Are Financial Markets Efficient Or Inefficient?
Business
Over short horizons markets work better than models.
There isn’t anyone in quantitative finance who believes otherwise – even those who find statistical arbitrages and make the markets more efficient than they would otherwise be. Participants who extract rent in this manner can fairly claim to “beat the market” – but this doesn’t make them “better than the market” as 24/7 predictors.
Historically, markets have been pretty hard to wield compared to models. That’s really the only reason to prefer the latter. A century ago it required an analog computer the size of a small apartment complex to determine probabilities of horses winning a race (the first autonomous realtime market – see Totalizator ( Totalisator ) History – A World’s First – Index (ozemail.com.au).
Market’s have also been clumsy in their design. At least as far as their role as a model replacement goes (they serve other purposes, obviously).
From that perspective, it seems ridiculous that inferring probabilities from options markets requires a subjective disentangling of linear combinations of probabilities that participants assign to different future outcomes (i.e. the inference of Arrow-Debreu prices).
Market’s have also been tied down by regulation, with further friction added where it is felt (rightly or wrongly) that staking is essential. There is a graveyard of would-be prediction markets. And markets are also the ultimate mis-mash of conventions, APIs, laws, code and feed conventions, like railways with inconsistent gauges.
But suspend your disbelief in the lightness of markets.
What if there was a domain (fast flowing data) and supporting software that made markets (maybe not as you know them) trivial to wield. Trivial for absolutely any purpose. How would this change the view of the modeling world? Would you need models at all? Or would it only be other people, on the other side of an API, that need to get their hands dirty?
Entire organizations will change. Model review starts to look like an expensive unnecessary undertaking, in large part. AutoML, as we understand it today, could easily be mistaken for the work of dithering incompetent committees. And even the poster child, reinforcement learning, might be a bit of a “gorilla holding a banana”. Moreover, to borrow the old ding at OOP, because it unnecessarily ties two quite different things together.
Well, you may not agree with me, but I hope you find the book thought provoking.
Microprediction: Building an Open AI Network: Cotton, Peter: 9780262047326: Amazon.com: Books
Are Financial Markets Efficient Or Inefficient?
