Modelling the asymmetric behaviour of property yields: evidence from the UK office market

Abstract

This paper examines the determinants of UK office market yields and their relative importance depending on overall monetary and financial conditions, with special attention given to the role of macroeconomic liquidity. To do so, we rely on a standard linear model as well as on non-linear one that allows for a transition between two possible regimes of liquidity conditions – accommodative or tight – both models accounting for possible trend reverting behaviour. The results of the study provide new insights to the discussion on property yield modelling. Whatever the type of modelling, linear or not, we find that in addition to its traditional drivers – notably risk-free interest rate and expected rental growth – money supply is a key factor of property yields. Moreover, depending on the evolution of the ratio of M2 to GDP, property yields evolve according to two regimes; in the one depicted as a normal liquidity regime, the property yields dynamics mainly obeys an error correcting mechanism which tends to counter excessive discrepancy between property yields and their fundamental value deduced from a Present Value type model, while in the second one, this mechanism is dominated by the impact of money supply growth which can induce increasing movements in the property prices, possibly turning to bubbles.

Keywords:

• Real estate yields
• global liquidity
• STR model
• nonlinearity
• asset pricing

Notes

1. We have retained the initial yield because it is widely used in the literature. However, another choice could have been to refer to the equivalent yield which is more common for market participants due to the particular nature of lease structures in the UK. Investigation of the dynamics of the equivalent yield is left for further research.

2. ‘Constant rates’ means: $\forall j\ge 0,g_{t+j}^e=g_t^e\text{and}\forall j\ge 0,r_{t+j}^e=r_t^e$; but, for both variables, the ‘constant’ values at date t differs from the ones at date s≠t.

3. The Gordon–Shapiro formula is generally written with the (expected) return rate and growth rate of dividends expressed in nominal terms. However, the formula holds for a constant retention rate of earnings if there is no inflation. In case of inflation, Lally (1988) shows that the Gordon-Shapiro equation has to include both rates in real terms with a retention rate defined in terms of inflation accounted earnings.

4. The series M2 has been smoothed before, by using a moving average of order 4 in order to best capture the persistent component of the series.

5. To conclude to the presence of a unit root, we refer to the results of different unit root tests and, simultaneously, to the shape of the dynamics of the series. If one observes long swings in the dynamics, we conclude that the series cannot be considered as I(0) even if some unit root tests may lead to the opposite conclusion.

6. Note that the IPD monthly index, from which we draw data at a quarterly frequency, tracks a specific subset of investors with requirement for monthly valuations, which could be a limitation for studies over a long period.

7. See the publication of Natixis for more details – http://cib.natixis.com/flushdoc.aspx.id=59495

8. We have not included the Asian part of the world, a choice which can be discussed, but we leave this extension for further investigations.

9. Note that other proxies such as short-run interest rates or broad money and credit outstanding M4 that are not discussed here are also theoretically pertinent as indicators for global liquidity and, more precisely, global funding liquidity. They were all tested empirically in our models and the final choice of M2 was determined empirically as it has greater explanatory power in explaining office yields.

10. http://www.jeanpierrelandau.com/wp-content/uploads/2013/05/Jackson-Hole-Print.pdf

11. A stochastic trend is characterised here as a random walk, according to a test for unit root.

12. According to MacKinnon tables, for a cointegration relationship with 5 regressors, a constant but no trend, the critical threshold is equal to −4.71 at the 5% risk level. With 3 explanatory series in the long-run relationship, two level shifts at the given dates and no trend, the ad hoc critical threshold obtained by simulation is −4.85 at the 5% level.

13. Note that the variable Δ(Δln(Real Rent Ratio)) has not been introduced in the model. Indeed, introducing it leads to a very instable specification with a non significant coefficient Γ, likely due to conflicting contribution of different variables.