Economic Complexity

Nowadays economic systems have evolved to become globalized, largely financialized and in- terconnected. This observation leads directly to another: any approximation that regards them as the sum of many independent parts is going to fail. In other words nowadays economic systems are Complex Systems and they are much more than the sum of their single parts, for the fact that they show emergent chaotic behaviours, that we are not able to predict nor to ex- plain within the mainstream framework. As the set of relevant interactions among economic entities is much wider than what has been used to build present-day’s economic theories, we can’t build up on these theories anymore. What we need is probably more similar to a brand new start. If we want our theories to be predictive in such a complex scenario we need to link them tightly to data and let our models go beyond overly simplified linear approaches. More importantly, we need to state clearly the limits of these theories and to discriminate situations where we can actually make predictions from those where we can’t. The methods, approaches, data and ideas proposed in this work and the way these ingredi- ents are combined together, embody our view on how a brand new start in Macroeconomic modeling should look like, for two reasons. The first is our concern in remaining constantly linked to quantitative data. Of course we make assumptions about some underlying mecha- nisms and general principia, but we constrain their validity to the extent of the ability of our methods to perform quantitative predictions. The second is the fact that we always try to use the right method for the right problem. Examples of this are provided in chapters 2 and 4 where methods derived from dynamical systems and machine learning are used and tested in the field of economic prediction. Mainstream economic modeling is often rigid and sim- plistic from a methodological point of view, with a predominant and pervasive use of linear regressions and descriptive statistics. These tools have strict validity limits and, from a con- ceptual point of view, provide a low level of falsifiability, unless a comprehensive theoretical framework is available. Thus, since reductionism is made hard in economics by the impossibil- ity to repeat experiments and control the environment, our goal is always to look at concrete predictions and we value our results on the basis of their accuracy. The philosophy that drives our approach is the similarity that exists between economic and ecologic systems. As discussed in chapter 3 economy and ecology are close not only for shar- ing the same root as a word, but also in the very essence of the interactions they describe: sets of individual competing and cooperating for the allocation of common resources. Once this connection is put in focus it is easy to move concepts and ideas among the two fields and start to observe similarities in the data that we have about economic networks and ecologic ones. In particular we explore the concept of nestedness in interaction networks, observed in ecology and economics: namely when specialized entities only interact with generalist ones. This means that only generalists have access to exclusive relationships. Both in economics and ecology this provides an obvious advantage and leads to the identification of the concept of diversification with a measure of fitness, intended as the ability to perform well and survive in the ecosystem. The presence of a nested structure also suggest the presence of a precise under- lying dynamics in the formation of such networks, and allows us to discover very informative taxonomic structures, as described in the final chapter. We open our discussion in chapter 1 with an empirical observation about the structure of the bipartite network defined by countries and the products they export. The nestedness of this network is exploited to build a quantitative measure of complexity for countries’ national economies and for products, in a self-consistent way. This measure of complexity is then used in chapter 2 to introduce a framework in which we are able to predict the Macroeconomic development of a set of countries. By using an approach derived from dynamical systems we are able to define quantitatively the level of predictability of a country’s development, sim- ilar in spirit to a weather forecast, where turbulent areas are much harder to predict. The parallelism with biological systems is studied more in detail in chapter 3, where we observe how the build-up of diversity seems to follow some universal features. We propose a minimal scheme to model this dynamics and we introduce the concept of Usefulness which reveals to be a crucial feature if we want to picture real-world diversity as the combination of smaller Building Blocks. Finally in chapter 4 we look at how the export data can be used to infer tech- nological relations among products. These relations are relevant for countries’ development, as we show that countries move in recurrent paths when developing. The accuracy of our methods in predicting new links in the countries-products network is quantitatively assessed and results to be as high as 16% in a-priori selected subsets.