Hi all,

Not sure if this question is super obvious or conversely not part of WLP's focus, so apologies if either of those is the case.

I've created and optimized some intraday strategies and am happy with the general direction it is going (i.e., % of wins going up, higher returns per profitable trade, etc). So my next step is to look at the successful trades vs unsuccessful trades and try to identify any characteristics (eg what's going on that day, the days before, premarket, etc) that might help me further improve the strategy.

I understand that if I knew what indicators I was looking for, I could throw those into the strategy and see how they do. But I honestly don't really know which indicators to prioritize.

In a machine learning world, I would take all my trades, label them (Eg profitable / unprofitable) and throw the kitchen sink of indicator values at various scales into some type of algorithm (whether deep learning or random forest or ensemble or even just a regression) to find the highest value features.

I have played with Neuro lab a bit, but I think I may be missing the boat in terms of how people analyze their winning and losing trades to identify features (ie indicators) automatically. Is this something done in WLP?

Thanks for reading all the way down! Any insights would be most appreciated.

Thanks,

Jon

I've been anxiously awaiting other people's answer to your question about how to best select trading indicators, but I see there hasn't been many takers. So let's "turn this question around" so there is a possible methodical answer to your question.

Wealth-Lab's strengths lie with measuring the performance (e.g. Profit-per-Bar, Net Profit, Recovery Factor, positive Equity Curve slope) of strategies, not trading indicators. That's because trading indicators do not determine performance--trading strategies do. So how does one pick the best indicators? Well, you work backwards. You let Weath-Lab identify the best trading strategies via their performance metrics, then identify which indicators those strategies successfully employ to be great. Not the answer you were hoping for, but at least it's the truth. Some related thoughts:

Wealth-Lab breaks its indicators into three groups: Standard (using Wealthlab.Indicators), TASC (using TASCIndicators), and Community (using Community.Indicators). The Standard indicators are the prototype indicators, which are time tested. You should master them first. The TASC indicators have all been publish in Technical Analysis of Stocks & Commodities magazine, so they have undergone some reviewing for usefulness. The Community indicators are either notable indicators not published in TAS&C, or specially requested by a Wealth-Lab user.

The bad news is that these indicators aren't grouped by purpose--and they should be--so you can intelligently pick which are most appropriate for the needs of your strategy. Probably the best advice here is to buy a book that groups all these indicators by purpose. (I've done some of that myself manually, and it's a "work-in-progress" personal project that has been ongoing for four years.)

What I do look for in an indicator is one that ...

1) ... divides out (or normalizes out) the market noise with its denominator. Yes, that means you need to get familiar with their calculation. Examples of such indicators are CADO and CMO. The Kalman filter indicator has a nice example of employing the CMO indicator. http://www2.wealth-lab.com/WL5Wiki/Kalman.ashx

2) ... that is adaptive in some way. The MACD, MACDEx, and MACDEx_Histogram are adaptive by design.

3) ... incorporates both price and volume action together since both are important. EMV, VWMA, VMACDH, MFI, NVI, VZO are examples.

QUOTE:
In a machine learning world, I would take all my trades, label them (e.g. profitable vs. unprofitable) and throw the kitchen sink of indicator values ... into some ... [statistical] algorithm ... to find the highest value features [i.e. to determine which model parameters are most significant].

It is possible to interface Wealth-Lab to R to do exactly this (say with a logistic regression, multi-variant regression, principal component, or discriminate analysis). You might want to followup in this discussion topic. https://www.wealth-lab.com/Forum/Posts/Accessing-R-in-WLP-39084

I'm hoping someone else can add something to my answer. Happy computing to you.

@superticker, thank you so much for your reply! I'm extremely grateful. And while it does not get me to the magic button in WLP that I just didn't see :), it provides a good systematic approach. I will begin working through the categories and criteria you mentioned and will update here on my progress.

If you have any book recommendations (until your magnum opus arrives on shelves!), definitely let me know.

Thanks!

QUOTE:
If you have any book recommendations ...

I was afraid you were going to ask that. I bought John F Ehlers book, Cycle Analytics for Traders, but that book is strictly for EE engineers that are into digital signal processing (DSP) like myself. Ehlers is into seasonal (cyclic) trading and identifying and tracking (phase locking) those cycles. But not all stocks are seasonal (although if they are, you can predict their future and trade effectively on those predictions).