One of my students asked me why we should worry about a Greek default or exit from the euro when the markets can already factor those risks into the price of any assets likely to be affected in the markets. This is the same question that Gillian Tett highlights in her column in the FT last Friday. Tett’s answer is that there are always institutional quirks in the markets that are hard to factor into different prices. That answer should send investors – and their lawyers – to look through the fine print of marketable assets to find arbitrage opportunities that have not yet been exploited.

They shouldn’t bother. They may well find something worth exploiting but that won’t make it easier to price in the risks around Greece. The reason market participants fail to price such risks efficiently has less to do with the completeness of their models than with the fact that many if not most of them are going to switch from one set of models to another in the event of a Greek default. It is this model switching – and not the discovery of new information – that will roil the markets. Moreover, no-one can predict just how big of an impact this model switching will have on prices. I suspect it will be large.

To explain this point, I should probably introduce (a model for) pricing models. The goal is to highlight discretion. First, models are supposed to be incomplete. If they included everything, they would be too hard to use. By implication, the people who use them necessarily have to choose what information to include and what information to leave out. Second, the people who build models have to decide which of the elements they choose to include are more important and which are less important to the overall end result. They can use empirical data to estimate their relative importance, but they have to choose what information to include and what to exclude from the models they use to make these estimates – and those choices will have an impact on the relative weights. Third, models have to take into account the interaction between different variables and estimates. In other words, structure matters and this is another area where the modellers get to exercise discretion.

All these choices – about the variables to include and exclude, about the relative weights or parameters, and about the structure of connections between them – rest on assumptions about how the world works. If you change those assumptions, then you need to think again about where to focus attention and why. That is the basic insight. And it has important implications for Greece.

Let’s start with a basic model for contagion. The first question is how to price in the risk that Greece will default. In doing this you look at the payment obligations, sources of cash, the prospects for political conflict and the possibility that someone will make a mistake along the way. This is the conversation you see most often in the press reporting and it implies a particular model for bond pricing. That model does not include the Greek financial system as a contingent liability. If the banking system were to risk collapse and the Greek government were to have to step in to bail out the banks, then this would make a fundamental change in our estimates of fiscal solvency. In fact, the experience of Ireland suggests that a possible collapse of the banking system could eclipse everything else. People who bought Irish bonds in 2007 were less likely to price in this possibility than people trading Greek debt today – but I suspect that a run on the Greek banks is not getting the attention it deserves in current pricing models for Greek government bonds.

A second question is where investors exposed to Greece will get liquidity to cover any losses or margin calls. If they are all selling different assets in different markets, then the actions these investors take to repair their balance sheets may not be relevant. If they are all hoping to raise liquidity by selling the same assets or in the same markets, then the consequences could be significant. Here it is useful to consider both the flight to quality and the structure of yields in Europe as a source of correlation. So long as investors are trying to protect themselves in the same ways while they are looking for liquidity, the market consequences are likely to be larger rather than smaller – particularly on the European periphery. I doubt that pricing models for peripheral bond markets are taking those patterns into account.

A third question is how much support the European Central Bank can offer as a buyer of last resort in European bond markets. There is a widespread presumption that the ECB’s commitment to purchase €60 billion per month through September 2016 should put a floor under bond prices. If anything, the concern at the moment is that the ECB will not be able to find enough supply to meet its targets. Hence any increase in selling pressure on the periphery should be easily absorbed. The challenge is to hold the proportions of ECB purchases relatively constant across countries. The current policy is to fix those proportions in line with the capital key for those countries participating in the euro. The ECB has some leeway to balance its purchases over time. The point is that things should average out across the program. That will be hard to achieve if market participants are primarily looking to sell peripheral country assets and to hold onto their core-country bonds. By implication, the constraints on ECB purchases could tighten suddenly in the face of a Greek default.

Of course it is conceivable that someone somewhere has built a model to fold the stability of the Greek banking system, the structure of portfolio requirements, and the constraints on ECB purchases into account when pricing in the risk of contagion. But it is more likely that investors are focusing on different models to use in different contexts – one where the Greek banks look relatively solid, another where the banks get into trouble, a third where portfolio responses start to have market implications, and a fourth where the ECB’s implicit commitment to do whatever it takes is found wanting. Each time one of these events is triggered, the markets will jump. The size of the jump is the measure of the extent to which market participants failed to price in ‘risks’ ex ante. We know that these things are possible and we can estimate the risk that they will occur, but I do not see a context where this model switching fails to move prices. And I would be reluctant to bet that the ‘market correction’ will be small.