Abstract: Learning is a convincing method to achieve coordination on a Nash Equilibrium (NE). But does learning converge, and to what? We answer this question in generic 2-player, 2-strategy games, using Experience-Weighted Attraction (EWA), which encompasses most extensively studied learning algorithms. We exhaustively characterize the parameter space of EWA learning, for any payoff matrix, and we understand the generic properties that imply convergent or non-convergent behaviour. Irrational choice and lack of incentives imply convergence to a mixed strategy in the centre of the simplex, possibly far from the NE. In the opposite limit, where the players quickly modify their strategies, the behaviour depends on the payoff matrix: (i) a strong discrepancy between the pure strategies is associated with dominance solvable games, which show convergence all the time; (ii) a substantial difference between the diagonal and the antidiagonal elements relates to coordination games, with multiple fixed points corresponding to the NE; (iii) a cycle in beliefs defines discoordination games, which commonly yield limit cycles or low-dimensional chaos. While it is well known that mixed strategy equilibria may be unstable, our approach is novel from several perspectives: we fully analyse EWA and provide explicit thresholds that define the onset of instability; we find an emerging taxonomy of the learning dynamics, without focusing on specific classes of games ex-ante; we show that chaos can occur even in the simplest games; we make a precise theoretical prediction that can be tested against data on experimental learning of discoordination games.

Abstract: The past decade has seen the development of new methods to infer causal relationships in biological and socio-economic complex systems, following the expansion of network theory. Nevertheless, the standard estimation of causality still involves a single pair of time dependent variables which could be conditioned, in some instance, on its close environment. However, interactions may appear at a higher level between parts of the considered systems represented by more than one variable. We propose to study these types of relationships and develop a multi-channel framework, in the vein of Barrett and Barnett (Phys. Rev. E, 81 (2010)), allowing the inference of causal relationships between two sets of variables. Each channel represents the possible interaction between a variable of each sub-system. Based on this new framework, we develop two different multi-channel causality measures derived from the usual Granger causality to account for linear interactions and from the concept of transfer entropy for nonlinear contribution. Our measures provide different information about the inferred causal links: the strength of the global interaction between the two sub-systems, the average frequency of the channel switches and the channel contributing the most to the information transfer process for each time step. After having demonstrated the ability of our measures to infer linear as well as nonlinear interactions, we propose an application looking at the U.S. financial sector in order to better understand the interactions between individual financial institutions, as well as parts of the financial system. At the individual level, the considered channels between financial institutions are expressed both in terms of spectral representation using wavelet transform and probability distribution using quantile regressions. Beyond the application presented in the paper, this new multi-channel framework should be easy to implement in other fields of complex systems science such as neuroscience, biology or physics.

Abstract: This study explores sleeping beauties, i.e. breakthrough inventions that experienced delayed recognition, by means of patent data. References in a patent signal the state of the art on which the patent is based, and they can limit the property rights established by its claims. A patent that is cited by many others, thus, includes some technology central to further developments. Patent citations can be used to study patented breakthrough inventions, identifying them as highly cited patents (Singh and Fleming, 2010; Castaldi et al, 2015). We add to this literature by analysing geographical determinants of the occurrence of sleeping beauties. A sleeping beauty is defined as a patent family that is both a sleeper (is not cited for at least x years after its priority date) and highly cited (receives at least x citations). Using this definition, with x=13, the database contains over 3,000 sleeping beauties. We hypothesize that the share of sleeping beauties in the output of a country and the share of sleeping beauties in the total of highly cited patents in a country is higher, the more geographically isolated the country is, reasoning that isolation renders diffusion and acceptance of new (radical) ideas more difficult. Geographical isolation is proxied by the mean geographical distance to all foreign patent inventors, measured both generally and specifically for each general technology class. We take into account the presence of international airports, average travel times, each country’s general proficiency in world languages, and the number of patents within each country, while controlling for technological and author variations across patents. Castaldi C, Frenken K, Los B (2015) Related Variety, Unrelated Variety and Technological Breakthroughs: An analysis of US State-Level Patenting. Regional Studies 49(5):767–781 Singh J, Fleming L (2010) Lone Inventors as Sources of Breakthroughs: Myth or Reality? Management Science 56(1):41–56

Abstract: In this paper we develop a testing framework for comparing the accuracy of competing density forecasts of a portfolio return in the downside part of the support. Three proper scoring rules including conditional likelihood, censored likelihood and penalized weighted likelihood are used for assessing the predictive ability of out-of-sample density forecasts, all closely related to the Kullback-Leibler information criterion (KLIC). We consider forecast distributions from the skew-elliptical family of distributions, as these are analytically tractable under affine transformations and projections onto linear combinations. We argue that the common practice to do forecast comparison in high-dimensional space can be problematic in the context of assessing portfolio risk, because a better multivariate forecast does not necessarily correspond to a better aggregate portfolio return forecast. This is illustrated by examples. An application to daily returns of a number of US stock prices suggests that the Student-t forecast distribution outperforms the Normal, Skew t and Skew Normal distributions in the left tail of the portfolio return. Additionally, the visualized dynamics of our test statistic provides empirical evidence for regime changes over the last thirty years. In a second application, techniques for forecast selection based on scoring rules are applied, and it turns out that the one-step-ahead Value-at-Risk (VaR) estimates from those dynamically selected time-varying distributions are more accurate than those based on a fixed distribution.

Abstract: Open innovation is a set of practices that enable organizations make direct use of external R&D to augment their internal research. Open innovation has received a lot of attention in the last decade, so it is of considerable interest to understand how widespread these practices are and how they affect the innovation process. Joint application for patents by multiple organizations is a form of open innovation that may result from joint R&D or other knowledge exchange between organizations. Interactions and collaborations affect the external knowledge potentially accessible by an organization, but they may also reduce the organization’s ability to appropriate the value of its internal knowledge. An optimal innovation strategy will balance these factors. We find that joint patent applications are relatively widespread and organizations utilise a range of strategies. To better understand some of the factors underpinning partner selection, we investigate the role of similarity between organizations and their impact on collaboration. We consider three dimensions of homophily, namely: technological proximity, geographical proximity as well as organization type (e.g. company, university, government agency). Here we construct a multi-level network in order to quantify these similarities between organizations. We define layers of the network as dimensions of homophily. These dimensions (layers) can be viewed as node attributes of bipartite networks. We use European Patent Office data dating back to 1978 for 40 countries with harmonized applicant names (OECD REGPAT and HAN databases) to construct four related bipartite networks relating organizations to patents, technological codes, geographic regions, and organization type. The respective one-mode projections can be combined as a co-organization network that is related by the different edge types: namely patents, technologies, geographical and organization type. This resulting network shows the structure of connections between organizations and the correlations between patent collaborations and the different dimensions of similarity under consideration.

Abstract: Asia is experiencing an unprecedented region-wide housing bubble right now. Should this bubble collapse, the economic and social fallouts are mind-boggling. As Asian governments race against time to defuse these ‘ticking bombs’, a deeper understanding of housing bubbles becomes necessary. By plotting the cumulative distribution functions (CDFs) of home prices per unit area in Singapore between 1995 and 2014, we found that these CDFs are stable over non-bubble years, and consist universally of an exponentially decaying body crossing over to a power-law tail. We also found in bubble years that dragon kings (positive deviations from the equilibrium distribution) develop near where the exponential body crosses over to the power-law tail. These were found in home price distribution of the Greater Taipei Area between Aug 2012 and Jul 2014, even though the two housing markets are structurally different. For the Singapore housing market, we also investigated the spatio-temporal dynamics of the bubble, and found price surges always start in a prestigious investment district, before propagating outwards to the rest of the island.