Commercial Property Price Indices – meeting your needs?

Commercial Property Price Indices – meeting your needs?

Article Type: Education briefing From: Journal of Property Investment & Finance, Volume 30, Issue 4

Introduction

Commercial Property Price Indices (CPPI) play an important role in the property market. They enable changes in prices, market trends and turning points to be measured over time. This is of considerable benefit to investors, valuers and policy makers in the absence of a central market place in which property is traded. Investors can compare the returns from their investments with the performance of the market as a whole. They can examine the volatility and risk of particular types of property, the correlations between them, and assess the merits of diversification. As no-one has personal knowledge of more than a fraction of the transactions taking place, CPPI can enable valuers to compare their experience with broader market trends. CPPI are also important for economic policy makers because of the role of real estate in fixed capital formation and the exposure of the financial system to property markets.

Most major markets now have at least one CPPI and some have several competing alternatives. Recently the RICS (2012, p. 7) has urged caution in their use by valuers.

Indices can be a guide to general trends in the market against which the performance of the property being valued can be judged. As such they are a potentially useful reference point when forming a judgment, though they cannot necessarily assist when markets are inactive or transactions are scarce. Thus, indices should be treated with caution, and valuers should have a clear understanding of the sources and reliability of the data from which the index has been derived.

The principal problem when using CPPI is that different indices from time to time produce different answers. These can be differences in the extent of peaks and troughs, the timing of market turning points, rates of change in prices, and even the direction of the trends themselves. The problems this can cause users is encapsulated in a May 2011 blog by Professor Richard K Green of the University of Southern California (Green, 2011). Green posed the question as to whether commercial property values in USA at that time were rising or falling given the contradictory evidence from two leading indices. The Moody’s/REAL Monthly CPPI reached a trough in April 2011 and in May 2011 values were 46 per cent below their peak, although by then they had started to rise. By contrast, Green Street’s CPPI showed the bottom of the cycle as having occurred in 2009 and values in 2011 were only 13 per cent below their peak. The implication was that whilst Green Street’s index suggested that recovery was well underway, the Moody’s/REAL index suggested that it was only just beginning and that negative equity was likely to be a serious problem for many investors. With such contrasting views of a market, the obvious question is which is correct? The answer is both are accurate within the limits of their respective methodologies. The indices differ in the data used and the ways in which they are constructed. Users therefore need to be aware of the differences in the ways in which CPPI are compiled and must be careful to select the ones that are most appropriate to their needs.

Users do not have identical requirements, for example, with some seeking information on trends in prices, others rental trends, and yet others total returns from income and capital growth. The difficulties in compiling CPPI and publishing them on a regular basis mean that few users are likely to produce bespoke CPPI that exactly meet their requirements. Rather, most users will rely upon available CPPI and will therefore have to form their own judgement about their suitability.

The article discusses the main issues raised by in the compilation of CPPI. It illustrates them by reference to some of the leading CPPI available in the UK and the USA. In the UK Investment Property Databank (IPD) and in the USA the National Council of Real Estate Investment Fiduciaries (NCREIF) use appraisals of their contributors’ properties. Moody’s/REAL CPPI and CoStar (both USA) use transactions data to produce repeat sales indices. The transactions-based index produced by MIT/CRE uses sales data for the NCREIF database. Many other organisations are capable of producing CPPI providing they have access to a portfolio of properties that can be regularly revalued or to a sufficient number of transactions. The key issues are the methodologies employed and the representativeness of the sample of properties used. There is a growing academic literature containing demonstrations of how CPPI can be improved, though these do not necessarily lead to the production of CPPI on a regular basis.

CPPI – some fundamental issues

A Property Price Index is a measure of the average change in prices. Many leading CPPI provide sub-indices in recognition that there can be variability from the market average for a sector, like retail or offices, or for a region or metropolitan area. The fundamental problem any price index has to solve is to ensure that the average price is measured on a constant quality basis. If the average price rises between two time periods, does this mean that prices have risen or that proportionately more of the more valuable properties were sold in the second time period? For this reason a simple average or the median is unreliable since they can be distorted by changes in the composition of transactions.

CPPI ensure that the mix of properties between two time periods remains constant so that any change in the weighted average price reflects a change in the market price rather than in the composition of the transactions. The Laspeyres index uses weights from the base time period but weights need to be updated periodically to reflect changes in investor and developer preferences. Alternatives are the Paasche index, which uses current weights and the Marshall-Edgeworth and Fisher indices, which use both current and past weights. In a chain index, two consecutive time periods are used to compute the weighted change in prices with the weights being updated for each successive pair of comparisons. In each pair, the earlier date is given the base value of 100 and the later index is calculated relative to this. The chain is produced by multiplying the index relative to the base period for period 1 by the index for period 2 relative to period 1.

In a mix adjusted CPPI properties are divided into groups or cells according to defined characteristics, such as property type and location, with the index being a weighted average of the cells. The more cells there are, the lower the potential variability within each cell and the lower the risk of average price changes being the result of a change in composition of the properties sold. However, more cells mean that there is a greater chance of there being no transaction in an individual cell during a given time period and having to impute a value, which reduces the accuracy of the index. Where the sample of properties is a sufficiently high proportion of the population, weighting may not be necessary as the composition of the index should be similar to that of the underlying population.

There are a number of alternatives to the mix adjustment approach. A repeat sales index measures the change in price between two sales of a property. As the same property is sold on each occasion, providing that there is careful filtering to exclude properties that have been redeveloped, the characteristics of the property between each sale should be similar so that the change in price reflects changes in the market price. The hedonic approach implies that the price of a property depends on a range of quantitative and qualitative characteristics. The approach seeks to find an underlying model using multiple regression techniques to determine the effect on price of these characteristics and to use these as the weights.

Weights can be by volume or by market value. The two approaches do not give the same answer. Expenditure weighting gives higher growth when proportionally more of the higher value properties are sold, whereas equal value weighting is likely to lower returns in periods of growth and inflate them in periods of slump, perhaps because smaller properties are more stable in value (Fisher, 1994). Transactions-based indices, like CoStar’s Commercial Repeat Sales Index, tend to uses equal weighting. Their indices are divided into institutional and general commercial segments because of the diversity of commercial property and due to different segments attract different investors with diverse investment goals (CoStar, n.d.). The appraisal-based indices, IPD and NCREIF, produce value weighted indices (IPD, 2011; NCREIF, n.d.). The rationale behind value-weighting is that an index measures the complete population. Consequently, a few expensive sales do not exert a disproportionate influence. IPD reports that it covers 58 per cent of the UK’s professionally managed property investment market, though it cannot include portfolios for which there is no end of calendar year valuation, such as some listed property companies. NCREIF’s membership is restricted to tax-exempt institutions and claims that it covers 60 per cent of the funds managed by its contributing members though there are some important populations of commercial investment properties that it does not cover (Geltner, 2011a, p. 4).

Seasonality affects house price indices which raises the question as to whether CPPI ought to adjust for seasonal trends. Annual comparisons are not affected by seasonal issues, though quarterly and monthly indices could be. Geltner (2011b) has suggested that there is a “February” effect in US transactions-based CPPI as this month reflects “renegade” or “opportunistic” deals done in December and early January rather than “strategic” ones, and there is recent evidence of more distressed deals being completed in February.

CPPI make use of the measuring rod of money but money does not have a constant value due to inflation. Generally CPPI do not report real inflation-adjusted changes. Those indices that cover a number of countries, such as IPD’s Pan-European Index, have to contend with the issue of fluctuating exchange rates potentially distorting the relative performance of different countries. IPD overcomes this by collecting data in the fund’s local reporting currency and converting it into common currencies at the end of the month. When reported in local currencies, the index is calculated by converting all markets to the same currency at a fixed rate over time in order to remove the impact of exchange rate movements.

Ideally a CPPI should be based on actual market prices. The problem in the commercial property market is that properties are traded infrequently so that only a relatively small proportion of the stock is traded in any given time period. Repeat sales and transactions-based indices do use actual transactions but an alternative approach, adopted by IPD and NCREIF, is to use valuation data. The use of such data raises the question as to how good a proxy valuation data is for market prices. A third possibility, not discussed in this article, is to make use of price data for shares in Real Estate Investment Trusts (REITs). As the owners of a REIT are the owners of the property, and if the shares are traded in an efficient market, then the prices should reflect traders’ knowledge and expectations about the property market.

Appraisal-based CPPI

Appraisal-based CPPI developed in the 1970s as means of producing a market index for property comparable to those for equities. They may owe something to the obligations many countries impose on certain groups of investors regularly to revalue the assets they hold (Devaney and Martinez Diaz, 2011). In the UK IPD was started in 1985 with data back to 1980 and, for a more limited sample, to 1970. It has grown to include indices for 17 European countries and a further seven worldwide, though the focus in this article is on the UK indices. In the USA, NCREIF began to produce an index in the mid-1970s, though the data collected before 1988 has been discarded due to the limited sample.

The IPD and NCREIF Indices are based upon the properties in the portfolios of their contributors. The implied assumption is that their share of the respective markets is sufficiently large that they can no longer be considered to be a sample but reflect the underlying population of properties. As discussed earlier in the UK IPD estimates that its sample includes 57.9 per cent of the total professionally managed property investment market by value though in some other markets, it is much lower, for example, 17.1 per cent in Germany (IPD, 2011). NCREIF reports that the tax-exempt institutions, on whose behalf the data contributing members report, have about 60 per cent of the assets managed by the contributing members. Once a certain threshold is reached, the addition of new funds is unlikely to change significantly the composition of the sample and require a restatement of the historic index. IPD under these circumstances freezes the historical index, as has happened with the annual indices for France and The Netherlands and the UK Annual, Quarterly, and Monthly Indices. NCREIF froze its index from the first quarter of 2003. IPD uses the term consultative indices where the sample of properties is likely to grow over time.

An appraisal-based CPPI has to define precisely the types of property for which it will accept valuations and which valuations it will accept. Both IPD and NCREIF are based upon on market appraisals for investment purposes of real buildings by property professionals. They use internal and external valuations, providing that the valuations meet the definition of market value, though NCREIF requires an external valuation every three years. Both include only completed and lettable properties and exclude occupational or predominantly owner occupied properties. They exclude properties purchased, sold or in the course of development during the measurement period, or have been subject to damage so as to avoid abnormal profits or losses generated by active management and changes in value due to non-market factors. IPD includes only directly owned properties and so excludes those held indirectly through an investment vehicle. Properties are included only if the whole portfolio is reported to IPD in its entirety, which prevents contributors being able to pick which properties on which to return data. NCREIF includes properties in joint ventures. Considerable detail is collected by IPD and NCREIF about each property and the lease conditions that go beyond the requirements for compiling an index. The data is checked for responses outside of the specified ranges, missing information and exceptional performance. Data from contributors is subject to random audit and there is external oversight of the quality assurance process. The nature of the contributors to both indices means that the properties tend to be of the type that institutions invest in. The trends they report may not reflect what is happening to other types of property.

The reliance of valuations raises the question of the extent to which these can be regarded as a proxy for prices. One of the main criticisms of appraisal-based indices are that they understate the true volatility in the market and are slow to recognise turning points because of the smoothing effect of valuations (Geltner, 2011a). Valuations tend to be bunched in certain times of the year. For example, IPD note that 36 per cent of their Australian properties and 70 per cent of their New Zealand ones were valued in the third quarter of 2011 (IPD, 2011), whilst NCREIF has a bunching of its reappraisals in the fourth quarter (Geltner and Goetzmann, 2000). This can present problems for an index that reports at intervals more frequent than once a year, particularly if appraisals are rolled over each quarter between annual reappraisals. It can mean that what purports to be a quarterly index is, in effect, a rolling annual index with some of the properties being revalued at different times during the year rather than having regular intra-year revisions. The effect is likely to be smoothing of the index and lags in the recognition of turning points because “stale” valuations are used, a criticism that has been levied at NCREIF (Fisher et al., 1994; Geltner and Goetzmann, 2000; Geltner, 2011a). Econometric methods have been developed to estimate what the unsmoothed indices should look like (Fisher et al., 1993; Cho et al., 2003). IPD responds to this challenge by using samples of properties for its quarterly and monthly indices which are revalued during the relevant period. The samples are smaller than for the annual index – the monthly index has 14 per cent and the quarterly index 47 per cent of the professional managed property market. The differences between samples could result in some incompatibility with the annual index.

For both IPD and NCREIF transaction based indices have been developed that make use of the sales within their set of properties (Geltner, 2011a; Devaney and Martinez Diaz, 2011). These show higher volatility and less autocorrelation than the appraisal-based ones, which is consistent with the appraisal-based indices smoothing trends. Devaney and Martinez Diaz (2011), though, found that the transaction based index derived from the IPD data set was not significantly different in its turning points from IPD’s indices, which may reflect how IPD produces its quarterly and monthly indices.

Valuations should be objective but could be influenced by clients. Clients have the motivation to influence valuations, for example, to reduce the value of funds when these are expected to fall so that investors are discouraged from withdrawing investments. They have the opportunity to do so through draft valuation meetings. This could be tacit rather than overt, though, for example, the choice of valuer, the provision of information or the instructions given (Crosby et al., 2010). Appraisal-based CPPI try to ensure that the valuations come from a cross-section of valuers by limiting the proportion of an index that can be contributed by a valuation firm, for IPD to 25 per cent of an index. Consolidation amongst valuation practices has reduced the number of firms so that the leading firms collectively undertake a high proportion of the valuations (Crosby et al., 2010).

The debate about the accuracy of valuations dates back to the 1980s when Brown (1985) argued that there was a high correlation between valuations and subsequent sales and between the valuations undertaken for different firms. Since 2003 IPD has undertaken a series of annual studies for the RICS into the accuracy of valuations. The most recent report used data from 2010 that was more limited than in previous reports because recession had reduced the number of properties sold (Shamsan, 2012). This also meant a more challenging environment for valuers with less comparable data being available. The weighted absolute difference between valuation and sales price increased from 11.4 per cent in 2009 to 12.5 per cent in 2010. It was found that 57 per cent of valuations were within ±10 per cent and 81.5 per cent were within ±20 per cent. The unweighted differences show a slight skew towards sale prices being greater than valuations, something which is suggested in previous reports. This might indicate that caution amongst valuers could be a factor behind smoothing in appraisal-based indices.

Transaction-based CPPI

In contrast to the appraisal-based CPPI, transaction-based CPPI make use of actual sales prices. Two methods are in use, repeat sales indices and hedonic indices, although there are other possible approaches. The repeat sales method has been used to create residential property price indices since the 1960s (Bailey et al., 1963). Although many of the repeat sales indices are for residential property, commercial property indices using this method have been created in the USA, including Moodys/REAL Commercial Property Index, produced by the MIT Centre for Real Estate (Geltner and Pollakowski, 2007) and CoStar’s (CoStar, n.d., 2010). Pioneering proof of concept work on commercial repeat sales indices was undertaken by Gatzlaff and Geltner (1998) using data from Florida’s Department of Revenue property tax records.

Repeat sales indices use the sale prices of properties from within a data set that have been sold more than once during a given period of time in order to estimate price changes over time using a regression equation. The method should solve the problem of constant quality as the price changes are for an individual property which is assumed to have the same characteristics at each sale, providing there have been no quality changes between sales. Geltner and Pollakowski (2007, p. 5) describe a repeat sale index as a “same-property-change index”. Repeat sales indices make limited demands for data other than price, when the sale occurred, and the address of the property. As additional properties are resold and contribute to the computation, the index for earlier time periods needs to be re-estimated and updated. If the index is used for derivatives trading, such as the Moodys/REAL Commercial Property Index, there can be no backward adjustment as this would undermine the basis of the contracts (Geltner and Pollakowski, 2007). Indices that are not adjusted provide less accurate historic data about market trends as they exclude information about price trends derived from second sales made after the index has been frozen. Geltner and Pollakowski (2007) argue that the exclusion means that the Moodys/REAL index reflects the price changes implied by realised investments as indicated by contemporaneous second sales (round-trip investment price change returns), but it does mean excluding the information content of price changes from other properties which is only revealed once they have sold for a second time, something they recognise is valuable for academic and historic purposes, though not for derivative trading.

The data set could be the population of properties if the data set is taken from a mandatory land registration system. Otherwise, the data set will be a sample of the population and will be subject to any potential biases resulting from how the data has been compiled. For example, the real capital analytics (RCA) data set used in the Moodys/REAL CPPI initially included properties with a sales price of $5 million or above, though since 2005 it has been extended to include ones above $2.5 million. Properties can only enter the data set with a first sale once they have risen in value above $2.5 million. A first sale during the time period of less than $2.5 million is not taken into account. CoStar reports that high value retail properties are under-represented in its database. These tend to be in multi-asset portfolios and it will only accept portfolio sales into its index if the whole portfolio is sold with no change in the inventory. In spite of the differences in data sets, CoStar has estimated the correlation between its index and the Moodys/REAL one at 0.96.

The sample of properties on which a repeat sale index is built is likely to be a relatively small proportion of the stock due to the limited number of sales each year. For example, in June 2010 CoStar’s database consisted of 1.3 million commercial property transactions, but included only 100,000 repeat sales since 1996 of which 85,428 pairs were in the indices. These included 17,323 retail, 13,781 offices, and 13,498 industrial pairs (CoStar, 2010). This does not prevent the construction of sub-indices for different types of property and regions, though it does limit the ability to produce regional indices for particular types of property. In periods of recession the number of sales on which to base an index declines. Repeat sales indices tend to deal with the problem of limited observations in two ways. One is through rules about the minimum number of observations needed to produce a sub-index. The other is through the use of ridge regression. This has the effect of dampening down the impact of colinearity in scarce data within an ordinary least squares regression model.

Each resale is treated as being an independent observation, though it is possible that the characteristics of properties that are resold more frequently could introduce error (Eurostat, 2011, p. 6). For example, better quality properties might be more saleable during a recession (Geltner and Pollakowski, 2007, p. 3). Adjustments may need to be introduced to correct for heteroskedasticity where the assumption of a constant variance of errors does not apply (Case and Shiller, 1989). The greater the time period to resale, the greater is the potential heteroskedasticity because of the risk of depreciation as well as improvements. The age of the property changes between sales and this could influence the transaction price (Case et al., 1991). The depreciation of commercial properties is potentially a more significant problem than for residential properties. Quality changes over time can be overcome by weighting the repeat sales observations by a declining function according to the time between sales.

Data cleaning is necessary to exclude properties for which there is risk of quality changes or for whom the prices may not be market ones. For example, the Moodys/REAL Commercial Property Index filters include: discarding data for properties that have been held for less than 1.5 years (and which may be subject to being “flipped”); portfolio transactions; first sales from before 1988; those with incomplete information; sales which are not at arms’ length, foreclosures and condemnations; and those where there has not been consistent usage, or where there have been major changes in size (though the rental area can deviate by 10 per cent), or where the year in which the property was built is not earlier than the first sale (indicating that they have been redeveloped) (Geltner and Pollakowski, 2007). Certain types of property are inevitably excluded, such as newly built properties which have to enter the data set before a repeat sale can be recorded.

The period of time over which a sale takes place between the property being first offered for sale, through the agreement of draft heads of terms, due diligence, the completion of the contract and transfers of moneys, and the registration of the change of ownership can be a lengthy one. Crosby and McAllister (2004) in the UK found that the median time from the decision to sell to completion of the contract was 190 days, which excludes the time taken for registration. Transactions-based indices harvest their data at different points in this cycle. This means that they can give different representations of the market. If data is harvested before deals have been finalised, the data set can include transactions that are subsequently aborted or for which the price is renegotiated. This is part of the difference in indices noted by Green (2011) in his blog discussed earlier; one index used prices from deals during the course of being finalised and the other final sales prices. When in the transactions cycle the data is harvested has an impact upon the timeliness of the index. There is a trade off between using more timely data that may be subject to revision, and using final prices that relate to deals struck some time earlier.

There are currently no commercial repeat sales indices in the UK. IPD has investigated the use of transactions-based indices but concluded that despite some encouraging results, they are less convincing than appraisal-based indices. The discretion over trading policies that fund managers enjoy can result in understatement of market downturns (IPD, 2011). CoStar is establishing a database of properties in UK, initially focussing on central London, Birmingham, Manchester and Liverpool, with the target of extending coverage to the whole of the UK. The London field research involved visiting every building in the City, West End, Midtown and Docklands areas, undertaking geo-coding, and ascertaining the size of 25,500 discrete commercial buildings (www.costat.co.uk/en.Research). This is facilitated by publicly available tax data on the lettable areas of commercial properties and by compulsory registration of sales and new leases of more than seven years.

An alternative approach to the use of sales data is the construction of a hedonic index. The hedonic approach argues that the price of a property depends on a range of quantitative and qualitative characteristics. The heterogeneous nature of commercial property makes producing a hedonic model very complex. However, recent valuations, providing that these are not too close to and influenced by sales, could be treated as a “catch-all” composite hedonic variable that could implicitly contain data relevant to a hedonic model that would otherwise be difficult to observe or measure (Geltner, 2011a). Since 2005 MIT has produced pro bono a transactions-based index using sales price and valuation data from NCREIF to complement the NCREIF indices (Fisher et al., 2007) and there has been a similar experiment using IPD data (Devaney and Martinez Diaz, 2011). Data availability means that such series tend to be limited in the sub-indices that can be produced, hence their role being seen as complementary to appraisal-based indices.

Conclusions

CPPI have developed considerably since the 1970s and most major markets are covered by them. They provide valuable information on trends within commercial real estate markets. Different methods can be used in their compilation and this can result in differences in the trends they report. The principle differences concern the type of data used in their compilation. The limited amount of data on transactions mean that some of the leading CPPI are appraisal-based, though there are also repeat sales and hedonic indices that use transactions data. The qualifications about appraisal-based indices are that they appear to give smoothed trends and draw on data from portfolios that may be representative of only part of the market, namely institutional grade properties. For transactions-based indices the principal problems are the more limited data available and when in the transactions cycle they harvest their data. For these reasons one might regard the different types of CPPI as complementary where CPPI of different methodologies exist in the same market. Ultimately it is the responsibility of the users to satisfy themselves that their use of the CPPI is appropriate to meet their needs.

Richard Grover and Christine Grover