Stijn Leupe shares his thoughts on defining models, and using data from models and different modeling techniques.
Lies, damned lies and … financial models
It is often said that everything can be proven with (misleading) statistics. The same could be said about financial models.
First we need to get out of the way the miss conception that statistics are required for modelling. Yes, there are statistical models and even more complex Artificial Intelligence models, but there are also expert rule models (eg if a customer had a payment default in the last year, we do not grant a new credit since we assume the risk on future default is too high) sometimes based on historical observations.
But what is then a model. The most easy definition in my eyes is: ‘A model is a simplification of reality’. So some kind of assumption/simplification is made. Without us being conscious, there are a lot more models around us than we think. For instance, the laws of motion of Newton are a model to look at mechanics. These laws work very well in everyday situations, but no longer ‘predict behaviour’ once you move near the speed of light. Also the BCG matrix is a model, used to prioritize between businesses in a clear but simplified way.
And even we people use mental models, we call them instincts, prejudice, fears, …
So instead of immediately diving into the data or the different ‘modelling’ techniques; the first and most important discussion should be: how will we use the model and what type of simplification do we accept.
The next time you thus see a not functioning model, please rather look at it as a model not fit for use. The applied simplification just does not hold for the situation you want to apply the model, but probably within that simplification the model was correct. A bit like those ‘damned statistics’, the math behind the statistic is correct, it is the use, the conclusion drawn that is not correct.
To end by paraphrasing Einstein “Everything should be made as simple as possible, but no simpler”
PS: image generated via Stable Diffusion, using the title of this post as prompt