Can a transaction-based approach track commercial property rental growth? Malcolm Frodsham and Steven Devaney explore

The trend growth in rents and the volatility of that growth are key elements of the expected return and risk from investing in commercial real estate. To measure historical trends and cycles in rents, rental growth series are constructed using periodic estimates of rental values produced during the portfolio valuation process. This type of series is not a direct measure of the change in rents on held property. This casts doubt on the reliability of such series as measures of the long-term trend and volatility in rents.

Market rent, or rental value, is produced in estimating capital value for properties in an investment portfolio. Market rent is defined by the Royal Institution of Chartered Surveyors as “the estimated amount for which an interest in real property should be leased on the valuation date between a willing lessor and a willing lessee on appropriate lease terms in an arm’s length transaction, after proper marketing and where the parties had acted knowledgeably, prudently and without compulsion”.

Changes in estimates of market rent can be tracked through time to produce an index. In the long term, it is reasonable to expect that the trend level of rental value growth will match the growth in rents themselves. However, capital value estimates are known to be subject to valuation smoothing, caused by anchoring on previous assessments of value. It is likely that rental value observations are one source of this smoothing and that rents are more volatile than rental values. There might also be client influence on valuations – rental values are an input that could be subject to such influence.

To estimate the degree of smoothing in existing rental value series and provide evidence on whether there are variations in turning points, we have constructed rent indices for different sectors and segments of the UK commercial property market using a sample of rent transactions. The results will have implications for modelling and valuing cash flows by investors and lenders. The rent series will also inform operational portfolio managers or occupiers looking to benchmark rent change on their portfolio with that of an index.

Transaction-based rent index
Transaction-based indices were developed for purchase/sale transactions, with many early examples relating to residential property. A key limitation of all transaction-based indices is their ability to track prices when there is a fall in sale volumes, usually in a weak market. However, rent observations are more common than sale transactions, and a certain degree of evidence is likely in all market conditions as leases expire.

It is recognised that a simple average of rents or prices in a given period is not an adequate basis for an index, as this does not control for variations in the quality of what is transacted. Therefore, two statistical approaches have been developed:

Repeat measures regression: this uses pairs of transactions recorded for the same asset (so, for instance, the change from the purchase price to the sale price). It assumes no significant changes in the attributes of the asset between events;

Hedonic regression model: this uses individual transactions including data on the attributes of what was transacted. The model can identify price movements through time for an asset with specified attributes (for example, change in price per square metre for a four-bed house).

A repeat measures approach is taken in this paper, using the change from one rent setting event to the next. A hedonic approach was not explored, owing to poor-quality data on building attributes. For example, there are significant definitional issues in relation to unit floorspace data – treatments of ancillary space, and so on.

The approach constructs an index from irregularly timed transactions by apportioning change in rent across the years between the events. To accurately identify the points where market conditions are shifting, many overlapping rent-change observations are required. A regression then finds the line of best fit that explains the variation in rent movements across the sample of evidence.

While the method assumes no significant changes to an asset between events, changes due to ageing are reflected. This means the effects of depreciation are incorporated in the measurement.

Many idiosyncratic rent movements are to be expected. Even in valuation data, there is a wide range in rental value movements on individual properties that is not evident from a published index. In a small sample, the effect of these idiosyncratic movements is to create gyrations (a sawtooth effect) in the transaction-based index. We use an outlier filtering process to mitigate this. We compute an annualised rental growth rate for each pair of rents in the dataset and we remove the 1% of cases with highest increases and the 1% of cases with the largest falls.

The effect of idiosyncratic movements in rents is amplified by weighting the observations. Therefore, an unweighted approach is likely to produce more robust indices. Yet published rental value series are constructed on a weighted basis. If there is a difference in trends between larger and smaller units, then the growth rate of a rental value index and an unweighted transaction-based series will diverge. Therefore, we also test repeat measures models that weight the observations according to the amount of rent paid.

The dataset we have used is sourced from the RES Lease Consortium. It comprises snapshots of rents and lease details such as commencement date, break date, rent review date and expiry date from valuation records of around 50 contributing funds. The contributors were a mixture of life insurance funds, pension funds, specialist funds and unitised funds.

The dataset in 2018 covered a rent roll of over £600m (€675m) in over 5,000 units with 80% of the rent roll and 94% of the units in either retail, office or industrial premises.

There is limited data on building characteristics such as age and condition, so we rely on the use of large samples to identify whether the typical change in rent differs according to the rent setting event. We measure the average change and the variation in rent changes for each type of rent setting event.

Care has been taken to exclude redevelopments, but minor refurbishment between leases, especially for offices, are retained in the sample. This treatment mirrors that of published rental value growth series (a degree of capital or improvement expenditure occurs in the assets tracked by these series).

Leases of less than one year have been excluded. These cases often relate to temporary lets over the Christmas period and subsidised lets (possibly associated with rate saving).

Headline vs effective rent value
An issue with rents recorded for new lettings or lease renewals is that there might be lease incentives underpinning the letting, the nature of which will vary between lettings. The economic impact of these incentives can be quantified in the form of an effective rent. This allows different types of rent event to be compared more easily – for example, a change from a rent review to a new letting where the former will not include incentives, but the latter could feature a rent-free period.

All recorded rents were converted to effective rents using a simple reduction in the headline rent proportionate to the length of the rent-free period within the overall term of the lease – measured to the earlier of the break or lease expiry date. This allowed us to compare trends in headline versus effective rent through time, a facility not offered by rental value indices.

The data on break incentives and penalties was so patchy that any data that was available has been ignored in these calculations. Data on capital incentives were also not available, so the effective rent change series may underestimate rent volatility.

Transaction Based Risk Index

Rent-change pairs
6,081 rent-change pairs from 2000-18 were identified. The mean length of time between rent events was 4.77 years, while the median length was five years.

As delays in rent settlement can be significant (for example, tenants can ‘hold over’ and rent reviews can be left unsettled for some time), data on the settled rent often did not emerge until a year or two after the end of the lease event, but it is attributed here to date of the event. Similarly, with new lettings, the rent might be negotiated in advance of the lease commencement date, but the rent is attributed here to the recorded lease start date.

Rent reviews in the UK are market-linked, but they are also usually upwards-only. This means that some rent review dates in the dataset are ignored in this exercise because they are not associated with a rent change. Where rents at review remain at the same level as for the preceding event, we cannot discount that the market has, in fact, fallen. Therefore, such instances are not used to form rent pairs, but a rent pair might span such a review using an event occurring before it and an event occurring afterwards.

Index results for individual years in the middle of the study period benefit from observations that start, span and end at the year concerned. Results for the later years will not have the benefit of observations where the second event occurred after 2018. This may bias the results if, for example, rent changes differ for poorer-quality units as their leases tend to be shorter and so contribute data more quickly as new rents are observed. As data on higher-quality units with longer leases becomes available in subsequent years, the shape of the data series may change.

In the results, we only present index results to the end of 2017.

Regression model
A simple repeat measures regression involves regressing observed changes in rent (the dependent variable) on to dummy variables that indicate the times of the first and second transaction in each case (the independent variables). Although our models are more complex, this summarises the essence of the model used.

There might be problems using rent review outcomes to form observations of rent change when the rent agreed is an estimate of the current market rent and not the outcome of an open-market negotiation. Any imperfections in the rent review process would therefore be reflected in the estimated index. They might also bias the transaction-based index upwards as there is a lack of corresponding observations where rent would have fallen were it not for the clause.

All the regression models we used include control variables for rent reviews that both tested and adjusted for any bias in index outcomes. This was preferable to excluding rent reviews from the exercise, as these still provide information on how rent is changing through time. In some models we include additional control variables to identify lease renewals. This allows us to test whether the rent movements produced by new lettings differ from renewals.

The transaction-based index results have been calculated on a headline rent and an effective rent basis, and also on an equal-weighted and a value-weighted basis. The results are compared with the MSCI rental value growth index.

The transaction-based index constructed from effective rents exhibit the same trend growth rate, 1.4% annually, as one constructed from headline rents.

Movements in the index based on effective rents exhibit greater volatility, 5% versus 4.6%, with a greater fall in the downswing after the 2008 financial crisis and more upside afterwards. This suggests that incentives tend to rise in a downswing and fall in stronger occupier market conditions.

Transaction-based index, equally The results for the equally weighted transaction-based index based on effective rents exhibit slightly higher trend growth that the MSCI rental growth index, 1.5% annually versus 1.1% annually, which could reflect the specific start and end dates used for the exercise.

The transaction-based index is more volatile than the MSCI rental growth index, 4.9% versus 3.1% annually, with a greater fall in the downswing after the 2008 financial crisis and more upside afterwards.

There is no obvious difference in turning points, which is counter to expectations. However, as noted, there are potential distortions from assuming that the rent change is attributable to the recorded event date. For example, a new lease is likely to be negotiated in advance of the lease start date, while a rent review might be settled in arrears.

If control variables are added isolating rent movements associated with new lettings, a similar overall picture emerges, but it can be seen that the turning points in the index are somewhat earlier. The index peaks in 2006, ahead of the MSCI index and our main transaction-based index. It bottoms out in 2010. 

Individual unit rent-change observations are subject to a high degree of idiosyncratic variation. The transaction-based methodology produces unstable results if based on a small number of cases with contradictory rent movements. This behaviour is amplified by weighting the results, particularly if there are large units with unusual rent movements. This effect can be seen in the more jagged results from the value-weighted index.

In summary, rent change estimates using a transaction-based approach exhibit similar average growth rates and a similar overall shape to the MSCI rental value growth series. This suggests that the approach taken to produce the transaction-based rent index is robust.

Changes in the transaction-based index are more volatile than those shown by a comparable rental value growth index, with stronger peak-to-trough movements, but not much difference on turning points. This supports the hypothesis that rental values are smoothed over time, with volatility increasing from 3.1% to 4.9% over the analysis period.

While there are some limitations to the transaction-based series, it is worth noting that the MSCI rental value growth index is constructed from a mix of headline and effective rental value estimates. It is not known how this composition changes through time and what effects this has on the index movements.

The transaction indices are work-in-progress, and more advanced modelling techniques to deal with deal timing, value weighting and higher frequency estimation are being pursued.

Malcom Frodsham is director at Real Estate Strategies, and Steven Devaney is associate professor at University of Reading