Please conduct research for a blog post on https://herox.com/IARPAGFChallenge
For a blog post introducing the Geopolitical Forecasting Challenge, we would suggest posting about current state-of-the-art techniques in geopolitical forecasting and setting up a framework for what contestants are trying to beat. Using this logic, we suggest posts on polling models and their failures in predicting elections, machine learning models used to quantify political risk, and The Good Judgment Project, which trains ordinary people to make accurate predictions. We have compiled ideas and background information for three possible blog posts below.
FUTURE OF POLLING
Currently, there is a lot of talk about how geopolitical forecasters failed to conduct polls that would predict both the Brexit vote in the UK and Donald Trump's victory in the U.S. presidential election. To discuss these turn of events where, in 2016, two major forecasts completely failed to provide appropriate predictions, we found three sources.
The first source titled "After 2016, Can We Ever Trust Polls Again?" talks about the U.S. election and why it was such a surprise. It talks about the failure to account for the correlated error when it comes to predicting the outcome of elections, and lists people who got it wrong. For example, FiveThirtyEight’s Nate Silver, "who successfully predicted the 2008 and 2012 presidential races, was among the few pollsters to at least account for correlated error", and gave Trump a 28% chance of winning. Nate Silver's approach could be worth mentioning and might spark some interest for the contestants to do additional research.
The second source is Science Magazine's article titled "Improving election prediction internationally". In the study, the authors developed prediction models for elections and covered 86 countries and more than 500 elections. Their models were able to predict 80-90% of elections in out-of-sample tests. This could also be a good starting point for potential contestants when it comes to giving them initial information on currently available research and methods.
The final source for this topic takes a different approach, showing that prediction polls are more accurate than prediction markets. A study called "Distilling the Wisdom of Crowds: Prediction Markets vs. Prediction Polls" the authors compared the effectiveness of prediction markets and prediction polls. They conclude that prediction polls were more effective for forecasting geopolitical events.
MACHINE LEARNING and political risk
Machine learning is currently one of the most-discussed topics when it comes to the prediction and analysis field. We were able to identify a press release that is introducing GeoQuant, a machine learning software that "fuses political science with computer science to address current inefficiencies in the field of measuring political risk for investors, executives and other business decision makers". The article talks about the most recent development when it comes to merging technology and geopolitical prediction, and introduces state-of-the-art technology.
THE GOOD JUDGMENT PROJECT
During our research about geopolitical predictions, almost all sources mentioned The Good Judgment Project. This project trains ordinary people to be superpredictors, allowing them to make accurate and confident predictions. The potential contestants can draw from the published results which examine "the development of confidence and accuracy over time in the context of forecasting" and can also be introduced to the project which uses the wisdom of the crowd to predict events. This approach can be taken in order to show what are the advantages of the Geopolitical Forecasting Challenge when compared to other competitions, such as The Good Judgment Project.
We list three different topics for blogs posts: future of polling, machine learning and political risk, and the Good Judgment Project. Each of these topics can be made into a blog post to introduce the Geopolitical Forecasting Challenge.