A1378
Title: Testing linearity in functional cointegrating regression
Authors: Xing Wang - Durham University Business School (United Kingdom) [presenting]
Abstract: Linearity testing in functional cointegrating regression is studied when the long-run conditional mean may be an unknown function. The null hypothesis specifies that the cointegrating relation is linear, while the alternative allows a flexible nonlinear link without committing to a particular parametric form. The proposed test is constructed to respect the uneven information generated by integrated regressors, so that inference is directed toward the recurrent part of the sample path where the data provide reliable evidence about the long-run relation. The resulting framework provides a practical diagnostic for empirical cointegration analysis: it helps distinguish a genuinely linear equilibrium relation from nonlinear long-run responses that may otherwise appear as persistent residual behavior, unstable adjustment, or poor fit of a linear specification. This distinction is relevant in applications where long-run elasticities, policy multipliers, or equilibrium pricing relations are interpreted from persistent economic and financial time series. The approach therefore provides a specification check before relying on linear cointegrating regressions for estimation, forecasting, or counterfactual analysis.
