Final Activity Report Summary - ADAPTUNPREDICT (Adapting to the entirely unpredictable, and other aspects of dynamic behaviour: beyond von Neumann's standard paradigm in games and economics) Homer's tale of Odysseus and the Sirens is often cited as the classical example of changing preferences, where naïve behaviour ends in disaster. Yet it may exemplify better a 'bounded decision model', with typically unwary sailors unable to predict the fate of those who come within earshot of the beguiling Sirens. According to the myth, however, Kirke alerted Odysseus to the danger. In effect she equipped Odysseus with an enriched decision model enabling him, but not his crew, to hear the Sirens without risking more than rope burn. Millennia later, on 9th May 2010, the 60th anniversary of the Schuman Declaration, the EU finance ministers needed to hold an extraordinary ECOFIN meeting in order to deal with the latest crisis afflicting Eurozone government bond markets. It was merely the latest manifestation of how bounded models can have major adverse consequences for financial markets. Indeed Narayana Kocherlakota, the President of the Federal Reserve Bank of Minneapolis, chose to write his Annual Report Essay for 2009 on 'Modern Macroeconomic Models as Tools for Economic Policy', and to recognise the significance of 'financial market frictions'. This physical analogy is significant. Just as designing lubricants may transcend ordinary physics, so redesigning financial markets to oil the wheels of commerce may be beyond macroeconomics. We can hardly expect the financial system to clear up any mess left by geological, meteorological, engineering, or even behavioural disasters. At best, it can provide insurance against some of these events. But a well regulated financial system should have explicit provisions, like 'living wills', for winding up failing institutions, and for specifying default clauses for all traded financial contracts. Financial disasters, or the threat of them, should be seen as entirely due to failures in the man-made institutional architecture of finance markets. In particular, contracts which fail to specify what should happen in any eventuality, foreseen or not, represent a form of bounded model of the future. Of course it would be futile to seek an unbounded decision or game model of even one contract, let alone the whole financial or economic system. Some eventualities have to be omitted from any tractable or even comprehensible model. As the statistician George Box has written: 'Essentially, all models are wrong, but some are useful'. Yet almost all economists' models are not only wrong, but 'hubristic': the models entirely ignore their own fallibility. Such models should be regarded as no less unprofessional in economics than they are in engineering. To discuss how to adapt to the unpredictable, the project has begun a pioneering investigation of 'enlivened models'. These avoid hubris with a branching structure of steadily unfolding sequences of bounded models. In any such sequence, each successive bounded model is enriched by features omitted from its predecessors. By definition, these bounded models are unable to model any later enrichments, which therefore become like unpredictable events. In statistics, enlivened models extend the fundamental concept of a stochastic process; in decision theory, they extend the standard notion of a decision tree; in game theory, they enrich the concept of the 'extensive form' of a game. Later Stefan Traub and I conducted a simple inexpensive experiment to observe features of experimental subjects' actual bounded decision models or attention processes in order to help explain instances of what standard economics regards as 'behavioural anomalies'. Finally, work has begun on meeting the key challenge of redesigning markets and other mechanisms that allocate both real and financial resources in order that they can function well even in economic environments so rich that they must be described by enlivened sequences of bounded models whose evolution, by definition, cannot be fully predicted within any one model.