In the last post I argued that prediction is flawed as unless we are Putin we do not know his preferences and thus cannot predict his actions. The conclusion was that assumptions of what these are are needed to be able to predict anything at all, but do it badly and all predictions will be wrong. So how do we go about modelling Putin?
The basic method often used is a ‘revealed preference’ approach – you rely on past experience of Russian responses to different events, looking for cases where Putin has revealed his preferences through decisions he has made. By looking at his response to previous situations, some estimates can be made of what Putin wants and cares about and what weight he places on different issues, based on what he’s done before.
This is the basis of most analysis of political decision-making (even if unacknowledged). Implicitly they are treating what leaders have done before as an indication of their preferences and thus a guide to future behaviour. In the case of the Russian responses to the events in Ukraine, many have looked back at Russian actions in Georgia in 2008. If Putin responded militarily to attacks on supported groups in Georgia, many argue, then this indicates the potential for him to do it again in Eastern Ukraine. The whole notion of working out a grander long-term picture of Russian foreign policy behaviour is based on looking at actions in the past, interpreting what they say about Russian aims and then applying this to predicting future actions. This is revealed preference prediction in action.
This form of analysis provides perfect information if we assume that:
(a) Putin is making all the decisions alone i.e. his preferences are the only ones that matter
(b) he has done since he first became President i.e. all decisions since 1999 were based on his preferences alone
(c) he has not changed i.e. his preferences have been stable since 1999
If these three conditions hold then have constant preferences – we can treat every decision since he came to power as a consistent rational response based on a fixed set of preferences, so each action shines some light on what these underlying values are. In this framework, with infinite decisions we could perfectly map Putin’s preferences and predict exactly his response to any circumstance: we would finally be able to conduct his cost-benefit analysis for him.
Now the problem is that these assumptions clearly do not hold – and that is why many predictions based on past experience are wrong. Clearly (a) is wrong but to differing extents at different times. Putin is no Stalin and he isn’t even comparable to many of the authoritarian Central Asian leaders. He sits atop a semi-authoritarian system rather than standing above it, the arbiter amongst competing factions rather than the autocrat imposing his will. But his power has ebbed and flowed so that at different times different peoples’ interests have been more represented, so the preferences we are looking for have been changing. As (a) doesn’t hold (b) doesn’t hold and (c) is also a debatable assumption – Putin may well have changed his views in the last 15 years. These factors make Russian behaviour inherently unpredictable as the underlying aims and motives can change in ways we cannot know in advance.
For example, looking at Russian action in Crimea I can’t think of a single article I read predicting annexation. So how did Kremlin-watchers get it so wrong? I’m arguing that it wasn’t their fault. Mark Galeotti has argued that Putin 2.0 has become Putin 3.0, with the Crimea annexation a watershed moment where Putin is displaying different patterns of response and a change of views. If this is the result of a change in Putin’s preferences then this was inherently unpredictable. All we had to work on were assumptions based on his former behaviour – revealed preference – with predictions based on entirely backward-looking information. By definition the only way to see a ‘new Putin’, a ‘regime change’, is to see him responding to events in a way we did not expect. If we had predicted his response then our model accurately captured him and there is no new Putin – we knew him already! We only recognise a change in Putin once we have failed to predict it. Of course change can be gradual and noticed over time – but this is still post hoc, based on Putin’s new response to a series of events rather than predicting these new responses. So actions based on changed behaviour can never be predicted in advance.
So can we model Putin? Has Putin changed? With formal modelling we in a quandary – it is impossible to say for sure. There is an infinite range of possibilities – Putin may be unchanged but other factions have more power or vice versa, all to differing extents – there is no way to pin this down formally and rule out anything. This is where ad hoc approaches come in. What we can do is, post-event, look at what has happened and try and interpret it. It has been said that the decision to annex Crimea was made by an unusually narrow circle – ‘the inner Putin’  (sorry) – rather than the normal broader group. This could suggest both the decision makers have changed in composition and influence, and Putin has changed in encouraging this move. But who knows for sure what underlies this? Can we definitively state that Putin has changed? No we cannot. Can we predict anything with certainty? No. All we can do is give our honest opinions while recognising the precariousness of our position. The important message to take from the failures over Crimea is that we have to be clear about what our assumptions are, what our predictions are based on and be honest about the level of doubt inherent within them. As long as we are honest about that then our ad hoc approaches will serve us as best as possible.
Modelling Putin is possible but it comes down to looking at the past and projecting this onto the future in some way through our filters of ad hoc views and analysis. We cannot formally prove anything in this area, but we can make claims supported by evidence as best as we can and still accept that they are likely to be proved wrong.
So analysts shouldn’t beat themselves up about not predicting the events in Crimea but should remember it as a warning of the inherent fragility of backward looking predictions in an environment where people can, and often do, change.
 Maybe coined by Brian Whitmore of the Power Vertical blog, but again but I’m not sure.