Simulations. I've seen some work done that gives very complicated (VERY COMPLICATED, even by stochastic calculus standards) results for even simple normal distributions.
This is a pretty new area where the boundaries are not hashed out. Any classic time series analysis text that introduces monte-carlo simulations should be more than sufficient to give the tools needed to start building simulations that can capture some measure maximum draw down distribution.
But, as i have said in the other few related threads that we are all a part of, the tools are sort of there - but without figuring out what what exactly you are trying to accomplish with that draw down measure, it's all useless. To wit, if you are asking me what's the largest amount you will lose, I can tell you -100%. I'd be right. There'd be zero under-estimation error (assuming not debtor's jails). But it's a completely useless answer.
This is a pretty new area where the boundaries are not hashed out. Any classic time series analysis text that introduces monte-carlo simulations should be more than sufficient to give the tools needed to start building simulations that can capture some measure maximum draw down distribution.
But, as i have said in the other few related threads that we are all a part of, the tools are sort of there - but without figuring out what what exactly you are trying to accomplish with that draw down measure, it's all useless. To wit, if you are asking me what's the largest amount you will lose, I can tell you -100%. I'd be right. There'd be zero under-estimation error (assuming not debtor's jails). But it's a completely useless answer.