The Basel Committee and the European Banking Authority1 have recently noted that banks' risk models do not give the same risk assessments for the same assets. In other words, banks’ risk models are not the same, making use of different estimation horizons and model assumptions, whilst reflecting different preferences and specific institutional situations. Both authorities find this problematic and have expressed a desire to move towards the harmonisation of models across the industry.
This is exactly the wrong analysis of the problem. Moves towards model harmonisation further destabilise the financial system by making it more procyclical and increasing moral hazard.
The problem arises because banks use internal models in evaluating the risk of their portfolios. This means that each bank tends to have its own risk model, related to, but distinct from the models used by other banks. This is especially prevalent for the largest institutions that tend to develop models in-house. Smaller institutions are more likely to buy models from consultants, implying a worrying degree of model homogeneity. Reported heterogeneity in risk assessments is also affected how the same exposure can impact differently on institutions.
From the point of view of the authorities, model heterogeneity is problematic. If regulation is risk-based, whereby capital charges are adjusted based on the perceived riskiness of an institution, the risk assessment must be accurate. If two banks assess the risk of the same portfolio very differently, one of them must be wrong. The solution is to harmonise the risk models so that everybody will report similar risk for the same exposures. Such harmonisation is essential for the maintenance of the integrity of the risk-based regulatory process. If regulation is risk-based, we must measure risk accurately, meaning that models should give more or less the same assessment of the risk of a particular exposure. Otherwise, risk based regulations are fundamentally unsound.
Model accuracy and model homogeneity
Unfortunately, this is based on a misdiagnosis of the problem. There are two factors at work here, model accuracy and model homogeneity.
Consider the first problem, model accuracy. Every model is wrong by definition. If the authorities pick one modelling approach over another, they may just as easily be backing the wrong horse, a model that is less accurate, affording financial institutions and the financial system less protection in the future.
The benefits of ‘in-house’ models
For this reason, it is generally better for financial institutions to develop their own models internally. This is more likely to lead to a healthy competition in model design and more protection for the financial system, because model quality will improve over time. A supervisory-mandated model is much more likely to stagnate and become ossified, leading to less model development, and ultimately less protection.
If the authorities end up backing the wrong horse, and some years down the road when the next crisis happens, analysts may find that a key contributor to the crisis was the wrong model promoted by the authorities. This means that responsibility has been transferred to the governments, making a stronger case for bailouts. Regulatory involvement in models design directly affects the probability of bailouts and increases moral hazard.
This means that it is better to leave model development to the financial institutions, let them take responsibility, whilst encouraging innovations in modelling.
This is not the main problem, it gets worse. Moving towards model homogeneity leads to procyclicality. If each bank develops its own models, and models are different across the industry, when the next shock arises some banks may view it positively and buy the underlying asset, whilst other banks take the opposite view and sell. In aggregate their actions cancel out, resulting in stable markets where extreme movements are unlikely.
If however the banks are forced to have the same models, they will all analyse the shock in the same way, and react in the same way, amplifying price movements. All buying or all selling. In a worst-case scenario it causes extreme price movements. It also undermines market integrity because it encourages predatory behaviour by other market participants not bound by the models. It doesn't really matter how accurate the models are. Even if the supervisor ‘lucks out’ and picks the best model, it will still harmonise bank reactions. Model homogeneity is procyclical and undermines market integrity.
The recent news that banks' risk models provide widely differing risk assessments is positive. It signals healthy competition among model designers and, more importantly, model heterogeneity and hence countercyclicality. Both the Basel Committee and the European Banking Authority have indicated that they are troubled by this news and are seeking to rectify the problem. Their conclusion is wrong. The problem is not model heterogeneity, the problem is excessive reliance of financial regulations on risk models.
Bank for International Settlements (2013), “Report on the regulatory consistency of risk-weighted assets for market risk issued by the Basel Committee”, www.bis.org, 31 January.
Danielsson, Jon and Robert Macrae (2011), “The appropriate use of risk models: Part I” VoxEU.org, 16 June.
Danielsson Jon, Hyun Song Shin and Jean-Pierre Zigrand (2009), “Modelling financial turmoil through endogenous risk”, VoxEU.org, 11 March.
European Banking Authority (2013), “Report on the regulatory consistency of risk-weighted assets for market risk issued by the Basel Committee”, website, 26 February.
1 See Bank for International Settlements (2013), European Banking Authority (2013).