Multifamily residential asset and space markets and linkages with the economy Alain Chaney ♣ Martin Hoesli ♦ ERES Conference Bucharest, June 25-28, 2014 ♣ GSEM, University of Geneva, Switzerland IAZI AG, Switzerland ♦ GSEM & Swiss Finance Institute, University of Geneva, Switzerland Business School, University of Aberdeen, UK Kedge Business School, France Outline Motivation Methodology Data Empirical Results 2 Motivation Methodology Data Empirical Results 2014 ERES Conference Theoretical background Property Market (DiPasquale & Wheaton) Macro Economy market rent demand=f(r, gdp…) price stock construction Real estate markets are influenced by macroeconomic factors through a variety of channels and these linkages have been documented both for housing (Kennedy, 2005; Cihák, Iossifov & Shanghavi, 2008; International Monetary Fund, 2008) and commercial real estate markets (Chaney & Hoesli, 2012; McCartney, 2012) Motivation Methodology Data 3 Empirical Results 2014 ERES Conference Previous work Asset market extensively researched by cap rate studies (early work includes Froland, 1987; Evans, 1990; Ambrose & Nourse, 1993) Limited number of recent studies applied more complex time series models, i.e. ECM that follow the strategy of Engle & Granger (1987) (Hendershott & MacGregor, 2005; Dunse et al., 2007; Clayton, Ling & Naranjo, 2009) Cap rates are found to depend on various economic forces Space market studies are mainly concerned with estimation of the rental adjustment process and explain (equilibrium) rents by employment, economic activity, interest rates, space supply, (natural) vacancy rate, construction costs, and lagged rental values. State of the art are ECMs that follow the strategy of Engle & Granger (1987) (Hendershott, MacGregor & Tse, 2002; Hendershott, MacGregor & White, 2002; Brounen & Jennen, 2009; Hendershott, Lizieri & MacGregor, 2010; McCartney, 2012) Methods ECMs of Engle and Granger (1987) are limited to a single cointegrating vector and the studies that have applied this approach treated economic variables exogenously Johansen (1988, 1991) and Johansen & Juselius (1990) developed a systems-based approach to cointegration which enables for more than one cointegrating vector This approach has been applied to the commercial real estate market only recently (Schätz & Sebastian, 2009; Kohlert, 2010) The cointegrating vectors derived from the popular Johansen procedure would allow for more than one long-run relation, but they are statistically motivated identifying restrictions No economic meaning: economic relations are not orthogonal Identification of short-run dynamics achieved with the recursive structure of Sims (1980) Results are not unique and depend on the ordering of the variables Methodologies applied do not seem to fully meet the complexity of the linkages Motivation Methodology Data Empirical Results 4 2014 ERES Conference Methodology To overcome some of these issues, we introduce a new modelling approach from macroeconometrics. It is based on Garratt, Lee, Pesaran & Shin (2003, 2006) and allows to incorporate long-run structural relationships, as suggested by economic theory, in an otherwise unrestricted VAR model. On the basis of the equilibrium framework provided by DiPasquale & Wheaton (1992, 1996), we model the commercial real estate market as a whole to account for the fact that construction, rents and cap rates are interrelated. We also model all series including core economic variables endogenously, therefore allowing for various contemporaneous linkages as well as for several long-run equilibrium relations. Standard VECM, but cointegrating vectors derived using economic theory (whose validity can be tested econometrically) Motivation Methodology Data Empirical Results 5 2014 ERES Conference Data Area Why? Switzerland Basic principles of macroeconomics and of real estate economics that underlie our empirical model are country-independent Several ‘building blocks’ which constitute the basis for our study have been applied successfully to various markets Availability and quality of required data in general and of transaction-based cap rates in particular Period 1974Q1-2013Q2 CR INF real estate cap rate; 0.25*ln(1+R/100) quarterly inflation rate; ln(CPI/CPI(-1)), whereas the CPI has been adjusted for inclusion of the end of year sales in 2000 R10 risk-free interest rate with a maturity of 10 years; 0.25*ln(1+R/100) LIB risk-free interest rate with a maturity of 3 months; 0.25*ln(1+R/100) M2 log real M2 6 RENT log real rent, s.a. before 1990 CON log real construction spending, s.a. 2014 ERES Conference Data Empirical Results Y Motivation log real Methodology gdp, s.a. Long-run analysis Theory predicts several long-run relations among these series Slope of the term structure (3m &10y interest rates) Fisher interest rate parity (3m interest rates & inflation) … Real estate excess return (cap rates & 10y interest rates and cap rate spread & construction / GDP) … augmented with money demand (M2, GDP & 3m interest rates) Real estate market equilibrium (rents / cap rates & constructions) Augmented fisher interest rate parity 7 Motivation Methodology Data Empirical Results 2014 ERES Conference Long-run analysis Economic theory suggests Slope of the term structure 𝑟10𝑡 = 𝑏10 + 𝛽11 𝑟3𝑚𝑡 +𝜉1,𝑡 Fisher parity 𝑟3𝑚𝑡 = 𝑏20 + 𝛽21 𝜋𝑡 + 𝛽22 𝑦𝑡 − 𝑚2𝑡 + 𝜉2,𝑡 Real estate market equilibrium 𝑐𝑜𝑛𝑡 = 𝑏30 + 𝛽31 𝑟𝑒𝑛𝑡𝑡 + 𝛽32 𝑐𝑟𝑡 + 𝜉3,𝑡 Real estate excess return 𝑐𝑟𝑡 − 𝑟10𝑡 = 𝑏40 + 𝑏41 𝑡 + 𝛽41 𝑐𝑜𝑛𝑡 − 𝑦𝑡 + 𝜉4,𝑡 Written compactly where 𝛽′ = 𝛽11 1 0 0 0 𝛽22 0 0 Results 0 𝛽23 0 0 0 𝛽24 0 𝛽44 0 0 1 𝛽45 0 0 𝛽36 1 Motivation Methodology 0 0 𝛽37 0 1 0 0 −1 The central bank tens to reduce libor almost one by one with inflation… … and some more when money velocity is low, i.e. if GDP growth is low (compared to M2). Financial crisis: no inflationary pressure central banks reduced interest rates and increased money supply to stimulate GDP growth and to avoid deflation Data Demonstrates the empirical validity of the D&W framework Cap rate spread indeed evolves with the evolution of the construction to GDP measure Empirical Results 8 2014 ERES Conference Error-correction equations Adjusted R2 all lie in the range of [0.17, 0.68] Adjusted R2 for benchmark models are much lower for most equations In line with this observation, the coefficients of the error correction terms make a significant contribution in most equations This shows that the error correction mechanisms provide for a complex and statistically significant set of interactions and feedbacks across the whole macroeconomy, including all real estate quadrants It also demonstrates that the benefits of the long-run structural modeling lie not only within the more structural interpretation and understanding based on economic theory but also within an improvement of the explanatory power of the short-run dynamics 9 Motivation Methodology Data Empirical Results 2014 ERES Conference Length of time to equilibrium Overall observations Length of time varies between 10 and 30 quarters depending on the shock some shocks eventually simply vanish while others, such as a shock in construction, M2, or GDP lead to oscillation Real estate excess return eq. Smallest influence exerted by a change in inflation Biggest changes caused by short- and long-term interest rates, M2 and GDP Length of time to eq. about 5 years Real estate market equilibrium Smallest influence exerted by a change in inflation Biggest changes caused by one standard deviation change in cap rates, long-term interest rates and rents 10 Length of time to eq. about 7 years Motivation Methodology Data Empirical Results 2014 ERES Conference Short-run dynamics The linkages between commercial real estate and the economic are bidirectional While it is well documented that economic variables influence real estate markets, GIRFs clearly show that real estate variables also exert some influence on the economy For example, if the monetary authority increases interest rates to reduce inflationary pressures, this will directly reduce GDP growth and inflation, but on top will also impact the real estate market through a reduction in construction and rents and through an increase in cap rates. This will feed back to the core economy, as lower construction and lower rents both reduce GDP. The ultimate outcome may be a recession and falling real estate prices These observations require modeling all variables (including the economic variables) endogenously 11 Motivation Methodology Data Empirical Results 2014 ERES Conference Summary & Conclusions Theory: predicts various linkages Previous studies: focused on a limited subset of these linkages, treated economic variables exogenously & included a single cointegrating relation We model the whole economy (including all four real estate quadrants) by incorporating equilibrium relations that are predicted by theory in an otherwise unrestricted VAR model and treat all variables endogenously We find four long-run equilibrium relations Due to their economic interpretation, these long-run equilibrium relations do not just improve the explanatory power of the models short-run dynamics, but they additionally help in the interpretation of economic conditions and identification of market disequilibria Short-run dynamics show that the linkages are bi-directional. This requires modeling all variables endogenously Results should also prove useful to investors, real estate developers, and tenants because a better understanding of the linkages can help them to prepare better for economic shocks and market disequilibria Researchers should benefit because the presence of bi-directional links and a variety of long-run equilibrium relations implies that it is likely that previous studies did not fully capture the whole error-correcting behavior Motivation Methodology Data Empirical Results 12 2014 ERES Conference Thank you for your attention!