Abstract
The STAR model is widely used to represent the dynamics of a certain variable recorded at several locations at the same time. Its advantages are often discussed in terms of parsimony with respect to space-time VAR structures because it considers a single coefficient for each time and spatial lag. This hypothesis can be very strong; we add a certain degree of flexibility to the STAR model, providing the possibility for coefficients to vary in groups of locations. The new class of models (called Flexible STAR–FSTAR) is compared to the classical STAR and the space-time VAR by simulations and an application.
Similar content being viewed by others
Notes
Necessary and sufficient conditions are much more complicated, so, as in Arbia et al. (2011), we impose only the usual necessary conditions.
We would like to thank an anonymous referee and the Associate Editor who called this problem to our attention.
The number of spatial units might seem small, but this choice is consistent with the selection of balanced space-to-time ratios in STARMA models illustrated in Otranto and Gallo (1994). They show, simulating data from STAR(1,1) processes in regular lattices, that, in correspondence of a time span \(T=100\), a number of locations higher than 25 can cause an ill-conditioning of the covariance matrix of the estimators, with a determinant close-to-zero. This problem is even greater in the presence of spatial autocorrelation between the disturbances because the collinearity problems could arise also in presence of a large T.
Other classical contiguity criteria are the bishop criterion (common vertex) and the queen criterion (common edge or vertex); see Anselin (1988), Sect. 3.1.2.
We have also used the bishop contiguity matrix and the results are similar and in many cases worse than the queen case. Results available on request.
In the wrong weight matrix case, we fix \(T=1000\) as in the previous subsection.
We are very grateful to Francesco Giorgianni and Gianluca Trifirò who have produced and made available this data set.
The other time series have a very similar behaviour; they are available on request.
Moreover, given the relationship between STAR and VAR models, as shown in Eq. (2.3), we estimate also a Sparse VAR (SVAR) model, shrinking some coefficients toward 0. The results are not good, compared with respect to the spatial models, particularly in terms out-of-sample forecasting, so we do not report the corresponding results. Details are available on request.
Data provided by Regione Campania—A.A.L. Caserta. The ageing index is calculated as the number of persons 60 years old or over per hundred persons under age 15. The old-age dependency ratio is the number of persons 65 years and over per one hundred persons 15–64 years.
References
Anselin L (1988) Spatial econometrics: methods and models. Kluwer Academic Publishers, Dordrecht
Aquaro M, Bailey N, Pesaran HM (2015) Quasi maximum likelihood estimation of spatial models with heterogeneous coefficients, CESifo Working Paper N. 5428
Arbia G, Bee M, Espa G (2011) Aggregation of regional economic time series with different spatial correlation structures. Geograph Anal 43:78–103
Besag J (1974) Spatial Interaction and the statistical analysis of lattice systems. J R Stat Soc Ser B 36:192–236
Cliff AD, Ord JK (1973) Spatial autocorrelation. Pion, London
Cliff AD, Ord JK (1975) Space-time modeling with an application to regional forecasting. Trans Inst Br Geogr 64:119–128
Cliff AD, Ord JK (1981) Spatial processes. Models and applications. Pion, London
Cressie N (1993) Statistics for spatial data. Wiley, New York
Fotheringham AS, Brunsdon C, Charlton ME (2002) Geographically weighted regression: the analysis of spatially varying relationship. Wiley, New York
Frühwirth-Schnatter S (2011) Panel data analysis—a survey on model based clustering of time series. Adv Data Anal Classif 5:251–280
Frühwirth-Schnatter S, Kaufmann S (2008) Model-based clustering of multiple time series. J Bus Econ Stat 26:78–89
Getis A, Ord JK (1981) The analysis of spatial association by use of distance statistics. Geogr Anal 24:189–206
Giacomini R, Granger CWJ (2004) Aggregation of space-time processes. J Econom 118:7–26
Haining R (1990) Spatial data analysis in the social and environmental sciences. Cambridge University Press, Cambridge
Hansen PR (2010) A Winner’s curse for econometric models: on the joint distribution of in-sample fit and out-of-sample fit and its implications for model selection. Stanford University, Mimeo
Holm S (1979) A simple sequentially rejective multiple test procedure. Scand J Stat 6:65–70
Hubert L, Arabie P (1985) Comparing partitions. J Classif 2:193–218
LeSage J, Chih Y-Y (2016) Interpreting heterogeneous coefficient spatial autoregressive models. Econ Lett 142:1–5
LeSage J, Pace RK (2009) Introduction to spatial econometrics. Chapman & Hall, Boca Raton
Liao T (2005) Clustering time series data: a survey. Pattern Recogn 38:1857–1874
Lütkepohl H (1993) Introduction to multiple time series analysis. Springer, Berlin
Maharaj EA (1999) Comparison and classification of stationary multivariate time series. Pattern Recogn 32:1129–1138
Mucciardi M, Bertuccelli P (2012) The impact of the weight matrix on the local indicators of spatial association: an application to per-capita value added in Italy. Int J Trade Global Markets 5:133–141
Otranto E (2008) Clustering heteroskedastic time series by model-based procedures. Comput Stat Data Anal 52:4685–4698
Otranto E (2010) Identifying financial time series with similar dynamic conditional correlation. Comput Stat Data Anal 54:1–15
Otranto E, Gallo GM (1994) Regression diagnostic techniques to detect space-to-time ratios in STARMA models. Metron 52:129–145
Otranto E, Mucciardi M, Bertuccelli P (2016) Spatial effects in dynamic conditional correlations. J Appl Stat 43:604–626
Pfeifer PE, Deutsch SJ (1980) A Three-stage iterative procedure for space-time modeling. Technometrics 22:35–47
Pinkse J, Slade ME (1998) Contracting in space: an application of spatial statistics to discrete-choice models. J Econom 85:125–154
Rand WM (1971) Objective criteria for the evaluation of clustering methods. J Am Stat Assoc 66:846–850
Steece B, Wood S (1985) A test for the equivalence of \(k\) ARMA models. Empir Econ 10:1–11
Stetzer F (1982) Specifying weights in spatial forecasting models: the results of some experiments. Environ Plan A 14:571–584
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Otranto, E., Mucciardi, M. Clustering space-time series: FSTAR as a flexible STAR approach. Adv Data Anal Classif 13, 175–199 (2019). https://doi.org/10.1007/s11634-018-0314-5
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11634-018-0314-5