Trade-offs between Asset Location and Proximity to Home: Evidence from REIT Property Sell-offs

To list a few: Ambrose, Ehrlich, and Hughes (2000), Gyourko and Nelling (1996), Hartzell et al. (2014), Campbell et al. (2003), Cronqvist, Hogfeldt, and Nilsson (2001), Ling, Naranjo, and Scheick (2018, 2019), Zhu and Milcheva (2018), Zhu and Lizieri (2019) and Milcheva et al. (2020).

REITs operate within a single asset class (because 75% of a REIT’s assets and income must come from real estate–related assets), follow regulated dividend payout policies (because they are required to pay out 90% of taxable income as dividends), feature high levels of institutional ownership (see Chan et al. 2003), and have similar antitakeover provisions (because of the 5/50 rule and excess share provision).

Riddiough et al. (2005) propose that adjusting the location differences might be important for reconciling the differences between private and public real estate returns. This is because the NCREIF index is biased toward larger assets located in first-tier markets while REITs hold a large percentage of their assets in lower-tier markets. Ling, Naranjo, and Scheick (2018) suggest that the lower risk profile of larger MSAs reflects the constraints that developers face in adding new supply. In other words, land supply elasticities (Saiz 2010) could be positively correlated with ex-ante required rates of returns. Gateway markets have relatively inelastic supplies and experience lower ex-ante risk premiums than non-gateway markets.

The FTSE NAREIT US Real Estate index contains all Equity REITs except those designated as Timber REITs or Infrastructure REITs. Updated annually, the list starts in 1993, which is deemed a symbolic year at the beginning of the modern REIT era (Feng et al. 2011), and runs until the present.

Although we are able to identify 1,943 firm-year observations of sell-offs based on S&P Global, in our event studies we rely on hand-collected sell-off events following Campbell et al. (2006) instead of using S&P Global for several reasons. First, while we are able to track changes in property portfolios, we do not know if a reduction in a portfolio is a real sell-off event because the purposes of these changes are not stated. For example, some portfolio reductions happen in cases of property exchanges, mergers, or acquisitions. Second, S&P Global includes asset sale dates but not announcement dates. Using the transaction (completion) dates from S&P Global for an event study is problematic because some transactions were announced several months before they were completed. For example, Highwoods sold 39 office properties in a transaction in 2005. The S&P Global transaction date for this transaction is July 22, while the announcement date is June 6. Third, the S&P Global database does not identify major sell-off events that are economically meaningful enough to influence trading. Small transactions convey little economic significance to firms. Lastly, as stated in Campbell et al. (2006), when examining CARs, it is important to control for deal-level information, which requires hand-coding from news searches.

Specifically, the arc length, d_ij, for each underlying property j sold (or held) by firm i is defined as: \( {d}_{ij}=\operatorname{arccos}\left({\mathit{\deg}}_{latlon}\right)\times \frac{2\pi r}{360}, \) where deg_latlon = cos(lat_i) × cos(lon_i) × cos(lat_j) × cos(lon_j) + cos(lat_i) × sin(lon_i) × cos(lat_j) × sin(lon_j) + sin(lat_i) × sin(lat_j). Lat and lon are property and headquarters latitudes and longitudes provided by S&P Global Real Estate Database, and r is the radius of the earth (≈3,959 miles).

The use of adjusted cost or book value in place of unobservable true market values may understate the (value-weighted) percentage of a REIT’s portfolio invested in MSAs that have recently experienced relatively high rates of price appreciation. Conversely, its use may overstate the percentage of a REIT’s portfolio in MSAs that have experienced relatively low rates of price appreciation.

Capozza and Seguin (1999) suggest that Property-level Operating Efficiency can be expressed as the sum of FFO, general administration costs, and interest expenses, all divided by total assets.

It is noted that the tendency among firms to hold nearby properties and dispose of distant properties does not imply that the average distance from the underlying properties to headquarters declines over time. As our paper focuses on dispositions, we do not examine acquisitions. It is possible that the average distance (in miles) between a firm’s underlying assets and its headquarters remains stable (or even increases) through mergers, acquisitions, and property exchanges.

In unreported results, there were 68 unique sellers (defined by their CRSP PERMNOs), of which 32 appeared only once while 17 appeared more than three times.

In our sample, the breakdown by property type is qualitatively similar to that deployed in Campbell et al. (2006), who examine equity REIT property sell-offs between 1992 and 2002. The breakdown by the use of sale proceeds is different, though, from that reported in Campbell et al. (2006), as the largest group in our more recent sample use sales proceeds to acquire funds.

In addition, we control for property age because, on average, older properties are more costly to run.

In fact, most non-core property types outperform core types in terms of returns and NOI growth. See the following Seeking Alpha link: https://​seekingalpha.​com/​article/​4276161-core-vs-non-core-reits-much-ado-nothing.

Here, the treatment sample is constructed based on sell-off events identified in Factiva news searches and differs from the sell-off sample based on the S&P Global full sample discussed in Section 3.1.

The reported results are based on a matched sample of treated firms that actually sold off properties and control firms with similar sell-off propensities. The control group (firms without sell-offs) should be similar to the treated group (i.e. firms with sell-offs) to estimate what would have happened to the treated group if it had not received the treatment (i.e. sell-offs). In other words, finding that CARs are significantly positive across all the models suggests that we failed to match the sell-off firms with the control firms for which sell-offs did not take place. Also, we do not include the gateway variable because it is used as one of the covariates in matching.

If the sell-off consists of multiple buyers/properties, we use the median distance to define DB and DS.

This exercise is performed based on a sell-off sample for the 2003 through 2013 period, which consists of 154 sell-off events. We retrieve buyer information from CoStar or the news articles. The sample size is further reduced to 56 because of missing buyer information. We are not able to include hedonic characteristics because in most of the cases, the characteristics of individual properties and their transaction prices were not disclosed.

In a related exercise, we test whether anchoring to the buyer’s home market price level explains the results. Following Ling, Naranjo, and Petrova (2018) and Liu et al. (2015), and using an approach similar to the approach we used to construct DB and DS, we construct a dummy variable indicating whether the buyer (seller) is from an expensive market, defined as the median price in the buyer’s (seller’s) home market that is above the sample median. By taking the difference between the two dummies and controlling for the sum, we find that the coefficients are never statistically significant. Results are not tabulated and but are available upon request.

We are not able to apply three tiers because of insufficiently many sell-off events in the secondary markets.