Complexity and industrial policy : Measuring the Intangible Growth Potential of Countries

The presentation focused on an attempt to test ideas about the complexity of products quantitatively. How a measurs for intangibles can be arrived at ? The philosophy starts from a simple premise. It begins by looking at 'fitness' or 'quality'. So if fitness is above average for people on the same income, the predict ion is that firm will grow. The first attempt to do this used the C OMTRAD E database, as there is no complete database of all production. COMTRADE , despite its weaknesses, was used as it provides homogeneous data . The analysis focuses on the quality of exports, rather than incomes. Complex products are differentiate d from simple products , where complex products are defined as pro ducts that can be produced only by few countries. However, competitiveness at country level is not due to specialisation, but diversification. Top countries produce both simple and complex products. Diversification breeds resilience. Pietronero and his tea m have developed metrics for both fitness and complexity. The relationships are not linear, it is limited by the lowest - income country that can produce the product. Fitness is defined as the number of products made ( proxied by exported products from COMTRADE ), weighted by the comp lexity of those products. Complexity is high if a product is made by a few countries of high fitness. The dynamic economic ecosystems start from countries, moving to (regional , etc. ) sub - systems , for which it is difficult to get homogeneous data though. The se ecosystems produce countries that are diversified , and companies that are specialised. From an analytical point of view a unified database on who produces what would be ideal . Even with available data though, the resulting algorithm can be tested by inserting it into a model that uses directly 10 measured data. The test turns out to be algorithm dependent, i.e. signal strength depends on choice of algorithm . The results can be used for predictive purposes. For instance, p lot ting log GDP against log fitness in 1995 shows China a s an outlier with both high fitness and low income , as opposed to the oil producers for instance . The clear prediction is growth for China. Examining the data using coarse grained analysis, it is obvious that economic complexity is highly heterogeneous. Regression methods would mix the dynamic properties of different systems and loose information as a result , whereas dynamic systems approaches retain this information . This i s a well - know n fact from weather forecasting. The data can be used to speak to other, non - trivial questions. For instance, one could ask whether corruption will hinder growth in Nigeria? Just looking at the place on the map, after a certain limit of fitness corruption seems not to matter. Nigeria is well below this threshold though , indicating that corruption will indeed tend to retard growth . The predictive power of the system increases better with higher levels of fitness. The data can also map product networks. To do this , one looks at whether two products are generally ma de together in many countries. One can then look at diachronic evolution and synchronic vicinity . Using the product space it is possible see that the evolution of product output at country level does not make jumps. Industrial policy should take this into account. The team plans to extend the database further over the course of 2014. Already the database has been extended to 60 years, and the product space is now more systematically defined. There are further interesting applications. For instance, one can construct a probabilistic model where capabilities are randomly assigned. The probability of a capability needed for a product follows a power law distribution . The resulting poverty trap model looks a lot like real data. Moreover, there appear to be overlaps between economics and ecology. The great divergence and the Cambrian explosion both are characterised by " triangularity " and " nestedness " . The method can also be used for growth de composition . For instance , one can look at the process of Japan catching up with the US . De composing countries into high and low fitness shows that countries with high fitness tend to grow faster. 11 Q & A Questions focused on how country size is taken into account, the weakness of data in capturing only exports, the linearity of the relationship between fitness and growth, and the possibility analysing the real technical content of products made. Pietronero clarified that while size is taken into account through the use of thresholds the model is slightly biased against the fitness of small countries. He made clear that the data is used only for lack of alternatives and that is of course has weaknesses. Sim ilarly, the linear m o del was employed as it is the simplest formulation that fit the data, and that data quality at present was not good enough to warrant a more complex formulation. Different algorithms should be measured against one another to see which is best, not against their data inputs.