Risk, Uncertainty, and Black Swans: Why Soviet Socialism Was Forever Until It Was No More
Talk given at the Annual Berkeley-Stanford Conference
UC Berkeley, March 3, 2017
A few words to begin with about the title.
Many of you probably recognize the reference to Alexei’s terrific and influential book about late socialism, “Everything Was Forever Until It Was No More.” The book is framed around a particular observation, which is that in the late Soviet period Soviet citizens assumed that Soviet socialism would last forever, but after the fact they looked back and saw all sorts of reasons why it had to collapse.
It wasn’t just Soviet citizens who felt that way, however – outside observers did as well. Indeed there has been a great deal of criticism of academic specialists, and perhaps more importantly of the U.S. and Western intelligence communities, for having assumed that “everything was forever” and for failing to predict the collapse of communism in the USSR and Eastern Europe. In fact, just yesterday I read a piece in Foreign Policy claiming that Kremlinologists are “ haunted” by their “fabled inability to foresee one of the most significant geopolitical events of the 20th century — the collapse of communism and the Soviet Union.”
My argument that if indeed they are haunted – which I don’t think is true, at least I’m not – they shouldn’t be. On the contrary, the early assumption of both Soviet citizens and outside observers that “everything was forever” was entirely reasonable, and ex-post claims that what happened was predictable are not only wrong but reflect a typical cognitive bias highlighted in some of the research in psychology and behavioral economics on cognition and irrationality in decision-making.
In short, my argument is that the collapse of the Soviet Union shares at least one thing in common with Donald Trump’s election as president, which is that it was a highly improbable outcome with enormous consequences. Which is to say, it was a black swan.
A few general points before diving in.
First, my remarks are influenced by some terrific books for non-specialists on decision-making, four of which I recommend in particular: The Undoing Project, by Michael Lewis; Superforecasting, by Phil Tetlock and Dan Gardner; The Signal and the Noise, by Nate Silver, formerly with The New York Times and now with fivethirtyeight; and The Black Swan by Nassim Taleb (which is rather more controversial than the others but full of provocative ideas).
Second, a word about the difference between risk and uncertainty.
Imagine someone tells you they think there’s a 50/50 chance of some outcome. How do you interpret that claim? On the one hand, the person may be referring to something like flipping a coin, in which case you can be very sure that the odds actually are, almost exactly, 50/50, and no additional information, or alternative mode of analysis, is going to change that conclusion. That’s risk.
Now imagine that my wife tells you that there is a 50/50 chance of the Patriots winning the Super Bowl. The fact is that my wife knows virtually nothing about professional football and even less about the Patriots. What she would be telling you is that she has absolutely no information about the game and no way to process any information that she might have anyway, so she has no idea what the odds are. Only a tiny bit of additional information, or some analytical knowledge about football, might change her assessment. That’s uncertainty.
The point, then, is that forecasts about most political events are always dealing with uncertainty, and usually a great deal of uncertainty, and while better information and better analysis may reduce uncertainty – for example, about electoral outcomes – if nothing else Trump’s election tells us that both the odds offered by statistical models and betting markets entail a lot of uncertainty.
Third, a quick word about so-called Bayesian reasoning, which in brief is a way to unpack how reason-based predictors go about making predictions (often self-unconsciously), whether we’re talking about gamblers, intelligence analysts, election modelers, stock pickers, or any other forecaster.
The idea is that you start with a “prior” – that is, an assessment of the odds based on past experience, or statistics, of the class of events that you need to predict. In the literature that is sometimes referred to as the outside view – outside the case at hand.
You then modify the prior with regard to the case at hand, so an inside view, and you keep modifying your take as you get new information or learn to process information better, which means your assessment of the odds is constantly changing.
So if you’re betting on the Super Bowl, you arrive at a prior from general knowledge about football and maybe Super Bowls, and you modify your assessment as you get more information about the Patriots and the Falcons, and then you keep modifying it as you get more information – for example, you learn that Brady has a broken finger. The same if you are betting on, say, the collapse of the Soviet Union, or the election of Donald Trump as president.
Finally, you can’t assess forecasting accuracy if people aren’t clear about what they’re predicting and what they think the odds are. The former is particularly important for political prediction markets these days – that is, places or internet sites where you can bet on political outcomes (like Predictit.com) – the reason being that they don’t want to be sued over vague or ambiguous language. So, if there’s a line on, say, Trump getting “impeached,” they have to be clear about whether they mean indicted by the House or convicted by the Senate, as well as by when.
Not surprisingly, one of Phil Tetlock’s points in the Superforecasting book, as well as in his earlier book, Expert Political Judgment, is that vagueness is a particular problem with “experts” and academics, because their tendency is to use vague and ambiguous language, if they’re bold enough to predict at all. That point was made very clear to me when I wrote a paper in the early 1990s assessing the record of academic specialists writing about perestroika. To cite but one example, there was a debate early on about whether Gorbachev would bring “radical reform” to the Soviet Union, but no one ever specified just what qualified as “radical reform.”
There is also an instructive section in Superforecasting on so-called nominal scaling – that is, using words like possible, probable, likely, very likely, and so on when making predictions. Tetlock tells a story about a 1951 National Intelligence Estimate that asserted an attack on Yugoslavia was a “serious possibility.” When an intelligence analyst later asked a colleague what he understood “serious possibility” to mean, the person told him a 65% chance. The inquirer was startled because he understood the implied odds to be much lower, so he asked a bunch of colleagues the same question and discovered that there was an even wider range of understandings of what the language implied – from 20% to 80%. Obviously how a commander-in-chief could interpret, or misinterpret, vague nominal scaling can have life-or-death consequences, as it later would during the Bay of Pigs fiasco.
The point, then, is that if you are going to lay out a betting line, or assess a field’s record in forecasting, it helps to be clear about what the prediction is and what the odds are, which is in fact rarely the case.
So let me turn to the case at hand: Why was it correct to conclude that regime change or state collapse in the Soviet Union was highly improbable?
An initial and obvious point is that it matters when the forecast is being made. Are we talking about January 1985, before Gorbachev took office, or early 1991, before the failed August coup? Clearly a great deal more had to fall into place between January 1985 and December 1991 to get the outcome we did than between January and December 1991. To put if differently, any decent Bayesian thinker would have been constantly reassessing the odds over that period. And indeed the record shows that’s what outside analysts were doing.
Another general point – analysts of perestroika were trying to assess risk in the face of a great deal of uncertainty. One factor was again time related: Tetlock’s research shows that even the best political forecasters – “superforecasters” – are unable to accurately predict anything important more than two years out.
Another is that there was very little in the way of a Bayesian prior to help assess initial risk. There had been very few efforts at real reform of Soviet-type regimes before, way less than would be statistically significant, and in many of those cases the effort had been snuffed out by an outside actor, Moscow, which obviously couldn’t happen with perestroika. Nor had there been any case of regime collapse or state fragmentation in the Soviet bloc. There of course had been lots of revolutions and regime changes historically, and certainly lots of efforts to draw lessons from those experiences, but those lessons were only marginally relevant to the first and largest socialist state and a global superpower.
So, let’s take March 1985, right after Gorbachev came into office, and consider whether it was reasonable to conclude then that regime collapse or state fragmentation in the USSR in, say, the next decade was the least likely.
To get at this, I’m going to do what Soviet analysts did during the perestroika era, which is divide obstacles to change into “subjective factors” – that is, things having to do with the mind, like interests or beliefs – and objective factors – essentially everything else.
To begin, we can divide subjective factors into the following three categories (and there are probably others):
- ideological constraints relating to the USSR’s core legitimizing narrative;
- entrenched interests of key decision-makers and power holders;
- and broad cultural factors that made change less likely than it otherwise would have been.
Although they disagreed over details, Western analysts mostly accepted that these kinds of subjective factors were going to make any effort by Gorbachev to reform Soviet socialism all the more difficult, assuming of course that he wanted reform in the first place. And the reasoning went that in the absence of reform from above, there was no reason to expect anti-system popular mobilization that would force change from below.
I don’t have time to go into each of the three subcategories, but let me say a few words about ideological constraints.
Unlike some, I think that the Soviet understanding of Marxism mattered a lot, albeit in complicated ways, in the Soviet Union, but the way it mattered most was its place in the regime’s legitimation narrative.
One way to think about this is to ask why China was able to grow into capitalism while preserving the leading role of the Party but the Soviet Union wasn’t. There’s a recent book – Shadow Cold War: The Sino-Soviet Competition for the Third World, by the historian Jeremy Friedman that I think suggests an answer. Friedman argues that Soviet decision-makers understood the Third World through a Marxist lens and cared about promoting socialism as they understood it, whereas Chinese decision-makers didn’t, or at least not nearly as much. And he ascribes this to his interpretation of the October Revolution as the overthrowing of an imperial monarchy that was led by people who took their Marxism seriously, while the Chinese revolution was primarily about anti-colonialism and national liberation. Those differences, he argues, affected how both states competed for influence in the Third World. And it also suggests why it was much easier for Deng and his colleagues to accept the reintroduction of capitalism into China than it was for their Soviet counterparts.
That of course simplifies a very complex story, but I think the essential insight is correct: the Soviet claims to being the world’s first and leading socialist state were critical elements in the Soviet legitimation formula, which in turn meant that any reform effort would have to stay within the limits of what was ideologically acceptable to conservatives and reformers alike. And that in turn meant that while some kind of reform was certainly possible, even likely, with Gorbachev’s arrival in power, truly “radical reform” that might undermine the regime’s core legitimizing narrative was much less likely.
As for objective obstacles, one way to get at this is through the lens of institutional path dependency, and again by contrasting the Soviet and Chinese cases. When Deng assumed power and started China down the reform path in 1978, Chinese GDP per capita was a little over $150, whereas Soviet GDP per capita was a bit over $6500, which is to say, more than 40 times higher than China’s.
Moreover, the Bolsheviks had taken power some 75 years before Gorbachev launched perestroika, and the basic institutions of the USSR’s fully nationalized, planned economy had been in place for over 50 years. The Chinese Revolution, by contrast, took place about 30 years before the launching of the Deng reforms, and China never really had a well-institutionalized planned economy, thanks to Mao’s constant disruptions.
So the Soviet system had been in place for a long time, it had produced major material gains for the Soviet people, it had turned the USSR into a global superpower, and it had an extremely complex industrialized economy. While the system wasn’t living up to the regime’s legitimizing claims, the basic institutional order hung together pretty well – the political kneebone was connected to the economic thighbone, so to speak. Changing it significantly was going to be difficult and fraught no matter what. And in particular, introducing market pressures into an economy that had been mostly impervious to them for many decades was going to be extremely costly and disruptive.
That is exactly what is meant by path dependency. You are on one path but getting onto a different and better one is costly. That’s not to say that there weren’t other, less costly ways to effect a transition than the one we got, but significant change that could overcome what Gorbachev called “the accumulating problems of the past” was going to be very costly one way or the other. Moreover, designing a more effective reform program gets us back to the ideology problem – no one in the leadership, including Gorbachev, was willing to accept reforms that meant giving up the Party’s monopoly of power or introducing private property in the means of production
In short, there were deeply-rooted subjective and objective obstacles to change, and a great many contingent and improbable things had to happen to get the outcome that we had.
Of course this is only one side of the story: there were also structural and contingent forces driving change. But to get my take on how the balance between those forces and the obstacles to change played out over time, you’ll have to read my dissertation.
To wrap up, probably the most robust finding in the literature on decision-making is that all of us, to one degree or another, suffer from “confirmation bias.” As one study famously put it, “Once formed, impressions are remarkably perseverant.” Which is to say, most people are not good at Bayesian reasoning: we get “anchored” to a particular belief, sometimes for entirely irrational reasons, and then refuse to change our minds when confronted with new evidence. That is one way in which we’re “predictably irrational,” and there are some pretty convincing arguments as to how evolution helps explain this and other hard-wired biases. And it also doubtless helps explain why some people continued to think Soviet socialism would be forever longer than they should have.
There’s another relevant, perhaps more important, finding in the decision-making literature, which is that after an unexpected event, people fool themselves about what they thought at the time, and remember themselves as having been more prescient than they actually were. More importantly yet for our purposes, they also see patterns, construct narratives, and offer explanations for why what happened happened, and indeed for why it had to happen. It turns out, not surprisingly, that it’s easier, and psychologically comforting, to narrate or explain ex post, and it’s easy, and comforting, to ignore or underweight uncertainty and contingency after the fact.
To return finally to Alexei’s book, my take is that Soviet citizens had very good reasons to assume that Soviet socialism would last forever, as did Sovietologists, at least before, say, early or mid 1987, by which time it was increasingly clear that some kind of dramatic change was underway. Even then, however, there was a great deal of uncertainty about how perestroika would play out. But it’s also not surprising that Soviet citizens – and many outsiders – look back after the fact and identify a host of factors that make what happened seem much more likely than it actually was, and to reinvent the highly improbable into something entirely foreseeable.
So the takeaway is: black swans happen, and they shouldn’t be reimagined as white ones, whether we are talking about the collapse of the Soviet Union or the election of Donald Trump as president.