Glitch8
 ( 12.08% )
- ago
Please enjoy this new blog post by our very own Eugene!

https://www.wealth-lab.com/blog/perfect-exits
4
636
6 Replies

Reply

Bookmark

Sort
- ago
#1
Thanks for the blog post. I have some questions:

1) When we are talking about setting 1 to 3 StdDev's, are we talking about StdDev's around the regression line? I just want to be sure (even if it's obvious to everyone else).

2) When one places a trade, they need to input a stop-loss price. So how does one get from from an MAEPct of -6.63 to a dollar figure so a stop-loss price can be entered in strategy code? Perhaps there's a MAEPct2Price datatype I don't know about. (One can hope.)
1
- ago
#2
1) The standard deviations relate to the Axis X values which in the case of Perfect Stop/Target are MAE% and MFE% of all trades the system made, respectively.

2) Assuming the backtest's position sizing is intact when applying to live trading, you can go from a percent stop loss to a dollar figure very simply, no?
0
- ago
#3
QUOTE:
1) The standard deviations relate to the Axis X values which in the case of Perfect Stop/Target are MAE% and MFE% of all trades

Now you've lost me. You're taking the mean of all MAE%, then shifting that mean value by 1 to 3 standard deviations to the left without out even consulting the regression line? But you have an excellent linear relationship with the regression line. Why not leverage that relationship somehow since it's predictive? The regression line is predicting what percentage trades succeed verses fail based on MAEPct. Why not employ that knowledge?

QUOTE:
2) ... you can go from a percent stop loss to a dollar figure ...

You mean ...?
CODE:
Position p = LastPosition; double stopPrice = p.EntryPrice * (1.0-0.0663); PlaceTrade(bars, TransactionType.Sell, OrderType.Stop, stopPrice, "-6.63 MAE%");
I was expecting something more complicated.

UPDATE: Fixed. It's now a Stop order.
0
- ago
#4
1) Could you explain in layman's terms what you're proposing to adjust? Maybe a formula. Thanks.

2) Yes, except that it should be OrderType.Stop.
0
- ago
#5
QUOTE:
1) Could you explain in layman's terms what you're proposing to adjust? Maybe a formula.

Absolutely. You have a beautiful linear relationship with your regression line. You have to take advantage of that. Consider the scatter plot below. It gives us an opportunity to control win rate very precisely by taking slices of the scatter plot at different MAE% values. For instance, say I want a win rate of 50%. The first vertical purple line shows a slice through the scatter plot at an MAE% of -4.0, which is yielding a 50% win rate.


Now if I shift the purple line slice to the right at an MAE% of -2.0, now you can see a win rate of 80% on that second slice. So we can program what we want our win rate to be, and the scatter plot will tell us what our MAE% must be to achieve that win rate. That's our goal.

Now you can use the regression line as the "center" of the scatter cloud to estimate this process. But it's probably easier to directly plot the Win rate vs. MAE% and read these values directly off this new, second plot. Try doing that in your example.

What would be really interesting is to plot (win rate)*(profit%) vs. (MAE%)
1
- ago
#6
I think the reason the Position Profit% vs MAEPct regression line is so linear is because one is modeling the losses if they don't sell early enough. That's a linear relationship, which is interesting. We fail to see this initially because we are thinking about profits when we should be thinking in terms of avoiding losses.

Another observation is that the market can't react to the MAE value because that's a function of the position taken by the trader (which will be different for each trader). But the market can react to the most recent Highest price (or more robustly, the higher percentile of the price), so from a modeling prospective--where you're modeling market behavior--using the most recent Highest price (or 95% percentile of the price) makes more sense than using the MAE.
0

Reply

Bookmark

Sort