The Cost of Competition in Mwanza, Tanzania

I subsidize the restocking cost of dona maize flour, a staple food, at retail firms that sell dona maize flour contingent on them selling it at a given, low price and making the price publicly known.

This experiment allows me to measure the firm-specific cost of competition and study how firms react to selling at below-equilibrium prices. I use this experiment in the first stage in a secondary study of how firms react when their competitors post below-equilibrium prices.

2024-08-28

2024-09-27

a. Firm-level profit margin needed to publicly post and sell at a low price relative to prices around them.
b. Price of dona maize flour, the subsidized food.
c. Number of customers per day who buy dona maize flour.
d. Volume of dona maize flour sold per day.
e. Input cost of dona maize flour
f. Profit margin of dona maize flour

I will measure heterogeneity in outcome variables with respect to baseline dona maize flour price, input cost, and profit margin, the combination of price and input cost.

a. Total revenue per day.
b. Total number of customers per day.
c. Whether prices of dona maize flour are publicly posted.
d. Change in beliefs about customer demand elasticity for dona maize flour.
e. Change in beliefs about prices of dona maize flour at nearby firms.
f. Price of sembe maize flour, a more-expensive substitute that is unsubsidized.
g. Number of customers per day who buy sembe maize flour.
h. Input cost of sembe maize flour.

I offer firms that already sell dona maize flour a subsidy contingent on them selling it at and publicly posting a given low price (subsidy price). The subsidy will last for two weeks. The subsidy price is relative to the firm’s belief of the lowest price per kilogram of dona maize flour within 100 meters of their firm; either it is 100 Tanzanian shillings or 200 Tanzanian shillings less than the lowest price. Whether a firm receives a subsidy is based on their elicited profit margin needed to sell at the subsidy price.

The subsidy amount is calculated based on the subsidy price, the profit margin the firm receives to sell at the given price, and the firm’s input cost of dona maize flour. The subsidy size reduces the firm’s input cost of maize flour such that they receive the profit margin at the subsidy price.

My subsidy allocation elicits the lowest profit margin a firm needs to publicly undercut nearby firms, which I call the firm-specific cost of competition. By giving every firm a chance to receive a subsidy, I measure the distribution of the cost of competition across the support of all firms in my sample.

I assign firms to receive subsidies with high probability or low probability to balance baseline characteristics along two comparisons. The first comparison increases similarity between subsidized firms and unsubsidized firms. The second comparison increases similarity between unsubsidized firms whose competitors receive subsidies and unsubsidized firms whose competitors do not receive subsidies. By balancing these two comparisons, I will use my subsidy assignment to study how firms perform when selling at below-equilibrium prices and study how firms react when their competitors post below-equilibrium prices.

Not available

All firms in the study have a chance to receive a subsidy. Whether firms receive a subsidy or not is decided in a Becker-DeGroot-Marschak (BDM) design or multiple price list (MPL) design, to be decided during piloting.

In the BDM design, firms are asked for the lowest profit margin they need to sell at each possible subsidy price. Firms then pull a bead out of a bag. Different possible profit margins are written on the beads, and the profit margin on the bead they pull is the profit margin offer. If the profit margin offer is the same or higher than lowest profit margin they stated to sell at the subsidy price assigned to the firm, they receive a subsidy to sell at the subsidy price for the profit margin offer. If the profit margin offer is lower than the lowest profit margin they stated to sell at the subsidy price assigned to the firm, they do not receive a subsidy.

In the MPL design, firms are asked, for a series of profit margins at each possible subsidy price, whether they would sell at that profit margin. Firms then pull a bead out of a bag. Different possible profit margins are written on the beads, and the profit margin on the bead they pull is the profit margin offer. If the firm said they would sell at the profit margin offer, they receive a subsidy to sell at the subsidy price for the profit margin offer. If the firm said they would not sell at the profit margin offer, they do not receive a subsidy.

I assign firms to either high-probability or low-probability bags of beads which differ in their distribution of profit margin offers. The profit offers in the bags are calibrated based on data collected during piloting. High-probability firms receive a subsidy with around 80% probability and low-probability firms receive a subsidy at around 20% probability.

I randomly assign firms to high- or low-probability using an algorithm that balances characteristics along two comparisons. The first comparison is between high-probability firms and low-probability firms. The second comparison is between low-probability firms near a high-probability firm and low-probability firms near a low-probability firm. Balance along the first comparison increases similarity between subsidized firms and unsubsidized firms. Balance along the second comparison increases similarity between unsubsidized firms whose competitors receive subsidies and unsubsidized firms whose competitors do not receive subsidies.

I run this double balance assignment algorithm on a computer. This assignment is done at the firm level and balances four variables collected during firm listing: The firm’s price of one kilogram of dona maize flour, whether the firm sells sembe maize flour, whether the firm is in or adjacent to a market, and the ward in which the firm is located.

Further, I randomly assign firms to be offered a subsidy price that is either 100 or 200 Tanzanian shillings less than the firm’s belief of the lowest price for one kilogram of dona maize flour within 100 meters of their firm. I do this at the ward level on a computer. Wards are randomly assigned to subsidy prices based on the number of firms they have, so that in expectation, 50% of firms are offered subsidy prices 100 Tanzanian shillings less than their belief of the lowest price near them and 50% are offered subsidy prices 200 Tanzanian shillings less than their belief of the lowest price near them.

The firm is the unit of randomization for whether they receive a subsidy with high probability or low probability.
The ward is the unit of randomization for the subsidy size offered to a firm.

No

500 firms

500 firms

300 firms will be offered a subsidy with high probability and 200 firms will be offered a subsidy with low probability. In piloting, 50% of firms were willing to accept a subsidy offer, so this allotment is intended create 150 subsidized firms and 350 unsubsidized firms.

Of all 500 firms, 250 will be offered a subsidy price that is 100 Tanzanian shillings per kilogram less than the lowest price per kilogram within 100 meters of their firm and 250 will be offered a subsidy price that is 200 Tanzanian shillings less than the lowest price per kilogram within 100 meters of their firm.

I calculate minimum detectable effects at the 95% significance level and 80% power for three primary outcome variables: Price of dona maize flour, number of customers purchasing maize flour per day, and total volume of maize flour sold per day. Data on prices of dona maize flour come from my firm listing exercise conducted July 2024. Data on customers per day and volume per day were collected in September 2022 in Mwanza from 207 firms selling dona. For comparing subsidized firms to unsubsidized firms, I assume 150 receive subsidies and 350 do not. I compare customers per day and volume sold per day, as the subsidy directly affects prices. My MDE for customers per day is 3.2 customers, which is 11% of the mean. My MDE for volume per day is 3.5 kilograms, which is 13% of the mean. All MDEs for this comparison are 16% of a standard deviation. For comparing unsubsidized firms whose competitors receive subsidies to unsubsidized firms whose competitors do not receive subsidies, I assume 350 firms in total are unsubsidized and 60% of them have their nearest neighbor subsidized. The 60% number is the comes from subsidy assignments yielded by the double-balance algorithm. I compare prices, customers per day, and volume sold per day. My MDE for price is 20 Tanzanian shillings, which is 2% of the mean. My MDE for customers per day is 3.6 customers, which is 13% of the mean. My MDE for volume per day is 3.9 kilograms, which is 15% of the mean. All MDEs for this comparison are 18% of a standard deviation.