Generalization of a hazard model

Question

I am looking for a statistical technique similar to a estimating a hazard model.

Suppose a person chooses between two actions, buy or sell. Given data on trading decisions, once a person has bought, say, a stock, I can calculate the time that elapses until they sell. Using a hazard model I can model this hazard as a function of other variables. This allows me to answer questions about what predicts a person's decision to sell.

But this ignores the fact that sometimes instead of selling a person buys more stock. This is the opposite of selling, and moves the person further away from "failure" (ie, selling). I can think of no analogy in epidemiology, which perhaps explains why I haven't been able find a method that is suitable for this situation.

Any suggestions for techniques that may be useful here would be greatly appreciated. Thanks!

Answer 1

Do you know about competing risk models? Rather than modeling the transition from a state A to a state B, you can model the transition from A to B or C or D, etc. Rather than one destination you have several "competing" destinations here.

The topic is coverd i.a. by Kalbfleisch and Prentice (2002). You can also have a look at van den Berg's contribution to the New Palgrave Dictionary of Economics.

Answer 2

The literature you are looking for is that of general point processes or marked point processes and what replaces the hazard is the general stochastic intensity function that depends on the past. One place to start that is of direct relevance to your question is Modelling Irregularly Spaced Financial Data by Nikolaus Hautsch.