Biased Estimates of Housing Price Developments

Abstract

This thesis intends to answer the question: Do house price indices based on the repeat-sales method possess sample selection bias? This is done by comparing different indices, where the repeat-sales index yields the highest appreciation rates for freeholder dwellings in Oslo in the period 2009-2020. Based on different methods we find evidence of a sample selection bias in the repeat-sales index. Furthermore, this bias is volatile, oscillates around 0.24%, and is classified as white noise. The most reasonable explanation for the bias is the wasteful use of transaction-data, leading to non-representative samples. The upwards bias aligns with the hypotheses of opportunistic buyers, home-improvements, downpayment and “starter homes”. Predicting a second sale proves difficult given available data, but investigation of attributes important for a second sale show that smaller apartments tend to sell twice in our given period. However, from an economical significance perspective this bias is not material.

This thesis intends to answer the question: Do house price indices based on the repeat-sales method possess sample selection bias? This is done by comparing different indices, where the repeat-sales index yields the highest appreciation rates for freeholder dwellings in Oslo in the period 2009-2020. Based on different methods we find evidence of a sample selection bias in the repeat-sales index. Furthermore, this bias is volatile, oscillates around 0.24%, and is classified as white noise. The most reasonable explanation for the bias is the wasteful use of transaction-data, leading to non-representative samples. The upwards bias aligns with the hypotheses of opportunistic buyers, home-improvements, downpayment and “starter homes”. Predicting a second sale proves difficult given available data, but investigation of attributes important for a second sale show that smaller apartments tend to sell twice in our given period. However, from an economical significance perspective this bias is not material.