An Empirical Method for Decomposing the Contributions of Land and Building Values to Housing Value

Abstract

This paper develops an empirical method that uses two separate housing related components to estimate housing value: land and building. The artificial neural network (ANN) technique is used to iteratively solve for two hedonic models simultaneously by minimizing the difference in the observed total value and the sum of the estimated land and building values. This method enables one to objectively separate housing value into land and building components. Using actual sales transaction data from Taipei City, we estimate the land value as a share of the total housing value. The results show that the land value accounts for a higher share with older properties. The share of the land value of low-rise buildings tends to be higher than that of high-rise buildings. The share of the land value can deviate by 20 percentage points between more or less expensive housing communities within Taipei City.