum factor was for the seller's speaking. It varied with each curve, and in one case was indeed 0.8 . He was very pleased with this.

Call me a physical scientist, but hold on here. Just 4 experiments? With just 11 bidders and sellers each time? Why hadn't he run his simulations hundreds of times, with different numbers of bidders and sellers? Why did he only have 4 supply and demand curves and not dozens of them?

I actually asked a question. I was in a conference full of EHIA specialists, and yet I asked a question. What I asked was whether he'd thought there might be some mathematical model that linked the supply and demand curves to the optimal seller speech percentage so that he didn't have to run all his simulations. "Yes!" he replied, excitedly. "That's just what we're hoping to do here!".

Sadly, I wasn't able to ask my follow-up question as the time was up and the next speaker had to start, but, for the record, it would have been: "So why the hell have you only done four experiments?!".

Latest entries.

Archived entries.

About this blog.

Copyright © 2005 Richard Bartle (richard@mud.co.uk).