1. When models are wrong we make bad decisions
2. Models are good at known knowns and known unknowns. They cannot model unknown unknowns
So things like the impact of pandemics are missed in various models that don't consider them, like possibly an insurance model or a sales forecast.3. “If data is the new oil, models are the pipelines”
4. You should choose a model that is “good enough “ and run it many times
This will stimulate your creativity as you see different outcomes. Ultimately you should be using the model as a tool, not to make decisions without considering alternatives.5. “All models are wrong but some are useful”
Models are metaphorical and need to be interpreted in context.6. Sometimes models can be helpful because they force you to make a decision
Even inaccurate models can be helpful for this. The author gave an example of lost explorers that had the wrong map but believing that it was correct caused them to start moving so that they eventually survived.7. To truly calculate the value of a model for making decisions…
You need to take multiple outcomes into account. Compare the action taken without the model, compared to what the model suggests, and then the difference between the two. This is a much smaller difference than giving the model 100% credit for your decision. Sometimes following a model can lead to worse outcome than not following it.8. Models can encourage people to take on more risk than they should
Examples from long term capital management where they were right for a while and then very wrong .
Another hypothetical example would be an insurance company that underestimates long term risk. They would get business in the short term and then be very under capitalized in the long term when a disaster strikes.
9. Underestimating risk can be very profitable in the short term
See LTCM and insurance examples above.10. Models can work for a long time and lull us into a belief that they are correct. They will frequently fail dramatically.
11. In an economic model used for trading simply simulating a crisis can cause it to happen in real life
12. The first weather forecast model was done by hand in the 1920s and it was dramatically wrong however…
It set the stage for accurate forecasts using computers 80 years later 13. Tiny errors early in early iterations in a model (for weather forecasts, for example) can compound dramatically and make the results way off.
No comments.