Why Sequential Trade System Statistics (like Drawdown) Are Meaningless. Becoming A Better System Trader part 3

In Blog

If you missed the first two posts in my Becoming A Better System Trader Series here are the links:

  1. Difference Between a Ferrari and A Mercedes Benz
  2. Why Not To Use Dollar Based Stop Losses


In this third installment, I am going to argue that the most popular measure of system risk, drawdown, is about as useless as it gets when looking at your trading system’s performance.


I know I am always stirring up controversy, but hear me out.


Drawdown assumes a certain sequence of trades that will and cannot be repeated. That is, the order of wins and losses your system experienced in the backtest will, 99.99% positive, not happen in that order in the future.


I will provide an example of how/why drawdown is useless, but first here is a list of sequential statistics and non-sequential statistics:


Sequential (bad information coming from the order of wins and losses)
-profit run up
-overall equity curve
-trade dependence (serial correlation of trades)
-consecutive wins or losses

Non sequential (good information coming from individual trades)
-average win to loss ratio
-average win
-average loss
-winning percentage
-expectancy (obviously)


There are many other useful and useless statistics that can be generated, but the main idea is to only make decisions based off statistics you can control.


For example, changing system metrics to reduce average loss size is a better idea than adding some rule that simply reduces the drawdown, but keeps average loss size the same. “But now the system is less risky, the drawdown is smaller”. Not at all the case. I know this sounds crazy, but if your system takes large losses and there just happened to be no string of large losses in a row during the backtest period then your system will “test” greatly.


However, once you enter live trading and new data is presented, your system may take a string of large losses that can destroy your account and you can easily realize double or triple the size of your backtest’s drawdown, but it could all be completely normal behavior and you may take a good system offline because the original analysis was flawed; not the system.


The system could also take a longer string of consecutive losses than occured in the backtest presenting the same problem. Another area out of our control, but can be appeased by proper decisions during the design and testing stage.
So a typical, and flawed, rule of thumb in trading books is to take 1.5x the backtest’s drawdown and use that as your “oh, no. the system is broken” rule.


This is arbitrary and random. Why 1.5x? Why not 1x, 2x or 5x? This is not proper system development. We cannot control drawdown size because it is product of trade sequence; that is, a non-repeatable series of events.


We can control individual trades which in turn provide you with an average win or loss size for the back test. Focus on this and you will see improvements in system performance. Focus on drawdown and you will undoubetly see the “unexpected” or “unseen” come true and become costly.
So great. You believe me. Now you probably want to know how to calculate a drawdown or risk measure that we can use.


There are two ways I use. First, you can look at average trade size (+ or – standard deviations, if you prefer) * losing percentage * number of trades in the future (you select how far you want to look).


For example, if you’d like to know what risk to expect in the next n trades use this formula: risk = losing percentage * average loss size * n
This won’t give you a drawdown for the entire system (because you do not need it), but will give you a measure to judge system health over the next n trades. In our ever evolving world this has to suffice. This is very basic and rudimentary assessment and not my preferred method.


The second way is run a monte carlo simulation in order to determine capital required to maintain a certain confidence level that your drawdown will not exceed a certain percent of your account size.


However, this method is too complicated and requires examples that would drag this post on. I will save it for subscribers or another post – maybe part 4?!?


Remember control what you CAN control. Individual trades = YES. Sequence of trades = NO.



As always, stay green.

David – TradeMyAlgos.com
p.s. here is my about me blog post in case you are curious who I am: Meet The Founder Of Trade My Algorithms

Recommended Posts

Leave a Comment