New Cross-Sectionally Dependent Panel Data Methods for the Analysis of Macroeconomic and Financial Networks

点击翻译按钮获取中文摘要

In the social sciences, it is common to use datasets in which information for a group of entities is recorded at multiple points in time. This is known as panel data and it forms the basis for longitudinal analysis. As with all statistical models, panel data models rely on assumptions. One common assumption of panel data models is that the residual variation in the data (i.e. that part of the variation in the data that the model cannot explain) is uncorrelated across entities. This is known as cross-sectional independence. However, this assumption is frequently violated in practice. The development of methods to control for cross-sectional dependence (CSD) is an active area of research.CSD can arise through two mechanisms. First, the data may exhibit spatial dependence, such that the behaviour of one entity may depend on the behaviour of its neighbours/peers. This is often called 'local' or 'weak' CSD. Second, the data for all entities may be influenced by one or more common factors. This is 'global' or 'strong' CSD. Often, both mechanisms may be jointly responsible for CSD. However, in practice, models that account for both spatial effects and common factors are rare, and those that do exist are highly stylised. We propose to develop a unifying framework for the estimation of sophisticated and realistic dynamic heterogeneous panel data models that account for spatial dependence and common factors.This project will generate three significant methodological advances. We will:(i) increase the flexibility and realism of spatial dynamic panel data models with common factors by developing techniques that allow for the model parameters to be heterogeneous across individuals, unlike most existing studies that assume parameter homogeneity.(ii) develop methods to exploit the network structure of spatial dynamic panel data models, opening new opportunities to use models of this type to understand the bilateral linkages among entities in the global economy.(iii) extend the methods discussed above from the common case of unilateral (or 2-dimensional) panel data to the more complex case of bilateral (3D) panel data, such as trade and investment flows.We will apply the methodologies that we develop to study three important aspects of globalisation. We will:(i) develop a new model to study the convergence of national business cycles onto a so-called global business cycle. Our model will allow us to separate convergence due to the effect of spatial linkages (e.g. trade and political relations, migration flows etc.) from convergence due to the influence of global factors. This model will help to guide the design of economic stabilisation policy in an interconnected world.(ii) develop a new model to study global trade flows and to separate the influence of spatial linkages (e.g. common borders, membership of free trade areas, common languages etc.) from global factors (e.g. the state of the global business cycle). The development of such models is of strategic importance to the UK, given the trade implications of Brexit.(iii) develop a new hierarchical model of global stock markets, where the performance of a firm may depend on spatial relations (e.g. linkages to other firms in its sector and/or in its geographical region) as well as a range of common factors (e.g. liquidity, investor risk aversion). Models of this type provide new insights into the globalised nature of economic activity and highlight opportunities and obstacles to economic growth for both the public and private sector.In sum, this project will make significant methodological contributions and will leverage these contributions to address pressing contemporary issues facing policymakers and professional economists alike.