Select Net Profit > Net Profit?
Author: Carova
Creation Date: 2/7/2015 3:45 PM
profile picture

Carova

#1
Hi!

Do I have a mistaken understanding @ Select Net Profit. According to all the descriptions that I have read Select NeT Profit is the Net Profit with all the >3-sigma trades removed, so Net Profit >= Select Net Profit. Unfortunately, I have an optimization going where the Select Net Profit > Net Profit (see attachment in following post).

Do I have a misunderstanding?

Vince
profile picture

Carova

#2
Attachment

profile picture

Eugene

#3
Most likely, all the outlier trades were negative outlier trades and there were no positive outliers. After removing the outliers, the resulting profit increased.
profile picture

Carova

#4
Thanks Eugene!

Your explanation makes "sense" but then the Metric does not make sense. :( Why would I EVER remove the trades that go against me? The "fat tails" will always produce these losing trades in abundance.

I need to read more about this Metric.

Vince
profile picture

Eugene

#5
QUOTE:
Why would I EVER remove the trades that go against me?

You probably wanted to say, "why would I ever LEAVE trades that go against me"? ;-)

Breaking changes are undesirable but if there's evidence on how its inventors calculated it, or there's some industry standard, like "SelectNetProfit = Math.Min(NetProfit, SelectNetProfit)", I'm open to suggestions on how to fix/enhance the metric!
profile picture

Carova

#6
:) Correct!

I think all of the positive >3-sigma winning trades should be removed and ALL losing trades should remain. When we are building trading systems I believe that we should be designing for the worst case (which is NEVER that good in real life ;-) ) and the positive outliers are few and far between in real trading IMHO.

Vince
profile picture

Eugene

#7
Unfortunately, I need something more than your or my thinking: some recognized examples or practice. According to the horse's mouth (RINA systems was acquired by Tradestation),

QUOTE:
Select Net Profit = Select Gross Profit + Select Gross Loss.
profile picture

Carova

#8
I agree, this is just me talking. RINA (and the rest of the metrics world) do not agree with me about this. :(

Vince
profile picture

Eugene

#9
Vince,

The good thing that the world can not stop us from devising our own performance metric e.g. "Select Net Profit Retaining Negative Outliers".

If everything goes as planned, it may appear in a future release of the library.
profile picture

LenMoz

#10
Vince,

QUOTE:
the positive outliers are few and far between in real trading IMHO ...
ALL losing trades should remain.

Perhaps, but I see the intent as to remove the effect of the outliers, both positive and negative, not penalize the strategy. Outliers in backtesting also occur if there are data issues, such as an incorrectly handled stock split.

You may also want to review the discussion in this related thread, http://www.wealth-lab.com/Forum/Posts/Performance-Outlier-Trades-Discussion-and-Request-33933. It's provides the background for percentage-based "Outlier Trades, Adjusted".

Len
profile picture

Carova

#11
Len,

Thanks Len! This is a multi-faceted issue, and as might be expected, there are many views (mine being just one) in how to handle it. I envy your strategies that produce primarily positive outliers - I am not so talented. :(

Unfortunately not all these metrics are available for optimizer access under the scorecards. I am trying to learn C++ (as an intro to C#) but my time for this is limited by other competing family priorities! :) Eventually I plan to code up some of my desired optimizer metrics into custom scorecards. Optimizers are merciless in displaying the limitations of using most current metrics as targets.

Eugene,

QUOTE:
The good thing that the world can not stop us from devising our own performance metric e.g. "Select Net Profit Retaining Negative Outliers".

If everything goes as planned, it may appear in a future release of the library.


Great! I can't wait!!

Vince

profile picture

Eugene

#12
QUOTE:
Perhaps, but I see the intent as to remove the effect of the outliers, both positive and negative, not penalize the strategy.

Well said. I'll agree with Len's point here. Let's not rush with adding this performance metric.