مقاله انگلیسی رایگان در مورد همبستگی های زمان گرا در بازارهای املاک و مستغلات جهانی ( الزویر )

مقاله سال 2017

The analysis of the linkages between the volatilities and co-volatilities of the global financial markets, especially the global real estate markets, is a critical issue on global portfolio diversification opportunities and risk management practices. Estimations of correlations between the asset returns are relevant for predicting time-varying beta coefficients in capital asset pricing model, obtaining an optimal estimation of hedge ratio, and forecasting the Value-at-Risk of a portfolio strategy. Thus, studies on the estimation of time-varying correlations across financial assets, mostly by modeling the time-series structure of financial asset returns and volatilities, have increased [1–5]. Notably, a popular approach to the modeling of multivariate asset volatility dynamics is the conditional variance–covariance matrix estimation method. Over the past three decades, various parameterizations of the conditional variance–covariance matrix have been developed in the multivariate generalized autoregressive conditional heteroscedasticity (MGARCH) model applied in the literature.

One of the most popular models for the estimation of multivariate asset correlation dynamics is that of Engle [6], who proposed a dynamic conditional correlation GARCH (DCC–GARCH) model in which the conditional correlation matrix is time varying. Similarly, Tse and Tsui [7] formulated the conditional correlation matrix as a weighted average of the past correlations. The basic specification of the DCC–GARCH modeling approach includes two aspects. One aspect is the univariate variance process, and another aspect is the time-varying correlation process. Several recent extensions of the DCC–GARCH model developed by Engle [6] have been developed to provide more flexibility to the modeling of the second aspect, i.e., the time-varying correlation process [6]. Several recent typical examples are the corrected DCC model by Aielli [8], the non-scalar DCC model by Bauwens, Grigoryeva, and Ortega [9], the volatility threshold DCC model by Kasch and Caporin [10], and the asymmetric DCC model by Tamakoshi and Hamori [11]. For recent reviews of MGARCH models, see Ref. [12] and for other applications of such models, see Refs. [13–15]. These MGARCH models provide a useful tool for understanding how financial volatilities move together over time and across markets. However, a regional financial market can be characterized by locality and segregation on the basis of the fixed location of countries. Financial applications typically consider the correlations between the pairs of returns of financial indices observed in different countries, without including the spatial effects in the model adopted. Thus, a natural extension is to incorporate the spatial effects in the appropriate MGARCH models. However, two kinds of problems must be solved before we can achieve the idea of such extension. One problem is on dimensionality and the other is the choice of an appropriate MGARCH model that is able to include spatial effects. If we directly incorporate the spatial effects into the MGARCH models, providing feasible estimates of the models would be difficult because they contain numerous unknown parameters and require the conditional covariance matrix to be positive definite [12]. To solve the dimensionality problem, Otranto [16] proposed a clustering algorithm to detect groups of homogeneous time series in terms of one extensively used DCC model. Following this line of thought, a natural choice of appropriate MGARCH models is the family of the DCC models. Moreover, as stressed by Ref. [17], the most appropriate DCC model spefication that can include spatial effects is the DCC model developed by Tse and Tsui [7]. Therefore, the starting point of this article is the contribution of Otranto, Mucciardi, and Bertuccelli [17], which introduced spatial effects to the analysis of dynamic conditional correlation models, and thus to the estimation and measurement of contagion effects. However, we take a step further in several aspects.

First, we considered a compound spatial weight matrix, which is a combination of the geographic distance and economic distance spatial weight matrices used by Case, Rosen, and Hines [18] and Zhu, Füss, and Rottke [19] to effectively demonstrate the effects of geographic and economic indicators on the global real estate markets. Our main motivation for studying spatial effects is that a country’s real estate market is prone to be affected by its nearby countries not only because of geographical proximity but also because of economic and financial similarities [19–22]. In terms of the specification of the geographic distance spatial weight matrix, we regarded all spatial units as the neighbor of each other and made use of a Gaussian kernel function form because we believe that the interactions between nearer neighbors are larger than that of those between farther neighbors. Moreover, we employed the Mahalanobis distance to construct the economic distance spatial weight matrix using relevant economic indicators, such as per-capita gross domestic products (GDP), population, national unemployment rates, and imports and exports of the country.1 We also believe that regions with similar economic development conditions will exhibit strong co-movement because the Mahalanobis distance considers the correlations of economic indicators. Notably, the kernel bandwidth is the key controlling parameter and can be specified either by a fixed bandwidth or by an adaptive bandwidth [19]. In contrast to Ref. [19], we chose an adaptive bandwidth rather than a fixed bandwidth. The optimal adaptive bandwidth is determined by the maximum log-likelihood function value and the significance of the regression coefficients (see Section 2.3 for details).