Updating Forecasts in Light of the Russian Invasion of Ukraine

Link to Google Doc

The development in the Ukraine-Russia conflict is terrifying and I wish health and safety to Ukraine and all Ukrainians. I do not have much to add other than I hope that it resolves quickly, with as few lives as possible spilled. I also hope that the suffering for the Russian people, that have not wished for this, will end soon. I do worry that the economic sanctions will lead to immense suffering in Russia. I’m not supporting Russia or condoning the actions of the Russian government, but I feel for the Russians that do not support the invasion. Seems like many of them are victims in this conflict as well.

I have made no public forecasts about the conflict, which is unfortunate. But given that there are a lot of unknowns for how this conflict will play out I thought that I could update my previous nuclear war prediction while also making another conflict-related prediction. 

First, why make these predictions? I believe that there’s real value in participating in prediction markets and other forecasting markets to convey the information and models one has. In fact, I have seen multiple accounts on Twitter of Ukrainians that could prepare for the invasion because they saw high probabilities of an invasion on Metaculus and other sites like it. I’m not only posting my predictions here, I’m participating in Metaculus as well trying to improve the accuracy of forecasts. Making a prediction that something is going to happen does not mean that I hope it is going to happen, the point is to be as accurate as possible with the available information. I’m also generally in line with the community predictions on Metaculus, but the following predictions are either not open questions on Metaculus or I’ve seen some people on Twitter arguing for significantly different probabilities.

In Predictions 2022 I made two predictions regarding world politics:

China attacking Taiwan: 9%, updated to 11%

Nuclear war: 1.2%, updated to 2.4%

I think both of these are affected by the conflict. (Although I will not describe my reasoning for the China-Taiwan estimate) The nuclear war estimate per year is very much based on Rodriguez (2019). I gave nuclear war a slightly higher probability, due to me being generally more pessimistic on the probabilities of significant conflicts. It is clear that the recent development in Ukraine significantly increases the probability of nuclear use in conflict, let’s now look at why.

What makes me update?

Any conflict involving military powers with immediate access to nuclear weapons immediately increases the risk. Conflicts lead to mistakes and resentment among other things which all increase the risk for nuclear weapon usage. Most recently and more specifically, Putin ordered his nuclear deterrence forces to be on high alert, indicating that nuclear weapons are being on the table.(Link) I also believe that the slower than expected progress for Russia in the invasion may lead to frustrations. It also seems like the economic sanctions increase the overall pressure, increasing the risk for mistakes and resentment etc. Thus, the risk of nuclear weapon usage has obviously increased significantly, even though it is still quite low. In total, I update my prior of 1.2% by a Bayes factor of 2x. Converting 1.2% to odds form gives 3:247 and updating this by a Bayes factor of 4x gives posterior odds of 6:247 or 2.4% probability. The resolution criterion is any offensive nuclear weapon detonation in 2022. 

This estimate is crazy. Not that I believe it is too high, I have seen some claim that the probability is closer to 4% and others that claim that it is just above 1% and the number of times we have been lucky to not have a nuclear war is also sobering, but rather that the expected life loss of such a probability is high. Having this estimate also begs other questions such as what the probability for retaliation is, conditional on the first strike, and the impact of a nuclear winter.

When will a peace or armistice be agreed upon? 

This is a really hard question and there’s significant uncertainty here as well. However, the recent peace talks give some hope that this might end. The best-case scenario is that they reach agreement in the next few days. That seems unlikely though since it seems unlikely that any side is willing to budge in their claims

Ukraine has done a better job at defending than Russia expected and they are of course unwilling to give up their independence. Putin, on the other hand, seems convinced that the invasion is the best course of action. And pulling out without anything to show would likely be aggravating elites supporting Putin’s expansionism. On the other hand, the harsh economic sanctions from the West will eventually force Putin’s hand unless he is willing to sacrifice his entire economy for this invasion. This leads me to predict a 75% confidence interval for an armistice or peace between March 31st and October 1st. (NB: I’m not saying anything about the terms of the deal.) The resolution criteria is official recognition of any armistice or peace deal from both Russia and Ukraine. 

(I note my usage of Putin in the above paragraph, it is like I’m assuming that he is acting like a madman on his own accord without support from anyone. This also seems to be the prevailing narrative, with people assuming that Putin has gone mad and rogue in some sense. I think that is a bit reductive, and I’m not at all sure that this war ends if Putin dies.)

Overall, I hope that this shows some of my current thinking about the conflict. This changes from day-to-day and the uncertainty is high, but I think it is important to at least try to quantify one’s beliefs. If nothing else, it forces one to evaluate the information available and perhaps forecasts of this kind can help in decision-making. And again, I urge you all to check out Metaculus and their forcasting questions—there’s a lot of information there and my own beliefs does not differ much from the community there. 

Leave a Comment

Your email address will not be published.

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