Clean & dirty energy: complements or substitutes?
Selling solar to small shops in Lagos

Emilia Tjernström

with: Lori Beaman Joshua Deutschmann Lauriane Yehouenou

Transforming Energy Markets Seminar Series

2 November 2026

Motivation

Motivation: three guesses

  1. Lagos shops on the grid: how many days a week with power all day?

    0.4 days a week

  2. What share would pay the retail price for a small solar kit?

    Almost none

  3. Cheaper solar: more or less generator use?

    More generator use

Context: on the grid, rarely powered

  • 91% of shops were on the grid
  • 90% of them lacked a full day of power
  • 25% also used a generator
  • Generator users spent 6× as much on energy
  • Only 4% already owned solar

Context: Nigeria runs on small generators

  • An estimated 22 million small petrol generators in use

    (Access to Energy Institute and Dalberg, 2019)

  • About USD 12 billion a year to buy & run them

  • Firms: electricity a top obstacle; 33 outages a month

    (World Bank, 2014)

  • Petrol price: NGN 238 \(\rightarrow\) 546 a liter in one month (2023)

    (National Bureau of Statistics, 2023)

Question: does cheaper solar replace generators?

  • Small solar: quiet, clean, no fuel to buy
  • If solar substitutes \(\rightarrow\) \(\downarrow\) generator use & fuel
  • If solar complements \(\rightarrow\) generator use could \(\uparrow\)
  • The answer matters for subsidies & for emissions

What we do & preview of results

  • We offered a small solar kit to 401 shops at random prices
  • Shops valued it far below the retail price
  • Take-up: 57% in low-price clusters vs 5% in high-price
  • Cheaper solar \(\rightarrow\) \(\uparrow\) generator use, not \(\downarrow\)
  • Clean & dirty energy: complements here?

Design

Context: the shops & the kit

  • 401 shops in four Lagos markets
  • Kit: 4 lamps, 15 W panel, 40 Wh battery
  • Retail: NGN 140,000 (\(\approx\) AUD 150)
  • Arms: Control, Information, Free trial

Information treatment

Design: a lottery that reveals what shops would pay

  1. Shop names the most it would pay
  2. Shop draws a sealed envelope with a random price
  3. Price \(\leq\) bid \(\rightarrow\) shop buys at the drawn price
  4. Price \(>\) bid \(\rightarrow\) no sale

\(\rightarrow\) naming your true value is the best strategy

Design: neighbors drew similar prices

Prices by arm All four markets

Take-up: price drives purchases

Take-up maps

Results

Demand: almost no shop would pay the retail price

Stated vs auction

Generators: cheaper solar, more generator use

Phone surveys Endline
Ran a
generator		 Generator
days		 Fuel
spending		 Has a
generator		 Energy
spending
Drawn price (NGN 10,000) −0.015 −0.076** −0.135 −0.009 −0.003
(0.009) (0.033) (0.114) (0.008) (0.056)
Bought the kit (2SLS) 0.239 1.260** 2.209 0.158 0.046
(0.152) (0.537) (1.804) (0.139) (0.957)
Control mean 0.144 0.593 1.413 0.136 1.765
Observations 648 648 643 348 321
Firms 371 371 370 348 321

Fuel and energy spending in NGN ’000 a week; standard errors clustered by price cluster
Stars from randomization inference: * 10%, ** 5%, *** 1%

Equations Permutation test Leave one out Specifications

Generators: the pattern shows up across energy measures

Drawn price (NGN 10,000) Bought the kit (2SLS)
Spending Generator use All five Spending Generator use All five
All shops −0.020 −0.046** −0.043** 0.363 0.836** 0.791**
(0.016) (0.022) (0.020) (0.278) (0.383) (0.357)
N 391 393 393 391 393 393
Confirmed payment −0.048*** −0.083*** −0.074*** 0.481*** 0.841*** 0.746***
(0.014) (0.022) (0.020) (0.157) (0.251) (0.224)
N 238 239 239 238 239 239

Indices in SD units; standard errors clustered by price cluster
Stars from randomization inference: * 10%, ** 5%, *** 1%

Mechanism: the kit powered lights; generators powered fans

  • 77% of shops powered a fan
  • 93% of generator users ran a fan on it
  • Buyers used the kit for lights (100%), rarely a fan (7%)
  • 83% would run a fan or AC on a bigger system

Takeaways for energy markets

  • Shops valued small solar far below its retail price
  • Subsidies raised take-up a lot
  • …but cheaper solar didn’t displace the generator
  • Generators powered fans; the kit powered lights

Appendix

Estimating equations

\[ \text{Bought}_i = \pi \, \text{Price}_i + \gamma \, Y_{i0} + \delta_{m(i)} + \lambda_{s(i)} + u_i \]

\[ Y_i = \beta \, \text{Price}_i + \gamma \, Y_{i0} + \delta_{m(i)} + \lambda_{s(i)} + \varepsilon_i \]

\[ Y_i = \theta \, \widehat{\text{Bought}}_i + \gamma \, Y_{i0} + \delta_{m(i)} + \lambda_{s(i)} + e_i \]

  • \(\text{Price}_i\): shop’s drawn price; \(Y_{i0}\): baseline outcome
  • \(\delta_{m}\): market; \(\lambda_{s}\): saturation arm
  • Standard errors clustered by price cluster

Back

Inference: permutation test for generator days

Back

Robustness: dropping one cluster at a time

Back

Robustness: alternative specifications

Back

Demand: stated vs auction willingness to pay

Back

Design: drawn prices by saturation arm

Back

Design: the information treatment

Back

Design: saturation arms in all four markets

Back

Take-up in all four markets

Back