You are here: Home Knowledge Base Economy Coping with Systemic Risk Proposals Multi-Agent Financial Network (MAFN)Models and Complexity Perspective for Policy Design
Symposium 2011

Proposal - Multi-Agent Financial Network (MAFN)Models and Complexity Perspective for Policy Design

The Challenge

The global financial crisis has revealed how systemic risk in the banking system can lead to huge economic downturns. Many of the problems leading to the crisis have been discussed and many suspects h ...

The global financial crisis has revealed how systemic risk in the banking system can lead to huge economic downturns. Many of the problems leading to the crisis have been discussed and many suspects have been identified (including investment banks, central banks, rating agencies, regulators and the economics profession).

1. Why did nobody see it coming?

Erosion of the science base in economics confining investigations for decades under the rubric that economic systems are closed and complete rather than complex and adaptive along with ‘group think’ evident in top economic journals, departments and key policy institutions such as the IMF, are major ingredients in the recent crisis. The paradigm shift needed and the skills gap among economists that has to be fixed is quite considerable to enable us to deal with a longstanding failure of economists and the regulatory bodies to keep abreast of the institutional and technological innovations in monetary and financial sectors. These have created unprecedented volumes of ‘inside’ money via securitization and other forms of pro-cyclical collateralized private credit creation, a shrinking of state supplied ‘outside’ notes and coins in circulation or M0 in so called cashless economies with an IT based payments technology which has changed payments habits irrevocably and also vitiated the monetary transmission of inflation in the CPI index, and a vast interconnected system of digital transference of financial liquidity in real time with very low latency along with ‘algo’ based hyper high frequency market valuation methods. 

As part of a fractional monetary system, the growth of privately created financialized debt and contingent claim leverage instruments (to over $12 Trillion in the non-bank sectors in the US and to a 600% bank assets to GDP ratio by 2006 in the UK) against a falling proportion of monetary base poses an ever bigger problem of convertibility to high powered money or a flight to cash in a crisis for which central banks and ultimately tax payers in perpetuity bear the cost. The lack of inflation overheating and the flat CPI price inflation-output trade off in advanced cashless economies where inflation fell to around 2% by1994 lulled monetary authorities into a state of complacency. This allows the use of the ‘Greenspan put’ and loose monetary policy with impunity to entrench asset price bubbles and structural imbalance in the US and European economies. Disintermediation from real investment has meant that GDP growth becomes correlated with the growth of the financial sector and asset/commodity price bubbles, contributing to ever growing systemic risk. This negative externality and public good aspect of the financial system is no different from over use and degradation of resources from pollution. Unless ‘clean up’ costs are factored into local decisions and the financial sector shrunk to a point commensurate to its social marginal benefits, the implicit and explicit socialization of financial sector losses will overwhelm the West. Holistic visualization techniques are needed to over come the fallacy of composition type errors that socio-economic negative externalities problems pose with the need to ‘cap’ the aggregate amount of an economic ‘bad’. The history of monetary institutions is replete with attempts to ‘cap’ or control both public and private debt creation- the question is can technology help us with digital mappings of financial flows between counterparties and with visualization and strategic tools to monitor, identify and respond to system-wide fragilities ?

2. Data base Driven Multi-Agent Financial Network (MAFN) Models

The crisis has shown that extant policy models and quantitative tools are too antiquated for the task at hand. Over reliance by economists on macro-econometric models based on highly aggregated data has impeded progress in the adoption of ICT (Information and Communication Technology) based modelling tools that can build digital simulation platforms from fine grained firm level data bases to monitor and manage systemic risk from complex interconnectedness of market participants and potential perverse regulatory incentives that may exist. While cross sectional statistical models of contagion can be useful, early warning signals such as the Sergoviano and Goodhart (2010) Banking Stability Index based on aggregated market data on CDS spreads were strongly rebutted by Rama Cont and myself at the IMF 2010 Operationalizing Systemic Risk Monitoring Workshop in that they are contemporaneous with the crisis and contain no more information than market indexes such as VIX.

With new ICT tools, the mapping of actual financial interconnections to analyse propagation of instability or contagion in the system can be done at a level of granularity and virtual depiction of reality not available in mainstream macro-economics tool box. The fundamental computational methodology for webbased visualization of complex data sets is object oriented programming and multi-agent technology with interconnections between agents modelled as networks. Hence, the acronym MAFN was coined by me. Nodes in a MAFN model represents financial intermediaries (FI for short), other end users and also regulatory authorities and they are imbued with the capacity for rule based behaviour or fully fledged autonomous intelligent behaviour. In a data base driven MAFN, agents and their interconnections are characterized by ‘as is’ quantitative constraints. Network structures are determined as in snap shots of bilateral obligations between counterparties. Both in the characterization of financial interconnections between agents and in terms of agents’ responses, MAFNs are not hampered as equation led macro econometric models are by the need for long time series or panel data. 

To remedy the lack of empirical work on the network structures of the CDS market, which in particular stands implicated in propagating the 2007 crisis, and also the off-balance sheet derivatives activities of FIs which mediate a large proportion of risk related liquidity provision in the economy, I recently characterized the topological and contagion spreading features of these system as being too interconnected to fail. The tiered central core of highly interconnected banks is part of a very sparse adjacency and flow matrix for the system. As few as 25 banks account for 99% of the $700 trillion gross notional derivatives market. In terms of financial flows and links in real world financial networks, we have high variance and fat tailed distributions which defy adherence to Nash equilibrium theoretical formulations for network activity. The failure of a highly connected FI will trigger an almost certain failure of other top banks with the contagion coming to an abrupt end when the ‘superspreaders’ have demised. As these banks account for the bulk of capital in the system – ipso facto no bank among the top tier can be allowed to fail highlighting the untenable implicit socialized guarantees needed for these markets to exist at their current levels.

Work at the Financial Stability Division of the Reserve Bank of India has been commenced by me (and Simone Giansante) on a MAFN style ICT based model of the Indian financial system. Automated access to both balance sheet and off balance sheet bilateral exposures of bank and non-bank FIs is being facilitated. Reflecting the highly dynamic nature of commercial banking in India, recent changes from one quarter to the next of network connectivity and of banks that over took others in terms of their eigenvector centrality, have revealed a potential HBOS or Northern Rock type situation. At the level of a firm, as in case of Northern Rock and HBOS, aggressive increases in market share which manifests as higher network centrality may appear worthy of awards and accolades. However, holistic modelling tools like network modelling can reveal how this can endanger system stability. The recent RBI Financial Stability Report has given the implicit socialized costs of systemic risk from the failure of these banks to be over 14% of total capital of the Indian banking system.

3. Complexity Perspective In Macro-economic Policy Design and Strategic Behaviour
The mathematical logic of Gödel-Turing-Post that underpins the new science of Complex Adaptive Systems(CAS) has three key components : (i)Meta representational systems and self referential or reflexive mappings that necessitates ' über' computational intelligence of a Universal Turing machine which can execute own and others' codes (ii) Contrarian or self-negating structures like the Liar (as in ‘this is false’) qua rule breaker who can negate what is computable or transparent, and (iii) The consequences of (i) and (ii) is a fundamental incompleteness of the system which can be represented by so called creative and productive sets with the latter depicting a Red Queen type arms race in novelty production or 'surprises' and structure changing dynamics.

The absence of a complexity perspective in game theory and macro-policy design has had far reaching deleterious effects in precipitating this crisis. Firstly, the complexity paradigm shows that claims made for rational expectations equilibria by neo-classical economics are flawed. Brian Arthur et. al. (1994) who pioneered the first agent-based computational economic models used the notion of reflexivity in the context of stock market prices to show that homogenous rational expectations equilibria is a logical impossibility. Steve Spear (1989) showed that the fixed points representing rational expectations have no computable solutions. Reflexivity refers to the fact that prices and other social phenomena depend on beliefs or meta-representations of outcomes, and has been popularized by George Soros: “Markets don't reflect the facts very well, partly because they create the facts themselves." Arthur (1994) also noted the inherent contrarian reward structure of asset markets, rewards tend to accrue most to those agents who are contrarian or in the minority, viz. it is most profitable to buy (sell) when the majority are selling (buying). Then, if all agents act in an identical homogenous fashion, they will fail in their objective to be profitable. Hence, heterogeneity in strategies becomes a rational response and endogenous non-predictability of prices and their boom-bust cycles follow not because the system is being buffeted by white noise or ‘news’ but because of activities of agents who can form meta-representations of the world. 

Recent advances in neurophysiology of the brain on mirror neurons, hailed as one of the great discoveries in science, gives an explanation for why the capacity for meta-representation with self as an actor in the mapping (leading to the recognition of hostile oppositional intentions) is essential for social and strategic behaviour. V.V. Ramachandran (2006) views Machiavellian behaviour, über intelligence, deceit and creativity as being part and parcel of this capacity of meta-representation. The significance of the mathematical logic of Gödel-Turing-Post is that any system incorporating such elements of self-referential calculation and meta-representatiion will imply incompleteness or in layman’s terms, the capacity to think outside the box and innovate.  As early as 1987, Ken Binmore in Modelling Rational Players had pointed out how extant game theory cannot incorporate rule breaking contrarian structures and novelty production. The only indeterminism is randomization between two known actions as the action set is fixed and given.

Likewise, the long legacy of the Theil-Tinbergen theory of macro-policy design based on the so called Linear- Quadratic- Gaussian model of optimal control assumes that the policy game is one against nature where policy targets are only buffeted by random noise rather than by intelligent agents who game the system. Though Robert Lucas seminally suggested the necessity of ‘surprises’ in the context of policy effectiveness (albeit in a context of inflation control) this was not seen as the generic need for strategic innovation implementable in an operational way in a game of competitive co-evolution to maintain status quo and stability. For almost two decades the rank and file of macro and monetary economists recommended precommitment to fixed strategies rather than one of co-evolution. First was in the form of currency pegs which were serially destroyed at great cost to the tax payer by the ilk of George Soros (who explicitly cited the Liar strategy). Then, especially in the UK, central bankers pursued a fixed quantitative inflation rule within a ritualistic framework that eschewed any attempt to co-evolve with a fast changing monetary and financial environment. 

In a remarkable study, Axelrod (2003) cites system failure in networks arises from a situation in which “coevolution is not anticipated”. A regulator-regulatee arms race (cf. parasite-host dynamic) requires monitoring and production of countervailing new measures (cf. production of anti-bodies) by authorities in response to regulatee deviations from rules due to perverse incentives or loopholes in place. Markose (2004, 2005) notes that if the Liar/contrarian or rule breaker cannot be eliminated then the inability of authorities to co-evolve with regulatees will lead to system failure. Thus, a complexity perspective upholds  monitoring for perverse incentives and regulatory arbitrage as key for the mitigation of systemic risk and we must avoid “official definitions of systemic risk that have left out the role of government officials in generating it” (Edward Kane). MAFN simulation platforms are well suited to conduct “wind tunnel” tests of policy both prior to implementation and to monitor it on an on-going basis for its capacity to generate race to the bottom type tactics or destabilizing carry trades.

    Related Proposals

    Proposal
    Symposium 2011

    We must pay much more attention on economic and financial history than we have done before

    The latest financial crises showed, once again, that "this time is not different". Many of the excesses and policy mistakes that contributed to the crisis could possibly have been avoided or at least ...

    The latest financial crises showed, once again, that "this time is not different". Many of the excesses and policy mistakes that contributed to the crisis could possibly have been avoided or at least alleviated if we had learnt and remembered the lessons of the history better. According to the history of financial crises, there are several common features in the most severe crises. The biggest crises are usually preceded, among other things, by excessive lending, leverage and maturity transformation, widespread risk illusion, ignorance of tail risks and group-thinking. How to avoid repeating making the same mistakes over and over again?

    Polity, Academia
    Proposal
    Symposium 2011

    Consistent macroeconomic policies that eliminate negative real interest rates and unsustainable debts are a precondition for a stable financial system

    It is well known that the banking system, by its nature, is particularly vulnerable to economic crisis; It is naive, however, to attribute these vulnerabilities solely to a lack of a more consistent r ...

    It is well known that the banking system, by its nature, is particularly vulnerable to economic crisis; It is naive, however, to attribute these vulnerabilities solely to a lack of a more consistent regulatory framework. There were important gaps: insufficient capital of poor quality, “shadow” institutions out of the reach of supervision, a remuneration structure that creates incentives to arbitrage and opacity in key financial vehicles. The main failure, however, has been the effective implementation of existing requirements and standards; What it is at stake is the quality and efficiency of Banking Supervision, particularly among industrialized countries; The focus should

    Polity
    Proposal
    Symposium 2011

    Regulating Leverage and Risk

    Economic theory shows us that there is an important externality in financial markets. Banks and other financial institutions have the incentive to make leveraged investments. That is, a bank can incre ...

    Economic theory shows us that there is an important externality in financial markets. Banks and other financial institutions have the incentive to make leveraged investments. That is, a bank can increase its expected return when making loans to an investment project by borrowing much of the money from other banks. However, by doing this, it imposes risk on those other banks -- if the project fails, not only is the bank jeopardized itself, but so are its creditors -- and it has no incentive to take account of this imposed risk. For that reason, there is a natural tendency

    Polity, Business
    Proposal
    Symposium 2011

    Macroprudential policy must curb excessive use of debt in the economy and in the financial system, taking account of various manifestations of excessive debt

    The key reasons for most financial crises are excessive growth in assets and excessive debt, which, in essence, are the different sides of the same coin. In the recent crisis, in addition to the mortg ...

    The key reasons for most financial crises are excessive growth in assets and excessive debt, which, in essence, are the different sides of the same coin. In the recent crisis, in addition to the mortgage boom in many countries, especially in the US, the excessive use of short-term debt within the financial sector itself turned out to be the most dangerous form of debt. At the same time, effectively very thin equity buffers were allowed in the banking sector. The true fragile state of the financial system was then largely missed by the authorities and investors. The crisis experience hence

    Polity, Academia
    Proposal
    Symposium 2011

    Coping with Systemic Risk

    Part 1: Why do regulators know so little about the empirical network structure of the interbank market? What data is needed to monitor the system? Is it possible to monitor systemic risk, based on mar ...

    Part 1: Why do regulators know so little about the empirical network structure of the interbank market? What data is needed to monitor the system? Is it possible to monitor systemic risk, based on market data only? An important reason for the lack of knowledge (interest?) in detailed interbank networks, is the usage of mainstream economic models, of the Dynamic stochastic general equilibrium (DSGE) type. DSGE models neglect the importance of complexity and heterogeneity and policymakers thinking in terms of these models essentially assume a very robust banking system having no feedback on the macroeconomy and vice versa. In

    Polity, Academia, Business, Civil Society
    Proposal
    Symposium 2011

    Macroprudential policy must be coordinated between countries in order to avoid inaction

    I argue that in the absence of strong regional or international coordination of national macroprudential policies, the combination of these policies will be dangerously weak in preventing the future g ...

    I argue that in the absence of strong regional or international coordination of national macroprudential policies, the combination of these policies will be dangerously weak in preventing the future global financial crises. There are two main reasons for my prediction. First, there is a well known problem of "taking the punch bowl away in the middle of the party", which makes the containment of financial excesses politically difficult at a national level. Second, there is an additional "first-mover disadvantage problem", which aggravates the "punch bowl problem" in an integrated financial market. The first-mover disadvantage problem is caused by an inability

    Polity, Academia
    Proposal
    Symposium 2011

    Macroprudential authorities need to learn to take the punch bowl away

    We need more analysis on the reasons, probably including political economy aspects, why policy makers often seem to fail to interfere with potentially dangerous economic and financial developments bef ...

    We need more analysis on the reasons, probably including political economy aspects, why policy makers often seem to fail to interfere with potentially dangerous economic and financial developments before it is too late. Part of the problem may concern mandates and available policy instruments. But it is probably the case that even if mandates and responsibilities are clear and effective instruments are available, the risk of inaction or under-reaction is still there. The problem may also result from the regulatory capture if the industry to be regulated has too strong an influence on the decisions of the regulator. If this

    Polity, Academia