![Sample our Geography journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5937/TOC-Subject-Sample-2021-Geography.jpg"})
Abstract
Investment theory dictates that capitalisation (cap) rates for freehold real estate should be determined by the risk-free nominal rate of return plus the risk premium (RP) less the expected growth rate, with an allowance for depreciation. However, importing the concept of the RP from the capital markets fails to guide investors through the complexities of the asset, or enable exploration of purchaser preferences and behaviour. A refined pricing model for real estate is proposed, based on a concept termed a risk scale, to distinguish between macro (market) and micro (stock) determinants of risk and growth within the RP. This pricing model is estimated for a major global investment market, using a cross-sectional inter-temporal framework, with a data-set of 497 transactions in the London office sector over 2010 Q2–2012 Q3. Average cap rates are estimated at just over 5%, with asset-specific attributes dominating yield determination, with submarket quality and tenant covenant most important; and unexpired term insignificant, surprising during the ‘flight to safety’ characterising the period. International investors bought at lower cap rates, despite the ongoing economic and financial instability of the study period. Improving understanding of pricing behaviour and market transparency is important and may be advanced through the pricing model.
Acknowledgements
We acknowledge the support of the Research Trust of the RICS who provided the funding to enable this project to take place and the help of a number of second year students from the University of Reading who helped to collate and refine the Co-star data used in this project. We also acknowledge the help and support of Co-star in providing us with advanced notification and download of their data-set inclusive of their building quality indicators, prior to general release and of CBRE who provided data from their Rent and Yield Monitor on London submarket prime rental values.
Notes
1. The Investment Property Forum (IPF) quarterly survey of advisors and managers was taken over by the Association of Real Estate Funds in 2013 and is now available to members of AREF.
2. It is worth noting that this is significantly before the 2016 UK Referendum on EU membership was mooted in the 2015 Conservative election manifesto.
3. CoStar actually reports net initial yield which is purely the rent passing divided by the transaction price plus any purchaser’s costs. Vacancies are therefore ignored causing properties with high vacancies to have low initial yields.
4. The net Equivalent Yield is the IRR of the expected cash flow with the outlay being the transaction price plus purchaser’s costs.
5. REML is one of two possible estimation techniques employed in most multi-level programmes and selects the model parameter values that maximise the likelihood function that is calculated from a set of data that has been transformed to focus on the parameters of interest. This transformation is achieved by removing the effect of the fixed variables.
6. This average is the common average across all the submarkets allowing for between and within-submarket variation and the bias generated by between submarket variations.
7. This is estimated as 1 − (.0285/(.0285 + .1672)).
8. Calculated as (.0285/(.0285 + .1672)).
9. For example, if the rent in a submarket is £20 per square foot above the average central London rental value and that difference grows by 10% to £22, assuming all else remains unchanged, the cap rate will fall by 3.6%; i.e. from 5.08% to 4.9%.
10. Selection between alternative non-nested multi-level models can be made using goodness-of-fit statistics that are relative estimates of the information lost when a given model is used on a given set of data. The AIC, AICC, CAIC and BIC are variants of a goodness-of-fit test that use a likelihood function with either a penalty for the number of estimation parameters included in the model (AIC and BIC) or sample size (AICC and CAIC). The model with the lowest AIC, AICC, CAIC and BIC are preferred (Bozdogan, Citation2000).
11. The Wald test is a parametric statistical test that in this instance is used to test the significance of the null hypothesis that the difference between the estimated sample variance and the true variance is equal to zero. If the test rejects the null hypothesis, then a statistical difference exists and this is assumed to be due to the model and its variance not capturing the true variance. The level of confidence in rejecting the null hypothesis is denoted by *** for the 1% confidence level; ** for the 5% confidence level and * for the 10% confidence level.
12. The influence of other submarket measures (absolute and grand mean-centred submarket vacancy rates and actual rental growth, adjusted for inflation) in explaining cap rates were examined. None of these results are reported in the paper as they were insignificant and failed to improve explanatory power.
13. These relatively more complex estimations are referred to as growth models.
14. One concern was that there is not a sufficient number of observations in each time period across 13 submarkets; this was therefore reviewed across alternative submarket definitions (first the three City, West End and Mid-Town submarkets and, second, the merging of contiguous submarkets into seven groupings down from the original 13). In each of these iterations, the time fixed effects and random effects remained insignificant, while the variation between submarkets became blurred and insignificant. The results are not, therefore, reported here.
15. Other random effects, tested in our mixed effects estimations as random slope effects, were absolute and grand mean-centred submarket vacancy rates and actual rental growth, adjusted for inflation and grand centred submarket rents, adjusted for inflation. The addition of these variables as random effects did not improve explanatory power. The results are not reported in this paper.
16. More complex covariance structures were investigated but they failed to significantly improve goodness of fit.
17. This finalised estimation was derived in a stepwise process from Estimation 6, retaining the key risk drivers that influence cap rates as specified in the conceptual framework.
18. For example, in the UK around the end of the millennium, falling inflation expectations caused a fall in medium dated Government bond rates from around a nominal 8% to 5% giving a significant increase in total returns within the bond markets. Reducing risk-free rates were set against reducing nominal rental growth expectations causing little change in property cap rates over the same period and therefore lower total returns compared to bond markets.
19. Tests show on the Level 1 residuals appear to follow a normal distribution. The residual histogram fits a normal distribution reasonably with a Skewness statistic of .321 although the Kurtosis statistic is a little high at 2.279.
20. These are contiguous submarkets.