Abstract:Selling a single item to $n$ self-interested buyers is a fundamental problem in economics, where the two objectives typically considered are welfare maximization and revenue maximization. Since the optimal mechanisms are often impractical and do not work for sequential buyers, posted pricing mechanisms, where fixed prices are set for the item for different buyers, have emerged as a practical and effective alternative. This paper investigates how many samples are needed from buyers' value distributions to find near-optimal posted prices, considering both independent and correlated buyer distributions, and welfare versus revenue maximization. We obtain matching upper and lower bounds (up to logarithmic factors) on the sample complexity for all these settings.
What problem does this paper attempt to address?
### Problems the paper attempts to solve
This paper aims to explore how many samples are required from the buyer value distribution in order to find a nearly optimal pricing strategy in the pricing mechanism of a single item. Specifically, the author studies the following problems:
1. **Sample complexity**: For independent and correlated buyer value distributions, as well as the two objectives of welfare maximization and revenue maximization, determine the required number of samples.
2. **Independent vs. correlated distributions**: Analyze whether there are differences between independent and correlated buyer value distributions.
3. **Welfare vs. revenue maximization**: Study whether there are differences between the welfare maximization and revenue maximization objectives.
### Research background
- **Auction theory**: Vickrey (1961) solved the welfare maximization problem, and Myerson (1981) solved the revenue maximization problem under independent distributions. Both won the Nobel Prize in Economics.
- **Practical application challenges**: Although there are optimal mechanisms, these auction mechanisms are often difficult to implement in practice, especially in online scenarios where buyers arrive one by one.
- **Effectiveness of simple mechanisms**: In the past two decades, it has been found that simple pricing mechanisms (such as the fixed - price mechanism) are very effective in practice and can provide nearly optimal results.
### Main problems
The core problems of the paper are:
- **Sample complexity**: How many samples are required from the buyer's value distribution to find a nearly optimal price?
- **Independent vs. correlated distributions**: Are there differences between independent and correlated distributions?
- **Welfare vs. revenue maximization**: Are there differences between welfare maximization and revenue maximization?
### Model description
- **Pricing mechanism**: The seller sets a fixed price \( \pi_i \) for each buyer. If the value \( V_i \) of buyer \( i \) is greater than or equal to \( \pi_i \), the buyer purchases the item.
- **Objective**: Maximize welfare (i.e., the item value of the winner) or revenue (i.e., the price paid by the winner).
- **Distribution assumption**: The buyer's value vector \( V=(V_1, V_2,\ldots, V_n) \) is drawn from an unknown but fixed distribution \( D \).
### Sample complexity problem
- **Learning algorithm**: Given a sample \( V \) independently and identically distributed (IID) drawn from an unknown distribution \( D \) and an error parameter \( \epsilon \), the learning algorithm needs to return a strategy \( \pi \) such that \( \pi(D)\geq\Pi(D)-\epsilon \).
- **Sample complexity**: The minimum number of samples required to ensure that the learning algorithm returns an \( \epsilon \)-approximate solution with a probability of at least \( 1 - \delta \).
### Main contributions
1. **Welfare maximization under product distribution**: The sample complexity is independent of the number of buyers \( n \) and is \( O\left(\frac{1}{\epsilon^2}\log^2\left(\frac{1}{\delta}\right)\right) \).
2. **Revenue maximization under product distribution**: Proved the lower bound of sample complexity of \( \Omega\left(\frac{n}{\epsilon^2}\right) \).
3. **Sample complexity under correlated distribution**: For correlated distributions, any learning algorithm requires \( \Omega\left(\frac{n}{\epsilon^2}\right) \) samples to return an \( \epsilon \)-approximate solution with a constant probability.
Through the study of these problems, the paper provides an important theoretical basis for understanding the sample complexity in the pricing mechanism.