Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
I think I will put this current model on the shelf, and start manually backtesting my next model
Reasons:
You can not set all the parameters for model1 in NT wizard> too complex
I do not know nearly enough C# > and I figure I want a tradable model before I am a half decent programmer
It challenging to manually backtest model1 since from 14 April to now it has only produced 53 trades. I remember Mike saying in one of his posts that he likes models that trade more frequently
at 57% win rate, it has huge DD and not worth paying someone to code my strategy
Like I mentioned in this previous post, I put the old model on the shelf and started backtesting a new model.
So the setup on my new model may occur a few times a day (compared to the old one which only setup like a few times a week), so I decided to take the three first trades of the setup of each trading day (sometimes it only setup 2 a day) for a current total of 52 trades.
I am loving this new model, I know its only 52 trades= N, but there is only a losing streak of 2 trades and winning streak of 3.
Honestly I don't think putting things on the shelf due to difficulty testing is the way to go. I know its harsh but what your doing is just delaying the inevitable, learning to code in order to test properly. Yes, with this system you have more trades (N) in a shorter period, yes it looks sexier. But the small 30 day window you tested means nothing, its no where near a complete market cycle nor is it representative of a financial year.
My advice stands from what i mentioned above. learn to code, and then you can test properly. This type of testing is going to be prone to finding small blips of activity with may not actually have an edge over the long run. in fact it might just be random phenomenon because of how short the duration is.
Why don't you just test the whole year instead of randomly testing months and "looking for correlation" . I emphasis that because you are putting implicit bias in to your testing and research. Yes it might not be overt but it is there once you begin doing it.
Paying for coding at first seems reasonable but usually prices get out of hand. In general look at a few hundred $$ just to get it started, even if the strategy test ends up unprofitable. The reason I say that, is your are going to have to pay for any debugging time, changes in code, or any other modifications. You will explain as best you can to the developer but chances that the signals and code comes out exactly as you want it first time, slim to none. So if your time is worth more then that go ahead and do it. But spending a few grand to test these 2 systems to me seems like a lot, but then again I can do it myself.
It also limits your flexibility, if you want to do isolated exit, entry or parameter sweeps. Again going to cost you and you are going to have to be very very specific on how you want to do those tests and view the results.
So I have to learn C# eventually. I am not spending a few grand on anything.
For Model #2 I can B.T. 12 trades in approximately an hour (thats four days of trading), so I could backtest an entire month of trading (20 days) in about 5 hours.
so would there be much error if I backtested the whole of 2014 (so far) assuming there is 55 trades a month? N= 9 x 55 = 495.........N=495 seems enough?
Hey Trey thanks for all the great advice, I owe you big time.
Model #1 is really hard to 'code' or implement in the wizard.
As with anything learn to walk before you can run. Its better to code some easy strategies that may not be profitable. But I can tell you from personal experience simple strategies can be profitable and are usually more robust. Then work up to your complex ones. Not only will it give you practice in the development cycle, which is as crucial to an algo tester as is trade evaluation cycle for a manual trader. But you will get a sense of what "works" and what "doesn't".
Yes if you did a year and had N=495 I would consider that a much fairer sample than what you have shown before. Honestly 5 hours isn't that long. I have spent many many more than that debugging alone.
No problem for the advice. Hope to see your progress soon.
Personally I learned C++ 15+ years ago in school but i thought this would help you with some learning resources. I have used the Lynda series to refresh myself and also Microsoft virtual academy. Lynda is paid but pretty good (or you can find it some other method wink wink). But the Microsoft virtual academy is really great for a free resource with high quality vids.
I would like to inquire who would be doing the backtesting (that I am paying), is it a quant developer? or just someone who knows how to code?
That leads me to my next question, do I have to hire a "professional backtester" , or can I just hire some kid who knows how to code and write a script with no trading experience?
Ave Loser = 0.72 (brought down after implementing new trailer rule)
Expectancy ratio = 0.18
This is probably one of my most statistically accurate models that I have backtested, with an accuracy score of 1/SQRT (106)= 0.097.
As you can see the metrics are horrible> probably wont continue backtesting this? Model one with an expectancy score of 0.54 is better (but I need to backtest that more, with N= 57)