Aren’t most firms using binaries for their internal prediction markets?

Chris F. Masse October 26th, 2007

It depends what you’re trying to predict. If you’re trying to predict an event probability, a binary contract will do just fine. But if you’re trying to forecast a variable, like a sales volume (or a box office result a la HSX), then using a set of binaries quickly become unwieldy. The so-called “Arrow-Debreu design” where you associate a binary contract with each of several potential ranges for the target variable is what H-P used in their classic printer sales experiment. But for traders that means trading several contracts for a single forecast, as if trading one contract wasn’t complex enough already. If you’re trying to forecast a half dozen variables and each of them is represented by, say, five contracts, then traders have to consider 30 contracts! It quickly gets impractical… Unless you really care about producing probability distributions for your variables, it’s just not worth it. H-P abandoned that approach long ago.

The bottom line is that if you want to use a prediction market to make forecast, you’re better off using a single contract whose payoff is a continuous function of the target variable’s outcome, like an HSX movie-stock, or an IEM vote-share contract. Alternatively, you can also choose a design that does away with trading entirely, like H-P’s BRAIN or NewsFutures’ Comptetitive Forecasting.

External Link: NewsFutures

All Best Posts Ever , Analysis (Meta) , Exchanges & Markets , Market Trading , Mechanism Designs
Comments(1)

One Response to “Aren’t most firms using binaries for their internal prediction markets?”

# Byrne Hobart on 26 Oct 2007 at 11:27 am

Given low enough transaction costs, each contract is a proxy for other contracts: let’s say we’re betting on voter turnout, and we have contracts in 5% increments: turnout will be above 40%, 45%, 50%, etc. When estimated turnout increases from 43% to 44%, all of these contracts go up — the only difference is the ‘delta’ (change in derivative price compared to change in price of underlying indicator), so a clever quant would just write a short script saying, roughly, “Find me the cheapest way to bet that turnout will be 47%, with a standard deviation of 3%,” which should yield a melange of long and short positions in various contracts, to be rebalanced as prices (and thus deltas) change.

The Arrow-Debreu framework is great, as long as computation is cheap (it is!) and transaction costs are minimal (we’ll get there!).