The deferred update technique is a widely used approach for building replicated database systems. Its fame stems from the fact that read-only transactions can execute locally to any single database replica, providing good performance for workloads where transactions are mostly of this type. In this paper, we analyze the deferred update technique and show a number of characteristics and limitations common to any replication protocol based on it. Previous works on this replication method usually start from a protocol and then argue separately that it is based on the deferred update technique and satisfies serializability. Differently, ours starts from the abstract definition of a serializable database and gradually changes it into an abstract deferred update protocol. In doing that, we can formally characterize the deferred update technique and rigorously prove its properties. Moreover, our specification can be extended to create new protocols or used to prove existing ones correct.