Thread by @JayUlfelder

I don't have a job right now, but I used to get paid to build models to try to forecast rare events emerging from complex systems. So, for what it's worth, the two best things I've read so far on forecasting this pandemic are...

.@robjhyndman's blog post on forecasting COVID-19, in which he warns that, "Forecasting pandemics is harder than many people think," and "fitting simple models to the available data is pointless, misleading and dangerous."
robjhyndman.com/hyndsight/fore…

.@zeynep's April 2 @TheAtlantic piece in which she argues that "right answers are not what epidemiological models are for."
theatlantic.com/technology/arc…

I'm thinking about this now because of an email exchange with my brother, @ulfelder, who was recently asked in a TV appearance if the apparent inaccuracy of many COVID-19 forecasts should undermine confidence in long-term forecasts of climate change.

IMO, what @zeynep said about epi models applies nicely to forecasting climate change, too: "The most important function of epidemiological models is as a simulation, a way to see our potential futures ahead of time, and how that interacts with the choices we make today."

With COVID-19, we're often seeing numbers of confirmed cases and fatalities that are lower than early predictions, precisely because people saw and are responding to those super-grim initial forecasts.

With climate change, we're seeing things get *worse* faster than early predictions indicated -- hotter temps, less ice, etc. -- precisely because people are not taking enough action in response to those forecasts.

In neither case do these inaccuracies prove that forecasting is doomed or pointless. Instead, they remind us that forecasting on topics like these is more about scenario development than soothsaying, and, when done well, the forecasts will often sow the seeds of their own demise.