We are now more than ten years on from the financial crisis, and in some ways much has changed. Banks have been re-regulated, resolution plans have been drawn up, and macroprudential authorities have been created.
One thing that has not changed is the fact that the financial system is large, complex, and interconnected. It is characterised by the interaction of many different types of entities and firms, and these interactions can give rise to unexpected risks. Indeed, in some ways the system may have become larger and more complex since the crisis. We think that there is a strong case for policymakers to think about the system as an interconnected whole, rather than as a set of distinct sectors to be regulated in isolation.
Why would we want to be able to think about, and analyse, the system as whole? Here are three possible reasons:
- Institutions being ‘too interconnected to fail’ was critical to the last crisis, and we should at least try to avoid running into the same problem again. We shouldn’t assume interconnectedness will sit neatly within sectors – AIG was critically interconnected with banks, but wasn’t a bank itself.
- We have a new resolution regime to deal with the failure of global systemically important banks (G-SIBs), which we hope is effective. But knowing the fault lines – and how to address them – before we need to resolve one of these banks would be hugely valuable.
- The system has and will continue to evolve. We need to keep pace with these changes in the structure of the system if we want to understand systemic risk.
All of this points to the need for system-wide analysis. Work on system-wide models is emerging (e.g. Bookstaber 2017, Hałaj 2017), but in many cases such efforts are hampered by an absence of system-wide data.
In principle we want to understand how shocks could propagate through the financial system. This could be via direct linkages between institutions – funding linkages that could ‘run’ or transmit losses, destabilising a major institution, which in turn spreads contagion through its reactions.
Or it could be via indirect contagion – fire sales of assets held in common with other institutions that face losses as market prices fall. This implies that the gold standard for monitoring risks would have two components: a map of common asset holdings, and estimates of funding connections across the financial system.
Putting this sort of information together in the UK – and we think most other jurisdictions – is still difficult. Work to improve official data is in progress (Jones 2018), which should help, although they may still be too aggregated to answer all the questions of interest (for example, to think about fire sales, one might want fairly granular data on asset holdings).
In the meantime, staff at the Bank of England have worked on a system-wide dataset for the UK (Aikman et al. 2019). While still far from a complete map, it can give us some useful insights, including beginning to answer some of our questions about contagion risk.
Comparing two financial systems
As an example of such an insight, and as a way of illustrating the benefits of zooming out to look at the system as a whole, system-wide data allow us to compare the structure of different financial systems. We can then start to think about different sorts of questions: Is there a common structure to which systems evolve, and if not, can we understand why? Where there are differences, are some structures more stable than others?
The figures below make this comparison between the financial systems of the UK and the US, as they were at the end of 2016. In particular, we show holdings of financial assets by different sectors: key financial sectors (pension, investment and hedge funds, insurers, banks and brokers), public financial institutions (the central bank and, for the US, government sponsored entities), other domestic – that is, non-financial – sectors and the rest of the world.
The size of the circles and numbers represent the percentage of total assets shown made up by that sector/asset type pair. Naturally, this abstracts from the overall size of each systems – in total, our data cover around $133 trillion of US financial assets and $36 trillion in the UK.
This is necessarily a very high-level view of a financial system, both for clarity of exposition and because of data limitations. Detailed notes to the figures provide sources, but in short we use data collated by Aikman et al. (2019) for the UK and a combination of Federal Reserve and SEC data for the US.
For the UK data, we also need to caveat that it is drawn from multiple sources with the potential for inconsistency that brings, and that in some cases – particularly hedge funds – our numbers are effectively rough estimates. Still, we think at this level of granularity, the overall picture they provide can be illuminating.
Sources: Bank of England: annual report, Bankstats, and regulatory reporting, Financial Conduct Authority: Hedge Fund Survey, International Organization of Securities Commissions: Hedge Fund Survey, Investment Property Forum: The size and structure of the UK property market, Office for National Statistics: MQ5 and United Kingdom National Accounts: The Blue Book 2017 and Bank of England calculations. Note that Bankstats data used here differ from those seen in the regular Bankstats publications due to differences in construction and methodology.
Sources: Federal Reserve: Z1 national accounts, Securities and Exchange Commission: Form PF
Our prior was that if ever there were two financial systems that we might expect to share structural similarities, it is those of the UK and the US. But we see some notable divergences:
- Bank lending looms large in the UK system, making up 15% of the assets in our data, compared with 9% in the US. Interestingly, the total share of system assets made up by loans is very similar (22% in the UK and 20% in the US), but the holding of loans is much more distributed in the US – including of course by Fannie Mae and Freddie Mac (‘GSEs’ in the figure), a clear structural difference between the two systems.
- The share of financial assets in equities is around 10 percentage points higher in the US than in the UK. This may not be entirely surprising given the size of each countries’ stock markets – the combined market capitalisation of the NYSE and Nasdaq is more than eight times that of the London Stock Exchange (WFE 2019). But we also see that the pattern of ownership is different, with investment funds and direct holdings by non-financial sectors playing a large role in the US, and proportionately greater foreign holding of UK equities.
These and the other differences visible in the figures could be driven by institutional differences – for example, in how retirement savings or mortgage loans are provided by each system – or by some path dependency in how the systems have evolved.
Absent a system-wide funding map, we cannot provide a full assessment of what these differences will mean for financial stability. Assembling the funding linkages is hampered by the lack of cross-border data reporting and sharing, and filling in these missing pieces should be a global priority.
Nonetheless, the asset holdings data point to some differences in the fault lines for the UK and the US. For example, notice the differences in the types of contagion risk present in the two economies. Insurance firms in the US are much more exposed to bond market dislocations than in the UK; while US pension funds carry more direct equity market exposure than their UK counterparts. On the other hand, UK pension funds are more exposed to the rest of the financial system through their larger holdings of investment fund shares.
This suggests that if stresses were to originate in these different asset classes, the propagation would work differently. Ideally, we would also want to characterise the funding structures and interconnections to know where the risks of forced sales could amplify any initial price moves.
The fact these differences are so easy to spot in the countries whose financial systems would be expected to be most similar suggests that completing similar mapping exercises for other advanced countries would be especially valuable. With further measurement, we could begin to ask what are the fundamental reasons that explain such differences, and what risks and linkages do we need to be most concerned about in the global financial system?
Aikman, D, P Chichkanov, G Douglas, Y Georgiev, J Howat and B King (2019), “System-Wide Stress Simulation”, Bank of England Staff Working Paper No. 809.
Bookstaber, R (2017), “Agent-Based Models for Financial Crises”, Annual Review of Financial Economics 9: 85-100.
Hałaj, G (2017), “Agent-Based Model of System-Wide Implications of Funding Risk”, ECB Working Paper No. 2121.
Jones, A (2018), “Economic Statistics Transformation Programme: enhanced financial accounts (UK flow of funds) – 2018 matrix update”.
World Federation of Exchanges (WFE) (2019), WFE Annual Statistics Guide (Volume 4).