Member-only story
What makes a good trading model in particular and models in general? Huge field of study in the philosophy of science with such words like epistemology, ontology, supervenience, counterfactual, ontic, lots of words with implications of $0.00, but I wanted to consider it in a very defined context to make money in trading or machine learning.
It’s interesting to consider the validity of models derived from empirical data and used to predict and act, in particular for trading the markets, and in general to minimize a cost function in supervised and/or maximize a utility function in supervised Machine Learning. There are three implicit assumptions that are usually never even considered.
Assumptions
1. Theoretical Sufficiency
There is sufficient data to create “any” model that can use empirical regularities in the past to predict and act on the future, at any level of precision, especially at a level of precision that can overcome trading costs and statistical noise.
2. Practical Sufficiency
Given the possibility of theoretical sufficiency, the assumption that the specific model you build or trained is good enough even if there is a theoretical mapping from past to future
