Essays in spatial panel econometrics

• Main Author: Palombi, Silvia
• Published: University of Strathclyde 2016
• Subjects: 330.01
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.693179

This thesis develops and applies state-of-the-art spatial panel econometrics methods in order to model and analyse labour-market outcomes within and across small areas in Great Britain, with respect to three particular aspects; namely the resilience of local economies to periodic shocks, the determinants of spatial disparities in local wages, and the relationship between output and unemployment over the economic cycle. The contribution is thus provided in four essays. The first essay (Chapter 2) explores the relative ability of local economies to preserve their long-run growth dynamics when faced by the destabilising effects of major shocks. It borrows concepts from the regional economic resilience literature to characterise the different reactions of different places to recessions. In particular, economies are distinguished on the basis of their ability to resist to and recover from shocks thus maintaining stability around their counterfactuals, a notion known as ‘engineering resilience’, and to resume (or improve) their underlying growth trajectory by the end of the recessionary period under consideration thus showing ‘ecological resilience’ (or ‘positive hysteretic effects’). Related to these notions are the ideas of adaptability and ‘path dependence’, which help explain why some economies are more vulnerable to shocks than others (over and above the static causes of interregional heterogeneity incorporated in the model via random effects). Taking annual wage series for nineteen British towns over the historical period 1871-1906, I fit a spatial panel data model to 1871-1890 data by Spatial Two-Stage Least Square / Generalised Method of Moments (S2SLS/GMM), and use estimated coefficients in combination with trend forecasts to obtain counterfactual predictions of wage levels after the 1890 shock through to 1906. This allows to analyse how actual wages in different towns performed in relation to their counterfactual paths, and to gauge their relative resilience to economic shocks. The key finding, and the main lesson that can be drawn from the historical experience of British towns, is that the sectoral composition of local employment is important for economic resilience; my evidence suggests that excessive and increasing specialisation in declining industries means lack of the structural flexibility needed to replace these industries with competitive and productive activities (shock-proneness), whereas economies with a diversified industrial mix have greater scope for restructuring and renewal (shock-resilience); moreover, towns dominated by mature, staple sectors but who have also developed new growth industries are more able to adapt to and tolerate shocks. The second essay (Chapter 3) considers the relative success of alternative, non-nested wage equations from the perspective of Great Britain’s 408 unitary authority and local authority districts (UALADs) over the period 1999-2009. The negative relationship between wages and unemployment, embodied within the so-called Wage Curve, has an extensive literature and has been referred to as ‘an empirical law of economics’. However there are newer theories that seek to explain regional wage variations without reference to unemployment, namely Urban Economics (UE) and New Economic Geography (NEG). The aim is to discriminate between competing models of wage determination in order to establish whether the wage curve can be accepted as superior to its non-nested rivals. To do so I adopt an ‘Inclusive Regression’ approach (Davidson and MacKinnon, 1993; Hendry, 1995), combining the wage curve and either UE or NEG within an Artificial Nesting Model (ANM); this incorporates a spatial autoregressive process involving both the dependent variable and the error components and is estimated by S2SLS/GMM. The main conclusion is that, at least when the level of geographical resolution is relatively low as in this sample, while being validated empirically the wage curve should not be taken as an outright ‘law’ governing the spatial wage distribution. Specifically, when using asymptotic P-values, the wage curve is not dominated by either UE or NEG, in the sense that unemployment retains its predictive power in the presence of either employment density or market potential, nor is it capable of falsifying its rivals, as each of these is also statistically significant under the ANM. Meanwhile, when using bootstrap P-values, the wage curve emerges as the leading statement when directly confronted by NEG whereas UE offers a seemingly adequate hypothesis. The third essay (Chapter 4) is an extension of the previous chapter, and provides conclusive evidence by implementing a more formal and rigorous approach to testing a null model against a non-nested alternative, i.e. the J-test. This is a well-established technique for choosing among non-nested rivals, and in this chapter I develop a version of the test for specifications (SARAR-RE models) which feature spatially correlated error components, thus accounting for interregional heterogeneity via random effects (also subjected, like the disturbances, to a spatially autoregressive process), as well as a spatial lag of the dependent variable and additional, potentially endogenous regressors. This chapter thus makes a valuable addition to the literature on non-nested hypotheses testing in the spatial panel context by extending the toolkit to random-effects models. I also provide Monte Carlo evidence showing that there are distributional issues associated with the asymptotic use of the J-test in small-to-medium samples, so another novelty of this chapter is the implementation of a Bootstrap scheme to construct a valid null reference distribution in finite samples when the null and alternative are SARAR-RE models estimated by S2SLS / GMM. In terms of the empirical application, bootstrap J-test results confirm the bootstrap ANM results from the previous chapter that the wage curve rejects NEG theory while UE theory is equally successful. Another finding, from the methodological angle, is that the bootstrap J-test is a reliable and effective procedure for correcting asymptotic reference critical values and distinguishing between competing hypotheses in all cases where one is not a reduced form of the other. The fourth and final essay (Chapter 5) is one of few to reconsider from a spatial panel econometric perspective an economic relationship - the ‘empirical law of economics’ known as Okun’s Law - which has been traditionally considered at macro level with no attention for sub-national phenomena; it is the first to do so for Great Britain, looking at the 128 British NUTS3 regions over the period 1985-2011. By means of specialist techniques recently devised for spatial data, I show that regional interdependencies have a prominent role in the unemployment-output relationship; the total Okun’s Law effect itself is close to the ‘law’ of -0.30 but more than two thirds of this are accounted for by the impact on local unemployment rate of real output variations in areas nearby, a finding suggesting that policy intervention at both national and regional level on a country’s labour market can be more effective if spatial effects are factored into the analysis and modelled / tested explicitly.