Backtest sorting by lowest drawdown
- This topic has 39 replies, 7 voices, and was last updated 4 years ago by
.
-
I’ve been thinking about drawdown and although PRT does not let us sort our backtest results according to which backtest has the least drawdown it is possible to manually check this. If you reduce the starting capital and then run an optimization backtest some results will result in losses if the money runs out when drawdown is greater than capital. Keep reducing the capital and repeating the backtest and it will soon be clear which variables result in more acceptable drawdown.
This may be what some on here already do but I thought that I would post it just in case it helps someone achieve better backtest results.
In this image the start capital is 100000 and all tests are profitable but as can be seen the drawdown is massive.
Whereas in this image the start capital is 1500 and some tests have resulted in so much drawdown that the money runs out – but there is clearly ranges where drawdown is more acceptable. Obviously if the drawdown all happens late in the strategy then this method may not work.
An alternative would be to code something that monitors drawdown from the start of any period of being OnMarket and quits the strategy (or makes capital zero) during a backtest if a level of drawdown is exceeded. Does anybody already do this and care to share their code?
1 user thanked author for this post.
At the moment I have just added the following to my strategies so that if drawdown is greater than desired that particular backtest is stopped with the QUIT instruction. This has the added benefit of reducing backtest time. I can then sort the results by Max Gain or Nbr Bets and the results at the top of the table are most likely to be the ones that did not quit during the backtest and so are of the most interest.
If anyone has a better way then it would be good try it.
12345MaxDrawdown = 1500IF OnMarket and (PositionPrice - Low)*CountOfPosition > MaxDrawDown THENQuitENDIFIf you use the above code on a strategy with money management then it won’t work very well. Best to use a percentage of equity as your maximum allowed drawdown if that is the case.
123456789Capital = 3000MaxDrawDownPercentage = 66Equity = Capital + StrategyProfitMaxDrawdown = Equity * (MaxDrawDownPercentage/100)IF OnMarket and ((PositionPrice - low)*CountOfPosition) > MaxDrawDown THENQuitENDIF4 users thanked author for this post.
My concern is usually not so much lowest drawdown, but highest ratio of gains/losses. Drawdown is an absolute number, and can be quite large for a system that makes even larger gains (system No. 1).
On the other hand, a system with only small absolute gains (system No. 2) may have small drawdowns as well. However, when you scale this system up to higher contract numbers, it may yield the same total absolute gain as system No. 1, but a much higher absolute drawdown.
This is why I prefer to look at the ratio of gains/losses instead of absolute drawdown. The ratio of gains to losses is independent of the number of contracts you trade.
So, if someone could find a way to sort backtest results by this ratio, it may be indeed quite helpful. However, usually, when you combine absolute gains and gain per position, you may come quite close to the best ratio of gains/losses.
Of two systems with the same absolute gain but different number of positions, the one with fewer positions will usually have lower drawdown and a better ratio of gains/losses. At least in my systems.
1 user thanked author for this post.
GraHal, here is another snippet I have posted today, and it may be interesting for your database. It sets the stop loss to break-even after a certain price movement into positive territory :
https://www.prorealcode.com/blog/learning/trailing-stop-max-favorable-excursion-mfe/#comment-5885
1 user thanked author for this post.
Weekly or daily charts for me usually come quite close to buy and hold.
If buy, hold a bit, sell, wait a bit, repeat = more profit than buy and hold after trading costs then it is a worthwhile strategy. Not exciting to trade but any profit is good no matter how it is come by!
No problem regarding the code for the snippet thing GraHal. In fact maybe the answer to the lack of others adding to your library can be solved by you personally hoovering code up when you see something you like. I think the process of adding may be too labour intensive /confusing for most people. Hopefully Nicolas will create a permanent library when he has some spare time! Bit off topic so back to drawdown and other closely associated subjects….
1 user thanked author for this post.
Find exclusive trading pro-tools on 
