R. A. Hettinga wrote:
For commodities, if you could somehow discover the demand and supply curves and predict how they were going to move, you could in fact calculate what prices were going to be. The problem is that you can only observe exactly one point on the demand or supply curve -- where it crosses the other curve. You can't observe any other point until at least one of the two curves moves. It's conceivable (although I'm not aware of anyone even attempting this) that if you had some (perhaps probabilistic) model for both curves as a function of some exogenous variables, that you might get some useful predictive information about prices.By the way, one piece of evidence that economics is maturing into a real science is that it is becoming usable by "engineers";
Well, finance, anyway, where it is possible to calculate some risk.
You can't calculate prices, though. You discover them.
All markets involve versions of supply and demand curves. However, predicting the future of market prices is notoriously difficult.
The problem is caused by a lot more than inability to see more than just the one point where the two curves intersect...that's just a statement of the market clearing price.
Whether the price of GE stock, for example, may go up next week, or down, or follows a shape described after the fact by some complex equation is unknown for a LOT of reasons.
(Friends of mine operate a medium-sized hedge fund, using as much knowledge as they can gather from tens of thousands of market values per day, using a whole panoply of buzzword math technologies (support vector machines, neural nets, Bayesian networks, agents, blah blah). They spent time with Doyne Farmer, formerly of Prediction and now at the Santa Fe Insitute. And yet they are only trying to gain a slight edge.)
Commodity prices are close enough to being like stock prices that the prediction problems are comparable. (And predicting commodity prices is a popular regime for trying these techniques.)
--Tim May
