"But no one knows!!"... Duh.
What are forecasting models?
- Not happened yet. You're forecasting because you don't have an answer—it's in the future. It's not model vs. truth. You don't have the luxury of truth. It is yet to be realized. So, it is always model vs. model.
- Not (uniformly) random guess. Some outcomes are more likely than others. Some outcomes are implausible. Work with that.
- Not certain. If you're mapping uncertainty to real numbers, you should have an explicit distribution. Encode your certainty in the distribution's spread because the mean already has a job.
- Not preference. The numbers you prefer may not be plausible numbers. Don't confuse them.
- Not reality. The model encodes reality. A one-to-one model of Phoenix would be a poor map. Focus on the salient; the signal.
- Not independent. All probability is conditional probability. You know things about the world.
- Not memorizing noise. Odysseus makes the terrible tradeoff of sailing between Scylla and Charybdis: underfitting and overfitting.
- Not unique. Other models work. More than one data-generating process could explain the data. The data cannot tell you if your model is right.
- Not useless. It explains past observations and generalizes to new ones. But don't intervene in ice cream sales to stop shark attacks.
Clayton, A. (2021). Bernoulli's Fallacy: Statistical Illogic and the Crisis of Modern Science. Columbia University Press.
McElreath, R. (2020). Statistical Rethinking: A Bayesian Course with Examples in R and Stan (2nd ed.). Chapman and Hall/CRC.
Smaldino, P. E. (2017). Models are Stupid, and We Need More of Them. In R. R. Vallacher, S. J. Read, & A. Nowak (Eds.), Computational Social Psychology (Chapter 14). Routledge.
The views and opinions expressed in this article are those of the author and do not represent the official policy or position of any employer, organization, or entity. This article is for informational and educational purposes only and does not constitute professional actuarial advice.
Generative AIs like Anthropic's Claude Opus 4.8 were used in parts of reviewing the writing.
