Darp8
- ago
Maybe this exists already. But here is the idea. You can have a strategy that signals 3 times more than the cash you have. At the same time when you look at charts of stocks and where they enter the human eye can see things not in the program.

So set it to Stage. Then look at the chart on entry and pick the 1/3 that is best. Am sure this is often done.

But how do you backtest?
Think possible, you run the backtest, and on each batch of signals on next bar, the backtester does what would happen live, in past You see 3 setups as example, you can only see the chart up to that bar and you select the best one out of three.

Thus the backtest stimulates precisely what would happen in real world, and you get your backtest result that is valid with your intervention and decisioning making. It would be very cool, I would love to do that.

Is there a way to do that now?

Ah, there is a way but takes time. You do it live, then a week later you compare results to what did happen, you put in more money into paper testing so all trades are taken. Then compare WL score between your picks versus all signals.
0
403
4 Replies

Reply

Bookmark

Sort
Glitch8
 ( 7.81% )
- ago
#1
The whole point of this is to quantify what you see with your eye to remove the subjective element out of it.
0
Darp8
- ago
#2
Glitch, yes I know. Am talking about what your mind thinks is the best of three but you cannot describe, so you cannot code something you cannot describe.

The first step would be to do it, and see whether your "intuition" was right or not via back testing it. If it ends up doing this, say you get 70% CAGR instead of 20% then it can be worth that extra human time for picking the best 1/3.

The next possibility is via some type of detector to quantify some way to program it. Such as detecting what are the differences between 2 groups? In general, that ability is super valuable ignoring the question at hand.

If you can just put say 400 setups/trades taken, then the best 100 and the worst 100 results, then have the software tell you these 6 indicators at these value ranges would take 80% of the best outcomes and not take 60% of the worst, and make sure it is not backfit, then wow would that be valuable.

You have added a lot to WL, am still checking that out. Per chance is there something like that in WL?
0
MIH8
- ago
#3
QUOTE:

Am talking about what your mind thinks is the best of three but you cannot describe, so you cannot code something you cannot describe.


Machine learning algorithms are able to handle these kinds of tasks. For example, if you collect enough data, you can train a neural network. Based on the decisions of many thousands of trades, such a network will learn what you cannot formulate. On the other hand, you will not be able to extract this information from it as a simple rule. (One therefore simply uses the trained net).

This problem can be found in other disciplines, as an example in chess programming. There, neural networks can be used for training on the basis of several hundred million chess games or expert games. Knowledge that cannot be formulated by experts but is used in practice can thus be abstracted and applied.

Although WealthLab already offers a lot of tools, I don't see this area of application in the near future (yet).
2
Darp8
- ago
#4
Thanks MIH. I just saw something in a video that maybe does this to an extent. https://youtu.be/PWFM6pZkCd4?t=1090 Price Grid. Seems to match existing price patterns that worked in past. It only has price action, but that is quite a bit.

Anyone having success using Price Grid Matching?

Edit: https://youtu.be/PWFM6pZkCd4?t=1378 at that time Volker uses exact same words I did, "your eye can see it".
0

Reply

Bookmark

Sort