macrocephalopod

For some self-criticism: we do have one class of agent (out of our 10) that I don't like and would like to get rid of eventually. It's a random/zero-info trader that either joins the top of book or crosses with some probability, and has a bias towards decreasing its inventory

More self-criticism: we made no attempt to be "parsimonious." Christian and the other sociologists and anthropologists did hundreds of hours of field interviews, so we made a qualitatively similar ecosystem that was also reasonable from my own experience, but we made it big

And more self-criticism: all the agents are too static. Sure they do things like volatility scale positions and adjust target exits, but there's no real optimization. In the future we want the agents to have more... agency

For the future: we need more real heterogeneity in strategies and not just having a bunch of things parameterized differently. You can't really escape high correlations doing that. The institutions are pretty dumb, so I'd like to actually generate fundamentals that they use

Other problem areas: because everything goes through a real matching engine, for every buy there is a sell. Sometimes classes of agents would collectively accumulate large net positions and the market makers would be stuck with it. Needs more willing longer term holders

The lovely part about building a very generalized platform is that you can add an arbitrary number of instruments and then in the agent configs just tell them which ones to subscribe to

From the correlation matrix, the first principal component isn't as high as it should be, and we don't have long-term positive drift or negative skew. But we've got a pretty clear strategy for getting those in the next version.