2.3 Main Real-Estate Indices Worldwide
2.3.1 THE INVESTMENT PROPERTY DATA INDEX
The Investment Property Databank (IPD) is a London-based entity that pro- vides information about commercial real-estate in the United Kingdom and worldwide. This commercial real-estate index is constructed on directly owned investments in completed and available for rent property assets held in invest- ment portfolios. Assets that were not fully owned or available for utilization during the period of measurement are excluded. The costs of trading, manage- ment fees, taxes, the impact of debt and cash are also excluded.
The IPD index family is now part of the MSCI group. The IPD family of indices are appraisal-based indices. The IPD UK Annual Property Index covered approximately 21,175 directly held UK property investments with a market value close to £152.7 billion as of December 2013. This index can be tracked back to December 1980, and, for a much smaller sample of assets, to December 1970. Reflecting the cumulative value of roughly 70% of the property assets held by UK institutions, trusts, partnerships and listed property companies, and almost 50% of the total professionally managed UK property investment market, the composition of the index is spread unevenly across various sectors of the real-estate market. Approximately 46.8% of the prop- erties (4,156 properties worth £71 billion) are retail properties, 26.5% office properties (2,663 properties worth £40 billion), 15.4% industrial properties (2,850 properties worth £23.5 billion), 3.8% residential compound proper- ties (9,127 properties worth £5.75 billion), and 7.5% (2379 properties worth
£11.5 billion).
Indices from the IPD family are value-weighted measures, this means that each asset contributes to the index according to its monetary weight. The investment performance used to assess the performance of the index which is the total return, calculated with the formula:
TRt= CVt−CVt−1−CEt+CRt+NIt
CVt−1+CEt−1 ×100 (2.1)
whereTRtis the total return in montht,CVtis the capital value at the end of montht,CEtis the total capital expenditure in montht,CRtis the total capital receipts in monthtandNItis the day-dated rent receivable in montht, net of property management costs, ground rent and other irrecoverable expenditure.
The total return index value starts with a base value of 100 and is calculated by multiplying the previous index value using the formula
It+1=It
1+TRt+1 100
(2.2) whereTRt+1is the total return in montht+1.
There is also an IPD UK Monthly Property Index started in 1986 that has a reduced representation given by a portfolio of 3,479 properties worth around
£43.3 billion at year-end 2014, covering about 40%-50% of the market. The IPD UK Monthly index is employed for marking to market positions on deriva- tives taken on IPD UK Annual Property Index. The latest IPD UK index is a quarterly index, IPD UK Quarterly Property Index tracking the performance of a portfolio of 9,712 properties worth £134.5 billion as of September 2014.
This index covers about 45%–55% of the market and the historical data goes back to 2000.
There is an entire family of IPD indices, all of which are exclusively deter- mined by the open market appraised valuations of actual buildings by property professionals. All properties that are directly owned by the organizations pro- viding appraisals to IPD are included; however, properties held indirectly through investment vehicles are excluded, as well as any bonds, cash or deriva- tive holdings.
In Figure 2.1 we show the tests for serial correlation for the returns on the IPD UK Monthly index. As with other real-estate indices analysed below, there is an alternation of serial correlation, positive short term followed by negative autocorrelation which appears after year three.
Figure 2.1 displays an upward trend up to the end of 2006 combined with a significant market crash in 2008 in the level series and positive serial corre- lation short term followed by negative serial correlation longer term. Serial correlation, also referred to as autocorrelation, is a very important feature of time-series data. In its presence, standard statistical models may produce unreliable estimators and standard statistical tests may become biased. In other words, if there is serial correlation but it is ignored, the data analysis con- clusions may be wrong. Typical examples are those of spurious correlations and spurious regression whereby an analyst finds statistically significant results when in fact the opposite is true.
Since real-estate property lacks granularity and the average time period to complete a real-estate transaction is three months, it is understandable that real-estate prices are sticky and one may expect a high degree of serial
0 200 400 600 800 1000 1200
IPD Monthly UK All Index Dec 86 Jun 88 Dec 89 Jun 91 Dec 92 Jun 94 Dec 95 Jun 97 Dec 98 Jun 00 Dec 01 Jun 03 Dec 04 Jun 06 Dec 07 Jun 09 Dec 10 Jun 12 Dec 13
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01 0 0.01 0.02 0.03 0.04
Returns Dec 86 Jun 88 Dec 89 Jun 91 Dec 92 Jun 94 Dec 95 Jun 97 Dec 98 Jun 00 Dec 01 Jun 03 Dec 04 Jun 06 Dec 07 Jun 09 Dec 10 Jun 12 Dec 13
0 20 40 60
−0.5 0 0.5 1
lag
Sample Autocorrelation
0 20 40 60
−0.5 0 0.5 1
lag
Sample Partial Autocorrelation
Figure 2.1. The IPD UK All Property Total Return Monthly Price Index between December 1986 and December 2014.
Notes: The IPD UK All Property Total Return Monthly Price Index level and percentage returns for monthly series between December 1986 and December 2014. The lower graphs show the ACF and PACF plots. The upper graphs show the index levels and percentage returns for the monthly series.
The lower graphs show the sample autocorrelation and partial autocorrelation functions plots over the same time period.
correlation. The sample autocorrelation function (ACF) and partial autocor- relation function (PACF) are statistical measures that can be used to indicate the existence of autocorrelation at individual lags. It is important to test for autocorrelation at various time lags to see how long it takes for this effect to disappear. The ACF and PACF plots help in realizing after how many lags the information on those real-estate indices is not relevant for the current index value and also see if the relationship with past values is one of positive or negative correlation. Fortunately, there are statistical tests that can be used to
test for the presence of serial correlation. The most commonly used statistical test is the Ljung-Box test.3The application of this test to the historical returns in Figure 2.1 using 6, 12, 30, 48, and 96 monthly lags clearly indicates that there is autocorrelation present even after eight years. The Q statistic for each lag and the corresponding critical value for determining if the computed Q value is statistical significant shown in parentheses is as follows: 6 months, 1064.5 (12.6), 12 months, 1245.7 (21.0), 30 months, 1470.6 (43.8), 48 months, 1643.3 (65.2), and 96 months, 1674.0 (119.9). Since the computed Q values are less than the corresponding critical value, we conclude that the observed serial correlation exhibited in Figure 2.1 is statistically significant. It is noteworthy that between January 1993 and September 2007,all IPD monthly returns were positive, representing one of the longest bull periods observed in an asset class, as pointed out by Tunaru (2013).
The IPDETRAI Monthly index represents the monthly estimates of the annual return of the IPD All Property Return. The values for this index for the period January 2005 to January 2014 are shown in Figure 2.2. Notice that the level of the index declined after the crisis of 2007–2008 to almost the initial level of 2005. The autocorrelation plots are very similar to the autocorrelation plots of the IPD Monthly UK All Return Index in Figure 2.1. Applying the Ljung-Box Q test to the monthly returns for this index indicates the existence of serial correlation at 6, 12, 30, 48, and 96 monthly lags. The Q statistic for each lag and the corresponding critical value for determining if the computed Q value is statistically significant shown in parentheses is as follows: 6 months, 299.00 (12.6), 12 months, 327.27 (21.0), 30 months, 504.11 (43.8), 48 months, 565.13 (65.2), and 96 months, 693.87 (119.9). Since the computed Q values are less than the corresponding critical value, we conclude that the observed serial correlation exhibited in Figure 2.2 is statistically significant.
From the autocorrelation graph in Figure 2.2 one can see that for the IPDE- TRAI monthly index the serial correlation is strongly positive for short-term periods of up to about one year, then negative for medium-term periods of length two years, and switching again to positive correlation for longer- term periods such as after five years. This type of behaviour of the real-estate index will pose some theoretical problems when we seek to develop pricing models for derivatives contingent on this index. For example, assuming a geometric Brownian motion for the continuous-time dynamics of such an index will be in flagrant contradiction with the stylized features revealed in this chapter.
3 The Ljung-Box Q-test is helpful to quantitatively test the existence of autocorrelation at multi- ple lags. The null hypothesis for this test is that the first d autocorrelations are jointly zero; that is, H0:ρ1=ρ2=. . .=ρd=0.
1000 1200 1400 1600 1800 2000 2200
IPDETRAI Index
Jan 05Jan 06Jan 07Jan 08Jan 09Jan 10Jan 11Jan 12Jan 13Jan 14
−0.06
−0.04
−0.02 0 0.02 0.04 0.06
Returns
Jan 05Jan 06Jan 07Jan 08Jan 09Jan 10Jan 11Jan 12Jan 13Jan 14
0 20 40 60
−0.5 0 0.5 1
lag
Sample Autocorrelation
0 20 40 60
−0.5 0 0.5 1
lag
Sample Partial Autocorrelation
Figure 2.2. The IPDETRAI Monthly All Return Price Index level and percentage returns for monthly series between January 2005 and December 2014.
Notes: The lower graphs show the ACF and PACF plots. The upper graphs show the index levels and percentage returns for monthly series between January 2005 and December 2014. The lower graphs show the sample autocorrelation and partial autocorrelation functions plots over the same time period.