Unsmoothing Real Estate Returns: A Regime Switching Approach Colin Lizieri, Stephen Satchell and Warapong Wonwachara Department of Land Economy / Department of Economics University of Cambridge European Real Estate Society Conference Milano May 2010
Not Another Paper on Smoothing? Many studies analysing impact of appraisal process on valuation-based real estate indices Standard method assumes time-invariant process Can we maintain this assumption? Transaction-based indices and adjustment for liquidity Tail dependence and behaviour in bubbles / crashes Intuition here: may be a regime-based structure For returns process (Lizieri et al., Brooks & Maitland-Smith) For smoothing process (Chaplin, informal model)
The Index Issue … FT Returns 2.02% 8.58% t,t IPD Returns 2.15% 3.25% t,t
Smoothing and Underlying Return Process Basic Smoothing Model Return Process Note the Time Subscripts …
Possible Models AR Model AR return process, single smoothing parameter AR-TAR Model Regime-based return, single smoothing process TAR-AR Regime-based smoothing, single return process TAR-TAR Regime-based smoothing and return processes
Defining Regimes and Modelling Regime Variables Tested Macro: GDP, Employment, Inflation, £/$ Asset & Cap Market: Equity Market, Interest Rate Endogenous: Cap Rate Analysis Quarterly, 1987Q1 – 2008Q4 Recursively Estimate and Minimise SSE Best Regime Models: AR-TAR FT Returns, LIBOR Best Regime Model: TAR-TAR FT Returns for Both Processes
AR-TAR
AR-TAR: FT Returns Regimes
AR-TAR versus AR: Fits The AR model appears too volatile – particularly across 2008 The AR-TAR model generates more plausible results
AR-TAR versus AR: Fits The AR model appears too volatile – particularly across 2008 The AR-TAR model generates more plausible results Still seems rather smoothed … Standard deviation compared to IPD 4.8% (3.1%) Serial correlation 0.51 (0.81)
TAR-TAR Model Focus on FT and LIBOR regimes All bar LIBOR-FT models improve SSE Best Model FT-FT All Coefficients Save 1 significant
TAR-TAR (FT Model) Return Processes Differ: FT Returns > -1.2% steady growth FT Returns < -1.2% negative, explosive autoreg. Implies very sharply falling underlying returns Low regime occurs 26% of time Regime not persistent … Smoothing Processes Differ: FT Returns < -13% very strong smoothing Information story? This state occurs 7% of time Smoothing parameter lower in other regime
TAR-TAR: FT Regimes Returns Regime: C 1 = -1.2% Smoothing Regime: C 2 = -13.2%
TAR-TAR: Model Fits
TAR-TAR versus = 0.8
Descriptive Statistics IPD Returns TAR-TAR = 0.8 Mean Mean Median Median Maximum Maximum Minimum Minimum Std. Dev. Std. Dev Skewness Skewness Kurtosis Kurtosis Serial Correlation Observations Observations
TAR-TAR versus = 0.8 Correlation = 0.92
Summary and Conclusions Aim: Extend Standard Desmoothing Model Take account of underlying return process Sensitive to asymmetries in return behaviour Sensitive to time-varying smoothing behaviour Approach Taken: TAR Models Behaviour based on indicator variable(s) Relatively simple to calculate and model AR-TAR and TAR-TAR outperform AR Best: TAR-TAR Model Based on FT Implications for Understanding of Risk Implications for Portfolio Strategy