Accounting for endogeneity in the demand for cooling

Accounting for endogeneity in the demand for cooling

Our work is motivated by the increasing demand for electricity to ensure adequate cooling, driven by the threat of global warming. The aim of our work is to investigate how households adopt and use air conditioning to adapt to climate change. Air conditioning is a mean to control for the home’s interior temperature and obtain the desired level of thermal comfort which helps to protect vulnerable people from high temperatures.

Using the OECD- EPIC survey, we focus on eight temperate, industrialized countries (Australia, Canada, France, Japan, the Netherlands, Spain, Sweden, and Switzerland). Our identification strategy exploits cross-country and cross-household variations by matching geocoded households with climate data.

Our outcome variable is the demand for electricity which we identify in electricity expenditures. We model it as a function of household characteristics and climate conditions. The adoption of air conditioning (AC) is our explanatory variable of interest. We study how much AC affects the electricity expenditure. 

The main challenge in quantifying the effect of AC on electricity expenditures is that even after controlling for a large number of household characteristics (e.g. composition, location, economic status) there are unobserved variables which could simultaneously affect the adoption of AC and electricity expenditures. In fact, the two decisions are related and share unobservable common determinants which represent a source of endogeneity. If we fail to take into account this unobserved heterogeneity we risk to measure the effect of AC on electricity expenditures incorrectly reaching misleading conclusions. In other words, results are biased when AC is an endogenous variable.

To overcome this issue, our methodology is based on a two-step estimate procedure called “Control Function approach”. The first step consists in estimating the adoption of AC, through a probit model, on a set of explanatory variables (household characteristics and climate conditions). The main challenge we faced was to find a valid exclusion restriction. This is a variable that has to be strongly correlated with the decision of AC adoption but must have no direct effect on electricity expenditure. We identify our exclusion restriction in past imports of air conditioners weighted by past climate conditions (‘past Cooling Degree days – CDD’). The validity of the exclusion restriction is essential for the implementation of the control function approach. 

After estimating AC on our explanatory variables and exclusion restriction we compute the generalized residuals. The second step consists in the estimation of our outcome variable, expenditure on electricity, on AC and the generalized residuals computed in the first step, in addition to household characteristics and climate conditions. The inclusion of the generalized residuals in the outcome variable equation corrects for the endogeneity of AC and allow us to conclude that, depending on the specification, households adopting AC spend on average 35-42% more on electricity compared to families that do not own such appliances. Households in the lowest income quantiles are the most negatively affected, as they spend a large share of their income on electricity.