RDP 2014-09: Predicting Dwelling Prices with Consideration of the Sales Mechanism 1. Introduction

The dramatic run-up in dwelling prices in many countries is generally accepted as playing a key role in the global financial crisis. Moreover, subsequent price falls have had large effects on economic activity and inflation. For these reasons, as well as more generally, policymakers are interested in understanding dwelling prices to help inform their views on the appropriate stance of monetary, fiscal and financial stability policies.

This paper asks whether the prices of dwellings traded under different sales mechanisms have different statistical properties and provide different information for understanding and forecasting dwelling prices. To answer this question, we investigate the time series properties of dwelling price indices in Sydney and Melbourne, which cover roughly 40 per cent of all Australian dwelling sales, and distinguish between the prices of dwellings transacted via bilateral negotiations (private-treaty sales) and the prices of dwellings that were auctioned.

There are several reasons why auction and private-treaty prices might provide different information and perform differently when forecasting future prices. First, auction prices could measure the common stochastic trend underlying all dwelling prices (hereafter the common trend) more precisely. One reason for this is that prices determined through auction have the potential to incorporate information (views about the value of a dwelling) from every bidder (hereafter buyer) that participates.

An example of such a case is when buyer valuations are correlated (or formally, affiliated), bids are publicly announced (known to other buyers), and bidding strategies are symmetric (the same across buyers).[1] With these assumptions, the English auction – the mechanism commonly used in Australia – yields a sale price that incorporates information from every buyer who actively makes a bid in the auction. This is because buyers use information contained in other participants' bids to help refine their own estimate of a dwelling's value.

In contrast, when prices are determined through a private-treaty sale, the number of views that have a role in determining the sale price for a dwelling is much smaller. In the case of a two-party bilateral negotiation between a buyer and seller, only information from those two parties may be directly incorporated into the price.

Price indices are, however, formed by averaging prices across a large set of transactions. In the data we have, there are seven (Melbourne) to ten (Sydney) times as many private-treaty transactions as there are auctions. This means that the disadvantage of fewer views being incorporated into each private-treaty price could be offset by having a larger set of views incorporated into the average price through more transactions. Whether average auction prices are a more precise measure of the common trend in dwelling prices is, therefore, an empirical question that we address in Sections 3 and 4.

A second reason that the sale mechanism could matter is that auctions and private-treaties weight buyers' and sellers' valuations differently. We argue that auctions are likely to place a relatively higher weight on buyers' valuations.[2] If buyers' and sellers' valuations evolve differently over time or the gap between the valuations of these two groups changes, this could be a second channel through which the type of sale is informative.

Indeed, a lag in the response of sellers' valuations to new information is consistent with a number of documented phenomena in the housing market including: that sellers' liquidity can be affected by changing prices (also known as equity lock-in: Stein (1995); Genesove and Mayer (1997)); that sellers may weight losses as compared to gains from a sale asymmetrically (for example, being more reluctant to incur a loss: Genesove and Mayer (2001)); and that sellers may have different information than buyers (Carrillo (2012); Genesove and Han (2012)). In contrast, buyers are less likely to respond with a lag to new information because they visit more properties and are less likely to be constrained by factors such as loss aversion or equity lock-in. Differences in the stickiness of valuations could result in differences in the autocorrelation of price growth from auction and private-treaty sales.

Using these ideas as our motivation, we investigate whether auction and private-treaty prices have different statistical properties and provide different information about dwelling prices and their forecasts. In particular, we investigate the extent to which alternative price measures are autocorrelated, can be used to predict one another, and can be used to predict average price growth overall. We also investigate whether the effects of shocks with differing degrees of persistence can be identified.

Using dwelling prices in Sydney and Melbourne, we find that auction prices can be used to predict both private-treaty prices and average dwelling prices (Section 3). For example, including lagged auction prices can reduce the one-quarter-ahead mean-squared forecasting error for average dwelling price growth by 10 and 18 per cent when compared with a simple forecasting benchmark, for Sydney and Melbourne respectively. In contrast, private-treaty prices are not useful for predicting auction price growth and have only limited predictive content for price growth in general. These results are quite remarkable, given that auctions are a small share of overall transactions.

We also find that auction prices have less autocorrelation in their changes (growth) than private-treaty prices, are less sensitive to transitory shocks, and converge more quickly to their new long-run or equilibrium value in response to a permanent shock (Section 4). As an example of the latter, at least 60 per cent of the adjustment to a permanent shock occurs within one quarter for auction prices, whereas less than 35 per cent of the adjustment occurs in private-treaty prices over the same time frame. In sum, our empirical results suggest that auction prices better reflect the common trend in all prices, incorporate new information more quickly, and are therefore more useful for forecasting.

We examine whether our results are consistent with two simple models of price determination (Section 5) – an English auction where buyers' valuations are linearly affiliated, and a bilateral Nash bargain. Our analysis points to a few core ideas that are required to link these two models with our empirical findings. First, even if an individual auction price incorporates more information (a larger set of valuations) than a private-treaty price, this by itself cannot account for the empirical results. In particular, once prices are averaged across a large set of transactions, any additional precision in the measurement of the common trend in price using auctions is unlikely to be large.

Second, a more plausible explanation of our findings is that there is a difference in the relative importance of buyers' and sellers' valuations across the two sale mechanisms. As discussed above, the theoretical models provide insight into why average auction prices weight buyer valuations more than average private-treaty prices. We further show that this can only explain the different autocorrelation properties observed in the data if seller valuations take time to respond to new information.

Third, we show that the average dispersion of sellers' valuations and the relative bargaining strength of buyers and sellers are important for the determination of private-treaty prices, but not auction prices. This is related to the differential weighting of buyers and sellers across the two price mechanisms. It is also reflects the fact that negotiation is crucial in bilateral trade, but less important for auctions.

Finally, although differences in the relative importance of buyers' and sellers' valuations, in the dispersion of valuations, and in relative bargaining strength are all important for price determination in the short run, they do not affect prices in the long run. In the long run, theory and the data suggest that both auction and private-treaty prices converge to a single common trend.

Bringing these ideas together, our results point to important differences in the short-term factors that drive changes in auction prices, compared with private-treaty prices. These differences are useful for identifying a common stochastic trend in dwelling prices, separating permanent shocks from transitory shocks, and improving forecasts of average price growth overall. Our results should prove useful both narrowly, in arguing for using auction prices (separately) in predicting near-term dwelling prices, and broadly, in demonstrating that the sale mechanism matters for price formation.


A model of affiliation is discussed in Section 5. In broad terms, affiliation and publically announced bids imply that if other buyers continue to actively make bids as the price of the dwelling rises during an auction, then each buyer upwardly revises their own assessment of the dwelling's value using the information contained in others' bids. [1]

In the absence of a reserve price, auctions will put all the weight on buyers' values. Even when reserve prices are used, we show that prices are still primarily determined by buyers when valuations are affiliated (Appendix D). In contrast, private-treaty prices typically reflect both buyers' and sellers' values, with weights equal to relative bargaining power for the Nash bargaining case. [2]