my trading system backtest
- Thread starter hanhao
- Start date
I think this is ok. You're returns are not a normal distribution. You probably have one trade that has a 250% return and a handful with returns that are very close to 0. The average is about 40%, but the standard deviation is larger because you have a few outliers to the upside.
Statistically, you don't have enough trades. 342 over 7 years is probably not actionable. It's a good start, but you probably need atleast 10 times more trades before you've got anything you can trade. I'd also be very suspicous if say 1/2 your trades were from a narrow block of time.
i further refined and curve fitted it with new conditions that give rarer but surer signals. Of course, the upside goes down. again, i am uncomfortable about the standard deviation of the winners being so high.
Winnings:
Average percent gain 25.66%
Standard Dev 27.40%
Total count 112
Losings:
Average percent gain -7.37%
Standard Dev 6.24%
Total count 105
Summary
Aveage percent gain 9.15%
Average Standard Dev 16.81863073
Total Count 217
Win/Lost count ratio 1.066666667
Percent gain/lost ratio 348.31%
Winnings:
Average percent gain 25.66%
Standard Dev 27.40%
Total count 112
Losings:
Average percent gain -7.37%
Standard Dev 6.24%
Total count 105
Summary
Aveage percent gain 9.15%
Average Standard Dev 16.81863073
Total Count 217
Win/Lost count ratio 1.066666667
Percent gain/lost ratio 348.31%
there are many securities and i shall give the average
average winning trades / losing trades = 1.3/1.5
average profit/average loss = 2.35 <== this result cannot really be accurate because when a stock has only no losses or no winners, the results come out as "N/A"
top 3 trades :
121%
104%
100%
bottom 3 trades:
-25%
-23%
-22%
average winning trades / losing trades = 1.3/1.5
average profit/average loss = 2.35 <== this result cannot really be accurate because when a stock has only no losses or no winners, the results come out as "N/A"
top 3 trades :
121%
104%
100%
bottom 3 trades:
-25%
-23%
-22%
Maybe you could model the account growth and shake it up with some monte carlo work. You would then be looking for a number of account models of growth that were close together and all rising smoothly. That is what I call the "comfort zone"... it is the same thing as reducing the std dev of gains probably..
I was reading up on Qigong and a master said that when we are relaxed and happy, when we have our inner smile, our mind and body works best. after that I added the idea of the comfort zone to the backtesting. It was the last piece of the puzzle and allowed me to let the mathematics guide me to where I want to be...
If you are making all your money on a few trades might as well do what you did and filter out the bad ones... personally I like "grinders", systems that make a lot of smaller gains and maintain consistency from day to day.
I was reading up on Qigong and a master said that when we are relaxed and happy, when we have our inner smile, our mind and body works best. after that I added the idea of the comfort zone to the backtesting. It was the last piece of the puzzle and allowed me to let the mathematics guide me to where I want to be...
If you are making all your money on a few trades might as well do what you did and filter out the bad ones... personally I like "grinders", systems that make a lot of smaller gains and maintain consistency from day to day.
Quote from hanhao:
haha qi gong
so there a software to do the monte carlo work? you are refering to doing the scan on an entirely random stock price right?
Not random data, you take all the real signals but you do a random ordering of them and repeat it. Each repetition will represent a different account with the difference being sort of which is a lucky trader and did not get a big drawdown and which is unlucky. Or just take all the worst signals and pretend they happened at the outset and the same for all the best signals.
If the monte carlo produces a set of account value lines that are bundled pretty close together your system is probably pretty good, providing you meet all the usual criteria regarding having enough samples to be meaningful, etc. I'm happy with 30 samples if just perusing a chart for patterns and 100 for a backtest, some insist it takes millions of samples, personally I did not learn that in my statistics classes...