A Framework for Measuring Consumer Preferences for Home Energy Efficiency Chris Bruegge Stanford University June 23, 2015 Bruegge (Stanford) Preferences for EE June 23, 2015 1 / 24
Research Questions Related Literature Context: Energy Efficient Homes Discrete home choice refects preferences for continuous choice of energy service consumption Questions Who purchases energy efficient homes high users or warm glow? How much energy would these people use in less efficient homes? Policy Connections How much energy do building codes save? How do homeowners respond to EE subsidies? Bruegge (Stanford) Preferences for EE June 23, 2015 2 / 24
Methods and Preliminary Results Discrete-Continuous Model Links preferences for home attributes (discrete choice) with preferences for energy consumption (continuous choice) Home purchase and energy usage data from Gainesville, FL Preliminary Results High energy users purchase the most efficient homes High users are most price elastic A simulated 5% building code efficiency increase reduces energy consumption by 3.5% if we account for sorting increases consumption by 1.9% if we don t account for sorting Bruegge (Stanford) Preferences for EE June 23, 2015 3 / 24
Roadmap Residential Energy Efficiency Standards Data Discrete-continuous Model Preliminary Estimates and Counterfactual Future Steps Bruegge (Stanford) Preferences for EE June 23, 2015 4 / 24
Setting (Florida) Building Codes Florida Energy Code gets updated every 3 years Each iteration results in about a 5-10% increase in efficiency requirements I m not modeling the supply side, although codes seem to have bite Bunching at code changes Code changes provide quasi-random variation across time in the energy efficiency of homes in buyers choice sets Bruegge (Stanford) Preferences for EE June 23, 2015 5 / 24
Codes Affect Efficiency of Homes in Choice Sets Density 0.2.4.6.8 1960 1970 1980 1990 2000 2010 Vintage of Homes in Choice Set by Purchase Date 6/1/2005 6/1/2000 6/1/1995 6/1/1990 Bruegge (Stanford) Preferences for EE June 23, 2015 6 / 24
Summary Statistics for 4K Homes in Gainesville, FL N mean sd min max Sales Price 3,922 128,248 146,625 10,250 2,400,000 Square Feet 3,922 1,907 755 466 8,286 Bedrooms 3,922 3 1 0 5 Bathrooms 3,922 4.3 1.4 1.0 13.0 Year Built 3,922 1986 14 1930 2006 U-value 3,922 0.31 0.05 0.26 0.56 p s 3,922 12.48 4.57 3.73 47.95 p elec 3,922 0.10 0.00 0.09 0.12 kwh 3,922 1096.29 478.39 146.45 5027.55 Therms 2,995 30.24 14.14 1.00 135.22 Income Level 3,922 7 2 1 9 Owner Age 3,922 58 14 28 80 An observation in the raw dataset consists of a homeowner / home pair Defining Choice Sets Bruegge (Stanford) Preferences for EE June 23, 2015 7 / 24
Indirect Utility for Agent i in Home j Agents choose (1) which home to purchase and (2) how much energy to use Monthly Mortgage pj H = f H (SalesPrice) Price of Energy Services ($/ F) Home Attributes that Affect Energy Service Demand pj S = f S (X j ) p elec (Square-feet, Number of Bedrooms) ( V ij = Vij D i, pj H, pj S, Xj S, X H j, ν ij ) + ɛ ij Demographics Income Bracket, Age, etc. Home Attributes that Only Affect Home Choice Location F.E. Functional Form Definining p S Bruegge (Stanford) Preferences for EE June 23, 2015 8 / 24
Behavioral Equations The probability that agent i chooses home j is given by Pr ij = exp(vij ) k Ω i exp(vik ) (1) and Roy s Identity implies that the optimal consumption of energy services ( F of heating / cooling) is given by log(s ij ) = α + β S i p S j + β H i (I i p H j ) + β Xs X S j + ν ij (2) Deriving Consumption Eqn. From V ij Likelihood Bruegge (Stanford) Preferences for EE June 23, 2015 9 / 24
Price Elasticity of Demand for Energy Services Age Low-Income Middle-Income High-Income Total 25-0.232-0.239-0.580-0.241 (0.010) (0.005) (0.002) (0.007) 35-0.303-0.357-0.517-0.372 (0.003) (0.002) (0.001) (0.002) 45-0.478-0.518-0.538-0.522 (0.003) (0.002) (0.001) (0.002) 55-0.389-0.441-0.515-0.469 (0.002) (0.002) (0.001) (0.001) 65-0.469-0.507-0.628-0.560 (0.002) (0.002) (0.001) (0.002) Total -0.384-0.465-0.576-0.498 Bruegge (Stanford) Preferences for EE June 23, 2015 10 / 24
High Users Purchase Efficient Homes Predicted Energy Services / Month at Mean Prices 10 15 20 25 30 35.25.3.35.4.45.5 Heat Transfer through Home Envelope (Btu / Ft^2 / Degree F / Hr) Bruegge (Stanford) Preferences for EE June 23, 2015 11 / 24
Preliminary Building Code Counterfactual Baseline Predict home choice and energy use for buyers who purchased a home in 2005 Building Code Upgrade Increase energy efficiency by 5% for 2005-vintage homes Restimate sorting and energy consumption Bruegge (Stanford) Preferences for EE June 23, 2015 12 / 24
Preliminary Counterfactual Policy Evaluation Method 1 Compare usage in efficient 2005 homes to baseline 2005 homes (with different occupants) Policy Evaluation Method 2 Compare usage by occupants of efficient 2005 homes to usage by same individuals in their repective baseline homes Baseline Method 1 Method 2 kwh / Month 1,026 +1.8% -3.5% Bruegge (Stanford) Preferences for EE June 23, 2015 13 / 24
Conclusion and Future Steps Today s Punchline: Discrete-continuous link seems to matter for policy evaluation Sorting of highest (+most elastic) users into most efficient homes Going Forward: Home-level building permit data => Richer controls Extensions of the model to solar subsidies, utility EE programs, etc Bruegge (Stanford) Preferences for EE June 23, 2015 14 / 24
Thank You! Bruegge (Stanford) Preferences for EE June 23, 2015 15 / 24
Related Literature Back Rebound Davis cash for appliances, Davis washer field experiment How Much are Codes Saving Jacobsen & Kotchen, Levinson Selection / Discrete-Continuous Dubin & McFadden, Bento et al Bruegge (Stanford) Preferences for EE June 23, 2015 16 / 24
Code Seems to Bind Back Frequency 0 50 100 150 200 Sep 2, 2001 Oct 2, 2001 Nov 1, 2001 Dec 1, 2001 Dec 31, 2001 Jan 30, 2002 Mar 1, 2002 Mar 31, 2002 Apr 30, 2002 Permit Application Date May 30, 2002 Jun 29, 2002 Jul 29, 2002 Aug 28, 2002 Bruegge (Stanford) Preferences for EE June 23, 2015 17 / 24
Price Response to Efficiency Standards Back Homeowners choose how much to heat and cool their homes Energy Services }{{} S= Indoor - Outdoor Temp = kwh of Electricity }{{} E Efficiency }{{} 1/η A policymaker cares about the impact an efficiency standard, x, will have on energy use, E: E x = S η + η S }{{ x} p S ps η η x }{{} Modeled Savings Rebound (3) I can estimate the term S p S, while the other terms come from engineering definitions or models. Bruegge (Stanford) Preferences for EE June 23, 2015 18 / 24
Model: Indirect Utility for Agent i in Home j Back Monthly Income Bucket Monthly Mortgage pj H = f (SalesPrice) Home/Neighborhood Attributes Square-feet, Location FE, etc. ( V ij = exp + exp ) βi H ( I i pj H ) ( α i + β S i p S j + γ X S j + ν ij ) + X H j β H i + ɛ ij Price of Energy Services ($/ F) p S j = f (X j ) p elec Home/Neighborhood Attributes Square-feet, Location FE, etc. Parameters are in BLUE, Data is in BLACK, Shocks are in RED Bruegge (Stanford) Preferences for EE June 23, 2015 19 / 24
Home-level Permit Data Back Bruegge (Stanford) Preferences for EE June 23, 2015 20 / 24
Measuring Energy Service Prices Back Engineers rate materials with a u-value Btu of heat transfer per hour x Degree F x ft 2 ( ( ) ( ) ) kwh therms p s = u ft 2 w elec p kwh + (1 w elec ) p therm btu btu Bruegge (Stanford) Preferences for EE June 23, 2015 21 / 24
Behavioral Equations Back By Roy s Identity, s j = V ij/ p S j V ij/ I i = exp(α i + βi S pj S + β Xs Xj S + ν ij )βj S exp(βi H (I i pj H ))βj H = exp(α i + β S i p S j β H i (I i p H j ) + β Xs X S j + ν ij ) βs j β H j ln(s j ) = α i + β S i p S j β H i (I i p H j ) + β Xs X S j + ν ij Bruegge (Stanford) Preferences for EE June 23, 2015 22 / 24
Likelihood Back The log likelihood is given by ln(l i ) = [ ] I(j chosen) log(pr ij ) + log(f ν j (ν ij )) i Bruegge (Stanford) Preferences for EE June 23, 2015 23 / 24
Defining Choice Sets Back -45 Days Home 1 Purchased +45 Days -45 Days Home 2 Purchased +45 Days -45 Days +45 Days Home 3 Purchased I observe when agent i purchases a home Agents choose from the set of homes that transacted within 45 days of their own home purchase We get choice set variation because homes sell at different points in time An observation is a homeowner / home pair For each homeowner, there is a observation corresponding to each home in the choice set Bruegge (Stanford) Preferences for EE June 23, 2015 24 / 24