NATURAL COMPUTATION IN FINANCES

Update August, 2022.

Natural computationalism, the subject of my PhD research, suggest the idea of discovering causal processes of information (algorithms/rules) instead to make adjustments on structures (neural networks for example) to approach a description of a behavior (datasets and its trends, for example). Here is one of my first tries of going further than approaching a data set trend. I thought then that it would be possible to approach the essence of the regularities instead the data points themselves, then I trained a neural network to predict change rates. This could be -I thought at that time- a sort of meta description of data.
Well, I was in a mistake. This experiment, even with a high level of accuracy predicting changes in prices of stocks (a methodology that is not new anymore) resulted strongly dependent on the data also. It was equivalent to calculate the first derivative. Even is provide some more information on the subject phenomena, it does not tell me much about its causality.

Recently I did similar approach using a much more sophisticated mode of artificial intelligence that considers the change of data in time (LSTM), but, even we obtained better results, still being not what we are looking for

I have to say that today this is not the case. I'm currently working on this processes of causality from a different perspective based on Algorithmic probability the basis of the natural computationalism, and also on the construction of agents that deal with regularities in datasets to take decisions.

I started to think that the forecasting is not the right path. Honestly, I almost want to abandon the idea

--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Application of Natural computation paradigm to prediction of financial rates.

(Details are attached to this article as a draft in pdf format)

Roughly, main premises of natural computation claims that the universe itself is the result of computations that take place in the most fundamental levels of physical reality. But in the context of natural computation to talk about computation does not means that the universe or computations in the universe are performed in a digital computer or in a Turing machine. No, to talk about computation in natural computation is to talk about to process information by means physical substratum.

Under a most general view, algorithmic complexity establish that there are strong reasons to think that despite the chaotic (non predictable) or non reducible behavior of the universe, it is possible to think that such complex behavior is the result of rules or algorithms well specified, because nature, the most of the times responds to them.

Under this view, all phenomena that happens in the universe, including finances, can be seen as the result of specific rules.

This research summarize natural computation premises to artificial intelligence methodologies in order to prove that is possible to predict the behavior of very complex phenomena by means finite and algorithmic computations.

To prove it, we use data available about behavior of finances (from yahoo finances), and using artificial neural networks we predict with a very good approximation rate the behavior of some of the main finances rates that characterize financial world as can be seen in the next picture.

Probably the most interesting aspect of this information is that, from the natural computation pint of view, we applied some of the principal axioms that characterize the algorithmic complexity, this is, that is better to find the rules that produce the behavior than to try to approximate data itself.

This last was possible focusing analysis in the differences of the the behavior of the financial rates than in data itself.

Note: This research is part of the INFOTEC research.

https://www.infotec.mx