“…These existing tests are roughly categorized into two types: time domain correlation-based tests and frequency domain 40 periodogram-based tests. The tests in the first category usually use the autocorrelations up to lag m (a user-chosen integer), so they are unable to detect serial correlations beyond lag m; see, e.g., Romano and Thombs (1996), Lobato (2001), and Horowitz, Lobato, Nankervis, and Savin (2006) for observable series, or Francq, Roy, and Zakoïan (2005) and Delgado and Velasco (2011) for residual series. To avoid selecting m, Escanciano and Lobato (2009) and Escanciano,45 Lobato, and Zhu (2013) derived a data-driven portmanteau test under the assumption that ε t is a martingale difference sequence (MDS).…”