Forecasts: Consumption, Production, and Behavioral Responses

How to cite

Rezaee, Arman; Gibson, Matthew; Ahmad, Husnain F.; Nasim, Sanval; Nadeem, Fatiq. 2019. "Forecasts: Consumption, Production, and Behavioral Responses." AEA RCT Registry. AEARCTR-0004343. https://doi.org/10.1257/rct.4343-1.1

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Arman Rezaee* , Matthew Gibson , Husnain F. Ahmad , Sanval Nasim , Fatiq Nadeem

* Primary investigator

Registry ID
AEARCTR-0004343
DOI
https://doi.org/10.1257/rct.4343-1.1
First registered
Last updated
Status
Completed
Trial dates
2019-04-01 – 2019-10-31
Countries
Pakistan ()
Registry record
http://www.socialscienceregistry.org/trials/4343

Abstract

Economic theory predicts that forecasts are an important determinant of welfare. In developing countries, however, agents may have difficulty forming accurate, precise forecasts because of limited information and human capital. This plausibly limits the scope for optimal responses to uncertain future events. We study the effect of two randomized interventions on forecast formation and behavioral responses. The first is the provision of day-ahead air pollution forecasts. The second is training in forecasting techniques aimed at reducing behavioral biases. We estimate impacts on forecast error in air pollution and travel times. Measured responses include willingness to pay for protective face masks and changes in time use. We examine effects on proxies for the variance of utility, which broadly reflects forecasting and responses to uncertainty. Finally we elicit willingness to pay for our forecasts, an important input to cost-benefit analysis of air pollution monitoring.

Intervention

We implement a randomized controlled trial to study people in their roles as consumers and producers of forecasts. The research design includes two orthogonal treatments. The first entails provision of day-ahead air pollution forecasts delivered by text message. The second entails in-person training designed to reduce behavioral biases in forecasts.

See pre-analysis plan for additional details.

Experimental design

Our sample is approximately 1100 households in two lower-middle-income neighborhoods in Lahore, Pakistan. Households are individually randomized into four equal-sized groups: a control arm, and three treatment arms (forecasts only, forecast training only, and both). Following a baseline survey, we buffer for approximately a one month period after which we implement our treatment. We develop two treatments: (i) an air pollution information clearinghouse that uses text messages and a forecast model to inform citizens about air pollution forecasts, and (ii) forecast training session. Our information treatment involves a system developed by the research team with support from OpenCodes that combines a forecast model, a GSM mobile connection and an API based system to treat our respondents with air pollution forecast information every day for a period of 3 months. Whereas the forecast training is implemented once to every househould over a period of one month following the SMS message intervention.

See pre-analysis plan for additional details.

Design details

Randomization method
Blocking and randomization will be performed in R using the tools in the blockTools package. See pre-analysis plan for additional details.
Randomization unit
Individual, blocked on several baseline outcomes. See pre-analysis plan for additional details.
Clusters
1100 households
Observations
1100 households
Arms
275 control households, 275 forecasts only households, 275 forecast training only households, 275 both treatments households.
Primary outcomes
Broadly, we are interested in three types of outcomes: 1) consumption, e.g. demand for our forecast product; 2) production, e.g. error in forecasting the time required for a future journey; and 3) behavioral responses, e.g. demand for particulate-filtering face masks. In theory these two treatments could be complements or substitutes. While our design will allow us to measure this interaction, it is not our primary focus. See pre-analysis plan for additional details.
Secondary outcomes
See pre-analysis plan.

Cited by

No citation data available for this registration yet.

Keywords