James LeSage , Manfred M. Fischer (p)

The focus is on cross-sectional dependence in panel trade flow models. We propose alternative specifications for modeling time invariant factors such as socio-cultural indicator variables, e.g., common language and currency. These are typically treated as a source of heterogeneity eliminated using fixed effects transformations, but we find evidence of cross-sectional dependence after eliminating country-specific effects. These findings suggest use of alternative simultaneous dependence model specifications that accommodate cross-sectional dependence, which we set forth along with Bayesian estimation methods. Ignoring cross-sectional dependence implies biased estimates from panel trade flow models that rely on fixed effects.

Katarzyna Kopczewska (p)

Spatial econometrics for big data point geo-locations has a limited possibility of forecasting with a calibrated model for a new out-of-sample geo-points. This is because of spatial weights matrix W defined for in-sample observations as well as the computational complexity for huge W. This paper proposes the novel methodology which calibrates both space and model relation using bootstrap and tessellation. Bootstrapping enables the calibration of the econometric model without the need for estimation on the whole dataset. Tessellation for the points in the best model selected allows for a representative division of space. New out-of-sample points are assigned to tiles and linked to spatial weights matrix as a replacement for original point. This efficient procedure supports the big data geo-located point data and makes feasible a usage of calibrated spatial models as a forecasting tool for out-of-sample data. This methodology will find its applications in real estate market forecasting as well as models of business location.
The first stage determines the bootstrapping parameters: the number of iterations and size of subsample to gain the efficiency of estimation. Two-dimensional analysis is conducted for both a-spatial OLS as well as spatial models, using random sampling and jackknife. In the second stage, the best bootstrapped model is selected as the best representation of the population. Spatial models are estimated with k-nearest neighbours W. One should note that in each model the W differs. Partitioning Around Medoids (PAM) algorithm together with CLARA - its big data equivalent are applied to find the best points representation which generates the medoids coefficients. In the third stage the Dirichlet tesselation is performed for the subsample selected in the stage two. This calibrates the space and allows for obtaining the calibrated spatial weights matrix, which is the best representation of the sample. The fourth stage is the forecasting for new out-of-sample geo-located point. The point is assigned to the tessellation tile in the overlaying procedure. In this controlled imputation, the basic subsample dataset is supplemented with the replacing vector of data for a new point, what deletes the variable values of original point in the selected tile. The forecasted values are produced with the calibrated model, selected with PAM / CLARA and for calibrated W. The quality of the forecast is tested for the different scenarios of this bootstrap procedure, including the selection of k neighbours, the model specification and the bootstrapping parameters.