Epistemological foundations for the assessment of risks in banking and finance

Abstract

This article presents the epistemological and conceptual foundations on which current attempts to model crises and assess financial risks are based. It draws a distinction between two research programs, in Lakatos' sense: on the one hand, crises understood as structural events within a cycle; on the other hand, crises seen as statistical tail events. The methodological, theoretical and practical consequences of such a dichotomy are exposed. A crucial difference lies in the assumptions about change in the causal processes generating economic outcomes, especially asset returns. Furthermore, this article insists on providing conceptual definitions of key terms that have distinct meanings within the two research programs.

Keywords::

• financial crises
• causal process
• forecasting
• research program
• risk

Jel Classification::

• B41
• C10
• E37
• G01
• G17

Notes

^1. Even though the corpus of economic methodology is rich (see Hands, 2001), the gap between economic methodology and the everyday practice and research in economics is wide.

^2. In one of their most famous articles on the modeling of bank runs, Allen and Gale (1998) distinguish between panics that are a product of the economic cycle and those that are ‘sunspot equilibria’. Gorton (1998) investigates empirically each of these two types of crises.

^3. Concerning the 2007–2008 crisis, it was vividly illustrated by the statement of David Viniar (quoted in Larsen, 2007), CFO of Goldman Sachs, who said that the bust consisted of ‘25-standard-deviation events, several days in a row’. According to calculations of Danielsson (2008), one of these events occurs on average once every 10¹⁴⁰ years, far more than the age of the universe itself.

^4. The crisis of 2007–2008 has also been interpreted in this way. The volume of US mortgages (subprimes) which initially failed was small compared to the balance sheets of banks and financial markets as a whole. On this subject, see e.g. Caballero and Krishnamurthy (2008).

^5. The use of the word ‘countercyclical’ in such a context is tricky, as it presupposes a cycle. But a statistical view of crises, as explained earlier, is precisely one that does not assume boom and bust cycles. For example, it would be meaningless (or at least out of context, i.e. out of model) to say that ‘measured risks are low during the bubble phase and then peak with the occurrence of the bust’.

^6. Another and potentially more fundamental question – which, once again, deals with the concrete results of particular indicators – relates to the ability of data themselves to mirror economic facts. If balance-sheet approaches must be criticized, it is on behalf of other recent developments, namely innovations that tend to make published balance sheets less representative of the actual activity of financial institutions than they were before. Although it is not my purpose to examine them in detail here, they include: (1) the growth of off-balance-sheet commitments through securitization or derivative markets, or (2) the increasing use of very short-term debt, whose rollover frequency is badly reflected in quarterly data. Allen and Saunders (2010) comment on these facts.