I have found that its fairly difficult to get an Mar much above about 1.3 over any time period on this system .. I think it has its attractions because theres a limited number of parameters which indicates your results are probably very robust .. has anyone else found anything different.. Im not asking for secrets just observations...
Regards
Chris
ATR Channel B/o System
CBO or Turtle?
If you are talking about straight ATR/CBO, that's probably correct. If you are talking Turtle, then other rules significantly enhance the MAR ratio (last trade rule, correlation rules, scaling rules, risk mgmt rules). I've seen as high as 1.8, depending on portfolio selection.
Cheers,
Kevin
Cheers,
Kevin
-
Bondtrader
- Roundtable Knight
- Posts: 101
- Joined: Mon Apr 28, 2003 11:07 am
- Location: Zimbabwe
Chris, I agree that it's possible to choose (1) a portfolio of markets and (2) a time interval of simulation and (3) a schedule of commissions and slippage, which prevent Veritrader's built in system "ATR Channel Breakout" from achieving MARs greater than 1.3.
However it is also possible to choose a portfolio and a time-slice and a C+S schedule, which allow Veritrader's ATR Channel Breakout system to yield MAR's in the 2.5 to 3.0 range.
It's all a matter of what problem you're trying to solve.
Chris, I infer that you aren't using the Veritrader presupplied data from Techtools, with the "Turtle Futures" runset, with the time period set to Jan 1993 thru Jan 2005, using the Veritrader out-of-the-box default commissions and slippage. If you were then you'd be getting an MAR of 1.7 for Veritrader's builtin ATR Channel Breakout, rather than the 1.3 that you reported. So again, it depends on what you are setting out to accomplish.
However it is also possible to choose a portfolio and a time-slice and a C+S schedule, which allow Veritrader's ATR Channel Breakout system to yield MAR's in the 2.5 to 3.0 range.
It's all a matter of what problem you're trying to solve.
Chris, I infer that you aren't using the Veritrader presupplied data from Techtools, with the "Turtle Futures" runset, with the time period set to Jan 1993 thru Jan 2005, using the Veritrader out-of-the-box default commissions and slippage. If you were then you'd be getting an MAR of 1.7 for Veritrader's builtin ATR Channel Breakout, rather than the 1.3 that you reported. So again, it depends on what you are setting out to accomplish.
Last edited by Bondtrader on Thu Jul 15, 2004 5:08 pm, edited 1 time in total.
-
Old European
- Roundtable Knight
- Posts: 144
- Joined: Mon Apr 19, 2004 3:04 pm
- Location: Old Europe
Chris,
I've also been playing around with the ATR Channel Breakout system.
It is certainly possible to come up with a quite robust system with a MAR of 1.75 or more in tests over more than 20 years, with a portfolio of approximately 30 futures markets and under very reasonable assumptions for e.g. earned interest rate (0%), RT commissions (USD 15) and slippage (7.5% / with roll-over accounting).
I consider such system certainly as quite decent and definitively as a candidate for mixing with other systems.
Old European
I've also been playing around with the ATR Channel Breakout system.
It is certainly possible to come up with a quite robust system with a MAR of 1.75 or more in tests over more than 20 years, with a portfolio of approximately 30 futures markets and under very reasonable assumptions for e.g. earned interest rate (0%), RT commissions (USD 15) and slippage (7.5% / with roll-over accounting).
I consider such system certainly as quite decent and definitively as a candidate for mixing with other systems.
Old European
Given the vagaries of computers generally and the fact that simulations are acutely sensitive to a number of variables (even the order of data files in the .SET file), does it make sense to post some standard results based on the VT default models and portfolios? Perhaps c.f. or Dan could do this if it seems worthwhile.
This doesn't involve the disclosure of anything proprietary and ensures that we are all working from the same page in our discussions as well as providing a test of the actual program installation.
Norwegian
This doesn't involve the disclosure of anything proprietary and ensures that we are all working from the same page in our discussions as well as providing a test of the actual program installation.
Norwegian
-
Forum Mgmnt
- Roundtable Knight
- Posts: 1842
- Joined: Tue Apr 15, 2003 11:02 am
- Contact:
The results should NOT be dependent on any particular computer unless there is a bug in VeriTrader.
Each of us running the same exact settings on the same exact data should get the same exact results every single time.
So I'm not sure what good this would do.
The results are very dependent on things like rolling algorithms start dates, portfolios, so most of the variations people see are due to those factors. Even something as simple as using the default settings in CSI's Unfair advantage results in different files depending on what version of that program you use since they have changed the defaults over the last year or so.
So to know if this is worthwhile, I need to know what problems you are trying to solve with the suggestion you made? Perhaps I don't understand the problem or we can solve that problem some other way.
- Forum Mgmnt
Each of us running the same exact settings on the same exact data should get the same exact results every single time.
So I'm not sure what good this would do.
The results are very dependent on things like rolling algorithms start dates, portfolios, so most of the variations people see are due to those factors. Even something as simple as using the default settings in CSI's Unfair advantage results in different files depending on what version of that program you use since they have changed the defaults over the last year or so.
So to know if this is worthwhile, I need to know what problems you are trying to solve with the suggestion you made? Perhaps I don't understand the problem or we can solve that problem some other way.
- Forum Mgmnt
Minor Variations
I think we can expect minor variations between implementations. Anyone who's done Numerical Analysis can relay how mathematical imprecision effects all computers. Throw in different numerical precisions and formats due to language or platform choice, and there are bound to be differences, even with the same data set. If we consider data variations per vendor, there are bound to be even more. All this is just a way of saying that IMHO, "ballpark" figures are probably good enough.
Cheers,
Kevin
Cheers,
Kevin
-
stancramer
- Roundtable Fellow
- Posts: 60
- Joined: Sat Feb 07, 2004 10:34 am
- Location: Washington DC
This was discussed on the Van Tharp chatboard last month, http://www.mastermindforum.com/phorum/r ... t=7969&v=f and here is a sample:
In quite a number of cases the motivation for "replicating the analysis" is timidity, lack of confidence. Re-doing something that a Guru has already done, allows you the luxury (seldom found in real trading) of knowing whether or not you got The Right Answer. You don't have to depend on your abilities as a programmer or your instincts as a trader or your skills as a troubleshooter/debugger, to decide whether or not to trust the results of your analysis. All you need do is compare against the Guru's results. It's easy. It's comforting. It's unambiguous. All the things that actual trading is not.
The amazing thing is how often these "replicators" spend all their time tiddling around with their Exactly Matching the Guru experiments, and never quite manage to create a new idea, or test an idea that no one has tested before, or -heaven forbid- actually open an account and trade a system in the real world with real Euros. They're too busy chasing down a software difference in the accounting of commissions on Leap Year, or other such distractions. I've seen it.
I too have come up against similar problems in trying to figure out whether my results are in the ballpark.
The biggest problem I find is not knowing the approximate values to strive for. I would find it incredibly useful to have a list of the key parameters for say the 3 main types of risk profiles (Trader, Investor & Banker) and the 3 (distinct) categories of markets, i.e. Futures, Stocks & FOREX. The parameters I am referring to here are not the system specific values but rather things like (using a hypothetical $1,000,000) the acceptable values for End Balance, CAGR%, MAR, Sharpe, Max TE DD & Longest DD, being the main reporting criteria for backtesting runs (the End Balance figure (using say $1mln acct) is would be useful simply to be able to gain a reference of point of Risk/Reward…)
For example I am testing a combination of ATR B/O, Triple MA & BB systems against the FOREX Sample portfolio, after having run a ~750 run test, adjusting the Risk% per Trade parameter for each of the systems I am presented with a set of possible End Balance and drawdown values that range from $4mln & MaxDD of 53% to $12.8mln & MaxDD of 78%. Naturally I have already excluded some of the obviously bad results/combinations that generated big DD & small End Balance… My question remains, what is OK or am I still completely of course here?
I’ve had similar dilemmas testing Stocks, Futures and combinations of markets & systems.
Any help/advise would be greatly appreciated…
The biggest problem I find is not knowing the approximate values to strive for. I would find it incredibly useful to have a list of the key parameters for say the 3 main types of risk profiles (Trader, Investor & Banker) and the 3 (distinct) categories of markets, i.e. Futures, Stocks & FOREX. The parameters I am referring to here are not the system specific values but rather things like (using a hypothetical $1,000,000) the acceptable values for End Balance, CAGR%, MAR, Sharpe, Max TE DD & Longest DD, being the main reporting criteria for backtesting runs (the End Balance figure (using say $1mln acct) is would be useful simply to be able to gain a reference of point of Risk/Reward…)
For example I am testing a combination of ATR B/O, Triple MA & BB systems against the FOREX Sample portfolio, after having run a ~750 run test, adjusting the Risk% per Trade parameter for each of the systems I am presented with a set of possible End Balance and drawdown values that range from $4mln & MaxDD of 53% to $12.8mln & MaxDD of 78%. Naturally I have already excluded some of the obviously bad results/combinations that generated big DD & small End Balance… My question remains, what is OK or am I still completely of course here?
I’ve had similar dilemmas testing Stocks, Futures and combinations of markets & systems.
Any help/advise would be greatly appreciated…
- Attachments
-
- worst, middle & best End Balance using a $1,000,000 account.
- Results Table.JPG (22.32 KiB) Viewed 13035 times
To stancramers point above I am reminded of a quote attributed to Keyne's (paraphrasing):
"It is better to be roughly right than precisely wrong"
ie just make sure you're making profits and not losses before worrying about comparisons with someone elses benchmark metrics
Whenever I start to delve to deeply into the minutiae of a problem this little phrase pops into my mind and I then make a conscious effort to step back and look more at the forest and less at the trees.
JamesK, perhaps you might find it helpful to come at the problem from the other side, YOUR side !!
What performance is reasonable/acceptable/desirable to you ?
What are your financial/trading objectives short, medium and long term ?
Try writing out a specific "business" plan and you may find that in the process of doing this you can answer many of the questions you raise.
Besides, does it really matter what the so-called industry benchmarks are ?
What's important is whether or not your trading is realising the objectives you have set for yourself.
I would not stop trading a system simply because some performance metrics were below a particular benchmark.
Equally, I would not stop trying to improve my system just because I had exceeded the benchmark.
TC
"It is better to be roughly right than precisely wrong"
ie just make sure you're making profits and not losses before worrying about comparisons with someone elses benchmark metrics
Whenever I start to delve to deeply into the minutiae of a problem this little phrase pops into my mind and I then make a conscious effort to step back and look more at the forest and less at the trees.
JamesK, perhaps you might find it helpful to come at the problem from the other side, YOUR side !!
What performance is reasonable/acceptable/desirable to you ?
What are your financial/trading objectives short, medium and long term ?
Try writing out a specific "business" plan and you may find that in the process of doing this you can answer many of the questions you raise.
Besides, does it really matter what the so-called industry benchmarks are ?
What's important is whether or not your trading is realising the objectives you have set for yourself.
I would not stop trading a system simply because some performance metrics were below a particular benchmark.
Equally, I would not stop trying to improve my system just because I had exceeded the benchmark.
TC
-
alerionsailor
- Contributing Member
- Posts: 9
- Joined: Wed Jan 05, 2005 8:20 pm