Methodology for the sequence analysis of building stocks

A new methodology is presented for analyzing longitudinal building data by considering building histories as sequences of states or events. This allows for the application of sequence analysis methods to any kind of building histories. A demonstration of this methodology is applied to two example datasets from a random sample of a stock of vanished buildings based on the records of a German building insurance company over a period of 56 years. In this sample, the diversity of the distribution of states increases with a slight fall near the end of the time period. Non-residential buildings remain longer in a given state than residential buildings, and private ownership is more stable than other owner types. The survival rate for buildings that undergo a change of their function is less predictable than for those without a change of function. The predictability of the states of buildings without ownership change has a greater variation than that of buildings with owner change. A clustering of building histories into groups of similar patterns can be used to calculate the probability of survival for a given building.