Construction of real estate price indices for developing housing markets: Does temporal aggregation matter?

1. Introduction

In Ghana, like many other developing contexts, access to reliable and adequate data on housing and housing-related transaction is largely difficult, and this situation has considerable implications for developing house price indices. House price indices are, however, critical in informing policy and investment decisions in the real estate industry. Well-established models for constructing housing indices have over the years evolved, although mainly from developed economies where relevant data, both in terms of quality and frequency abound. The importance of house price indices, however, requires that despite the acute dearth of information in developing countries, there is the need to explore for models that respond to peculiar contextual dynamics.

Houses are transacted continuously and on a daily basis. However, houses that are typically sold in a day are very few and infrequent (Englund et al., 1999), as compared to other financial assets such as stocks. As a result, studies that rely on transaction data often find it necessary to pool data together across time so that the sparse data set problem could be overcome and also to preserve degree of freedom. The aggregation of time, however, involves an implicit assumption that indices generated from broader aggregated samples are statistically the same as those generated from less aggregated constituent sub-samples (Owusu-Ansah, 2013). For example, when housing transactions are combined to construct a quarterly index, it is assumed that a house transaction in January has occurred at the same time as one in March. Similarly, a house transaction in January is assumed to have occurred at the same time as ones in June and December when constructing semi-annual and annual indices, respectively. To estimate a quarterly index therefore, the monthly coefficients are restricted to be equal within the quarter. That is, the quarterly price function is assumed to remain constant through January, February and March and then jumps to a different value for April, May and June (Owusu-Ansah, 2013). Because demand and supply relationship in the housing markets vary over time, this assumption is questionable. Temporal aggregation may therefore bring about bias in the construction of house price indices and returns.

Even though broader level of data aggregation may not be an effective solution to the small sample size problem, at the finest levels of...

$\begin{array}{l}\\ \end{array}$ $\begin{array}{l}\\ \end{array}$ $\begin{array}{l}\\ \end{array}$ $$ $$ $$