Abstract
We examine the possibilities and boundaries of the sample-based valuation of residential real estate portfolios using upstream principal component and cluster analysis. We seek to increase estimation accuracy without increasing the total sample size. We use a model portfolio of about 2,400 properties in Germany and discuss different stratification methods prior to drawing fairly small samples, extrapolating the portfolio value, and calculating bootstrap confidence intervals to estimate the associated precision. The models described in practice-oriented literature are thus refined using a theoretical statistical process.