Testing for Bubbles in Housing Markets: A Panel Data Approach

Transcription

1 Testing for Bubbles in Housing Markets: A Panel Data Approach GDN, Prague 2008

2 Outline Introduction Bubbles in Housing Markets Data Panel Data Tests Comparison with Other Housing Market Indicators Summary

3 Introduction Is There a Bubble? National data - Mc Carthy and Peach (2004), Galin (2004) Panel data - Malpezzi (1999), Galin (2006) Regional data - Smith and Smith (2006), Himmelberg, Mayer, Sinai (2005) Stationarity and Earnings Campbell and Shiller (1987), Wang (2000)

4 Introduction Panel Data Tests Cross section dependence - Pesaran (2004) Unit roots - Im, Pesaran, and Shin (2003), Pesaran (2007) Cointegration - Pedroni (1999, 2004) Causality - Hurlin (2004), Hurlin and Venet (2004) Application - Malpezzi (1999), Galin (2006) Research Objective Combine panel data unit root and integration tests - bubble indicator US real estate market, house prices and rents Changes in house prices and rents

5 Present Value Model House Prices and Cash Flows [ ] Ci,t+1 + P i,t+1 P i,t = E t, i = 1,..., N 1 + D [ Ci,t+1 P i,t = E t 1 + D + C i,t+2 (1 + D) C i,t+k (1 + D) k + P ] i,t+k (1 + D) k [ ] lim E Pi,t+k t k (1 + D) k = 0 P F i,t = j=1 1 (1 + D) j E t[c i,t+j ]

6 Present Value Model Cash Flow Process I(1) and No Bubbles Rational Bubbles S i,t = P i,t 1 D C i,t = 1 D E C i,t+j+1 t j=0 (1+D) j = 1 D E t [ P i,t+1 ] B i,t = D E tb i,t+1 P i,t = P F i,t + B i,t

7 Present Value Model Stationarity of house prices and cash-flows Case 1: P i,t stationary and C i,t stationary Case 2: P i,t stationary and C i,t non-stationary Case 3: P i,t non-stationary and C i,t stationary Case 4: P i,t non-stationary and C i,t non-stationary Bubble Indicator =0 in Case 1, 2 =1 in Case 3 =pvalue of a stationarity test of P/r in Case 4

8 Aggregate Data Aggregate Price-Income and Price-Rent Ratios quarters price-income ratio price-rent ratio

9 Panel Data Dataset 1-23 Metropolitan Statistical Areas (MSA), 1978:1-2006:2 House price index (HPI) from the Office of Federal Housing Enterprise Oversight (OFHEO) - the weighted repeat sales index based on mortgage contracts Rent of primary residence index (RI) estimated by the Bureau of Labor Statistic (BLS) Regional CPI from (BLS) Dataset Metropolitan Statistical Areas, HPI from OFHEO The fair market rent, US Department of Housing and Urban Development (HUD) Aggregate CPI from BLS

10 Panel Data Price-Rent Ratios in 23 MSA price-rent ration :h1 1982:h1 1986:h1 1990:h1 1994:h1 1998:h1 2002:h1 2006:h1 New York Boston Philadelphia Los Angeles San Francisco

11 Panel Data Price-Rent Ratios in 23 MSA price-rent ratio :h1 1982:h1 1986:h1 1990:h1 1994:h1 1998:h1 2002:h1 2006:h1 Cincinnati Kansas City Milwaukee Minneapolis St. Louis Atlanta

12 Panel Data Price-Rent Ratios in 23 MSA price-rent ratio :h1 1982:h1 1986:h1 1990:h1 1994:h1 1998:h1 2002:h1 2006:h1 Pittsburgh Chicago Miami Denver San Diego Seattle

13 Panel Data Price-Rent Ratios in 23 MSA price-rent ratio :h1 1982:h1 1986:h1 1990:h1 1994:h1 1998:h1 2002:h1 2006:h1 Detroit Dallas Houston Honolulu Portland Cleveland

14 Panel Data Average Price-Rent Ratios in 23 MSA price-rent ratio :h1 1982:h1 1986:h1 1990:h1 1994:h1 1998:h1 2002:h1 2006:h1

15 Panel Data Average Price-Rent Ratios in 273 MSA Average price rent ratio

16 Panel Data Average Absolute Price-Rent Ratios in 273 MSA Absolute price-rent ratio

17 Panel Data Price-rent Ratios Using Indices, 273 MSA s P/r>1.5 P/r>1.3 P/r<0.9 P/r<

18 Panel Data Price-rent Ratios Using Levels, 273 MSA s P/r>22.5 P/r>20 P/r<13 P/r<

19 Panel Data Tests Pesaran (2004): Diagnostic tests for cross section dependence in panels CD = s 0 X NX N 1 Corr(ˆɛ i, ˆɛ j ) A N(0, 1) N(N 1) i=1 j=i+1 1 price-level price-diff. rent-level rent-diff. CD CD CD CD 23 MSAs, 1978: :02, semi-annual data *** *** *** MSA, , annual data *** *** *** 0.63

20 Stationarity Panel Data Unit Root Tests - Pesaran (2007) Cross-sectionally Augmented Dickey Fuller (CADF): y it = µ i + ω i t + α i y i,t 1 + υ i ȳ t 1 + p i j=1 λ ij y i,t j + pi j=0 ϖ ij ȳ i,t j + ε it H 0 : α{ i = 0 αi = 0 for i = 1, 2,..., N H 1 : 1 α i < 0 for i = N 1 + 1, N 1 + 2,..., N. CADF: t i,ti,n(p i ) = t i (T, N) CIPS: t = 1 N N i=1 t i (N, T )

21 CADF, 23 MSA price rent p/r p/r MSA CADF CADF CADF rank Atlanta-Sandy Springs-Marietta, GA Boston-Quincy, MA (MSAD) Chicago-Naperville-Joliet, IL (MSAD) *** *** ** 23 Cincinnati-Middletown, OH-KY-IN *** Cleveland-Arkon, OH (MSAD) *** *** Dallas-Plano-Irving, TX (MSAD) Denver-Aurora, CO Detroit-Livonia-Dearborn, MI (MSAD) Honolulu, HI Houston-Sugar Land-Baytown, TX *** Kansas City, MO-KS Los Angeles-Long Beach-Glendale, CA (MSAD) Miami-Miami Beach-Kendall, FL (MSAD) Milwaukee-Waukesha-West Allis, WI ** Minneapolis-St. Paul-Bloomington, MN-WI *** New York-White Plains-Wayne, NY-NJ (MSAD) ** Philadelphia, PA (MSAD) Pittsburgh, PA Portland-Vancouver-Beaverton, OR-WA San Diego-Carlsbad-San Marcos, CA San Francisco-San Mateo-Redwood City, CA (MSAD) Seattle-Bellevue-Everett, WA (MSAD) * 22 St. Louis, MO-IL

22 CADF, 273 MSA, Top 25 price rent p/r p/r MSA CADF CADF CADF rank Ann Arbor, MI Athens-Clarke County, GA Barnstable Town, MA Bend, OR Bridgeport-Stamford-Norwalk, CT Canton-Massillon, OH Deltona-Daytona Beach-Ormond Beach, FL Des Moines-West Des Moines, IA El Paso, TX Flint, MI Jackson, MI Lakeland, FL Las Vegas-Paradise, NV Lexington-Fayette, KY Mansfield, OH Monroe, LA ** New Haven-Milford, CT Olympia, WA Poughkeepsie-Newburgh-Middletown, NY Rochester, MN Saginaw-Saginaw Township North, MI Sherman-Denison, TX Spokane, WA St. Joseph, MO-KS Toledo, OH

23 CADF, 273 MSA, Bottom 25 price rent p/r p/r Abilene, TX *** ** 271 Amarillo, TX *** ** 268 Atlantic City, NJ ** Beaumont-Port Arthur, TX * 262 Boulder, CO ** 269 Burlington-South Burlington, VT *** * 259 Colorado Springs, CO ** Corpus Christi, TX *** * 256 Dayton, OH * Fort Lauderdale-Pompano Beach-Deerfield Beach, FL (M Greeley, CO * 261 Houma-Bayou Cane-Thibodaux, LA *** Knoxville, TN * 260 Los Angeles-Long Beach-Glendale, CA (MSAD) *** Madison, WI ** 270 Ocean City, NJ * * 264 Oxnard-Thousand Oaks-Ventura, CA *** * 265 Philadelphia, PA (MSAD) ** *** 273 Phoenix-Mesa-Scottdale, AZ * 263 Reading, PA *** ** 267 San Diego-Carlsbad-San Marcos, CA * * 258 Santa Cruz-Watsonville, CA * * 266 Scranton-Wilkes-Barre, PA *** 272 Shreveport-Bossier City, LA *** * 257 Waterloo-Cedar Falls, IA

24 Panel Data Tests Unit Roots Tests 23 MSA, 1978: :02, semi-annual data Method price-level price-diff. rent-level rent-diff. p/r IPS *** *** 5.48 CIPS *** *** MSAs, , annual data Method price-level price-diff. rent-level rent-diff. p/r IPS *** *** CIPS *** *** -2.12

25 Panel Data Tests Pedroni (1999,2004) Cointegration Tests y i,t = µ i + ω i t + ψ i x i,t + ζ i,t for t = 1,..., T, i = 1,..., N. The null hypothesis of no cointegration: H 0 : γ i = 1 for all i where γ i is the autoregressive coefficient of the residual ˆζ i The alternative hypothesis H 1 : γ i < 1 for all i i.e. no common value for the autoregressive coefficient is presumed The critical values were generated using 50,000 simulations by bootstrapping to preserve cross-sectional dependence and including autocorrelation. Stat. 23 MSA 273 MSA group adf-stat

26 Panel Data Tests Panel Data Granger Causality - Hurlin (2004) and Hurlin and Venet (2004) Let y i and x j be two stationary variables. y it = µ i + L l=1 ϕ(l) i y i,t l + L l=1 δ(l) i x i,j,t l + ξ it ξ it are normally i.i.d. with zero mean and finite heterogeneous variances and ξ i = (ξ i1,..., ξ it ) are independently distributed across groups. Homogeneous Non Causality (HNC): H 0 : δ i = 0, i = 1,..., N, δ i = (δ (1) i,..., δ (L) H 1 : δ i = 0, i = 1,..., N 1, δ i 0, i = N 1 + 1,..., N, where N 1 [0, N) is not known. i )

27 Panel Data Tests WNT HNC = (1/N) N i=1 W it where W it denotes the Wald statistic associated with the individual test of H 0 for each i = 1,..., N = N Z HNC NT 2 L [ ] (T 2L 5) (T L 3) (T 2L 3) (T 2L 1) W NT HNC L converges in distribution to N(0, 1) as N for a fixed T > 5 + 2L

28 Panel Data Tests Hurlin Tests for Homogeneous Non-Causality in Panel Data H 0 Z HNC NT P-value 22 MSAs, 1978: :02, semi-annual data price-diff. does not Granger cause rent-diff rent-diff. does not Granger cause price-diff MSAs, , annual data price-diff. does not Granger cause rent-diff rent-diff. does not Granger cause price-diff

29 Bubble Indicator, 23 MSA price price-dif. rent rent-dif. coint. p/r bubble Year CIPS CIPS CIPS CIPS Pedroni CIPS indicator * * * * *** ** * ** ** *** * * * *** ** *** *** *** *** *** *** * *** ** ** ** *** * ** *** *** *** *** * *** ***

30 Bubble Indicator 273 MSA price price-dif. rent rent-dif. coint. p/r bubble Year CIPS CIPS CIPS CIPS Pedroni CIPS indicator *** *** ** * * *** *** **

31 Comparison Bubble and Other Indicators 1/ Bubble Indicator 273 MSA Existing Single-Family Home Sales Composite Affordability Index Bubble Indicator 23 MSA Sales Price of Existing Single-Family Homes HMI

32 Comparison Bubble and Other Indicators 2/ Bubble Indicator 273 MSA Bubble Indicator 23 MSA Total Housing Units Started New Houses Sold Houses for Sale Value of Constraction

33 Summary Panel Data - Stationarity Relationship between house prices and fundamentals is not stationary over the whole sample. Bubble Indicators Rational bubbles in several periods but mainly in the late 1980s and the early 2000s up to Our bubble indicators coincide fairly well with the pattern of price-rent ratios, the HMI and the value of construction index. Panel Data - Causality Changes in prices predict changes in rents and viceversa