Proposing policy by analogy is risky Neil Joh N so N A paper plane is a wonderful toy model with which to explain how real planes fly, and water flow is a great analogy for teaching about electrical flow through circuits. But without business-class seats, a paper plane can never be used to explain why two people pay vastly different prices for the same flight. Likewise, nobody unplugs a television to get a glass of water. By cross-checking against our everyday experience and intuition, we can quickly see the limitations of such a toy model and analogy. However, when it comes to the com- plexities of the financial sector, our intuition (and arguably that of many financial experts) is so limited that rigorous statistical validation of any toy model or analogy is essential before policies are suggested. This is the potentially dangerous shortcoming of Haldane and May���s paper1. In models of complex systems and net- works, tiny changes in the model���s assumptions ��� or changes in what it means to be a node, a link or ���infectious��� ��� can inadvertently invert the emergent dynamics, for example by turn- ing a stable output into an unstable one. Such changes can therefore amplify the inherent risk in any resulting policy suggestions. There is already substantial consensus that policy-makers need to embrace financial- market risk within the framework of complex dynamic systems2,3. However, markets contain many heterogeneous objects, the interactions of which may change in any number of ways in the blink of an eye (or the click of a mouse). This new dynamic regime, in which the char- acter of both the links and nodes can change on the same timescale4,5, lies well beyond standard models of ecological food webs, dis- ease spreading and networks. The resulting dynamic interplay can generate unexpectedly large market fluctuations ��� and it is these that invalidate the financial industry���s existing approach to the pricing of financial derivatives and the management of risk2,3. The financial model6 borrowed by Haldane and May is an interesting, abstract, complex- systems toy model. However, even the model���s original creators6 emphasized that ���In order for these kind of models to be more realistic, some improvements certainly are needed���. They state that their focus was on ���theoretical concepts��� whose ���relevance for real markets requires quantitative estimates of the parameters. Given the abstract nature of the model, this appears to be a non-trivial task.��� They are absolutely right. Would you fly in a paper plane that had been scaled up to the size of a 747? Policy-makers may never fully appreciate a model���s limitations, so policy suggestions are potentially dangerous unless accompanied by a quantified health warning of a model���s robustness and underlying assumptions, based on rigorous statistical testing against state- of-the-art financial data sets. Otherwise we simply increase risk, rather than reduce it. Neil Johnson is in the Department of Physics, University of Miami, Coral Gables, Florida 33146, USA. e-mail: njohnson@physics.miami.edu FORUM Financial systems Ecology and economics A growing body of literature deals with the application of theories developed in other disciplines to financial institutions, to which a paper in this issue now adds. As outlined here, however, views differ as to its relevance. See Perspective p.351 C. Gillon/Getty The PaPeR in bRieF paper1 ������the is entitled ���Systemic risk in banking ecosystems��� and is co-written by an expert in banking and an expert in theoretical ecology and science policy. was prompted by events underlying ������it the international financial crisis that began in 2007. focuses on the network dynamics of ������it financial institutions and, in particular, on the influence of the pricing of ���derivatives���. are ���Derivatives ��� financial instruments that have become fiendishly complex, and that allow investment houses to hedge against, and bet on, price movements of commodities, bonds, shares and currencies without needing to hold the underlying asset. authors apply models from ecology ������the and epidemiology to explore, by a simplified ���toy model��� analogy, how an initial bank failure can propagate through such institutions. offer suggestions on how overall ������they system stability can be achieved while ensuring that individual banks can maintain their necessary economic functions. 3 0 2 | N A T U R E | V O L 4 6 9 | 2 0 J A N U A R y 2 0 1 1 NEWS & VIEWS �� 2011 Macmillan Publishers Limited. All rights reserved

Network theory is sorely required Thomas l ux Hecosystem aldane and May1 argue that models from research can offer valid insight for understanding the financial sector. But is this too far-fetched an analogy? Can one really imagine the regulation of financial markets being based on their similarities to networks such as food webs? My answer, in a sense, is ���yes��� ��� we should take these similarities seriously. This is not to say that we should equate banks, and their depositors and hedge funds, with some type of schematic predator���prey model. It is rather the potential similarities between the structural, system-wide prop- erties of these very different areas of study that we should be interested in. As Haldane and May point out, research in biology has arrived at quite clear-cut results on the deter- minants of the robustness and vulnerability of ecosystems. By contrast, the modelling of ���representa- tive agents��� in economics has led to a delusive neglect of the effects of interaction between those agents. By focusing mainly on individ- ual optimization of utility or profit, economics has lost the perception that ���more is differ- ent���7 ��� namely, that higher-level aggregates (for instance, the global financial system) can have properties that cannot be understood solely on the basis of their constituent units on a lower hierarchical level (the single bank or investor). Built upon this extreme form of reductionism, the established framework for bank regulation has been exclusively micro- oriented and has lacked any consideration of systemic risk factors. As recent history has shown, system-wide effects are important. The default of Lehman Brothers in 2008 had the contagious effects of a ���super-spreader��� disease, and the subse- quent domino effect brought the entire finan- cial system to the verge of collapse. Systemic, ���macro-prudential��� regulation is now an issue on the political agenda, and the structure of the financial sector has to be scrutinized for its stabilizing and destabilizing feedbacks. However, the micro-based banking literature has little to say on such issues. We know from the natural sciences that structurally similar connections between micro units might lead to similar system behaviour in very different areas. It seems essential, therefore, to take stock of the accumulated knowledge on net- work structures when studying systemic risk in the banking sector. The few available phe- nomenological studies of particular segments of the interbank market in the ���econophysics��� literature have already identified network structures that are known to be vulnerable to shocks8,9. The near-collapse of the overnight interbank market at various stages of the recent crisis provides an empirical confirmation of how susceptible this particular structure is to disturbances. Connections between financial institu- tions are, however, multi-faceted, and only part of this complex man-made system has been mapped in existing toy models. Going beyond toy models, an empirical assessment of how trading in complex derivatives affects the network topology of the banking sector, and how that interacts with other linking factors (such as interbank credit lines), is urgently required10. There is, of course, also a need to go beyond the first step of analogies, and relatively simple mechanical models, to examine the behavioural micro-foundations of how the agents involved choose their connections in this financial ecosystem. ��� Thomas Lux, currently a visiting professor at the International Christian University, Tokyo, is in the Department of Economics, University of Kiel, and the Kiel Institute for the World Economy, 24105 Kiel, Germany. e-mail: lux@bwl.uni-kiel.de 1. Haldane, A. G. & May, R. M. Nature 469, 351���355 (2011). 2. Bouchaud, J.-P. & Potters, M. Theory of Financial Risk and Derivative Pricing: From Statistical Physics to Risk Management (Cambridge Univ. Press, 2009). 3. Johnson, n. F., Jefferies, P. & Hui, P. M. Financial Market Complexity: What Physics Can Tell Us About Market Behaviour (oxford Univ. Press, 2003). 4. Gross, t. & Sayama, H. (eds) Adaptive Networks (Springer, 2009). 5. Zhao, Z. et al. Phys. Rev. E 81, 056107 (2010). 6. Caccioli, F., Marsili, M. & Vivo, P. Eur. Phys. J. B 71, 467���479 (2009). 7. Anderson, P. W. Science 177, 393���396 (1972). 8. Soram��ki, K., Bech, M. l., Arnold, J., Glass, R. J. & Beyeler, W. e. Physica A 379, 317���333 (2007). 9. iori, G., De Masi, G., Precup, o. V., Gabbi, G. & Caldarelli, G. J. Econ. Dynam. Control 32, 259���278 (2008). 10. Markose, S., Giansante, S., Gatkowski, M. & Shaghaghi, A. R. Too Interconnected To Fail: Financial Contagion and Systemic Risk in Network Model of CDS and Other Credit Enhancement Obligations of US Banks (Univ. essex, 2009). track���s arm to allow the animals to run across the entire L. Surprisingly, they found that when the mice were resting before gaining access to the second arm, some of the firing sequences in their brain matched those subsequently recorded on the new arm. They refer to this predictive activity as preplay (Fig. 1). One might argue that, on exposure of the animal to the extended segments, preplay merely reflects replay of activity associated with the familiar first part of the maze, the shape of which was similar to that of the new part. But mice with no prior experience on any track also showed preplay. Together, these observations suggest that, in a new environ- ment, activity sequences involving a specific assembly of cells are selected from already- existing sequences in the network. So why have other investigators using similar experimental procedures not detected preplay? After all, in the most common design, the firing sequences that occur during a behaviour are compared with resting sequences both before and after the behaviour ��� just as in the present study. Dragoi and Tonegawa10 propose that their method of comparing spike sequences NeuRosCieNCe Seeing into the future The resting brain recapitulates activity patterns that occurred during a recent experience, possibly to aid long-term memory formation. Surprisingly, corresponding brain activity also occurs before an event happens. See Letter p.397 e DVa RD i . mose R & maY-BR i T T mose R Tregionof races experience can be found in the activity of the sleeping brain. One in which such traces are detected is the hippo campus, which is required for episodic memory in mammals. In rodents, for example, most hippocampal neurons fire selectively when the animal is in a particular location1,2. When these neurons ��� called place cells ��� are active during a particular experi- ence, they also tend to be active during subse- quent sleep3,4. The order of firing is also often preserved5,6. Such subsequent replay of brain activity also occurs in the awake state, when an animal rests between bouts of running7,8. The recurrence of experience-related firing is thought to contribute to the reorganization of synaptic connections between neurons during memory consolidation4,9. However, Dragoi and Tonegawa10 write on page 397 of this issue that hippocampal reactivation is not merely a reflection of prior experience. They initially recorded sequences of firing in place cells when mice that had been running on one arm of an L-shaped track were resting at food locations near the ends of that arm. As expected, place-cell firing sequences were replayed during pauses at the food loca- tions. The authors then opened the other 2 0 J A N U A R y 2 0 1 1 | V O L 4 6 9 | N A T U R E | 3 0 3 NEWS & VIEWS RESEaRch �� 2011 Macmillan Publishers Limited. All rights reserved