Last year, I did some public one-shot forecasts for the first time. I modeled my attempt at this from Scott Alexander who has done this for a time and I think it is a fun way to make sure that I keep track of what is happening in the world. It is also good calibration practice. Below are the predictions, with the ones that happened in bold and those that did not happen in regular text. (If the outcome of an event is ambiguous, the question is italicized and I’ve not counted that question in my scores or calibration.)
To start with, my Brier score for the predictions ended at 0.1641. With Brier scores, a lower score is better. My score on these one-shot forecasts can be compared with my Brier score on the full-predictions I have made where I’m allowed to update my prediction as new information comes up, there I have a Brier score of 0.1048. Metaculus average Brier score is 0.098, where they are also able to update their forecasts as new information comes in. Using the log-scoring formula (x * ln(p) + (1-x) * ln(1-p)), where x is the outcome and p is the prediction), my negative log score was 0.4762, where, again, lower scores are better. Excluding the personal questions, where I can affect the result and spend different amounts of effort depending on the prediction, my scores decreased somewhat to 0.1775 (Brier) and 0.5083 (Log).
I’m satisfied with these results, though they could of course be better. One factor that contributed to a worse score is that I included some sports predictions, where I did not cover my entire probability distribution with questions. For example, in my Formula 1 predictions, I correctly predicted Max Verstappen to win. But since I did not include my probabilities for the other drivers, I got a bad 0.67 Brier Score from that prediction. Excluding the sports-predictions decreases my Brier Score to 0.1500, which is better.
Calibration
This plot shows how well calibrated I am in my predictions, for simplicity I’ve converted all predictions below 50% to the top bins by taking 1-p. The red line shows perfect calibration and the blue line is the fraction of predictions in that bin that were correct. Being under the red line means over-confidence and being over the red line means under-confidence.
I’m pretty well calibrated on the upper bins, but it is clear that I’m over-confident on the lower bins. Though I should note that I only did three predictions in the 50%-bin which means that I would have been poorly calibrated on that bin either way. It thus seems that I should become less confident in my predictions around 60% and 40% for coming predictions.
Predictions 2022
Swedish Politics
Liberalerna not getting into riksdagen: 75%
Miljöpartiet not getting into riksdagen: 35%
Socialdemokraterna getting the most votes: 87%
Sverigedemokraterna getting the most votes: 4%
Moderaterna getting more votes than Sverigedemokraterna: 54%
Any party currently not in riksdagen getting more than 2% of the votes: 5%
Magdalena Andersson being the prime minister after the election: 35%
More shootings in Sweden in 2022 than in 2021: 44%
Unemployment rate in Sweden lower than 8% at the end of 2022: 35%
US Politics
Anyone in the US (inside the borders) being killed by a drone: 15%
Joe Biden dying: 8%
World Politics
China attacking Taiwan: 9%
Nuclear war: 1.2%
Crypto
Bitcoin over $70 thousand at the end of 2022: 30%
Bitcoin below $10 thousand at the end of 2022: 8%
Ethereum over $7.5 thousand at the end of 2022: 25%
Ethereum less than $1 thousand at the end of 2022: 5%
Ethereum overtaking Bitcoin in market capitalization anytime, and for any duration, during 2022: 5%
Market capitalization of the entire crypto-market, according to CoinMarketCap, spending an entire week over $3 trillion in 2022: 65%
Bitcoin dominance, also according to CoinMarketCap, being reduced to less than 30% for any amount of time during 2022: 75%
The floor price of Bored Ape Yacht Club above 80 ETH at the end of 2022: 25%
Higher volume on OpenSea in 2022 than in 2021: 90%
Personal
Posting more than 12 blogposts during 2022 (including this one): 60%
Posting more blogposts in 2022 than in 2021: 85%
Post this list of predictions before January 10 2022: 95%
Post a mid-year update to this list of predictions: 15%
Posting at least two blogposts in each month of the year: 15%
Running a combined total of more than 350 kilometers during 2022: 60%
Running a 10k faster than 45:00 minutes: 25%
Going to bed earlier than 22:00 for 10 or more nights in a row: 1%
Finish more than 40 books: 90%
Getting at least 4 excellent (highest grade at SSE) in the 8 courses I will take next year: 65%
Being happier in 2022 than in 2021: 55%
COVID
Me catching any variant of Covid during 2022: 45%
Over 6.6 thousand dead from Covid in Sweden during 2022: 30%
A new variants appearing that leads to a significant amount of cases worldwide (> 2 million cases): 70%
Science/tech
The JAMES WEBB Telescope finding something groundbreaking: 5%
Clear evidence for life outside of earth: 0.00001%
AI solving hard math or science problems by itself: 2%
Over $2 billion in venture capital funding for the development of fusion: 65%
Elon Musk or SpaceX announcing a formal mission to Mars: 60%
Sports
Winter Olympics 2022 getting canceled: 15%
Sweden getting more than 10 medals in the Winter Olympics 2022: 72%
Sweden getting more than 15 medals in the Winter Olympics 2022: 35%
Norway winning the most medals of any country in the Winter Olympics 2022: 33%
Liverpool winning the Premier League: 26%
Chelsea winning the Premier League: 27%
Manchester City winning the Premier League: 30%
F1 winners
Max Verstappen: 15%
Lewis Hamilton: 13%
Nikita Mazepin: 2%
A winner that is not from Mercedes, Ferrari, or Red Bull: 22%
Merceds winning the Constructor’s: 23%
Other
The queen of England dying: 15%
Kendrick Lamar releasing another album: 65%
Another period of only online teaching at SSE: 10%
Over $400 million moved by GiveWell during 2022: 75%