How do errors in the construction process of real estate indices bias returns of these published indices ?
||How do errors in the construction process of real estate indices bias returns of these published indices ?
||Book of Abstracts: 15th Annual European Real Estate Society Conference in Kraków, Poland
||At present numerous academic papers including Fisher, Geltner and Webb (1994) and Bond and Hwang (2005) leave little doubt that by institutions published real estate indices are biased. This concerns especially the calculated return volatilities of real estate indices, that seem to be downward biased. Also significant autocorrelation may be introduced by biases in returns. We reproduce the single steps of an index construction process. Therefore we simulate “true” return time series for individual properties which are latent in reality. Thereon we induce particular smoothing causes in the return series. Our contribution to the unsmoothing literature is to show the influence of these particular smoothing causes on the volatilities and the autocorrelation structure of individual property return series and index return series. Our key finding is that simulated biased index return series show similar autocorrelation structures as the NCREIF property index. Further our results show that the “stale appraisal”-effect, which is an anomaly of the NCREIF property index, is a crucial factor in smoothing. We discuss the four smoothing effects that presumably bias real estate returns: Appraisal-smoothing, non-synchronous appraisal of single properties, cross-sectional aggregation of property return series, and the stale-appraisal-effect. Our model of smoothing effects is adjusted to the US-American NCREIF Appreciation Property Index. Since it is thinkable to construct indices in different ways and in fact there exist different index construction methodologies, the impacts of biases on real estate return values depend on the index construction methodology. So our empirical results of the chain-linked smoothing effects can not directly carried forward on other real estate indices. For example, the stale-appraisal-effect is a particular characteristic of the quarterly NPI Index: Every quarter only a fraction of the properties in the NCREIF-index portfolio is re-valued. The values of the other properties are taken over from the respective past quarter. For this reason we do not only report our empirical findings for the chain-linked smoothing effects, but also for the individual smoothing effects. In our paper we review the analytical and empirical results of the real estate literature concerning the magnitudes of different smoothing parameters. This concerns for example the appraiser-alpha, a parameter which describes the fraction of the current property transaction values an appraiser will account for in his appraisal process. Further we offer a description of the technical details of the modelling of smoothing effects in real estate returns. Thereby we refer only partly on considerations which can be already found in the real estate literature. This concerns for example the just called appraisal-model of Quan and Quigley (1989, 1991) which we use to model the appraiser-behaviour. Also we employ new considerations, for example for the modelling of the non-synchronous appraisal process. We find that not only the volatilities of returns are downward biased but also the mean returns in a time series with only few quarterly observations. We suggest to account for this downward biased mean returns in the application of unsmoothing procedures and report our results for the modified Zero-Autocorrelation unsmoothing procedure, originally proposed by Fisher, Geltner and Webb (1994).
Post discussion ...
These pages are best viewed with any standards compliant browser (e.g. Mozilla).