Approximation in Complex Pricing Mechanisms

How to cite

Michler, Jeffrey; Slade, Peter; Nuño-Ledesma, José; Wu, Steven. 2022. "Approximation in Complex Pricing Mechanisms." AEA RCT Registry. AEARCTR-0008966. https://doi.org/10.1257/rct.8966-1.0

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

Jeffrey Michler* , Peter Slade , José Nuño-Ledesma , Steven Wu

* Primary investigator

Registry ID
AEARCTR-0008966
DOI
https://doi.org/10.1257/rct.8966-1.0
First registered
Last updated
Status
In development
Trial dates
2022-09-15 – 2022-12-09
Countries
United States of America (Arizona)
JEL codes
L11, C91, D9
Registry record
http://www.socialscienceregistry.org/trials/8966

Abstract

Recent research in mechanism design has shown that simpler pricing strategies can closely approximate the profits obtained by more complex profit maximizing pricing mechanisms. But these simpler strategies are highly sensitive to the firm’s knowledge of the distribution of demand for each good. This has restricted empirical research on the topic to case studies and limited the application of these pricing strategies. The question remains, how well do simpler pricing strategies approximate the profit maximizing mechanism? We propose to use a lab experiment to determine how simpler pricing strategies perform when gaps exist in firm knowledge regarding the demand for goods and the correlation between consumers’ reservation values for goods.

Intervention

Participants in the experiment act as multi-product monopolies selling goods or bundles of goods into a market of automated computer buyers (consumers). Our interest in the experiment is how less complex pricing mechanisms perform relative to the profit maximizing mechanism when firm knowledge of market demand is incomplete.

Our intervention randomizes what information a participant has regarding market demand. We have four different interventions/treatments:

1) Perfect information: Participants are provided with information regarding both the underlying distribution of consumer valuations and the correlation between consumers' valuation for the goods.

2) Unknown distribution: Participants are provided with no information regarding the underlying distribution of consumer valuations.

3) Unknown correlation: Participants are provided with no information regarding the correlation between consumers' valuation for the goods.

4) No information: Participants are provided with no information regarding both the distribution and the correlation.

Experimental design

A player in the lab experiment will act as a firm trying to decide the prices to set for the various goods that they sell. The marginal cost for all goods is zero. The "consumers" that the player is trying to sell to are a set of 10,000 computer buyers or bots that are pre-programmed to purchase goods or bundles of goods if the price is less than or equal to that bot's preset valuation. Firms are monopolistic, so each player is the sole seller in her market and does not compete against other players. We constrain the prices a player can set and the reservation values for the bots to be integers. The goal for the player is to set prices in order to maximize revenue (profit). Play takes place over 40 of rounds. Payouts will be calculated as a fixed percentage of average revenue earned across three randomly selected rounds.

There are four parameters which we will experimentally manipulate in order to induce random variation in the markets that a given play i will face.

1) Pricing Strategy. Through the course of the game, all players will play each of the four pricing strategies and be asked to make decisions about how to price their goods using that strategy. The game will take place in 4 sub-games with 10 rounds in each sub-game. In a sub-game the player is assigned, at random, one of the four pricing strategies. They then use that strategy for all 10 rounds in the sub-game. After the sub-game, the player is then assigned, at random, to a second of the pricing strategies. This continues until the player has played with all 4 pricing strategies.

2) Number of goods sold by the firm. In the experiment, there will be four possible goods that a firm could sell, denoted by color (blue, green, yellow, and red). Before the start of the game each player will be told the number of goods they are selling (2 through 4). For a given number of goods, all players selling that number of goods will be selling the same color goods.

3) Distribution of consumer valuations. At the start of the game, a player will be randomly assigned to a market in which consumer valuations are either uniformly distributed or follow a beta distribution. For the uniform distribution, valuations will follow U ~ (0,100), which has a mean of 50 and a standard deviation of 29. For the beta distribution, valuations will follow B ~ (5,5). Scaling the beta distribution by 100 gives the distribution a mean of 50 and a standard deviation of 15.

4) Correlation between consumer valuations. At the start of the game, a player will be randomly assigned to a market in which every consumer has the same correlation in their valuation for the goods on sale. Consumers will express either a negative correlation in their valuation for the goods or their valuations will be perfectly independent. Regardless of the distribution, the value for all goods will be drawn from distributions that have a correlation coefficient of -0.25.

With these four parameters, we can define four different markets which an individual player will be randomly assigned into: uniform-independent, uniform-negative, beta-independent, and beta-negative. We can also define three different production schedules which an individual player will be randomly assigned into: two goods, three goods, and four goods. This gives us 12 market-schedule combinations.

At the end of every round the participant will be asked to rate how they currently feel using a slide bar that displays different emojis.

Design details

Randomization method
Randomization of treatment is done by the oTree interface when participants log in to play the experiment.
Randomization unit
Randomization is done at the participant (individual) level
Clusters
1200 individuals.
Observations
1200 individuals.
Arms
300 individuals for perfect information treatment, 300 individuals for the unknown distribution treatment, 300 individuals for the unknown correlation treatment, and 300 individuals for the unknown distribution and unknown correlation treatment.
Primary outcomes
We have six primary outcomes: 1) Profitability: the strategy that attains the most money in the experiment. 2) Adaptability: the most profitable strategy across markets. 3) Effectiveness: the share of the maximum achievable profit obtained across all rounds and all markets. 4) Convergence: the most profitable strategy across rounds. 5) Maximin: the strategy that limits losses (has the largest minimum). 6) Variability: the strategy that minimizes the variance in profits.
Secondary outcomes
We have two sets of secondary outcomes. First is the decision-making process, measured as a) response time and b) number of values entered in a price field. Second is emotional response measured after each question using a slide bar and emojis.
Sponsors
Name: Social Sciences and Humanities Research Council (SSHRC) Location: Canada Url: https://www.sshrc-crsh.gc.ca/

Cited by

No citation data available for this registration yet.

Keywords