Controlling for the Impact of Variable Liquidity in Commercial Real Estate Price Indices

Article excerpt

Liquidity in private asset markets is notoriously variable overtime. Therefore, indices of changes in market value that are based on asset transaction prices will systematically reflect intertemporal differences in the ease of selling a property. We define and develop a concept of 'constant-liquidity value" in the context of a model that is characterized by pro-cyclical volume of trading. We then present an econometric model that allows for estimation of both a standard transaction-based price index and a constant-liquidity index. Our application to the NCREIF database reveals that, in the case of institutional commercial real estate investment, constant-liquidity values tend to lead transaction-based and appraisal-based indices in time, and also to display greater volatility and cycle amplitude. The differences can be significant for strategic investment policy viewed from a mean-variance portfolio optimization perspective.


Measuring and monitoring changes in investment values is fundamental to understanding any investment asset class, including those traded in private markets. This problem has received particular attention in the private real estate investment industry, where there has long been a recognized need to compare real estate risk and return to that of other asset classes, such as publicly traded bonds and stocks (including REITs). Yet, such measurement of private asset market price changes or capital returns faces serious problems, both conceptual and empirical.

The basic problem is the difficulty of measuring market value movements in an environment where whole, heterogeneous assets are traded infrequently and irregularly over time, typically between a single selling party and a single buying party. Individual asset sale prices provide asynchronous, idiosyncratic, and noisy indications of market value. Another potential problem is posed by the fact that typically only a fraction of all the assets in the market population transact during any given period, and those that transact may not be a random sample of the population. This nonrandomness may cause sample selection bias in empirical analyses. A third fundamental problem is posed by the phenomenon that private asset markets typically display highly variable liquidity over time. During "up" markets, capital flows into the sector, there is much greater volume of trading, and it is much easier to sell assets. Just the opposite typically occurs in "down" markets. This intertemporal variation in the ease of selling an asset affects the interpretation of transaction prices. An important implication is that transaction-based price indices do not hold constant the liquidity in the market.

The first two of these problems have been addressed in the search, real estate, and statistics literature, and to some extent more recently in the financial economics literature. Econometric procedures for estimating transaction price-based indices of periodic market value changes have been developed and honed over the past several decades, including the hedonic value model developed by Rosen (1974), and the repeat-sales regression pioneered by Bailey, Muth and Nourse (1963). These procedures allow the estimation of a periodic market value change index from noisy, asynchronous, heterogeneous asset prices.1 The problem of identifying and correcting sample selection bias has been addressed in general by Heckman (1979), and, more recently, applied specifically to real estate markets in several studies, including Gatzlaff and Haurin (1997, 1998) and Munneke and Slade (2000, 2001).2 While the present paper will include these solutions, our primary focus is on the third problem of private asset market price indices noted above, that of understanding the impact of variable liquidity on the observable transaction price data. Although the phenomenon of pro-cyclical variable liquidity in private asset markets has been widely noted in the practitioner and trade literature, the only previous attempts that we know of in the academic literature to quantitatively control for this problem in the construction of market value indices have been so-called "de-lagging," or reverse filter, procedures that have been applied to appraisal-based indices of commercial property values. …