1
$\begingroup$

Some (stock market) traders have the ability to produce a high percentage of winning trades (80%+, positive return) over years. I had the chance to look into real money trades of two such traders and I also got trading instructions from them for research.

Now the interesting part is that if you strictly follow their rules then you usually end up with more losers than winners on the long run. But after a while you get some kind of subconscious "feeling" for winners which also shows in the results. I assume that this "feeling" is a hidden function which can be modeled.

My question is: Is there work about how to model such "gut feeling" and subconscious knowledge by means of machine learning (especially with little training data sets)? Is there relevant literature about this topic?

Regards,

$\endgroup$
  • $\begingroup$ Claude Shannon has published something about the entropic market hypothesis and build some analog circuits for stock market prediction. The problem is, that the exact source is unknown, because it was done in the time before the Internet has emerged. $\endgroup$ – Manuel Rodriguez Feb 19 '19 at 12:48
1
$\begingroup$

You could perhaps model gut feeling or subconscious bias as a prior in a Bayesian context, and then try to learn from the data how much to modify/moderate the bias in each individual case.

I think there is another issue with the problem you outlined. We might expect it to be normal to see more losers than winners in the long run: trading is a zero-sum game where the house always takes its cut. The trick to being a successful trading algorithm seems to be to make the losers small (cut them early) and the winners big (let them run).

| improve this answer | |
$\endgroup$
  • $\begingroup$ Thanks for your answer. The trading rules take care of the amount to lose/win (planned winners are always a multiple of the losers). For most traders I agree that money management is the main key to being profitable. But the interesting thing here is that the two traders have exact entry and exit rules based on price levels (in combination with order flow) which have a less than 50% chance of being correct when naively applied. But some alleged hidden function tells them when to take the trade and when it's better to stay out. $\endgroup$ – Hyndrix Feb 19 '19 at 12:57
  • $\begingroup$ More interestingly is, that I can also profitable trade with their rules after some training (with increasing success) by picking the "good" occurrences of all signals. $\endgroup$ – Hyndrix Feb 19 '19 at 13:00
  • $\begingroup$ How do you know which are the "good" occurences? If this is subconscious, you could try to bring it to consciousness through a psychological technique, like free-association. Alternatively, you could backtest your 'good'/'bad' calls - work through historical data, classifying each call from the algorithm as 'good' or 'bad' based on your intuition. When you have scored every call for a couple of years or whatever, you can make an objective assessment on how good your intuition is, and build a predictive model that uses the data on price, volume or whatever to predict your intuitive calls. $\endgroup$ – Jonathan Moore Feb 20 '19 at 11:10
0
$\begingroup$

It sounds like a supervised learning from not-so-many samples. Representation/metric learning could be of some help.

There are two books by Bob Volman that help quite a bit to make subconscious conscious. "Understanding Price Action" and "Forex Price Action Scalping". Despite sales-driven titles, this is the best non-scam attempt to formalize discretionary trading I ever saw. (Even though, one will become a one-trick pony at most, but on the other hand it's enough). If you google a bit, you'll find a dropbox with countless price charts commented by author.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.