The problem with treating a financial network like a financial market

Ten years after the financial crisis, the risks associated with securitisation are yet to be fully appreciated

It is almost precisely ten years since BNP Paribas froze three investment funds linked to subprime securities, marking the start of one of the most significant financial crises and deepest recessions of our time. At the centre of this crisis was an obscure credit derivative called a ‘collateralized debt obligation’ (CDO). The crisis raised many questions about our financial system (to name a few: banking culture, remuneration, risk models, political connections and regulation), but one issue was largely left largely untouched: the risks associated with the supply side organisation of products like CDOs.

This oversight should concern us all. After all, for a bubble to become a bubble it must have organisation. And if these securities had been safe, then the well-known problems around bank exposures and interconnectedness would not matter at all – and we might instead be praising the responsibility of our bankers for taking risk so seriously.

But history shows that financial innovation on the supply side follows a fairly predictable path: from the building of simple arbitrage positions in the early stages of an activity, to bricolage as complexity adds earnings capacity, then sabotage as gaming proliferates. As values outstanding of ‘collateralised loan obligations’ (CLOs) – a close cousin of CDOs – currently approach the $1trillion mark, and as the value of assets under management in Exchange Traded Funds (ETFs) exceeds $4trn, it is perhaps time to revisit the last crash to consider what supply side risks may continue to lurk in the finance sector.

We put together our own database of actors involved in the structuration of US$ denominated CDOs between 2001 and 2007 from a variety of industry documents. You can read the full results in our paper ‘Governed like a market, structured like a network’ (available here).

Our analysis proceeds from the observation that the early policy and academic work on CDOs tended to treat the activity like an orthodox market. As various ethnographies in banks have noted, mitigating risk is viewed as important within banks (see research by Vincent Antonin Lepinay and Donald MacKenzie and Taylor Spears), and in the case of CDOs risks were understood in traditional market-based terms as those arising from adverse selection (the risk that banks would cherry-pick the best assets and securitise the riskier ones) and moral hazard (the risk that banks would be disinterested in asset quality and contract enforcement because the default risk fell on investors).

Various risk-mitigating devices were therefore introduced, but the key organisational intervention was the use of independent collateral managers acting on behalf of investors to select and manage CDO assets. The independence, skill and experience of collateral managers featured heavily in the CDO marketing literature sent to investors, and Credit Rating Agencies often made discretionary adjustments to account for their track record. Collateral managers were also understood to create market efficiencies if they sold underperforming assets (on a risk weighted basis) and bought better performing assets, so that the good (companies) would drive out the bad (companies).

We asked a fairly simple question: how effective would those risk-mitigating features be if CDO structuration was not organised like a market?

We found that the banks involved in the structuration of those CDOs at an early stage of the development of the practice remained centrally involved at its peak in 2007. This structural and relational ‘embeddedness’ is a classic feature of networks rather than markets, because it implies that (social) relations formed across organisations endure and act as a barrier to entry to others. In other words, CDOs are not put together through the impersonal interaction of atomised individuals mediated by the price mechanism, but through a set of relations forged in the process of ‘doing business’ which become resilient.

We also found that law firms were surprisingly central, and that individual banks formed preferred attachments with specific law firms, again reflecting the relational aspect of the business. A small number of finance-law assemblages were at the core of the network. Other law firms in Delaware and the Cayman Islands which provide the infrastructure of regulatory arbitrage were even more central. One small law firm in Delaware, for example, was involved in 79% of all US$ CDOs in our sample.

The efforts to signal the independence of collateral managers resulted in a dramatic expansion of their number after 2004, with the majority managing the assets of either only one or two CDOs. They form the majority of a highly fragmented periphery, with the overall network exhibiting a strong core-periphery structure.

What this shows is that CDO structuration was organised like a network, not a market. And the risk mitigating features designed to meet market-based risks were therefore inadequate. In a network, risks present quite differently – they tend to be more structural so that information asymmetries or incentive distortions between two isolated actors may be of lesser risk than the pattern of relations across a network at the aggregate. The capacity for individual actors, such as collateral managers, to change norms and behaviour in that network may therefore be hampered by their peripheral position. And because relations elsewhere are not mediated fleetingly through price, but are embedded in social systems of repeat and reciprocal interactions, collateral managers ability to influence the practice of others may be constrained by historic interactions within the network which homogenise norms, resulting in shared behavioural expectations that become difficult to adjust.

There are therefore strong pressures to conform for any new actor entering the activity, where resistance may lead to reputational damage. These are the conditions under which groupthink and risk-blindness arise, particularly in a network which nurtures permissiveness rather than responsibility and trust.

Two immediate issues arise. The first is relatively straightforward: there is no benefit to the independent selection and management of assets by collateral managers when a structurally peripheral actor must select assets produced by a small number of finance-law assemblages, often using the same models. Indeed our network analysis highlights a structural paradox: the peripheral position of collateral managers is a function of the need to be seen to be distant from the banks (there are very few instances of repeat relations between banks and individual collateral managers).

This may mitigate market-based risks, but in a network it robs collateral managers of the structural power to exercise voice or exit pressures on banks. They have neither the authority to encourage banks to offer better quality collateral or discuss hidden risks, nor the capacity to influence collateral quality by refusing to buy, because with so many other collateral managers in the activity area, one will always take on the job.

The second is that innovation in this sector did not have the Schumpeterian effect often suggested. It is often thought that product innovation opens up opportunities for new entrants with different or more specialised skills, thus altering the structure of the network. But in this case, pre-existing relationships at the core seemed to endure, despite the growth of new entrants. The structure of the network, in fact, became more similar over time. Those relations may simply have been formed as a matter of convenience: working repeatedly with the same actors may reduce time and resources spent. Relations may also congeal and exclude others as systems of trust and reciprocity build.

There are, however, darker explanations. Relations may form amongst groups, leading to a culture of favours, insider group formation and potentially misconduct if breaches of established norms become routinised (as explored in research by Blake Ashforth and Vikas Anand, and Joseph Lampel). Our paper offers no view on this, but the US Securities and Exchange Commission enforcement actions brought to date suggest that gaming the process of collateral selection was common, if not systemic. Our next paper, on ‘social network risk’, will explore some of these relations in more depth.

Professor Adam Leaver will be joining the University of Sheffield’s Management School in October 2017 and will be affiliated to SPERI.