Spatial and Temporal Dynamics of the Singapore Housing Market Tay Jiajie, Darrell Department of Physics School of Physical and Mathematical Sciences 7 May 2014 Complexity Institute, Innovation Center 1
4 years Have not returned to former highs Dow Jones Industrial Average Picture Taken from Google Finance 2 years Took about 5 years to return to its previous highs 2
Housing Markets Many previous studies focus on liquid markets Slow to correct or exploit inefficiencies Inefficiencies results in Deadweight losses Need to understand non-liquid markets such as the Housing market Involves the livelihood of people Leverage Prolonged Dynamics (Time scales) Effects of the of two crashes are different depression of are amplified; home prices Market becomes stagnant Identify housing bubbles Determine susceptibility to crashes Measure effectiveness of policies 3
Overview and Previous Studies Determinants of Home Prices Supply and Demand Income and Wealth Equilibrium distributions Wealth and Income Distributions Housing Market Distributions Time Series Analysis Critical Transitions? 4
Theory of Wealth and Income Pareto (1897) estimated that income distributions followed a power law Mandelbrot (1960) proposed that only the tail end follows a power law Other empirical evidence Klass, Oren S., et al. [2006] (Forbes 400): 1.49 Souma [2002] (High net worth Japanese): 2.05 5
Theory of Wealth and Income Theoretical models done by Chakrabarti and Yakovenko Exchanges in fraction of total wealth (E) Exchanges with saving propensity (E-PL) Additive/Multiplicative Processes (E-PL) Exponential body and power law tail Empirical Study of Yakovenko et al. showed similar features Draulescu, A. A., V. M. Yakovenko, 2001b, Physica A 299, 213-221 Yakovenko, Victor M., and J. Barkley Rosser Jr. "Colloquium: Statistical mechanics 6 of money, wealth, and income." Reviews of Modern Physics 81.4 (2009): 1703.
Home Price Distributions Hedonic Model Home price is a product of N factors, P = N i=1 F i The log-price is a sum of random variables N log P = i=1 log F i By the Generalized CLT, in the limit where NOhnishi, Takaaki,, et log al. "On P the is evolution normally of the house price distributed distribution." (2011). Empirical studies showed that the tail is better fitted with a power law 7
Time Series Analysis Treat bubbles a precursor to critical transitions Length and time scale divergence Spectral Reddening Discrete Fourier transforms to analysis the frequencies Tan P. L. J., S. A. Cheong, Critical slowing down associated with regime shifts in the US housing market. Eur. Phys. J. B (2014) 87: 38 Power concentrated in lower frequencies Increased in autocorrelation and variances FT ω 1/2 8
Preliminary Works Data on the Singapore Housing Market Equilibrium Distributions and Deviations Spatial and Temporal Dynamics of the Singapore Housing distributions Time Series Analysis 9
Dwelling Types in Singapore HDB Flats Housing Development Board (HDB) Condominium (Condos) Cost Private Properties Landed Properties 1-5 Room Flat Executive, HUDC Studio Apartments Condominium Executive Condominiums (EC) Terraces Semi Detached/ Detached Bungalows Grants by HDB Highly regulated Singaporean/PRs No grants Foreigners allowed to purchase No grants Singaporeans/PRs 10
Private Sale Housing Data (1995-2012) The Dataset 1. Address of Property 2. Price (Total, psf, psm) 3. Transaction Date 4. Type of Dwelling 5. District (Numbered) 6. Sectors (Colored) 11
HDB Housing Data (2000-2012) The Dataset 1. Address of Property 2. Transaction Price 3. Floor Area 4. Transaction Month Unit Price is calculated Sectors chosen using address/town 12
Data Processing Segregated into the different types HDB Properties Condominiums Landed Properties Sorted into Postal Districts/Sectors Psf Price discounted using Historical CPI (Base Yr:2009) Psf Data is fitted with: Exponential Pareto Distribution 13
Landed Properties Pareto Distributed Statistically significant power law with α 5 Significance Testing using Clauset-Newman p-test (p = 0.058) 14
Results: Condominiums Fits well to an exponential distribution. (T = $444psf) Poorly fitted to a Pareto Dist. Empirical evidence that housing price is related to income No Hump Hump Appears Hump at the region $3k to $4.6k a Dragon King (DK). Hump Persist Appearance of Hump from 2007 and persists 15
Possible Explanation Investment class districts (D9, 10) contributed to the hump Upper Quartile Price Movement generally in tandem followed by wild swing in 2006-07 Greatly affected by 2008 correction Deviation of Prime, Investment Grade Properties Price as start of bubble formation 16
Spatial and Temporal Dynamics 17 Prices starts to increase during 2006 in agreement with the stationary analysis The bubble starts in District 9 and 10 and spreads out radially
Sliding Window Time Series 1 month Analysis DFT to detect Spectral Reddening Autocorrelation Sliding 2 years window Every slide = 1 month 18
Discrete Fourier Transform Power evenly spread out during quiet years Power concentrated at lower frequencies during possible bubble years Asian Financial Crisis 19
Lag1 Autocorrelation Autocorrelation low at quiet years Autocorrelation spikes up during the possible bubble years Asian Financial Crisis can also be seen 20
HDB Properties Agree with Income/Wealth Exponential Distributed with crossover at $600psf Second regime appeared only after 2009 Price in the different districts move in tandem 21
Comparison across types Housing distribution emulates Income/Wealth distribution Lead Lag Relationship Pareto Exponential Bubble in Condominiums, but not in Landed and HDB Bubble spreading spatially, but not across housing types 22
Future Works Comparison to Taiwanese Data Not as highly segregated as Singapore Exponential body and power law tail The segregation should be in locations Expect to pick out Universal features Discover non-equilibrium features Emulate income/wealth distribution 23
Future Works and Summary Build Agent Based Models Expand on the computation models proposed by Chakrabarti and Yakovenko Develop a model for housing that can be used for scenario testing Bubble spreading across the different housing type 24
Future Works and Summary 8 cooling measures from 2009 to 2013 Critical slowing down in Singapore Housing Market Round 1 (14 September 2009) Round News 2 (20 reports February in Singapore 2010) suggesting that Round cooling 3 (30 measures August 2010) are working Round We 4 determine (14 January that 2011) stability of the housing Round market 5 (7 December 2011) Round Treat 6 cooling (5 October measures 2012) as perturbations Round System 7 (11 is January stable if 2013) recovery rates are Round increasing 8 (28 June and converging 2013) 25
References 1. A. Dragulescu and V. M. Yakovenko, Physica A: Statistical Mechanics and its Applications 299, 213 (2001). 2. V. Pareto, Cours d Economie Politique, Lausanne, 1897 3. V. M. Yakovenko and, J. Barkley Rosser, Jr., REVIEWS OF MODERN PHYSICS, VOLUME 81, OCTOBER DECEMBER 2009 4. B. Mandelbrot, Int. Econom. Rev. 1 (1960) 79 5. Klass, Oren S., et al. Economics Letters 90.2 (2006): 290-295. 6. Souma, Wataru. Springer Japan, 2002. 343-352. 7. SingStats, 2013 8. Urban Redevelopment Authourity, 2013 9. Tan, James Peng Lung, and Siew Ann Cheong. "Critical slowing down associated with regime shifts in the US housing market." The European Physical Journal B 87.2 (2014): 1-10. 26