Bdixon - I haven't really read up about efficiency measures so unfortunately I can't point you any further. My interest in efficency is to free up time otherwise spent in random expectancy trades - which allows you to try to fit other trade ideas into the freed up time, thereby improving your overall return on capital on any given period. eg - combining trade ideas 1 + 2 by allocating risk capital between them yields a weighted average of the expectancy in trade ideas 1 & 2. But if you were able to find a way to remove enough dead spaces in the trading period such that you could fit trade ideas 1 & 2 together without overlap, then you are able to make your risk capital work twice as hard, thereby yielding the sum of the expectancy. A simple example is trading US markets in the day and Japanese markets at night. Another way might be to try to take out, say Fridays out of one strategy without changing the expectancy and add another strategy that yields a positive expectancy on Fridays. Anyway, I've hijacked this thread enough ... it is Acrary's thread and I just wanted to add my 2 cents on profit factors.
Alrighty, thanks for your thoughts and I agree about hijacking this thread, but if Acrary is trading the night hours he's gonna need something to read and think about while he's sitting there. 'Night to you.
"The market can stay random infinite amount of time compare to each trader has a lower absorbing barrier, that is, a finite amount of capital, the point of which the capital is wiped out constitutes the end of that traderâs participation in the trade, and that, if not wiped out, the market continues indefinately. The probability of ruin, P, that is the point where our traderâs capital is reduced to itâs absorbing barrier, in a 50% trading system: (i.e. arithmetic mathematical expectation = 0) is bounded by: (a - z - (alpha -1)) / (a - (alpha - 1)) <= P <= (a - z) / (a - (beta - 1)) In the above equation, z represents the traders initial capital, beta, the amount he bets and loses on a losing trade, the upper absorbing barrier (a>z+beta), and alpha, the amount he wins on a winning play. From this, as a -> infinity, P -> 1. Thus, in a trade where alpha x p <= beta x q where p is the probability of winning an amount alpha, and q the probability of losing an amount beta, the trader must eventually lose his capital z". So, do we need a system of more than 50% winning? I thought even money system can have positive expectancy using money management rule. Or do we need edge in a trading system? And, by how many percentage?
I think using expectancy as a measure of efficiency has some value. My point about the profit factor was that it is a valuable tool for improving CONSISTENCY. Thanks for the thoughts.
No, I tried to demonstrate how the frequency of trades and profit factor contribute to describe a universal truth. I believe a formula can be derived that can be used to describe the likelihood of profitability within a period of time. I never spent much time on trying to derive one (Monte Carlo estimates are good enough for me), but I'm sure someone with a math background could do it relatively quickly.
I read his interview in the Oct. 97 issue. Doesn't seem to offer any value from what I see: http://www.traders.com/Documentatio...97/Abstracts_new/Interview/Interview9710.html I ordered his book that was published in 1997. I'll check it out when I receive it. I have the book he wrote with Kroll and I didn't find any value in it. In an earlier post you said you learned and forgot about this material. I was just looking for some insight into what you replaced it with. I've been using it since 1988 and haven't found anything better so I was deeply interested.
Good stuff, thanks I'll look into it. The big picture idea about correlation was that combining non-correlated (or weak correlated) methods can provide some benefit toward improving the modified sharpe ratio. One of my future research projects was to try to determine the size of the benefit.
Could you elaborate? If the material is unclear I can go over the parts that you may not be getting. I'm sure others are in the same position but just don't want to ask.
Doesn't look anything like what the edge test was designed to do. I'm including a summary for those that don't want to review some of my old posts. To do the edge test you use a single method at a time. First you backtest on the data you're using to develop the method. Then, when you're satisfied with the overall results you separate the trades by long and short by year. It'll look something like this: 1996 Long +3.00 hold 1 day Short -2.00 hold 2 days Long -1.00 hold 2 days Short +4.00 hold 1 day etc. Then you process the year of 1996 and pull out random individual trades with the same length of hold (being careful to avoid reuse of any one day). Ex. Long -2.00 hold 1 day Short +1.00 hold 2 days Long -2.00 hold 2 days Short +1.00 hold 1 day etc. When you get done you total up results of the long and short trades for the random pass for the entire period. ex. Long -4.00 Short +2.00 You do this random pass thousands of times (Monte Carlo) and rank each each of the passes so that you have a distribution from 1% - 99% for both longs and shorts for each year of the tests. Ex. Long 1% -16.00 etc. Long 99% +21.00 Then you compare the total you have for your tested trades versus the distribution to rank where your trades are as compared to the random trades. (do this for both longs and shorts for each year). If both longs and shorts rank 70% or better (20%+ better than random) then you might be looking at a edge. Do the same test with out of sample data and shorten the time period to 3 months (so that you can view multiple forward time periods). If the numbers continue to be 70% or better on both longs and shorts then you probably are trading with a edge. You do this test every 3 months after you start trading it to make sure the edge is not deteriorating. If it drops below 70% then stop trading it. My experience with it has been very good and worth looking at during the development of a trading method phase.
i build a sheet in which i have daily data for sp futures in the first columns followed by a column with daily gains as calculated by ln(Ct+1/Ot+1), so i am calculating tomorrows return. now i derive all sorts of things ending with the close of today (Cto). i did indicators as well as different structures of highs and lows in combinations. and i correlated them with the dailyGain of tomorrow. in addition i made a reference column with random numbers only. no i found that many things were correlated higher than random, which is always about 0.01, and some seemed to be siginificantly higher correlated with 0.1 or above (one going beover 0.5). so far i have not found anything which i had not known in advance from other types of analysis. but i am very aware that i just spend an hour here and there and that the strength of your concept might essentially be that you can cover a universe of potential variableCandidates quicker than with other techniques. i did something else. i build a spreadsheet were i can combine different variables and really simulate trading the next day. i calculate a ttest to determine the relation between my selected sample of trades with the universe of all days. and i calc hit ratio, payoffratio, sharpe ratio and mdd. i require at least 100 trades out of ten to fifteen years data history. and i analyse my patterns in terms of robustness against parameter changes. i do not worry if they change the outcome (necessarily they do), but i prefer that they do it in a linear way, not in a spiky manner. i think this latter requirement of mine goes well with the correlation technique. i programmed a montecarlo drawing of returns as well and ranked my sharpe ratios against the random, but with the things i liked anyways, i was always on top of the list, so i skipped that. i think this might be because i am ranking sharpe ratios against sharpe ratios. and since i am already fitting towards higher sharpe this result is imbedded. now i found that doing the correlation work is not faster than my excel-"testingSystem". since i have to program the "indicator" anyways, i can immediately do that in the testing system and i have all my system stat at once. my best finding so far is on nasdaq futures uses one parameter, holds positions for three days, makes about 150 trades since 1996, has a mod sharpe ratio of 1.32, hit ratio of 68% and payoff ratio of 1.6. it's been in paper trading over a year now, delivered very similar stats since then and wil go live within the next weeks. do not know how this will turn out. peace