← Back to abinael.xyz

REPS — Residential Retrofit Policy Simulator PROTOTYPE

Which policy gets Indonesia's Type 36 households to adopt the “Cooling Survival Kit”? Import data, set a policy, see adoption, cost-effectiveness & equity. · A. S. Lumempouw

1 · Data

10 income deciles loaded (illustrative, Indonesian ranges).

2 · Policy levers

Consumer subsidy
Green financing (subsidised loan)
Construction-standard mandate (new builds)
Behavioural nudge + boost

Behavioural & engineering assumptions

More assumptions (cost, savings, grid…)

Outcomes of this policy

Equity — who adopts & who benefits

Adoption rate by income decile
D1 = poorest, D10 = richest. I use this chart to see whether subsidy and financing actually reach the poorest deciles.
Welfare gain as a share of household income (per adopter)
Energy-burden relief. The poorest gain most relative to income — and a larger share of their gain is recovered comfort vs cash. (Absolute rupiah favours big users; income-share is the equity view.)

Compare the four instruments core of this research

Each instrument is run on its own at a representative setting. “Best” is highlighted per column. No instrument wins on every criterion — the point is to make the trade-off visible. Two honest footnotes: the mandate’s fiscal cost is ≈ 0 because its real cost is off-budget — borne by developers and buyers (the incidence is itself a research question); and the nudge setting is optimistic — field-experiment nudge effects are typically a few percentage points, so treat the nudge row as an upper bound.
About the model

I built this simulator to answer a question my engineering thesis could not: which policy instrument actually gets Type 36 households to adopt the retrofit? It links the Energy3D savings from my final degree project to a discrete-choice (logit) model of household adoption, then scores subsidies, green finance, mandates, and nudges on effectiveness, fiscal cost, and equity. This is a decision-support prototype — transparent and adjustable — not an econometrically estimated model.

For each income decile, I compute yearly net benefit as:
NB = cash saving + value of recovered comfort − annualised cost of the retrofit

  • Cash saving = (energy saving % × baseline use) × (1 − rebound) × tariff.
  • Recovered comfort = comfort-value × (energy saving × rebound × tariff). I model poorer deciles as taking more of the gain as comfort rather than cash — they were rationing use before the retrofit. Concretely, the rebound slider is scaled by decile: ρ_d = min(0.6, ρ × (1.4 − 0.08·i)), so the poorest decile rebounds ~1.4× the slider value and the richest ~0.7×. This rationed-comfort welfare term is central to how I read equity in the results.
  • Annualised cost = (retrofit cost after subsidy) × capital-recovery factor, at the loan rate when green financing is on, otherwise at the household's own (present-biased) discount rate. I annualise over a 15-year horizon — conservative, since the envelope measures last for decades.

Adoption follows a logit on a benefit-vs-affordability index z: P(adopt)=1/(1+e^(−z)). z rises with net benefit (scaled by income), falls with an adoption-friction term that captures behavioural barriers — split incentives, hassle, distrust, non-take-up — which is why uptake stays low even when the retrofit pays off on paper (the energy-efficiency gap). A nudge partly offsets that friction; paying upfront without financing triggers a liquidity penalty; a mandate forces adoption on the new-build share.

Aggregation rolls up across the housing stock: adoption, kWh and CO₂ saved, fiscal cost, cost per kWh and per ton CO₂, and a social benefit–cost ratio. I add the social cost of carbon on avoided emissions; subsidies count as transfers, not social costs.

Data and sources: the 57% energy saving and retrofit cost come from my Energy3D thesis (Lumempouw 2026); tariff IDR 1,444.70/kWh (PLN 2024); grid emission factor from IEA/ESDM for Indonesia's coal-heavy grid. The income deciles loaded by default are illustrative — Indonesian urban ranges I set for exploration. You can import real household data via CSV.

Caveats I want visible: the logit scale is assumed, not estimated from revealed-preference data (that would be the next research step); deciles are representative, not from a survey; comfort valuation is a parameter, not measured directly. I left every assumption on a slider so the sensitivity shows up in the outputs.