Custom optimization criterion - Profit to Max Relative Drawdown

Avatar
  • updated
  • Completed

Hello everyone,


I wanted to share a method I've been using manually to evaluate trading strategies. During optimization, I typically choose "Max Relative Drawdown on equity" and then extract the results into Excel. The next step involves dividing the Profit by the Max Relative Drawdown on equity using the formula:

= Profit / Max Relative Drawdown on equity

This calculation provides me with an insight into which strategy generates the most profit per 1% of Drawdown.

The only downside to this manual approach is the inability to utilize genetic optimizations. Consequently, I find myself resorting to brute force for most of my sets.

Despite this limitation, I find the metric valuable as it allows me to reconsider sets that would usually be discarded. For instance, one practice I enjoy is identifying the set with the best profit/drawdown ratio and then normalizing the lot size to achieve the desired drawdown.

Let me illustrate with an example: I have a set that boasts the highest profit/drawdown ratio, earning $20k with an 80% drawdown. Considering a 2% risk per trade, I adjust the risk to 0.5%, resulting in a 20% drawdown. Now, I have a set with the highest profit for a 20% drawdown, a set that would typically be overlooked.

Thanks for reading!
Avatar
2
Andrey Khatimlianskyi
  • Completed

Will be available in 2.58.4 (soon)

Avatar
1
lytnin
Quote from Hannes Waser

While this method eq/maddDD is intuitive, I would like to bring a minor note to your attention.

if we calculate equity/mean(DD) the optimizer can use it much l, much more effectively.

(mean values are easier to derive during mathematical processing).


Fair enough.

However, i like looking at the max to have a sense of how bad things can get, and if its worth the risk.

Avatar
1
Hannes Waser

While this method eq/maddDD is intuitive, I would like to bring a minor note to your attention.

if we calculate equity/mean(DD) the optimizer can use it much l, much more effectively.

(mean values are easier to derive during mathematical processing).