1/ This thread inspired by @ndehouche response to my thread on forecasting. Once upon a time, I believed forecasting was just a matter of grinding away at assumptions and perfecting the "model". I figured all was needed was working harder on the math and when confronted with my
2/ limitations with math, we hired statisticians, PhD's in math (even MIT IYIs)...Moved well beyond regression. Wasted hundreds of thousands of dollars on failed models. I did not know why they failed. They added more and more variables and forecasts got even worse.
3/ We decided to stop running our business BASED ON FORECASTING DEMAND. We still projected inventory needs (demand), but we KNEW they would be wrong. Our business started to flourish because we placed clipping left tail risk ahead of maximizing right tail profit.
4/ Our business has always involved fat tailed distribution. I never heard of fat tailed (extremistan) formalized until reading Nassim Taleb. All of his books articulated not only my failures but the solutions we tinkered or discovered long before reading his brilliant Incerto.
5/ Forecasting demand for every business I've been in was an exercise in futility and in retrospect, we had mistakenly assumed demand followed a Gaussian distribution curve when in fact, we operated in a fat tailed environment. I did not have the language to describe the problem
6/ I reject using Amazon's methodology for forecasting or any company that has devised a multitude of ways to transfer risks of poor forecasting. I maintain the single best way to approach demand forecasting to assume you'll be wrong and plan accordingly.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
