Monitorización de la coyuntura económica regional a través de un indicador sintético

Resumen:En el presente trabajo se obtiene un indicador sintético que pretende proporcionar una herramienta para el seguimiento de la coyuntura económica de una región. Siguiendo la literatura existente el indicador sintético se construye mediante la aplicación a los datos coyunturales del modelo factorial dinámico que permite reducir la dimensionalidad de los datos iniciales. De este modo se obtiene una interpretación más simple y compacta a través de un conjunto reducido de factores comunes. El indicador obtenido se utiliza para el caso de Galicia. Los resultados obtenidos permiten seguir la evolución, con periodicidad mensual, de la economía gallega de manera consistente con la evolución trimestral del PIB calculado por la oficina de estadística regional. Además, el estimador sintético calculado se puede utilizar para predecir el PIB a través de un modelo aditivo generalizado (GAM) en el que se tiene en cuenta la estructura de correlación de los datos.Abstract:This paper builds a synthetic indicator that aims to provide a tool for monitoring the economic situation of a region. Thus, a synthetic indicator is constructed by applying to the economic data a dynamic factorial model that allows reducing the dimensionality of the initial data. To guarantee its technical solvency, the synthetic indicator developed considers the methodological developments pointed out by Stock and Watson (1991). In this way we proceed by applying the dynamic factorial model to the economic series. This sort of indicators constructed using dynamic common factor models aims to represent a relatively large set of initial series by means of a smaller set that achieves a simpler and more compact interpretation. These indicators must fulfill the characteristics pointed out by Burns and Mitchell (1946), Artís et al. (1997) or Cuevas and Quilis (2010, 2015). Those are the following characteristics: i.- Length of the series; ii.- Economic significance; iii.- Statistical quality; iv.- Soft profile; v.- Speed in the availability of information, and vi.- Monthly frequency. By doing this, a simpler and more compact interpretation is obtained through a reduced set of common factors. The variables finally included in the indicator are: 1.- Consumption of gasoline and diesel; 2.- Exports of goods deflated by the Export Unit Value Index; 3.- Imports of goods deflated by the Export Unit Value Index; 4.- Surface of the houses to create a new plant; 5.- Registration of tourism vehicles; 6.- Production of tourism vehicles; 7.- Air transport of passengers; 8.- Maritime transport of goods (loaded and unloaded); 9.- Housing visas; 10.- Index of turnover in the industry deflated by the Industrial Price Index (IPRI); 11.- Affiliations to Social Security; 12.- Industrial Production Index deflated by the IPRI (linked series); 13.- Retail trade sales index; 14.- Travelers entering the hotels; and 15.- Index gives turnover in the services sector deflated by the services CPI. The MARSS R package is used to estimate the indicator (Holmes et al., 2014). The functions of the package estimate the parameters of the model (Z, Q and R) by maximum likelihood using the Expectation Maximization (EM) algorithm. The EM algorithm of Dempster et al. (1977) was used for this type of models by Shumway and Stoffer (1982) and Watson and Engle (1983). For the estimation of the factors, the Kalman filter is used once the model is represented in the state space. The results show the number of common factors that are needed to represent most of the variability of the original series. To do this, the eigenvalues of the sample autocovariance matrices are calculated. Likewise, with the objective of finding the co-integration range “s”, the Johansen test (1991) was carried out to the nine series in which there is information since 1995. This test concludes that, with a level of significance of 5%, the co-integration range is 8 and, consequently, the series share a total of 1 trend in common. Then, the factorial loads are obtained in the series used, which are all positive and are used to calculate the common factor. The variable with the highest burden is relative to the number of social security affiliates. Once the estimation of the synthetic indicator has been carried out, it is important to analyze the explanatory capacity of the indicator on quarterly GDP growth as a commonly accepted benchmark indicator. This analysis shows the existence of cyclical coherence and stability in the relationship that will allow accepting the common factor as a valid synthetic indicator. Finally, once the indicator has good properties to be a good predictor of GDP, a model is established that relates these two indicators and allows GDP to be predicted from the synthetic indicator. The results obtained allow us to monitor the monthly evolution of a regional economy in a way consistent with the quarterly evolution of GDP. In addition, the estimated synthetic estimator can be used to predict GDP through a generalized additive model (GAM) in which the correlation structure of the data is taken into account. This work has several advantages when comparing to similar proposals in previous literature. We can summarize this advantages in the following four. Firstly, it is necessary to point out the good performance of the indicator when compared with the evolution of GDP. The high degree of coupling between the evolution of both magnitudes (synthetic indicator and GDP), determined by the correlation coefficient, shows their cyclical coherence and stability. The second advantage lies in the simplicity of the procedure used. In this paper, a simpler procedure is used and, therefore, with fewer hypothesis assumptions than in most works with similar characteristics and, on the other hand, the empirical results are very similar. A third advantage of the calculated synthetic indicator is that it serves as an indicator of the evolution of the economy, at this time the periodicity of GDP is quarterly, so in any quarter, two months in advance and by calculating the synthetic indicator and the establishment of the function of relationship between the two indicators by the proposed GAM model, it will be possible to anticipate what the evolution of GDP will be in that quarter. Finally, the adjustment of the polynomial of degree three that, in principle, is slightly worse than that made by “splines” has the advantage that allows, intuitively, give an approximation of the inflection point, in this case when IS = 1, 75, which can be used as a control value in the monitoring of the conjuncture. Therefore, an easily replicable methodology is presented, both with the data available in another region and with the data available at national level or even with other countries.