TabPFN used for Trading, part 2.

Here is the follow-up to my previous post which shows out of sample results over the last week, 19th to 23rd of October.

Again a somewhat mixed bag, but with one huge caveat: this is the latest model of TabPFN over a smaller train data set. In my previous post the train data set was 17 week's worth of data and the TabPFN model used was version 3.0. In between then and now I updated to version 3.5, which in general is a bigger model, and I found that I was pushing the memory limits of my GPU with that much training data and then predicting on a following week. As a result, I found I had to reduce the amount of training data to be able to use my current framework, which currently means training on 1 week's worth of data to predict on the following week, using one model for all types of price behaviour.

This has two profound implications for my use of TabPFN:

• I will need to be much more parsimonious with my feature set to reduce the memory footprint so that I can include a longer history of training data
• I will probably need to train a series of more specialised models, e.g. one model for cycling price data, one for trending price data etc.

More in due course. 

TabPFN used for Trading

It has been quite a while since my last post, and during this time I have been investigating a couple of complimentary things: TabPFN and revisiting my Kalman-EMA filter. I say complimentary as I have used the Kalman-EMA as a feature input to TabPFN, among other features, and the results seem promising so far. For example, here are a few screenshots of equity curves on a few, selected forex pairs:

and finally, one that is not quite as good as the above.

I won't go into much detail about this now as I plan to write a post at the weekend with true, out of sample results, but I will note that the white equity curves are the equivalent of a buy and hold, the cyan and accompanying red ema is the raw signal, and the yellow, blue and green are some simple, practical trading applications.

 

More soon. 

Expressing an Indicator in Neural Net Form, Final Update

Just a short update. Over the last few months I have been investigating the idea of optimising an indicator by applying the apparatus of neural net training to the calculation of weights for the indicator values rather than simply varying a look back period length. It has been an interesting investigation from which I have gained some insight and also had occasion to program some useful back testing infrastructure, but ultimately I have come to the conclusion that the light isn't worth the candle and I have now ceased this work, with nothing of real value to report.

I now intend to move on to other more promising projects, which I think will be a better use of my time. More in due course.