Mainstream economics and finance is dominated by models of decision-making under risk. Modern macroeconomics has its analytical roots in the general equilibrium framework of Kenneth Arrow and Gerard Debreu (Arrow and Debreu (1954)). In the Arrow-Debreu framework, the probability distribution of future states of the world is known by agents. Risk can be securitised and thereby priced and hedged.
Modern finance has its origins in the portfolio allocation framework of Harry Markowitz and Robert Merton (Markowitz (1952), Merton (1969)). This Merton-Markowitz framework assumes a known probability distribution for future market risk. This enables portfolio risk to be calculated and thereby priced and hedged.
Together, the Arrow-Debreu and Merton-Markowitz frameworks form the bedrock of modern macroeconomics and finance. They help explain patterns of behaviour from consumption and investment to asset pricing and portfolio allocation. This has been a well-trodden path for the past 50 years.
The path less followed has been to study optimal choice under uncertainty – the inability to form priors on the distribution of future outcomes – rather than risk (Knight (1921)). Neither the Arrow-Debreu nor Merton-Markowitz frameworks admit such uncertainty. Instead, modern macro and finance has been built on often stringent assumptions about humans’ state of knowledge and cognitive capacity.
For the past 40 years, the most popular of those informational assumptions has been rational expectations (Muth (1961)). That, too, has dominated modern macro and finance for a generation. In its strongest form, rational expectations assumes that information collection is close to costless and that agents have cognitive faculties sufficient to weight probabilistically all future outturns.
Those strong assumptions about states of knowledge and cognition have not always been at the centre of the economics profession. Many of the dominant figures in 20th century economics – from Keynes to Hayek, from Simon to Friedman – placed imperfections in information and knowledge centre-stage. Uncertainty was for them the normal state of decision-making affairs.--Andrew G Haldane and Mr Vasileios Madouros (2012) "The dog and the Frisbee." (1-2) [HT Lisa Herzog]
At the time of the speech Haldane was the "Executive Director, Financial Stability, Bank of England" (he is now "Chief Economist at the Bank of England and Executive Director, Monetary Analysis and Statistics."; Madouros is now "a Senior Manager in the Financial Stability area of the Bank of England.") Given the centrality of London's financial service industries to the financial system, their views are very important.
Models of decision-making under risk have dominated economics, finance, and epistemology for much of the second half of the twentieth century. Their limitations as (a) guides to evaluating systemic risk have been exposed during the financial crisis of 2007-9 (and its aftermath). Moreover, during that same crisis (b) they failed to hedge individual firms against potentially catastrophic losses.* While (a) and (b) are distinct in 1996 "commercial and regulatory risk judgements" became more or less identical when the bank regulators started to rely on "the acceptance of banks’ own models." (6-7; with references to the "Basel Committee on Banking Supervision.")+
One response to these events has been (I) to double-down on models of decision-making under risk with ever more subtle model-selection techniques, feeding them more data, and crunching these with ever more powerful computers. Haldane and Madouros note the main problem with this approach: "collecting and processing the information necessary for complex decision-making is costly, perhaps punitively so." (3) It's not clear how to internalize this cost to the modelers. So, in reality, the relevant information is likely to stay outside the modeler's perception. Moreover, in practice, there is also a tendency of over-fitting. (5) One may add that in the real world lots of agents have strong incentives not to be forthright about their strategies and positions, so it is an open question if all the relevant information will ever be available to the right agents in real time. Even if one were to grant, heroically, that all the relevant information is contained in relative prices (and their movements), it is by no means obvious that the modelers will recognize it in a timely and cost-efficient fashion. While simulations may help to reveal some salient information, it is by no means obvious that even when individual firms (and agents) can take precautionary measures, their joint measures will not make systems more fragile.**
Haldane and Madouros offer a second response to the failure models of decision-making under risk: (II) move to simple heuristics to regulate complex firms in highly uncertain environments. They argue in favor of this approach based on ideas inherited from Herbert Simon and developed more fully by Gigerenzer, who is cited copiously in their essay (and whose work gave it its catchy title). Moreover, in support of this approach they use data from the crisis to test in various ways performance of firms and various decision rules. A further argument is that the non-trivial compliance costs to regulation under simpler rules could be reduced. (10)
While I have considerable sympathy for approach (II), it is noteworthy that Haldane and Madouros do not reflect on possible shortcomings of this approach. One is fairly obvious: that one is relying on the wrong simple rule(s) which worked well in the past, but turn out to be less than robust in new or very different environments. This is a known problem for psychological heuristics (which have evolved under considerable trial and error). A hint of another such a problem is mentioned, but not really discussed in the paper. Financial agents will try to exploit weaknesses in and discrepancy among regulatory systems by so-called 'regulatory arbitrage.' One way to prevent this is to have universal regulatory regime, but that day has not arrived yet.
I'll mention one further problem with (II). This one is ad hoc because it is a bug of Haldane/Madouros's approach not a general feature of working with heuristics. In passing, they also embrace regulatory transparency (17). The problem is, it has not been shown that such regulatory transparency actually aids in making the simple heuristics/rules more robust in practice. In fact, Haldane/Madouros partially and tacitly recognize this problem because they advocate more supervisory discretion (14-15). But they never explain how such discretion is compatible with either a (simple) rules based framework nor transparency about such rules. When economists talk about rules-based frameworks they intend to reduce discretion, so more work is needed here.
Finally, it is worth noting that neither (I) nor (II) aim, what I would argue ought to be part of our focus: (III) to change the ethos and norms within the financial system such that collective learning and collective responses to unfolding crises are possible. So, unless we are willing to treat financial crises as natural events outside the scope of control, ultimately handled by tax-payers, we need to create a set of practices and generate incentives in which some measures of market stability or health are monitored and adjusted to by market participants and regulators (recall here and here, for background here, and here). About such utopian proposals more before long.
* Recall the famous remark (early in the unfolding crisis): “We were seeing things that were 25-standard deviation moves, several days in a row,” said David Viniar, Goldman’s chief financial officer. “There have been issues in some of the other quantitative spaces. But nothing like what we saw last week.” Financial Times.
+ Sadly Haldane and Madouros do not discuss to what degree this regulatory change was a consequence of s-called regulatory capture, financial industry lobbying, and/or economists' professional consensus.
**This is by no means a philosophical point in financial regimes in which trading risk for banks is, in part, socialized.
Comments