Track-oriented multiple-hypothesis tracking (MHT) is an effective paradigm for multi-target tracking. Much of the research emphasis in MHT in recent years has been on effective hypothesis management in order to contend with a potentially large number of competing track hypotheses. Here, we identify track hypotheses that are generally not considered in MHT implementations. These hypotheses include a target birth event in the absence of a sensor measurement. We show that inclusion of these hypotheses leads to improved maximum a posteriori (MAP) estimation results. In the time-invariant case, the targets that are unobserved at birth remain unobserved: we refer to these as ghost targets. In the general case, the MAP solution includes some targets unobserved at birth and subsequently detected.