Forecasting Tenure Durations From Survey Responses On Current Spells

In various survey data, the tenants report the spell from past entry to the current date. Although that gives useful information about the mobility pattern, it does not provide any immediate forecast how long the tenants will stay in the future or eventually remain in probability. Information of that sort would be important for planning processes, but the data signal only the interoccurrence time from which we can estimate the backwards recurrence time, what underestimates the total stay. Inferring about the latter needs strong assumptions on the nature of the underlying stochastic processes, discussed by Feller, Cox, Karlin and Taylor, Bhat and Miller, Prakasa Rao and others. By means of simulations we first illustrate the size of underestimation. Then we use the simulations as a benchmark to estimate and forecast the total spells from Austrian census data.