Complexity theory and "New Economics"

Eric Beinhocker tries to go beyond the left-right distinction and economic orthodoxy to describe what he calls "economics for the real world". He describes the 'new economics',

Rather new economics is best characterised as a research programme that encompasses a broad range of theories, empirical work, and methods. It is also highly interdisciplinary, involving not only economists, but psychologists, anthropologists, sociologists, historians, physicists, biologists, mathematicians, computer scientists, and others across the social and physical sciences. It should also be emphasised that new economics is not necessarily new. Rather it builds on well-established heterodox traditions in economics such as behavioural economics, institutional economics, evolutionary economics, and studies of economic history, as well as newer streams such as complex systems studies, network theory, and experimental economics... The common thread running through this broad research programme is a strong desire to make economic theory better reflect the empirical reality of the economy. New economics seeks explanations of how the economy works that have empirical validity. Thus behavioural economists run painstakingly crafted experiments to explain actual human economic behaviour. Institutional economists conduct detailed field investigations into the functions and dysfunctions of real institutions. Complexity theorists seek to understand the dynamic behaviour of the economy with computer models validated against data.

He points to the use of the likes of agent-based models to analyze various markets. On policy making, he advocates eschewing the traditional mechanistic approaches (incentives, interventions to correct market failures, costs-benefits analysis etc) and embrace an approach that accommodates complexity, unpredictability, and reflexivity.

He outlines three principles of this approach. One, instead of going in with one design, experiment with multiple approaches and iterate and refine the ones that work best. Second, policies and institutions should be as adaptable as possible. Third, instead of being policy engineers, policy makers need to see themselves as stewards who "create the conditions in in which interacting agents in the system will adapt towards socially desirable outcomes". This would involve an evolutionary approach to execution. 

This is an addition to a growing body of literature that advocates an experimental and iterative approach to program design and implementation. 

Visualizing Kiva flows

David Roodman has a superb graphical visualization of the cross-border flow patterns of over 4.3 million different types of microloans - education, health, food, agriculture, retailing etc - channeled through Kiva, the ostensibly person-to-person microlending site.

Intercontinental Ballistic Microfinance from Kiva on Vimeo.

The graphical illustration is stunning in its ability to convey the emergent dynamics of such activities. The flow patterns show how the numbers of lenders and borrowers build-up with time. Once a small trigger initiates the flow, (presumably) demonstration effects tap into latent demand among the massive pool of borrowers and generates confidence among lenders. It is also an excellent illustration of how social systems develop and the importance of initial patterns in consolidating the final outcomes.

The borrowers are concentrated in West Africa, Kenya, Latin America, Peru, Chile, Philippines, Indonesia, and parts of Eastern Europe whereas the lenders mainly come from US, West Europe, Australia and Japan.

On a more general note, I believe that a lot of complex public policy challenges can be more effectively communicated using vidualization graphics. For example, a time series trajectory of visuals of an area can highlight how specific infrastructure or other interventions there impacted the area's development in a cognitively striking manner. In fact, such visualization can beautifully capture the emergent dynamics of social and economic systems in response to specific triggers.

Complexity economics and the global financial system

Andrew Haldane of the Bank of England has an excellent speech highlighting the complex and adaptive nature of the global financial markets. He compares the reaction of global financial market to market events as similar to the "flap of a butterfly’s wing in New York or Guangdong generates a hurricane for the world economy". He writes,

"Complex because these networks are a cat’s-cradle of interconnections, financial and non-financial. Adaptive because behaviour in these networks are driven by interactions between optimising, but confused, agents. Seizures in the electricity grid, degradation of ecosystems, the spread of epidemics and the disintegration of the financial system – each is essentially a different branch of the same network family tree."

Conventional wisdom on complex systems like eco-systems and financial markets was that they were self-regulating and self-repairing. It was though that complex systems tended to exhibit greater stability, and complexity strengthened self-regulatory forces in systems, so improving robustness. However, the events of the past 18 months have revealed a financial system which has shown itself to be neither self-regulating nor self-repairing. He uses four mechanisms to explain complex adaptive systems

1. Connectivity and stability - robust-yet-fragile character

Interconnected networks exhibit a knife-edge, or tipping point, property. Within a certain range, connections serve as a shock-absorber. The system acts as a mutual insurance device with disturbances dispersed and dissipated. Connectivity engenders robustness. Risk-sharing – diversification – prevails. But beyond a certain range, the system can flip the wrong side of the knife-edge. Interconnections serve as shock-amplifiers, not dampeners, as losses cascade. The system acts not as a mutual insurance device but as a mutual incendiary device. Risk-spreading – fragility - prevails. The extent of the systemic dislocation is often disproportionate to the size of the initial shock.

Another feature of connected networks is their 'long-tailed distribution' - the histogram formed by the number of links to each node. Unlike the randomly configured network with its symmetric and bell-shaped distribution, many real-world networks do have a thin middle and long, fat tails. There is a larger than expected number of nodes with both a smaller and a larger number of links than average.

Long-tailed distributions have been shown to be more robust to random disturbances, but more susceptible to targeted attacks. Therefore, long periods of apparent robustness, where peripheral nodes are subject to random shocks, offers little comfort or assurance of network health. It is only when the hub – a large or connected financial institution - is subject to stress that network dynamics will be properly unearthed.

Another feature of connected networks is their 'small world' property. In his famous chain letter experiment, Stanley Milgram showed that the average path length (number of links) between any two individuals was around six – hence 'six degrees of separation'. He found that certain key nodes can introduce short-cuts connecting otherwise detached local communities. This property will tend to increase the likelihood of local disturbances having global effects – so-called 'long hops'. A local problem quickly turns into a global one.

Haldane examines the global financial system and finds several interesting changes over the past two decades. First, the scale and interconnectivity of the international financial network has increased significantly - nodes have ballooned, increasing roughly 14-fold, and links have become both fatter and more frequent, increasing roughly 6-fold. Second, the international financial network exhibits a long-tail. Measures of skew and kurtosis suggest significant asymmetry in the network’s degree distribution. Third, the average path length of the international financial network has also shrunk - between the largest nation states, there are fewer than 1.4 degrees of separation.

2. Feedback and stability

The sub-prime crisis generated panic hoarding of liabilities (counterparty risk meant that banks hoarded liquidity rather than on-lend it) and distress sales of assets (to meet margin calls or reduce exposures). Individually-rational actions generated a collectively worse funding position for all. These rational responses by banks to fear of infection added to the fragility of an already robust-yet-fragile financial network.

3. Uncertainty and Stability

Through widespread counterparty uncertainty, networks have important consequences for the dynamics and pricing in financial markets. Given the multiple levels of splicing and dicing of derivative instruments, it was impossible to even trace back counterparties, leave accurately alone pricing those risks. Links in the chain are unknown and determining your true risk position is thereby problematic. The network chain was so complex that spotting the weakest link became impossible.

4. Innovation and stability

Another dimension of network stability was the role of complex financial instruments. Financial engineering unleashed into the markets an alphabet soup of instruments whose range of real risks were often impossible to assess with any reasonable degree of certainty.

He draws insights from network theory in areas like ecology, epidemiology, biology and engineering, to explain the emergence over the past decade of a financial network characterized by complexity and homogeneity (pro-cyclical and exposure to similar types of instruments and areas). The trend towards slicing and dicing risk and diversifying them through securitization and derivative instruments dramatically increased the system inter-connectedness and complexity. he says,

"Follow-the-leader became blind-man’s buff. In short, diversification strategies by individual firms generated heightened uncertainty across the system as a whole... a strategy of changing the way they had looked in the past led to many firms looking the same as each other in the present. Banks’ balance sheets, like Tolstoy’s happy families, grew all alike. So too did their risk management strategies. Financial firms looked alike and responded alike. In short, diversification strategies by individual firms generated a lack of diversity across the system as a whole. So what emerged during this century was a financial system exhibiting both greater complexity and less diversity. Up until 2007... complexity plus homogeneity equalled stability."

The impact of these trends was that the financial network,

"... was at the same time both robust and fragile – a property exhibited by other complex adaptive networks, such as tropical rain forests; whose feedback effects under stress (hoarding of liabilities and fire-sales of assets) added to these fragilities – as has been found to be the case in the spread of certain diseases; whose dimensionality and hence complexity amplified materially Knightian uncertainties in the pricing of assets – causing seizures in certain financial markets; where financial innovation, in the form of structured products, increased further network dimensionality, complexity and uncertainty; and whose diversity was gradually eroded by institutions’ business and risk management strategies, making the whole system less resistant to disturbance – mirroring the fortunes of marine eco-systems whose diversity has been steadily eroded and whose susceptibility to collapse has thereby increased."

He then draws on the experience of other network disciplines and provides some tentative policy prescriptions to manage the financial network and avert systemic dislocations. He discusses three areas,

"1. Data and Communications: to allow a better understanding of network dynamics following a shock and thereby inform public communications. For example, learning from epidemiological experience in dealing with SARs, or from macroeconomic experience after the Great Depression, putting in place a system to map the global financial network and communicate to the public about its dynamics... Part of the answer lies in improved data, part in improved analysis of that data, and part in improved communication of the results;

2. Regulation: to ensure appropriate control of the damaging network consequences of the failure of large, interconnected institutions. For example learning from experience in epidemiology by seeking actively to vaccinate the 'super-spreaders' to avert financial contagion; and

3. Restructuring: to ensure the financial network is structured so as to reduce the chances of future systemic collapse. For example, learning from experience with engineering networks through more widespread implementation of central counterparties and intra-system netting arrangements, which reduce the financial network’s dimensionality and complexity."

Agent-based models and complex adaptive systems

Agent-based models have been at the centre of the emerging field of complexity economics, which explores the interaction between economic agents under varying constraints and rules.I have blogged about such models in earlier posts on school education and residential segregation.

Rajiv Sethi has an informative post on such models which, as he rightly claims, provides "microfoundations for macroeconomics in a manner that is both more plausible and more authentic than is the case with highly aggregative representative agent models". He defines them as "computational models in which a large numbers of interacting agents (individuals, households, firms, and regulators, for example) are endowed with behavioral rules that map environmental cues onto actions". He also writes that they generate "complex dynamics even with simple behavioral rules because the interaction structure can give rise to emergent properties that could not possibly be deduced by examining the rules themselves".

In a recent essay in Nature, Doyne Farmer and Duncan Foley make a strong case for the use of agent-based models in economics on the grounds that the existing econometric and DSGE based models suffer from the fatal flaw that they are fitted to past data and do not account for outlier (tail risk) events. They write,

"An agent-based model is a computerized simulation of a number of decision-makers (agents) and institutions, which interact through prescribed rules. The agents can be as diverse as needed — from consumers to policy-makers and Wall Street professionals — and the institutional structure can include everything from banks to the government. Such models do not rely on the assumption that the economy will move towards a pre-determined equilibrium state, as other models do. Instead, at any given time, each agent acts according to its current situation, the state of the world around it and the rules governing its behaviour.

An individual consumer, for example, might decide whether to save or spend based on the rate of inflation, his or her current optimism about the future, and behavioural rules deduced from psychology experiments. The computer keeps track of the many agent interactions, to see what happens over time. Agent-based simulations can handle a far wider range of nonlinear behaviour than conventional equilibrium models. Policy-makers can thus simulate an artificial economy under different policy scenarios and quantitatively explore their consequences...

Agent-based models potentially present a way to model the financial economy as a complex system, as Keynes attempted to do, while taking human adaptation and learning into account, as Lucas advocated. Such models allow for the creation of a kind of virtual universe, in which many players can act in complex — and realistic — ways. In some other areas of science, such as epidemiology or traffic control, agent-based models already help policy-making."

As Rajiv Sethi writes, one of the major reasons why agent-based models have so far failed to take off relates to the difficulty of defining decision rules for agents under differing conditions, and in evaluating the effects of different factors. Further, creating agent-based models for the whole economy "requires close feedback between simulation, testing, data collection and the development of theory", which in turn demands "serious computing power and multi-disciplinary collaboration among economists, computer scientists, psychologists, biologists and physical scientists with experience in large-scale modelling".

A few popular agent-based models include John Conway's Game of Life, Thomas Schelling's segregation checkerboard, Leigh Tesfatsion's ACE.

Robert Axtell and Joshu Epstein have their silicon-based 'artificially intelligent agent-based social simulation' called the Sugarscape model. The Sugarscape includes the agents(inhabitants), the environment (two-dimensional grid) and the rules governing the interaction of the agents with each other and the environment. Eric Beinhocker provides an simple exploration of complexity economics and adaptively emergent systems in his book, The Origin of Wealth.

Update 1 (25/7/2010)

Economist has a nice summary of the research on ABMs. Unlike conventional models which use "representative agents (identical traders, firms or households whose individual behaviour mirrors the economy as a whole) and where interaction happens only indirectly through pricing, ABMs use a bottom-up approach which assigns particular behavioural rules to each agent (for example, some may believe that prices reflect fundamentals whereas others may rely on empirical observations of past price trends) and agents’ behaviour may be determined (and altered) by direct interactions between them. In an agent-based model you simply run a computer simulation to see what emerges, free from any top-down assumptions. It writes

"ABMs, in contrast, make no assumptions about the existence of efficient markets or general equilibrium. The markets that they generate are more like a turbulent river or the weather system, subject to constant storms and seizures of all sizes. Big fluctuations and even crashes are an inherent feature. That is because ABMs contain feedback mechanisms that can amplify small effects, such as the herding and panic that generate bubbles and crashes. In mathematical terms the models are “non-linear”, meaning that effects need not be proportional to their causes."