OmniVest User Forums
User-Created Portfolios
The 90% Hit Rate Experiment
10 replies, 1482 viewings
|
OmniVest User Forums
User-Created Portfolios
The 90% Hit Rate Experiment
|
Last Activity 7/22/2015 8:11 AM 10 replies, 1482 viewings |
|
|
|
| Printer friendly version |
Steve Mayo Legend  Posts: 414 Joined: 10/11/2012 Location: Austin, TX  User Profile |
Here's what I did: First, I created several new lists of leveraged, sector and speciality ETFs (foreign, real estate, high dividend, bonds, etc.) - knowing I would have few trades I wanted the added diversification of ETFs. I tried to add as many as possible to increase the number of trades. I then ran those lists (one at a time) against all the strats using Strategy Lab, selecting one of the canned condition filters (mostly conservative trend and the EMA3/EMA7 one, and "added" any that had a 90% or better hit rate and a reasonable return (i.e., more than just a trade or two). I then simply sorted by %Wins in the strategy listing, selected all that were above 90% and saved it as a new portfolio. Unfortunately, I didn't keep good notes and now I have no idea exactly what conditions each of the 28 strats uses (hope that naming issue is fixed soon!) It's mostly trending strats with leveraged inverse ETFs. Anyway, here's the graph. It has a Calmar of 3.6 which is the best I've been able to do, at least with a low drawdown, over a 7-year period without the benefit of Port Switching...and it's only 22% invested. My first attempt had a return of just under $1MM. To get the return up a bit more, I added a few more ports that were in the 80+% accuracy range and it pulled my hit rate down, but it increased the return without increasing the MDD....nice! [Edited by Steve Mayo on 5/7/2014 1:10 AM]  Attached file : graph1.png (91KB - 895 downloads) Attached file : list.png (796KB - 281 downloads) | ||
| John W Elite   Posts: 654 Joined: 10/11/2012 Location: Sydney, NSW, Australia User Profile |
Just a general question - what are the Settings you used? | ||
| CR Miller New User Posts: 2 Joined: 3/3/2014 Location: W. Bloomfield User Profile |
I've been trouble by using strategies that are too thinly traded for the possibility of being pseudo optimizing. Any suggestions on the lower limit of TPM? I've been waiting to do further testing until better strategy annotation because I also loose track on what the strategies are. I've been trying to determine what is the desired metric. The bottom line is CAR with a high calmar. Hit rate is very nice to have but I sweat when the value drops before recovering and making a profit. I wish there was a metric for that. | ||
| JimB Member Posts: 36 Joined: 3/7/2014 Location: Rogers, MN User Profile |
| ||
| Steve Mayo Legend Posts: 414 Joined: 10/11/2012 Location: Austin, TX User Profile |
With stock trading, returns are serially correlated (markets really DO trend/cycle because of the economic cycle, news, crowd psychology, etc.) so, again in theory, you CAN have non-equal expectancy of return (something better than 50/50 odds) and a prolonged 'winning streak' is quite possible. The question, as you pointed out, is what IS the true expectancy of your trading system and how much data do you need to determine it with sufficient accuracy. In drug trials, we do power calculations to determine how many patients we need to test to get a specified level of confidence in our findings. The statistics looks impressive, but in the end, it's still a guess, and it is often wrong -- that's why so many clinical trials fail to reach statistical significance. If you can't do it for a well-controlled clinical trial, can we really figure it out for something as complex and unpredictable as the stock market? Likewise, how many trades does a system have to generate before it is considered robust? It's a hard question that gets into statistical things like mean, standard deviation and whether the the return from your trading system follows a normal bell curve. But, rule of hand, I would say at least 20 to 30 trades across your timeframe. (My example had as few as 3 to 5 trades in some strats which is why I expressed caution. However, is the sum of the parts -- 6.5 TPM for the port -- different after you combine 28 strats? Probably yes, but it still seems like too few trades to be robust.) Hit rate, Profit Factor, and %Allocation are mathematically related: optimum hit rate = 1/(SQRT(1+PF). Thus with a profit factor of 1.2 (gains 120% as much as it loses), you need a hit rate of 66%. Increase your profitability to 2 and the hit rate needed drops to 58%. There is similar math for the probability of having consecutive losses (at 3 sigma confidence): n=log(0.0027)/log(1-HR). At 70% HR, you can expect about 5 consecutive losses in a row with odds of 1 in 370 or about 1 such run per year (1/370 = .003 = a 3 Sigma tail); at an 80% hit rate, expect about 4 sequential losses, at 90%, expect about 3 in a row once a year. But, as we know, the stock market doesn't always follow the "normal" rules. | ||
| kmcintyre Elite Posts: 890 Joined: 10/11/2012 Location: Portland, OR User Profile |
You are such an asset to this group. Thanks! Have you considered writing a "Statistics for Stock Traders" book (or video series). I think there would be a large audience for it. Kudos! Keith | ||
| Steve Mayo Legend Posts: 414 Joined: 10/11/2012 Location: Austin, TX User Profile |
The problem is I know JUST ENOUGH statistics to get myself into trouble but not enough to get myself back out. LOL But, it's an interesting idea -- maybe just something for our group, though. | ||
| SteveJ Veteran Posts: 105 Joined: 10/11/2012 Location: UK User Profile |
Thanks Steve & Mark. | ||
| Steve2 Elite Posts: 750 Joined: 10/11/2012 Location: Annapolis, MD User Profile |
Things would be so much easier if market position outcomes were like flipping coins. Sadly, I think it's more like flipping a weighted coin where the market gods vary the weight on each flip. I routinely capture max winning and losing streaks on the simulations I do. Here's an example of how sequentially correlated position outcomes can be. My live portfolio consists of 23 strategies (20 RTM, 3 Trending), has a hit rate of 66.3%, TPM of 105.9, and a CAR of 98.6%. A 6+ year simulation (2008 to present) contains a maximum winning streak of 39 positions and a maximum losing streak of 30 positions. Let me add my thanks for all the contributions you've made to the forum. Your stuff is always an interesting read. Steve | ||
| Steve Mayo Legend Posts: 414 Joined: 10/11/2012 Location: Austin, TX User Profile |
I hear you! The 2008 one-day drop was 6 sigma -- that's only supposed to happen once every 1.4 million years in a normal distribution! (The 1987 crash was 21 sigma drop. The biggest gain was 11 sigma in Oct-2008) In reality, a 6 Sigma drop occurs about once every 6 years (a 6 sigma one-day gain occurs about every 20 years) so the market clearly doesn't follow a normal bell-curve distribution, at least at the two tails. One hopeful thought: you are undoubtedly counting TRADES not mark-to-market end-of-day equity which is what that equation presumes. :-) PS: Someone asked me to explain the statistical jargon better. Sigma basically means standard deviation. The average daily return on the S&P is around 0.03% and the standard deviation of that return (a measure of variability) is 0.98% (read: a lot of volatility for very little average return). A 6-sigma loss would be a day when the market loses 0.03 - (6 x 0.98) = 5.85%. Do the compound return calculation on that 003% gain and you get the 8% average return (slope of the equity line) of the market over the long haul (assuming no change in market characteristic!) -- if you can stomach the fluctuations in the short term. In comparison, the ARM4-Margin port has an MONTHLY average return of 5 with an SD of 8.87 (I haven't calculated it on daily) meaning that 95% of the time (in theory, at least), the monthly return should be between +/- 2 Sigma or -13% and +23%. That's over a 7-year period...it can be much different in a shorter timeframe, of course. | ||
| Flying Dutchman Member Posts: 24 Joined: 3/17/2014 Location: The Netherlands User Profile |
Thanks for sharing the optimum hit rate etc. Could you direct me to some background articles on such statistics? |
|
|
|
|
Administrator
Forum Moderator |
Registered User
Unregistered User |
 E-Mail this thread to a friend |
Toggle e-mail notification |
