Responsiveness of low-income households to hybrid price/non-price policies in the presence of energy shortages: evidence from Colombia

Abstract

At the beginning of 2016, Colombia was experiencing an energy shortage, and in order to avoid mandatory power cuts, the government launched an unexpected hybrid price/non-price energy-saving policy. In this paper, I evaluate how low-income households in a major Colombian city respond to this policy. Using hourly household electricity consumption data, I find that, on average, households reduce electricity consumption by 4.5% as a result of the policy. It is striking that even low-income households, who consume relatively small amounts of electricity, respond to energy-saving policies and engage in conservation behaviors in the short term. In my analysis, I also find that the effect is stronger the higher the household pre-treatment electricity consumption levels and smaller among poorer households. However, the heterogeneity in terms of income level vanishes once I control for household pre-program electricity consumption levels. Finally, my point estimates are comparable to the impact estimates of policies that are similar to the one I analyze in this paper.

Appendices

Appendix 1. Example of electricity bill

Fig. 6

Photo of a bill from EMCALI

Appendix 2. Alterative empirical approach

Another way to check my findings is to estimate the following regression for “T = 0” and “T = 1” separately,

$$ {e}_{iwdh}={\alpha}_{idh}+{X}_{wdh}\beta +{\varepsilon}_{iwdh} $$

Then, using the estimates from each period separately, I use data from “T = 1” to calculate the average of the predicted consumption of electricity. That is, I use data from “T = 1” and the estimates of the model at “T = 0” to predict ${\widehat{e}}_{iwdh}$ and then calculate $\overline{{\widehat{e}}_{wd}}$ for each “w” and “d” of “T = 1.” Then, I do the same but using the estimates of the model at “T = 1”. Finally, I plot them in Fig. 7, which shows that the predictions for “T = 1” using the model from “T = 0” are, for all the days after the policy implementation day, higher than the predictions for “T = 1” using the model from “T = 1.” Actually, the mean difference in the predictions is 0.0063 kWh while the median is 0.007 kWh, which are equivalent to a reduction in electricity consumption of 3.6 and 4%, respectively. These results are a little bit lower than the estimates from both of my empirical approaches, but completely in line with them.

Fig. 7

Predicted consumption for each day after the treatment

Appendix 3. Additional figures and tables

Fig. 8

Percentage changes by pre-consumption decile (I do not include the estimated percentage change for the first decile because, given the pre-treatment consumption was close to 0, it was so large)

Fig. 9

Percentage changes by % of missing values

Fig. 10

Number of light-days-hours

Fig. 11

Temperature in Cali

Table 16 Estimated effect of the policy by income level