Date post: | 14-Dec-2015 |
Category: |
Documents |
Upload: | chester-simpson |
View: | 214 times |
Download: | 1 times |
Does Property Transactions Matter in Price Discovery in
Real Estate Market: Evidence from the US firm level data
William Cheung and James LeiUniversity of Macau, Macau ChinaERES 2014 Bucharest University of
Economics
Motivations• The results of public market (REITs stock) V.S. private real
estate market are mixed. - Hoesli et al. (2013) find that public real estate market leads
the private real estate market. - Yavas et al. (2011) find that there are variations across firms
within each property type. For any given property category, REIT returns could be leading NAV returns for some firms while NAV returns could be leading REIT returns for some other firms.
- Tuluca et al. (2000) find that private market seems to informationally lead the public one.
• Ross (1987) defined a market as efficient if there is a lack of arbitrage opportunity. Therefore, private real estate market makes itself as a compelling case for efficiency because of illiquidity.
• Duffie, Malamud, and Manso (2010) find that private information sharing promotes the effect of public information sharing.
Main Findings• Significant contributions to price discovery
from the private markets.• Price discovery from the private markets
increase further relative to the public real estate market, when employing transaction windows, as compared to full samples.
• Impulse response analysis shows that private real estate market converges even faster than public market real estate market around transaction windows.
• The results are robust to length of transaction windows and property types.
Our Uniqueness• A unique dataset of daily property transactions
covering 01/02/2001 to 12/31/2013.• Synchronized public and private pairs around
transaction windows, not by regular calendar days as in the earlier studies in the literature.
• Estimate long-run relation between public and private real estate markets with respect to information generated by property transactions in the underlying spot market.
• Unique environment of property markets and transaction data allow us to provide empirical evidences on private and public information sharing.
Contributions• We provide a new angle to test the relative
contributions to price discovery between public and private real estate markets: the comparison between full samples and transaction windows.
• Transaction windows matter because either the appraisal-based or transaction-based values of the underlying properties should react to the new information of property transactions and incorporate it into new values.
• Though public real estate market dominates the price discovery with respect to private real estate market, as stated in literatures, we fill the gap that private real estate market can become informative when transaction windows are taken into account.
Data
• Source: SNL financial• Full samples: from 02 January 2001
to 31 December 2013Full samples:
Property type
Diversified Office Hotel Industrial
Number of firms
24 19 26 8
Sample Size
106,801 69,661 46,762 21,245
Number of Transactions 1,116 1,316 717 633
Data• Transaction windows:
– lead_lag 25 days, based on each transaction date t, we include [ t-25, t+25 ] observations• Example 1: there was a property transaction on 04/25/2013 of
Starwood Hotels & Resorts Worldwide. To construct the transaction window of lead_lag 25 days, based on 04/25/2013, we include the observations of [ t-25, t+25 ]. Therefore, the transaction window will be from 03/20/2013 to 05/30/2013, only weekdays included.
– lead_lag 30 days, based on each transaction date t, we include [ t-30, t+30 ] observations
– transaction_date_lag 5 days & lead_lag 25 days, based on each transaction date t, we set t-7 as each new transaction date, denoted as t7, and include [t7 -25, t7 +25] observations
• Example 2: there was a property transaction on 04/25/2013 of Starwood Hotels & Resorts Worldwide. To construct the transaction window of transaction_date_lag 5 days & lead_lag 25 days, we first lag the transaction date back to 5 days which should be 04/18/2013. Then, based on 04/18/2013, we include the observations of [t7 -25, t7
+25]. Therefore, the transaction window will be from 03/13/2013 to 05/23/2013, only weekdays included.
DataProperty
typeDiversified Office Hotel Industrial
Number of firms
24 18 21 8
Sample Size
26,162 32,727 21,179 11,210
Number of Transactions 1,116 1,316 717 633
Lead_lag 25 days
Lead_lag 30 daysProperty
typeDiversified Office Hotel Industrial
Number of firms
24 18 21 8
Sample Size
28,199 35,207 23,321 11,935
Number of Transactions 1,116 1,316 717 633Property
typeDiversified Office Hotel Industrial
Number of firms
24 18 21 8
Sample Size
26,170 32,717 21,202 11,188
Number of Transactions 1,116 1,316 717 633
Transaction_date_lag 5 days & lead_lag 25 days
Methodology• Vector Error Correction Model (VECM)
where Total_return and NAV are the change of total return index and net asset value (NAV) in period t, respectively, Z = Total_return bNAV is the long-term relationship between total return index and NAV, and are i.i.d. innovations.
Methodology• Gonzalo and Granger ratios (common factor loadings)
• Gonzalo and Granger's (1995) price discovery focus on the error correction process. The model estimates the common factor weights that reflect the permanent contribution to the common factor (efficient price). The common factor weights are derived from each market's error correction coefficients.
• Superior price discovery is attributed to the market with the higher GG ratio.
Tables of GG Ratios
Property type
Full sample Lead_lag_25_days
Total_return
NAVTotal_ret
urnNAV
Diversified
29% 71% 18% 82%
Office 16% 84% 11% 89%Hotel 28% 72% 19% 81%
Industrial 39% 61% 6% 94%
GG ratios between full samples and lead_lag 25 days
Tables of GG Ratios
Property type
Full sample Lead_lag_30_days
Total_return
NAVTotal_ret
urnNAV
Diversified
29% 71% 7% 93%
Office 16% 84% 25% 75%Hotel 28% 72% 23% 77%
Industrial 39% 61% 18% 82%
GG ratios between full samples and lead_lag 30 days
Tables of GG Ratios
Property type
Full sampleTransaction_date_lag
_5 days & Lead_lag_25 days
Total_return
NAVTotal_ret
urnNAV
Diversified
29% 71% 20% 80%
Office 16% 84% 11.5% 88.5%Hotel 28% 72% 22% 78%
Industrial 39% 61% 28% 72%
GG ratios between full samples and transaction_date_lag 5 days & lead_lag 25 days
• The GG ratios (common factor loadings) of private real estate market increase further relative to public real estate market, when considering transaction windows.
• The reaction of NAV to shocks of the three transaction windows converges faster than that to shocks of full samples• The slopes of the dashed lines are
steeper than those of solid lines• The distance between two dashed
lines becomes narrower than solid lines
Conclusions
• Consistent with Oikarinen et al. (2011), Hoesli et al. (2012), we find that public and private real estate market exhibit long-term cointegrating relationship
• We also find that public and private real estate market exhibit long-term cointegrating relationship with samples of transaction windows
• We test the relative contributions to price discovery between public and private real estate markets around transaction windows and find that the information content in the real estate market increases further, as compared with that of full samples. Private real estate market does matter in price discovery around transaction windows
Robustness – Normalized Co-integrating Vector
Property
type
Full
samples
Lead_lag
25 days
Lead_lag
30 days
Transactio
n_date_lag_
7 days &
lead_lag 25
days
Diversifie
d0.143 0.130 0.130 0.130
Office 0.111 0.091 0.091 0.096
Hotel 0.136 0.130 0.130 0.130
Industrial 0.127 0.127 0.127 0.127
Comparison of the normalized cointegrating vector between full samples and transaction windows. The normalized cointegrating vector of transaction windows show the robustness.