Bullshit, bullshit everywhere

The whole literature of using machine learning or advanced statistics for time series forecasting is just sectarian bullshit.

It takes some time for the realization that almost everything you read is bullshit to sink in. This is because writing itself degraded from a method of transmission of knowledge to profit and attention seeking activity.

Ok, why this is bullshit? Well, time series are logs of numerical measurements of some outcomes of some unknown set processes with unknown characteristics. Ok, partially known.

Any serious analysis require a regular closure - all the factors are enumerated and known (nothing is missed, nothing is imaginary or redundant), all weights are correct (no overshoots), no unknown factors could emerge, no known factors could disappear or lose its weights.

Do you know any such systems? Well, in classical mechanics and basic non-organic chemistry, perhaps.

Let me formulate the principle. Any forecasting based on logs instead of analysis of factors which produced these outcomes is a modern day alchemy or just highly sophisticated bullshit. There is even a meme for this - the past performance does not predict the future returns, which is exactly the same notion.

As long as the whole system continuously evolves while you write down your time series, they will always be outdated and already irrelevant by definition. But this is a second order realization.

The first order, at least in this universe, the Causality Principle, and unless you can close on a certain set of multiple causes (multiple causation is the universal principle) you cannot predict or forecast anything at all. Any correlations would be due to a chance and will be imaginary and wrong.

The factors which will influence the further events are either not here yet, or won’t be there, or will lose its significance or will gain an enormous weight.

One more time - time series, which are just logs of already happened outcomes, do not contain any actual relevant data to base the forecast on. It is just like predicting a fire by smoke instead of what is actually burning and other physical factors.

So, everything you read about forecasting financial markers based on time-series is bullshit. Correlations may emerge due to the fact that millions of people (and bots) are looking at and reacting on the same emerging patterns on an essentially the same chart, so there is always some synchrony and some consistent actions (like FOMOing in) out of it.

Why do people buy this crap? Well, because most of them think that it cannot be just all nonsense, and something must be working, and one should give it a try (without understanding, of course).

In reality, however, sectarian “teachings” and supporting literature is much more common and general phenomena, and yes, it is all sectarian nonsense.

Trading systems definitely could be made, but they must have nothing to do with forecasting or predicting (which is impossible due to the principles outlined above), but being reactive and ahead looking instead.

By the way, no animals forecast anything. They just know how to react to changes.