Acceso usuarios
Energy poverty is affecting a growing number of households in France, and a certain number of subsidies such as the energy voucher have been introduced to help households in difficulty. The problem, according to the Cour des Comptes, is that these aids are not necessarily paid to the households that need them. This paper addresses the issue of predicting which households are in a situation of energy poverty. The empirical analysis is based on data from the Statistics on Resources and Living Conditions survey, a component of the EU-SILC survey. The sample is made of 15,644 household observations and the “inability of a household to keep home adequately warm” is used as an indicator of energy poverty. In the representative sample of the French population, 10.5% of households are concerned. The XGBOOST machine learning method was used to predict these households, based on predictors of housing characteristics, material deprivation declared by the household, income and personal characteristics of the household. The results show two income thresholds per equivalised consumption unit: €10,000 and €20,000, which could be used to identify 3 classes of household. The SHAP value increases up to €10,000, then decreases but remains positive up to the €20,000 threshold. The change in predicted probability is therefore positive and increasing up to €10,000, positive and decreasing between €10,000 and €20,000, then negative and decreasing. The paper also highlights the need for France to define a clear indicator of energy poverty, and the value of machine learning methods for predicting the households concerned.
