I have get two capital curves from two strategies on the same stock as below pictures: note: the red numbers on the left-up of the pictures is total orders and win ratio. Now how shall I confirm which one is better, such as comparing their final profit, biggest backward, win ratio and so on. Whether there is a general method which can quantitatively evalue different strategies' effect, which is better and how better? Thank you!
I think your better off researching this topic then asking on a forum. Theirs many ways to measure a system. I'm an amateur in this world and maybe someone will have more to add. I use Amibroker myself and have just recently read a book by Howard Bandy called Quantative Trading Systems. Great book and I got it a lot just for examples of working through Amibroker AFL. I'm not an expert on this and some of these terms might not be recognized in the industry and may pertain to Howard or Amibroker. One of the first parts he goes over in system development is having an objective function to optimize your system around. This function could be Max System Drawdown, Max Trade Drawdown, Net Gain, Risk Adjusted Return, Sharpe Ratio, K-Ratio(basically a slope of the equity, smoother equity=higher k-ratio), Win/Loss Ratio, %Winners, perhaps the number of trades, the holding period. You need to have some sort of goal for optimizing or designing a system. Optimizing is basically trying to add/change criteria to make the system reach a higher mark for your objective function. Could be changing a stop loss, holding days, some variable in the system, etc. Not sure what your using to back test but this kind of comes into play sometime when I use Amibroker. They have an optimize function where I can enter some variable and have it be optimized through a range that I set. In a regular back test it will show me all the results for every step of the variable I asked. It will show about 40 columns for the different objective functions I'll call it or basically stats. So in basic optimize, I could have a variable saying price is above 50 day moving average. I want to see what happens if I use a different moving average. So I use the optimize function and ask it to go from a 5 day moving average to a 50 day moving average in steps of 5. It'll provide me a list of the results for each moving average in that system. Well you can go through this list, and kind of look through it and pick one to put in the system. This is all with In Sample data which I'm using to optimize(which you can easily over do if not careful). Maybe one provided a very nice annual return but it had a 30% drawdown in the system at some point, could you handle that drawdown live? Maybe it had a sharpe ratio of 7 but it only had 10 trades over 10 years, you probably couldnt' handle waiting for trades like that. They also have a walk forward optimize. I use the optimize feature, and then instruct it what objective function I want it to use for the In Sample data to maximize and then walk it forward into a out of sample data range and it'll provide me a report of how it performed in the insample data and out of sample. So say I wanted to optimize CAR/MDD which is compound annual return/ max draw down. It will go through the in sample with my optimize parameters, choose the best result for my variable and then run it through the out of sample and compare how it performed. This is just an example as to how theirs many ways to measure a system I believe. It needs to fit your style and risk tolerance for it to work. I often manually optimize as the walk forward optimize isn't sometimes what I want to use but that was just an example of how you need to have the paramaters you want in the in sample data to build the system. I've had some systems I've tried designing that could provide astounding results, but could produce a drawdown of 30%, is this something I could handle for a year live trading? Maybe it has a very sporadic equity curve. Maybe its profitable but its only right 30% of the time, is that something you can trade? Maybe that curve is from a backtest result you have above, how does the forward test of out of sample data look like with that system? Do they match each other? Maybe your system just fit the in sample data you we're using. Then you could get into the Monte Carlo analysis, what are the possibilities of these trades if they we're in different orders. It's a very complex world that I've kind of started studying into and I find fascinating. The above is just what little knowledge I've acquired and maybe its completely bogus to any professionals here. I'm just trying to show that you have to use what works for yourself. A few metrics I look for are Sharpe Ratio, Max Drawdown, Annual Return, and Percent of Winners and Exposure. These just fit what I'm looking for and you could have very different goals. Theirs many great resources and books out there for system development.
Hi Metamega, it is a long reply. I think it must waste your much time, so thank you for your sincerity , and what you say is very important for me. Yes, the two figures are come from my strategy which I have tested it on many stock by Matlab codes. Most of them are nice, at least get much profit; I test this strategy for about 10 years data on Day lines, some ones is 15 years, some ones are 5 years(not each stock have long history). I think this strategy is ok, but I should do more details about open orders, especially closing orders on which I have spent much time, but just get a little. After testing much method of closing order, I find it is still better that close order when price is below the latest buttom. Maybe it is because I still know a little about evaluate the effect, so now I am seeking for good evaluating tools. Thank you for your kind advice. Have fun and good trade!
MDD is a good starting point, ie. the biggest drawback in % of the account value. Of course the smaller this value is, the better it is. Usually this number is a negative percent number. Small then means towards zero. One can also smooth the equity curve by using MA etc, and then calc the MDD, but this method is seldom used. Additionally the ROR (Rate of Return) on a per period basis is a good measure. For example the average ROR% per month.