Doesn't that just mean that the definition of risk is "returns adjusted by variance" or "returns adjusted by drawdown"? But, it seems like you are talking about theoretical risks as opposed to realized risk. Maybe I'm misreading your comment, but I take it to imply that a risk model would only be close to accurate if it took into account the maximum realized risk historically plus some additional quantum of risk, which would be determined in real-time via some statistical techniques. If a trader makes 10% in a year with minimal realized risk are we not able to say that his realized risk was his actual risk for that year? Do we have to assume that at any time he was in a trade a black swan could have occurred, wiping out his 10% gains and more? Every time I enter a trade in 2012, do I have to assume my risk is as extreme as the day of the 1987 crash? While I think that may be theoretically correct, it seems like an extreme position.
Please do, I would love to learn from a seasoned trader like yourself. I am sure you have been trading for a long time and maybe even managing a group of traders, so there is a lot for you to share here.
in the mid 80's, there was a stat at CBOE,that 95% of all new traders, those that bought seats or leased and took the test to get one,when tapped out, you have to return your badge so you can't walk on the floor and make erroneous trades,so from those returned badges and the CBOE stats,95% of new traders blew out within the 1st year, i would imagine the number is higher now,being that you can open an acct easier than getting a real estate license,in other words, no knowledge whatsoever..that is the 1st time i heard the stat...so if you wanted to ask the CBOE,they might be able to give you detailed info
At the point of the knife You never see anyone How the strong will survive At the end of their gun We run by Strange Advance
No, it's ok to apply even an inaccurate risk model. You only should know its limitations and when to apply it and when not. Risk is the sum over the outcomes of all possible negative events multiplied with the probabilities of those events. Of course you don't know those events and their probabilities in advance, so you try to approximate risk. The most frequently used proxy of risk is the maximum historical drawdown, which you normally use to calculate the profit percentage of a strategy. Problem is that the drawdown depends on the order of trade results, which is random. For this reason, historical drawdowns have little correlation to risk of a trading system. You can try to improve that by getting a drawdown distribution with a Monte Carlo simulation, but then other problems arise. I just want to make a point that all methods to adjust returns by risk have their merits, but you must know what you're doing. Just waving the buzzword "risk-adjusted return" and dismissing all research that does not use it is ignorance.
I fear you greatly overestimate my trading. In fact, so far I have never traded in my life. Only my PC trades. I guess that makes my recommendation obsolete for you, but if you want it nevertheless, the fundamental paper about covariance based risk calculation is by Edward Thorp: http://www.edwardothorp.com/sitebuildercontent/sitebuilderfiles/KellyCriterion2007.pdf and an excellent book about applying statistical methods to adjust trading system returns is "Evidence based technical analysis" by Aronson.
(A) You recommendations would not be obsolete, but would rather be irrelevant - you do understand the difference, don't you? There was a point in my life when I too thought that I know everything because I programs in S+ and has read a few relevant books. That was when I started in the silly business in 1996. Later I discovered that the reality of the world is very tricky and that statistical models are just models (as it is said, "the best model of a cat is a cat itself"). This relatively obvious fact cost a few people I know a few billion dollars. (B) Oddly enough I have read the first one (nothing fundamental about it, in my view) and can tell you a few things - (1) it has nothing to do with performance measurement (2) it does not work for convex instruments and (3) is a pretty good paper otherwise. There is way more to measuring performance and strategy allocation then simple bet sizing, of course. (C) I have never seen the second one, but I doubt if it is applicable to any of the strategies that I run (i trade volatility arbitrage exclusively). Also, I do not believe in back-tests unless I know who's lunch I am eating - and it is the case with most technical analysis strategies that you rarely could explain why strategy is making money.
It's kind of interesting that you mention Thorp, a proponent of Kelly, which is a bet-sizing approach relying on relatively fixed values to its parameters, i.e. if your win percentage and your winner/loser ratio are X and Y historically, that implies bet size Z. I have to say that it strikes me that this kind of position-sizing approach would go against what you've been advocating in this thread. Yes, I know there are some more specialized versions of Kelly which do not necessarily rely on fixed values, but that is definitely its main thrust.