Registered: Apr 2003
12-25-11 12:04 AM
Quote from feng456:
How does a system all of a sudden do that?
Let's imagine for a second that I am your manager, e.g. a head of the desk. A good thing to have sometimes, keeps you thinking straight.
In the situation described, I would be less concerned with that fact that a model stopped working and is losing money, but with the fact that the losses are unmanageable (according to you). My conclusion would be that the problem is in risk management, not in the system design.
So, a couple questions:
(a) is it the only strategy you are running? If the answer is yes, it's probably worth developing another few to add diversity to your book. If you insist of running a single strategy, you must decrease your leverage.
(b) how do you think your strategy mix performs ex-condition? For example, let's say your model buys SPY every time the President of the US has a bowel movement and holds it to the close of the day. What would be your mean, median, 95% worst and the worst daily P&L be ex-condition - meaning, if you just bought SPY in the morning? If you have multiple systems trading at the same time, try to build a history of asset allocations in the portfolio and see it's ex-condition P&L.
(c) are you using some sort of a system for strategy allocation? In general, if you have a few years of backtests on multiple systems, you want to build a quantitative allocator that takes into account correlation between strategy performances, recent performances etc. For example, I know guys who like looking at percentile of consecutive losers for each system (vs the back test) and decreasing capital allocation as that percentile increases.