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)
so I have completed my manual backtesting of 101 trades
First sample is from 1 Jan - 23 April 2014.
Second OS is from 1 Aug - Oct 30 2014 .
My first question is should I add the begining equity amount to the end of the first sample?
ie if the ending equity balance at 23 April was $5.00 should I add it to 1 August and start the second sample with $5.00? and build on that?
(My opening balance for my second sample was set to $1 , commissions was set to .0019 (based on the spread charged by AAAfx at the time I started the backtest)
Here are my stats (calculated based all the trades in both samples):
Win Ratio 41%
Loss Ratio 59%
Ave Win = .0013 (EUR/USD)
Ave Loss = .0008
N = 101
Max DD = 89%
Please let me know if I am not following proper backtesting procedure.
Would you trade this model (in conjunction with another pair using the same model (Portfolio Backtesting)?
What other metrics would you require when deciding to implement this model?
Thank You to those that responded, have a great weekend!
If this is a walk forward testing model yes you should carry your previous P/L and current position inventory with you. Honestly your max DD is too high, 89% is really bad. When you take this to monte carlo you will bust out almost guaranteed.
No this was not walkforward testing (not quite sure how you define it), but like I said in the previous post two individual samples (of 50 trades each), First sample = Jan 1 - April 23, 2nd sample = 1 Aug - Oct 30 2014 .
The second sample was my "out of sample" sample.
So you still want me to add my closing account balance from April 23 to the opening balance at 1 Aug? (if this is the case I dont think I will have such a huge drawdown)
What do you mean by 'previous P/L and current position inventory with you' sorry it sounds interesting but do not understand the fancy jargon.
I would study your equity graph (if you have one) and honestly consider if you can stomach the ups and downs of your account's journey. See this example.
Look at this equity curve, and then tell me you wouldn't love to have these results, right?
Great, now send me $1,999 :)
Just kidding. OK, let's look at this equity curve a littler more, and see if we think we can realistically handle …
Along those lines I like to find streaks (both winning and losing) to prepare myself what kind of emotional turmoil I may have to endure during a long losing streak. With a 41% win rate you're likely to have longer losing streaks than winning streaks.
Also, are you considering slippage? And, depending on the strategy, your actual fill prices may not be what an actual market will give you, or you may not get filled at all but your model assumes you always get filled. This can change things dramatically.
1) Well walk forward testing is defined by Robert Pardo in his book The Evaluation and Optimization of Trading Strategies. But back to my statement. If Sample 1 is your in sample, and Sample 2 is your out of sample then NO you do not add your equity from 1 period to the next. Sample 2 starts again at initial equity.
If it was walk forward, and Sample 1 was the first testing window and parameter set and Sample 2 was the second testing window and parameter set, then yes you would add the equity from the first period to the second.
2) In short, Monte Carlo is a stress testing method and through sampling without replacement allows you to generate confidence intervals for trading statistics like MaxDD, Net Profit, avg trade value etc... It is basically a type of statistical bootstrap. I will not delve any deeper since there are actually many methods that can use this type of testing. But on futures.io (formerly BMT) trade resampling without replacement is the most common.
If you are having trouble with my "jargon" and the definitions here. I suggest you watch some of kevin daveys webinars I am pretty sure he covers this material. As well as pick up a few algorithmic trading / development books. Not to be a ball buster but most of this is extremely common and essential to building and testing good systems.
In regards to including slippage into your model which was posted above. I take the average spread or the max spread (if i am playing more conservative) and subtract that from every transaction/trade. That will help make your tests more realistic. So you have your spread sheet of trades, look at the maximum spread your broker has and subtract that from every trade and recalculate your stats. Thats a worst case scenario when you have to get out in high volatility, this is especially true in spot OTC FX. spreads can be in excess of 10-15 pips and your stops may trigger doing those times. So it should be built in. At worst you underestimate your profit, but its better to do that than under estimate your risk and exposures.
What if I tested the same model with the same parameters (during the same timeframe) on the ES or GBP/USD......would you estimate or expect similar metrics?
Thanks, I will read Pardo's book. and I nearly done the lynda course, sort of went through all the lectures but not entirely understanding everything.
I would not expect similar metrics. There is a caveat, that is that this system was developed from stage 1 with the intent to trade multiple instruments or a basket of instruments. Then you would want to see similiar metrics but you may not actually have that result.
It depends on what statistic your trying to find the std error of. For example you want to find the standard error of your average trade. Take the StdDev of all of your trades / sqrt(nTrades). This can be used as a std error approximation for pretty much anything, assuming a normal distribution.