Forecasting the collapse of Silicon Valley Bank

Guest post by Resident Contrarian

The collapse of Silicon Valley Bank is a particularly prime territory for retrospective analysis for a few important reasons. The largest of which being that it was unusually relevant to Manifold users, who skew heavily towards being involved in the tech industry. Manifold Markets was lucky enough to not have ties to SVB (and thus was saved from a world of stress), but many of our users would have had real, practical concerns about SVB’s default and its effects on both their ability to pay and be paid.

The power of our markets increases not only with the number of participants in the market, but also in proportion to how much skin in the game each participant has. The value of a prediction market is in its ability to artificially create interest in questions through a betting mechanic; in this case the question came pre-packaged with a level of interest even the best market doesn’t usually attain.

Given this perfect storm of interest, how did our markets perform?

How well did Manifold predict the collapse?

The timeline of SVB’s collapse began on March 8th, when the bank announced it had sold a number of its immediately marketable assets to bolster its financial position and was seeking to raise even more capital through outside investments. This triggered some significant financial-market worry about their positioning, and Manifold’s markets responded quickly. The first question regarding SVB’s possible collapse was posted in the early afternoon of March 9th:

In the period between the creation of the market and the March 10th bank run resulting in SVB’s collapse, “yes” predictions ranged from 26% and 42%. At first glance, this appears low - a casual observer who trusted the market might have looked at the numbers and concluded that SVB was safe. 

The picture gets a bit more complex once you consider the mechanics of bank runs and bank failures, namely that in every way that matters, a bank run mimics the mechanics of a prediction market. SVB’s depositors represented a large pool of parties with real interest in the question of the bank’s possible collapse, voting their lack of confidence through withdrawals; the largest difference being that the real-world depositors predictions both had the potential to become a self-fulfilling prophecy, and did.

The amount of uncertainty the market’s participants expressed as a group was well above a level that would spell disaster if mirrored in the bank’s actual depositors. The actual events of the collapse substantiated this; their concern about the bank’s stability closely mimicked the sentiment of the market with a finger on the actual-outcomes trigger - an accurate prediction of actual outcomes, one step removed.

How well did Manifold predict depositor outcomes? 

The second phase of SVB questions shifted from the issue of the collapse itself to focusing on depositor outcomes. Given the FDIC’s anemic $250k depositor protections, would depositors caught up in SVB’s collapse be made whole? 

The market’s predictions here were right in a much more straightforward way than in the previous question. After the March 10th creation of the market pictured above predictions corrected fairly rapidly towards positivity, maintaining 71% or better confidence that depositors would be made whole until a joint statement from the US Treasury, Federal Reserve and FDIC put a stake in the matter late on March 12th.

Slightly more complex questions were predictably trickier to foretell:

On the harder question of whether or not depositors would be able to withdraw the majority of their money the first business day after the collapse, market participants were significantly less confident and hovered in an uncertain 45%-56% range until the FDIC and government response gave a definite answer. 

While the market was always reasonably confident the government would step in to make things right, it was a lot less sure it would be able to do so fast enough to keep the experience seamless from a business-day-to-business-day perspective. 

Markets as stress management

As we mentioned at the top of the article, this incident was a real issue with real implications for real people. For some people, this represented a lot of stress - for instance, it would have been stressful to be a CEO who wasn’t sure they could access the funds they needed to make payroll, or worrisome for an employee who wasn’t sure if their employer would be able to continue paying them if their VC cash was locked up in some eternal government limbo.

The understandable impulse in those situations is to assume the worst; since there was so much at stake, it would have made sense to worry about losing a job or a company in the chaos.

The wisdom of crowds might have been of help here, at least for some. Each of these markets painted a consistent picture of caution, but one which fairly quickly settled on a realistic and broadly accurate prediction of something less than total ruin - of a bank that stood a good chance to fail, but of failsafes that would very likely kick in to prevent worst-of-all-worlds outcomes.

Prediction markets often are used for clearly practical purposes, and that’s good - it’s useful to be able to rely on the wisdom of crowds to plan your next move. But with stories like SVB, we see some additional utility waiting to be picked up. It’s the opportunity to be able to access dispassionate evaluation of something an individual might not be able to be dispassionate about themselves - to get a 10,000’ foot view on something you yourself might be too close to to evaluate in an accurate, unbiased way.

This stress-managing, situation-clarifying function of Manifold is obviously just one of dozens, but it’s one we look forward to more people learning about and using.