In this post, I am interested in understanding how insights from the energy efficiency industry contribute to understanding the demand for photovoltaics. I argue that the energy efficiency industry, through its performance contracting business, provides insight as to what are the key determinants of whether residents will invest in renewable technologies. In this memo, I show that housing turnover is a key predictor of solar demand in California.
Energy Efficiency and Solar Power
Energy efficiency is the business of Energy Service Companies (ESCOs). Their job is to reduce energy usage and share in energy savings. They do this by installing new technologies that use less energy to power the same amount of stuff, e.g. new boilers or thicker insulation that seals the building envelope from energy leakage, or by better controlling energy usage, e.g. through the installation of monitors and sensors. The hallmark of an ESCO is a performance guarantee. The ESCO assumes the risk of installing energy-reducing and energy-efficient measures, often fronting the capital for energy efficiency projects. ESCOs then earn a return on the difference between the amount of energy consumed prior to the ESCO project and the amount of energy consumed after the ESCO project holding energy demand levels constant.
ESCO contracts are typically budget-neutral, meaning that all the measures the ESCOs put in will pay for themselves over the duration of the contract. The length of the contract is a function of client sector, building type and age, and plans for occupying the building. Client requests, like the installation of solar panels, which often have longer payback periods than other technologies, require the client to contribute extra funds or make other compromises in terms of the scope of the contract. As such, whether or not photovoltaics are adopted as part of the energy efficiency project depends in no small way on the duration of the contract, and thereby the client sector, building type, building age, and plans for occupancy.
Building Turnover Influences Performance Contracting
One factor, identified above, that shapes the demand for performance contracting is building turnover. Traditionally, private sectors are less likely to take advantage of performance contracting than Municipalities Schools and Hospitals (MUSH) because the length-of-stay in a building is a source of uncertainty that shapes the likelihood of contracting. According to a representative from TAC:
“…I think the government would accept longer paybacks because their institutions are institutions and they know they are going to be a school district for 20 years. They know they are going to be a government building for 20 years. I think private industry looks at their real estate assets as something that can fluctuate. Or, a lot of industry leases space. From a business standpoint, they credit themselves with a fixed lease and if they went out on a 10 year program that might exceed their lease. They feel like that’s not a good investment. That’s particularly true in the commercial real estate office market.”
The basic intuition, here, is that the more transient people are, the less likely they are to renovate their properties.
The dataset used to assess whether building turnover impacts the demand for photovoltaics is constructed by matching photovoltaics data from the California Solar Initiative (CSI) 2007-April 2010 and census data on housing characteristics from the Federal Financial Institutions Examination Council (FFIEC) 2009. The CSI data provides detailed information about solar installations—including total installed cost, rebate levels, photovoltaic models, and county and zip code of installation—by sector—government, commercial and residential. The FFIEC data provides census tract level data on housing characteristics—number of housing units, whether the unit is renter or owner occupied, when housing units were occupied, income of residents, and various characteristics of the housing stock. To match the CSI data, which is organized by zip code, to the FFIEC data, which is organized by census tract, a map of census tracts by zip code provided by American Fact Finder was used to identify census tracts that overlapped with each zip code. Because there is not a one-to-one mapping of census tracts to zip code, when multiple census tracts overlapped with zip codes, a census tract was selected at random. The implied assumption is that housing characteristics do not vary substantially within zip code and that a randomly selected census tract would approximate the housing characteristics of the zip code.
Annual housing turnover is proxied by dividing the number of housing owners who moved into their housing units between 1999 and March 2000 by the total number of owner occupied housing units in the zip code (census tract). The same was done for renters who moved into their housing units between 1999 and March 2000. The imperfection of using this proxy is that the housing units being occupied might actually be newly constructed housing units—indicating new home ownership/rentals rather than housing turnover. However, this problem is partially addressed by controlling for the median house age in each census tract.
The dataset is restricted to residential solar installations and to those with completed solar installations. For the data extract that I use in the analysis, solar systems must be complete by April 2010 in order to be included in the analysis. Table 1 shows summary statistics of the dependent variable (Watts), independent variables (Owner Occupied Housing Turnover and Renter Occupied Housing Turnover), and control variables.
The results demonstrate that owner occupied housing turnover is significantly negatively correlated with the size of solar installations as measured by watts. Model 3 in Table 2 shows that a 1% increase in owner occupied housing turnover equates to approximately a 13 watt decrease in watts installed. The same cannot be said for renter occupied housing turnover, however. Although Model 3 shows that renter occupied housing turnover is negatively correlated with watts of solar installed, it is largely insignificant. It is assumed that endogeneity does not play a large role, here, in that the demand for solar is unlikely to drive housing turnover. No further experiment, however, has been conducted to determine the direction of causality.
Other interesting results include median house age. A one year increase in median house age correlates with a 30 watt decrease in solar installed. One potential rationale for this finding, which emerged from conversations with solar installers has to do with the structural integrity of the roof. Older roofs may need additional support to sustain a large solar array, so solar installations on older homes tend to be smaller. As expected the incentive paid per watt positively correlates with the demand for solar and the price per watt negatively correlates with the demand for solar. These variables are included as controls in this analysis and no further work has been done to assess whether they are predictive of solar demand.