Hello.
I’m a masters student.
My dissertation is to do research on the use of ML algorithms in Trading(Finance),
I’ve been advised by my supervisor to consider CLA as an alternative for Deep Learning/RL/Q Learning, but having read some pdfs ,documentations and examples of HTM/CLA implementations I feel like there is no way to make it work in my dissertation or in case if there is it is not suitable for Masters dissertation maybe it is for PHd.
Am I right ?
Here is a PDF with links a sample codebase of a mobile applications for stock analysis using HTM.
I am a little put off by trying to predict stock.
Other than spotting that the market has some general trend it is essentially a random number generator.
I could see a case where you dumped in massive data for all aspects of the economy to spot leading indicators but I have to point out that this has been tried for a long time and to be best of my knowledge, nobody is doing much better than guessing a coin toss.
Yeah this seems about right. It requires a large amount of domain knowledge and natural language understanding to interpret what’s going on and how it might affect market sentiment. For example, how do you predict geopolitical events like the killing of Qasem Soleimani a couple of weeks ago?
These realities must make it harder to get good signal/noise ratios from stocks – which makes it into a feature engineering problem. How to get metrics with enough periodicity to at least distinguish predictable from random activity. Maybe use multivariate anomaly detection to look for general shifts in an asset class like currencies? It seems all about sussing out signal and not going for anything too specific.
We have centuries of market training data in place. Many people have done the equivalent of autocorrelation with virtually every conceivable factor thrown into the mix. Even standard leading indicators are not very reliable; for example - according to the baltic dry index industry should be on it’s knees right now.
The problem is that the major influence factors are essentially black swans that are by definition unpredictable. These random factors filter through the system and are reflected as random price moves. Garbage in - garbage out!
Just wanted to point out that here at https://intelletic.com/ we are using HTM technology to trade futures and we are definitely doing it better than coin toss. As mentioned on the website, we have back-tests that have produced 47.3% simple annual returns and our live tests conform to these metrics.
Not surprising in a rising market.
Get back to me when the market goes into bear territory.
Our technology works regardless of whether the market is going down or up. We are able to profit from both the market directions.
Keep up the good work!
Btw, if I was getting anywhere near 47% annual return, I wouldn’t be looking for investors!