Misperceptions and Product Choice�Evidence from a Randomized Trial in Zambia

Jie Bai, Harvard Kennedy School

David Sungho Park, KDI School of Public Policy and Management

Ajay Shenoy, U.C. Santa Cruz

This document is an output from the research initiative ‘Private Enterprise Development in Low-Income Countries’ (PEDL), a programme funded jointly by the by the Centre for Economic Policy Research (CEPR) and the Foreign, Commonwealth and Development Office (FCDO), contract reference PEDL_LOA_9507_Shenoy. The views expressed are not necessarily those of CEPR or FCDO.

This project was supported in part by a grant from the Institute for Social Transformation at the University of California, Santa Cruz.

We also gratefully acknowledge financial support from:

Information is crucial for firm productivity

• Even large firms in the US lack / will pay for information
• And make costly mistakes
• Dickstein and Morales, 2018; Tanaka et al., 2020

​

Information constraints bind even more for informal micro-enterprises

• May not know basic information about market demand
• Namely: which products are profitable
• May hold inaccurate beliefs
• I think onions are unprofitable
• I never stock onions
• I never observe true profits
• I never update my beliefs
• [Repeat]

This Project

Do inaccurate beliefs keep shops from stocking profitable products?

1. ML algorithm: identify products stocked by some but not all otherwise similar shops
• Those who don’t stock believe returns are lower than in reality
2. Run RCT to reimburse shops for 1 or 2 weeks of stocking
• Do they keep stocking after reimbursements end?
• Do they earn similar returns to shops that stock all along?

​

Preview of results: yes.

Outline

1. Motivation
2. Identifying Potential Suboptimality in Stocking
3. Experimental Design
4. Results
5. Discussion & Literature

Goal: Choose Very Specific Products Stocked by Some, but Not All

• “Target” products
• Stores similar in main products
• But target product is only stocked by ~30%

​

• Can compare onions to onions
• Harmonize goods and units
• Prices are comparable
• Supplier to one shop can supply another

​

Data: Samaniego de la Parra and Shenoy (2025)

• Sept 2022: Screener (3000 shops)
• Locations
• Images of Inventory

​

• Jan—Mar 2023: In-Person Survey (1000 shops)
• Sourcing (regular suppliers and markets)
• Main products
• Target products
• Lots more….

Identifying Product Choice

2 Zambia-based enumerators working separately & reconciled by third

​

Identifying Product Choice

Vectorization, Step 1: Full-Dimensional Dummies

Vectorization, Step 2: Dimensional Reduction through Principal Component Analysis

Credit: Casey Cheng, Towards Data Science

Vectorization, Step 2: Dimensional Reduction through Principal Component Analysis

What are the component axes?

Clustering Shops by Product Choice

Credit: java T point

Challenges

There is considerable variation in products stocked within each cluster

Survey Module on Target Products Shows Evidence of Misperceptions

• If stock: retail price, order price, conversion => Markup
• If not stock: what do you think you would charge/pay for ____ of product?

Shops that don’t stock expect

• Lower returns
• Returns comparable to main goods

​

Shops that DO stock:

Target product is more profitable

Could be consistent with misperceptions…

BUT…

Is this a misperception…or just rational sorting?

True potential markup

Belief about markup

Outside option

If they stock: Observe truth, update beliefs

But why stock? I believe it’s less profitable than the outside option

Misperception hypothesis:

Is this a misperception…or just rational sorting?

Rational Sorting Hypothesis:

True distribution of potential markups

Stock

Don’t stock

Outside option

People who don’t stock accurately believe they would earn lower markups

Experiment: Induce people to stock. Do they continue?

True potential markup

Belief about markup

Outside option

Experiment: Induce stocking, see if they keep stocking after reimbursements end

Misperception hypothesis:

Outline

1. Motivation
2. Identifying Potential Suboptimality in Stocking
3. Experimental Design
4. Results
5. Discussion & Literature

Intervention

Control

90

1-week subsidy

97

2-week subsidy

84

Recommend & Offer Reimbursement

Survey:

• Ask about 3 goods (including target good)
• Do NOT specify which is the target good

1 Week later: return to reimburse purchases

1 Week later: return to reimburse purchases

Experimental sample: 271 shops who

1. Are in one of the target product clusters
2. Report not stocking at baseline

“We are interested in studying the sales and profitability of [name of target product] for shops like yours. We would like you to stock the product for one (two) week(s), and we will reimburse you for the cost

​

…you may continue to stock the product at your own expense if you believe it is profitable, or you may stop stocking the product if it is not. We encourage you to do whatever is best for your business. It will not affect your eligibility for any future study.”

Data

Overall Timeline

Timeline of the Experiment

Outline

1. Motivation
2. Identifying Potential Suboptimality in Stocking
3. Experimental Design
4. Results
5. Discussion & Literature

Treatment has a persistent impact on stocking the target product

Pre-registered primary outcome: stocking in last rounds

Treatment has a persistent impact on stocking the target product

Pre-registered primary outcome: stocking in last rounds

The effect is not just persistent unsold stock

Expenditures on target product

Theory: Rational Sorting Implies Treated Shops will have lower markups

Distribution of Markups

Control

Treated

CDF

Markup

Excess mass at the bottom

No evidence that treated shops earn low/unprofitable markups

Shops that didn’t stock at baseline are bad at predicting returns when they do start stocking

Suggestive Evidence that Treatment Improved Store-Level Outcomes

Alternative Hypotheses

• Strategic Response to Market Competition
• Market can only support N shops selling the good
• We change which shops are part of N
• Saturation test: no evidence that more nearby treated reduces stocking
• Fixed cost of stocking
• Continuous fixed cost can’t explain persistence
• One-time fixed cost (e.g. information) hard to reconcile with data
• 90% who don’t stock at Round 0 know where they could buy the product
• Most already visit the market where they ultimately buy the product

Outline

1. Motivation
2. Identifying Potential Suboptimality in Stocking
3. Experimental Design
4. Results
5. Discussion & Literature

Summary

• Evidence of misperceptions about profit
• Shops that don’t stock believe profits are lower than reported by shops that do
• RCT that induced shops to stock with reimbursements
• Persistent stocking long after reimbursements end
• No evidence of low profits for shops induced to stock
• Shops that don’t stock before the experiment can’t predict subsequent returns

​

Literature

• Information frictions
• Search: Jensen, 2010; Aker, 2010; Bergquist et al., 2024; Cai et al., 2024
• Large firms: Dickstein and Morales, 2018
• Small firms face information constraints on even basic choice: products to stock
• Learning
• Hindered by noisy signals (Conley and Udry, 2010; Bold et al., 2017)
• May not use available info (Hanna et al., 2014)
• Our setting: pessimism => no signals at all
• But small intervention has persistent impacts
• Different results from Banerjee et al. (2022)
• Different design, different setting

Thanks for listening!

​

Email questions / comments to

azshenoy@ucsc.edu

Bonus slides

Reimbursement Amounts

Main Results

Main Results (cont)

Can’t reject difference in means

Shop’s ex ante claim that the product is unprofitable does not imply lower markups

No Evidence of Strategic Incentives