Bayesian Spatial Autoregressive Model Research Articles

Contagion, Not Only Interconnection: Measuring the Transmission of Financial Distress

This paper proposes a novel approach to understand contagion of financial distress in the banking system, which takes into account the spatial nature of the phenomena. We use a Bayesian spatial autoregressive model that treats the likelihood of default of each bank as endogenous, and dependent on the network formed by all the other banks. Identification is achieved by controlling for bank fundamentals, latent macrofinancial and bank specific shocks that have similar consequences to contagion and act as confounding factors. Through the lens of a simulations exercise we study the importance of the structure of financial networks for financial stability, shedding light on the empirical adherence of important theoretical propositions that remain untested.

Flexible shrinkage in high-dimensional Bayesian spatial autoregressive models

Several recent empirical studies, particularly in the regional economic growth literature, emphasize the importance of explicitly accounting for uncertainty surrounding model specification. Standard approaches to deal with the problem of model uncertainty involve the use of Bayesian model-averaging techniques. However, Bayesian model-averaging for spatial autoregressive models suffers from severe drawbacks both in terms of computational time and possible extensions to more flexible econometric frameworks. To alleviate these problems, this paper presents two global–local shrinkage priors in the context of high-dimensional matrix exponential spatial specifications. A simulation study is conducted to evaluate the performance of the shrinkage priors. Results suggest that they perform particularly well in high-dimensional environments, especially when the number of parameters to estimate exceeds the number of observations. Moreover, we use pan-European regional economic growth data to illustrate the performance of the proposed shrinkage priors.

A bayesian spatial autoregressive model with k-NN optimization for modeling the learning outcome of the junior high schools in West Java

Increasing the human capital development index of Indonesia is needed to realize the country’s dream to become a developed country in the world. Quality education is needed for that purpose, and this should start from an early age. School is a formal institution for knowledge transfer, which is very useful in building the quality of an Indonesian’s character. Since 2000, Indonesia has made enormous effort to improve the quality of education, which is measured by increased learning outcome, which is measured by mean national examination score. Indonesia has focused on three major aspects, namely, improving equity and access, enhancing quality and relevance, and strengthening management and accountability. These three aspects are translated into eight standards accreditation score. Education quality is believed to have spatial characteristics that follow the Tobler law. In general, schools close to each other, especially in one administrative area, have the same quality characteristics. The spatial characteristics need to be included in modeling the national examination score. Because of the normality assumption problem, we use a Bayesian spatial autoregressive model (BSAR) to evaluate the effect of the eight standard school qualities on learning outcomes and use k-nearest neighbors (k-NN) optimization in defining the spatial structure dependence. We use junior high schools data in West Java. West Java is one of the largest provinces in Indonesia with the highest number of junior schools. The result shows that the national examination score of the junior high schools in West Java is significantly influenced by the standard of graduate competence, and the standard of assessment. We found that the spatial effect also significant which means the average of the national examination score of the nearest schools influences the national examination of the junior high observed.

Assessment of the hurricane-induced power outages from a demographic, socioeconomic, and transportation perspective

Natural disasters have devastating effects on the infrastructure and disrupt every aspect of daily life in the regions they hit. To alleviate problems caused by these disasters, first an impact assessment is needed. As such, this paper focuses on a two-step methodology to identify the impact of Hurricane Hermine on the City of Tallahassee, the capital of Florida. The regional and socioeconomic variations in the Hermine’s impact were studied via spatially and statistically analyzing power outages. First step includes a spatial analysis to illustrate the magnitude of customers affected by power outages together with a clustering analysis. This step aims to determine whether the customers affected from outages are clustered or not. Second step involves a Bayesian spatial autoregressive model in order to identify the effects of several demographic-, socioeconomic-, and transportation-related variables on the magnitude of customers affected by power outages. Results showed that customers affected by outages are spatially clustered at particular regions rather than being dispersed. This indicates the need to pinpoint such vulnerable locations and develop strategies to reduce hurricane-induced disruptions. Furthermore, the increase in the magnitude of affected customers was found to be associated with several variables such as the power network and total generated trips as well as the demographic factors. The information gained from the findings of this study can assist emergency officials in identifying critical and/or less resilient regions, and determining those demographic and socioeconomic groups which were relatively more affected by the consequences of hurricanes than others.

The Role of Contagion in the Transmission of Financial Stress

I examine the relevance of contagion in explaining financial distress in the US banking system by identifying the component of bank level probabilities that is due to contagion. Identification is achieved after controlling for macrofinancial and bank specific shocks that have similar consequences to contagion. I use a Bayesian spatial autoregressive model that allows for time-dependent network interactions, and find that bank default likelihoods depend, to a large extent, on peer effects that account on average for approximately 35 per cent of total distress. Furthermore, I find evidence of significant heterogeneity amongst banks and some institutions to be systemically more important that others, in terms of peer effects. JEL Classification: E44, G01, C11, G21

Risk of bladder and kidney cancer from exposure to low levels of arsenic in drinking water, Nova Scotia – Canada

Introduction: Arsenic (As) in drinking water affects the health of hundreds of millions of people worldwide. Although West Bengal, Bangladesh and Taiwan are among the most affected regions, with As concentration as high as 4,700 μg/L and where levels >300 μg/L are common, As contamination of well water is also prevalent across the US and Canada. Studies of populations exposed to such high levels show strong associations and dose-response relationships with As in drinking water and a wide range of illness, including, skin, lung, bladder, and kidney cancer. The threshold at which these diseases develop is uncertain at lower levels of exposure. Recent evidence suggests possible health effects at levels as low as 10 μg/L, current World Health Organization advisory limit. This study models the risk of developing bladder/kidney cancer in those exposed to As around current guideline levels. Methods: A Bayesian spatial autoregressive model was used to model risk at 3 levels of As exposure (0–2 μg/L; 2–5 μg/L; >5 μg/L—based on 10,498 private well samples) in 864 bladder and 525 kidney cancer cases diagnosed in Nova Scotia Canada, between 1998-2010. The model accounted for spatial dependencies and included covariates (e.g. smoking proxies). Results: The risk of developing bladder cancer (both sex) was on average 16 % (2–5 µg/L) and 18 % (> 5 µg/L) greater than that of the referent group (< 2 µg/L), with associated posterior probabilities of 0.88 and 0.93 for these risks being greater than the reference rate. Effect sizes for kidney cancer were lower; the risk being on average 5 % (2–5 µg/L) and 14 % (> 5 µg/L) greater than that of the referent group, with associated probabilities of 0.66 and 0.87. Stratified analyses by sex, showed larger posterior probabilities for an increased risk amongst males– a result likely due to low female case counts. Conclusion: This study provides evidence for health effects at low level of As exposure in drinking water.

The Effects of Congressional Staff Networks in the US House of Representatives

Standard accounts of legislative behavior typically neglect the activities of professional staff, who are treated as extensions of the elected officials they serve. However, staff appear to have substantial independent effects on observed levels of legislator productivity and policy preferences. In this article, we use a novel data set of comprehensive longitudinal employment records from the US House of Representatives to estimate the effects of congressional staff on legislative behavior. Specifically, results from a series of heteroskedastic Bayesian spatial autoregressive models indicate that members of Congress who exchange important staff members across congresses are more similar in their legislative effectiveness and voting patterns than we would otherwise expect. These findings suggest that scholars should reconsider the role of staff in the legislative process.

Jurisdictional spillover effects of sprawl on injuries and fatalities

There is a considerable literature on the relationship between sprawl and accidents. However, these studies do not account for the spatially correlated effects of sprawl on accidents. In our analysis of 122 jurisdictions in Southeast Michigan, we use a Bayesian spatial autoregressive model to estimate how injuries and fatalities in one jurisdiction are associated with sprawl in that jurisdiction and sprawl in neighboring jurisdictions; we also correct for heteroskedasticity in the data. Using principal component analysis, we create a sprawl index from five underlying land use characteristics. Our results show that the number of injuries and fatalities in a jurisdiction increases with the magnitude of sprawl in neighboring jurisdictions. We believe that this is because more drivers per capita in sprawled jurisdictions traverse similarly sprawled neighboring jurisdictions for daily activities. Furthermore, driving habits attuned to less defensive driving in sprawled jurisdiction are transferred to similarly designed neighboring jurisdictions, contributing to accidents in the latter.