Thursday, 15 December 2011

More on Algos

I had two questions to yesterday's post and thought they merited a post as an answer. The questions were:
Anonymous said...
Hi, what minimum level of Profit factor would you consider the system usable, assuming all other statistics are okay?

Anonymous said...
Hi EL,

I thought you liked high win ratios (70% or better). These algos have win ratios less than 50%. What changed?

Everything in an algo is a trade off. It's like a chair with four legs. If you cut an inch from one leg then it impacts the other 3 legs.

Profit Factor is just one metric. Remember that Profit Factor is:

(PW * AW) / (PL * AL)
  • PW = Probability of a trade being a winning trade

  • AW = Average win size

  • PL = Probability of a trade being a losing trade

  • AL = Average loss size

I look at win rate, average trade size (not just average winning trade size), targets winning and losing trades, and, probably most importantly, how the equity curve looks on in sample and out of sample results. As a general rule of thumb, a PF between 1.3 and 2 is OK of the other metrics fit. I am wary of PFs that are too high asthey are usually the result of either too small a sample or curve fitting. That is not to say I reject it, but I look more deeply at the results. In the end, the SIM trading validates it all.

Nothing has changed in my desire for high win rates. In my discretionary trading it's a must. In algo trading I may buy a trade off by accepting a lower win rate if I get something for it. The issue is that the bigger the average profit is, usually the lower the win rate. So it comes down to the picture that a particular algo is designed for. Issues such as the average trade size become important. The smaller the profit, the bigger the impact of slippage. However, the higher the win rate the smaller the impact of slippage as slippage occurs on losing trades not winning trades.

As you can see, designing an algo requires answering many questions and making lots of decisions. What makes it easier is that you press the "go" button and have a whole bunch of statistics so you can make an intelligent decision. And if you have done the proper work including the Walk Forward Analysis, you have a good chance of ending up with a robust algo fit for purpose.

The chart below is a MultiCharts DAX chart running that algo. In the first 3 hours of trading there were 3 trades. 1 loser and two winners. I also have the ability to have the algo stop trading once a daily profit (or loss) target has been met. This can increase the win rate of the algo too. Operating an algo on this basis can be a good idea for many traders as it can ensure the probability to finish green on more days.

1 comment:

  1. Thank you for your explanations! I asked the first question.