This article presents a bottom-up, data-driven method using multi-output Random Forest regression to quantify household-level income-related rebound effects and their greenhouse-gas impacts. It trains on the Swiss Household Budget Survey (HBS, ~38,975 households, aggregated to 19,000 after preprocessing) and predicts induced consumption changes for a cooperative housing dataset (COOP, ~1,500 households) by simulating a 500 CHF rent-derived increase in disposable income. Rebound expenses are translated to kg CO2-eq via a hybrid LCA (ecoinvent, EXIOBASE, Agribalyse). Key findings: induced spending concentrates on travel, housing and food; travel and housing have the highest emissions per CHF (~0.3–0.6 kg CO2-eq/CHF); feature importance identifies disposable income as dominant (FI≈0.66); model test R² averaged ~0.39 (test) and overall plausibility check for total income prediction gave R²≈0.81. The method is extensible to other sectors (energy efficiency, food-waste prevention, textiles/furnishing).
Questions this document answers
2022-02-15
Swiss Household Budget Survey (HBS): 38,975 households (processed to ~19,000 for training); COOP (housing cooperative): ~1,500 households.
500 CHF per month (average cooperative rent reduction vs free market in Zurich).
Multi-output Random Forest Regression; averaged R² ≈ 0.48 (training), 0.39 (testing); plausibility check for total income prediction R² ≈ 0.81.
Disposable income (feature importance FI ≈ 0.66).