Testing for ARCH in the presence of additive outliers

摘要: In this paper we investigate the properties of Lagrange Multiplier (LM) test for autoregressive conditional heteroskedasticity (ARCH) and generalized ARCH (GARCH) in presence additive outliers (AO's). We show analytically that both asymptotic size power are adversely affected if AO's neglected: rejects null hypothesis homoskedasticity too often when it is fact true, while has difficulty detecting genuine GARCH effects. Several Monte Carlo experiments these phenomena occur small samples as well. design implement a robust test, which better than conventional AO's. Applications to French industrial production series weekly returns Spanish peseta/US dollar exchange rate reveal that, sometimes, apparent effects may be due only number and, conversely, can masked by outliers.