The base case doesn't exist in Ukraine
Why predicting the outcome of the Russian invasion is so hard
In a class last week, my students and I debated the probability that Vladimir Putin remains President of Russia for the rest of the year.
It’s a question that is obviously difficult to answer definitively. We’re in unknown territory and there are good reasons to think that his position is both unshakable and also hanging by a thread. Personalist autocrats like himself rarely lose power; he has thrown his country into recession, isolation, and a war that seems unwinding. One could make strong arguments for both sides of the equation.
But when expressed in probabilities, it was more manageable. Putin might be ousted and he might stay. All we needed to do was come up with a number that seemed reasonable given the balance between those two outcomes.
If the war hadn’t started, we would have assumed a >95% chance of him lasting the year. Since the war had started and he was personally responsible for the destruction it was causing, we thought there was a chance that the security services remove him from power. We estimated that that would be a 10% chance, giving us a figure of 85% chance that he is in power on January 1, 2023.
That is a little bit higher than what prediction markets say, but broadly in line with them.
Bayesian predicting
The exercise thinking through Putin’s time in office was made easier by the fact that we had a base case. If nothing else changed, we would assume to see him in power. We needed only to measure how far we thought the world could deviate from that path.
In most political analysis, this is the foundational logic that exists in the background of our writing.
If we’re talking about the chance of Build Back Better passing, the default is that the status quo persists, no legislation advances, and so our job is to forecast how likely and what needs to happen to push Congress from that trajectory.
For an election, we can assume that the incumbent or challenger wins (depending on our perspective of who seems more likely at the moment) and then see if news events signals a shift.
This way of thinking can lend itself to various biases, but it also has solid grounding as a real-life demonstration of Bayesian inference. We have a starting point of what we think will happen. We get a new piece of information. We update our priors based on that piece of information.
In predicting how the Russian invasion of Ukraine will end, this isn’t possible. At least for myself, I don’t have a starting point for a forecast. As two NYU professors wrote in reviewing Nate Silver’s book in 2013:
the Bayesian approach is much less helpful when there is no consensus about what the prior probabilities should be. For example, in a notorious series of experiments, Stanley Milgram showed that many people would torture a victim if they were told that it was for the good of science. Before these experiments were carried out, should these results have been assigned a low prior (because no one would suppose that they themselves would do this) or a high prior (because we know that people accept authority)?
What’s the end game?
Since the start of the war, attention has rightly been on the human suffering and keeping track of the latest developments.
There have been a number of surprises. Kyiv has held out despite pessimistic warnings. The Russian military seems to be less capable than thought. The US and Europe have responded with harsher sanctions than anticipated.
For the analyst, a surprise can only exist where there is a default case for how things will end. Kyiv will fall to greater military forces; the Russian military is capable of conducting complex operations; the US and Europe will fail to agree on sanctions.
That’s what makes forecasting the end of the Ukraine conflict so difficult. After nearly two weeks of war, we have different possible endings, each with a very clear reason why it won’t happen. One of them must come true, but it’s difficult to identify the prior probability for each with any clear sense.
Ending 1: Putin calls off the invasion due or Russian forces are routed. But Putin has committed himself so deep with rhetoric of “de-nazification” and killed so many innocent civilians that it’s hard to see how he stops rather than continuing to bomb cities, even if all ground troops are depleted.
Ending 2: Russia occupies Ukraine and installs a puppet government. But the Russian military doesn’t seem capable of capturing major cities and is vulnerable to an insurgency.
Ending 3: A diplomatic settlement is reached. But the Russian demands are so high that Ukraine could not agree to them.
Ending 4: The war continues indefinitely as Putin refuses to withdraw and Ukrainian forces prevent greater incursions. But sanctions on Russia’s economy are so great that this may not be feasible for more than a few months.
Ending 5: Russian security services depose Putin and end the war. But they don’t seem willing to do this and autocrats are rarely ousted.
Looking at the possible endings, it’s hard to have a good sense about which is most likely and how each day’s events brings us closer or further away to each one.
Beyond what happens on the battlefield, the paradox of a secure Putin committed to a war that he can’t seem to win is one that’s hard to unravel with any clear probability of the answer.
What’s the solution?
The job of a risk analyst depends on having a sense of what is more or less likely to happen in the world. For Ukraine, that seems nearly impossible.
We could assign probabilities to each ending and update them periodically. That would be better than nothing and certainly better than mere guessing. But I would have low confidence in the number I’m giving to any ending.
Ultimately, the only approach that seems feasible is to give up trying to predict, since events on the ground and in the Kremlin will dictate what happens. With limited and patchy information, I wouldn’t want to place too much stock in any single data point shaping my opinion and am even wary about predicting an overall trajectory.
For political risk analysts and those whose organizations rely on their predictions, sometimes it’s best to admit when the world is too uncertain and when the base case no longer exists. In that case, we have a number of ways to manage, from scenarios to contingency plans to building monitors. But forecasts are sometimes too brittle and the events on the ground too fast-moving to give any weight to someone who confidently says they know where this is all going.
Thanks for subscribing. This is a free newsletter so feel free to share it.