A spatial approach in analyzing the structure and dynamics of the Romanian GDP Renata-Graziela BOAR, Alexandru IOVANOVICI and Horia CIOCARLIE Abstract—Modern economies are highly dynamical systems with states described by a vast amount of parameters having a lot of uncertainties and in a state of precarious equilibrium. Our research is geared towards applying techniques at the cross roads between econometrics, statistics and graph theory in order to improve the tools for analyzing the key factors that influence the state of the economy as seen through the gross domestic product. The novelty of our approach is characterized by decomposition of the key sectors of the GDP and building a spatial representation of the relationships between them in the form of a complex graph. Index Terms—econometrics, graphs, scale free networks, GDP, Romania I. I NTRODUCTION E CONOMY plays a key role in our modern world. What- ever we do and however we choose to live our lives we have to take into consideration some economical aspects of some sorts. From a simple payment for groceries and up to starting a new business and buying a factory every decision we make is influenced by our perception of what is generally known as economy. The layered view of this mostly abstract system can be seen at the highest level in what it’s called macroeconomy. Of particular importance in the field of macroeconomic analysis is the possibility of finding the relationships between various key players in the economic exchange and more importantly the cross influences caused by this players at the large scale. Classical solutions to this problems exists in the literature and we are going to present them but suffice to say that they usually relay on statistics. In the same time we witness the growing interest and large variety of applications of what is called complex networks or in a more specific approach social networks. Literature exists on the application of what is called the “new network science” [1] in various fields of human activity, including economics. As presented in the Barabasi’s seminal paper this approach allows us to model and represent anything which can be expressed as a relationship between some entities - this being actually one of the underlying powers of the graph data structure - and provides us with novel means of exploring and understating this newlly structured data. Instead of simply analyzing a set of numbers we are now capable of visualizing the relationship between data and in the same time, by applying the newly proposed metrics we can obtain a better perspective at the influence of various nodes to the network at large. This rest of the paper is organized as follows: in Section 2 we make a survey of the most important research approaches All the authors are with the Department of Computers, POLITEHNICA University, Timisoara, Romania. Their e-mails correspondingly are: re- nata.boar@cs.upt.ro, iovanalex@cs.upt.ro, horia@cs.upt.ro. in this field, Section 3 presents the application of our analysis on the data set regarding the structure and evolution of the Romanian GDP with pertinent charts and graphs and in Section 4 make a discussion of the results and draw some conclusions. II. STATE OF THE ART Applying mathematical analysis and statistics in order to interpret economic data is not a novelty. The entire field of econometrics is placed at the crossroads between economics and mathematics, especially statistics. In the same time there is a great deal of efforts involved with predicting the outcome of a time series in our case the GDP of a particular nation. Most of this was carried with tools and methods designed for statistical analysis.. Predicting the future is one of the desires of mankind and in our case it has an even greater importance being the key factor between fortune or loss. Our approach is mostly inspired by the research carried by Mantegna [2] which involved constructed a correlation matrix based on stock trade and transactions. His methodology involved building the corresponding graph of trade interaction on which he afterward computed the minimum spanning tree. The continuation of his research was presented in [3] and involved the generation the planar maximal graph which was able to transport even more information to the analyst. The key aspect in building the graphs is representing by finding the correlation between the various data sets. Con- sidering the problems of the standard correlation the authors present in [4] a solution of building the correlation matrices based on partial correlation. One interesting aspect we adopted from their paper is the use of a so called threshold in order to cut out the uncorrelated data. Still, their research is based on stock data which is seen a very good testing platform mostly because there are a large number of observations and in the same time they are considered a correct representation of the state of economy. A invigoration of the research in this field was carried after the financial crisis. Kennet et al. [5] proposed after analyzing the stock dynamics between 1999 and 2010 a new metric called index cohesive force and which is claimed to be able to predict a market crash. Stock trade was the target of another interesting and important research carried also by Kennet and his team and presented in [6]. By using the Dow Jones index seen as a dependency matrix they were able to identify some episodes of crisis. Going back to our study, reflected in this paper, there is a number of similar initiatives presented in [7], [8] and [9] especially the possibility of using what is known as complex – 259 – 9th IEEE International Symposium on Applied Computational Intelligence and Informatics • May 15-17, 2014 • Timişoara, Romania 978-1-4799-4694-5/14/$31.00 ©2014 IEEE