Forecasting Update

Given some recent news and developments, I thought it would be appropriate with a quick forecasting update where I give some of my recent forecasting estimates and the reasoning behind them. I hope that this can help others update their forecasts and bring some clarity to the situations. Unfortunately, media does not really do forecasting with clear probability estimates which often makes it seem like some situations are more probable than they likely are.

Will there be a terrorist attack, labeled as such by the Swedish government, with more than 5 confirmed deaths in Sweden before Jan 1st 2025? 7.65%

The most salient example of this in recent times in Sweden is the likelihood of a terrorist attack. In the past week, both The Swedish Security Police (SÄPO) and the US and UK embassies have warned of a higher threat level than usual against Sweden. This sound incredibly grave and scary, but one should never forget the base rate. It seems like it is quite common that embassies, especially the US embassy, warns their citizens of higher risks of terror attacks. For example, in their limited archive, the US embassy in Sweden has at least 2 security warnings since 2020. There are likely many more than that in the previous 20 years since 2001. And that is not counting all the warnings issued in other countries. Thus this warning should not update you much on the risk of a terrorist attack in Sweden.

Another base rate to take into account is the base rate of terror attacks in Sweden. In about 115 years, there have been 14 terror attacks or uncovered plots of terror attacks. Of these, only one has killed more than 5 people. Using Laplace’s rule of succession, this gives us a naive base rate of 1.2% . But perhaps this is the wrong time span to use in constructing the base rate. If we instead look at the time since 2000, there have been 5 terror attacks and none of them resulted in more than 5 casualties. Using this for the base rate, we get ~4%. In my forecast, I use a mid point between the 2 base rates, i.e 2.6%, which gives me a total base rate risk of 5.1% over 2024 and 2025.

From here, we need to figure out how much to update based on the statements from SÄPO and the embassies. There should definitively be an update towards the upside based on the comments, as they are more likely to be made in a world where we are about to see a terrorist attack then not. I’m updating the forecast by 1.5x leaving me with a final estimate of 7.65% probability of a terrorist attack that leads to more than 5 confirmed casualties. 1 The yearly estimated probability is therefore 3.9%.

Will there be more than 10k confirmed human cases of H5N1 globally before Jan 1st 2024? 2%

Will there be more than 10k confirmed human cases of H5N1 globally before Jan 1st 2024? 4%

(Since I posted this on Feb 15, I’ve updated my forecast twice. First on the news of a girl in Cambodia dying from H5N1, a 1.5x update. And then on the news that none of the people that had been in contact with the girl have been infected, back to the original estimate.)

Another story, that I’ve not seen covered much in Sweden but which has gotten a lot of coverage in the parts of Twitter and the internet that I frequent is the transmission of H5N1 from birds to mammals in many parts of the world. For example, Science has reported that minks in Spain were infected. So far, human infection has been very rare and human-to-human (h2h) transmission hasn’t happened yet. Still, news like these raise concerns since it seems likely that fewer mutations are needed for h2h when minks have already started transmitting it.

Yet again though, the concerns make it seem like a second pandemic is right ahead of us. And I think recency bias is very strong here, it is easy to imagine a second pandemic close to the last one. But being able to imagine it does not affect the probability of it happening. It does also seem quite unlikely given very low base rates. According to Wikipedia, there have been about 20 outbreaks of avian flu since the beginning of the nineties. So far, none of them has led to widespread h2h transmission, even though some of them has affected other mammals before. Given Laplace’s rule of succession, this gives us a naive base rate of about 4%.

I think the current news is not serious enough to warrant any large updates. So my estimate for the probability of 10k cases before 2024 is 4%.

Then there is the question of mortality. Since there are so few cases, we cannot know for sure how deadly H5N1 is to humans. But from studies on other mammals and from the few human cases, it seems like H5N1 is likely to be much, much deadlier that Covid-19. Potentially the mortality rate is over 30%.

In my estimate, I use a very conservative 50% mortality rate. And I estimate a 2% likelihood of over 10k confirmed deaths from H5N1 in 2023.

Given the low risk, but potentially very high damages from a pandemic, authorities should be spending lots of time and money on preventing an outbreak. Like Zvi Mowshowitz, I believe that there are few changes one could and should do privately, apart from perhaps stocking up on masks.

As is often the case, the lesson in forecasting is that one should not take an increase in news coverage or concern to always mean that the probability of an event happening has changed. That can be the case, but it also very often only means that people have just started noticing the risk.

  1. Intuitively, this seems pretty high. But looking at the wiki it seems like terror attacks happen more often than I would’ve previously thought.

Leave a Comment

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.