Pastcasting: a very cool project to get better at forecasting the future

One of the main challenge when trying to improve one's forecasting skills is to get as much feedback as possible.

But getting prediction feedback is rarely easy. If you focus on economic growth, you will get a feedback every quarter or so. Thats is not many opportunities to test and adjust your assumptions.

If you focus on geopolitics, it is even worse. Many forecasting questions take years to resolve and you are not even sure to be able to face similar issues in additionnal years or decades.

You are in a better position if you forecast data that is frequently updated - say your company's sales, a website traffic or stocks. You may actually get really good at anticipating short-term fluctuations. But it is likely that your data will eventually be affected by broader trends (macroeconomic, geopolitics...) for which you are not getting frequent feedback.

Pastcasting: a cool idea to get lots of feedback

Pastcasting is a project I found on the website https://www.quantifiedintuitions.org/.

It consists in answering forecasting questions that has already happened, from a vantage point further in the past. Those questions are sourced from resolved issues in known prediction markets like Metaculus and Good Judgement Open.

Here are some of the questions you may find when "playing":

• Will American Airlines file for bankruptcy protection before 2021? [The date is 2 sept. 2020]
• Will the S&P 500 close higher in 2020 than 2019? [The date is 25 oct. 2020]
• Will Montenegro become a full NATO member before 1 January 2018? [The date is 20 avr. 2017]

As a big advocate of introducing much more practice when training forecasters I like the idea of faking the future in a more or less realistic environment. This echoes to approaches we are implementing at Centolla to train our staff and refine our techniques.

Contactez nous

Envoyez nous un message ci-dessous et nous vous répondrons.