Switchback Price Experiments with Forward-Looking Demand

Yifan Wu,Ramesh Johari,Vasilis Syrgkanis,Gabriel Y. Weintraub

2024-10-19

Abstract:We consider a retailer running a switchback experiment for the price of a single product, with infinite supply. In each period, the seller chooses a price $p$ from a set of predefined prices that consist of a reference price and a few discounted price levels. The goal is to estimate the demand gradient at the reference price point, with the goal of adjusting the reference price to improve revenue after the experiment. In our model, in each period, a unit mass of buyers arrives on the market, with values distributed based on a time-varying process. Crucially, buyers are forward looking with a discounted utility and will choose to not purchase now if they expect to face a discounted price in the near future. We show that forward-looking demand introduces bias in naive estimators of the demand gradient, due to intertemporal interference. Furthermore, we prove that there is no estimator that uses data from price experiments with only two price points that can recover the correct demand gradient, even in the limit of an infinitely long experiment with an infinitesimal price discount. Moreover, we characterize the form of the bias of naive estimators. Finally, we show that with a simple three price level experiment, the seller can remove the bias due to strategic forward-looking behavior and construct an estimator for the demand gradient that asymptotically recovers the truth.

Computer Science and Game Theory,Econometrics,Theoretical Economics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the problem of bias in estimating the demand gradient due to consumers' forward - looking behavior when retailers conduct price experiments. Specifically:

1. **Problem Background**:
- Retailers learn consumers' price sensitivity (i.e., the demand gradient) by setting different prices to adjust the reference price and optimize revenue.
- In the traditional two - price switching experiment, retailers switch between "high" and "low" prices and calculate the discrete derivative to estimate the demand gradient.

2. **Impact of Forward - Looking Consumer Behavior**:
- Consumers can foresee possible future discounts, so they may choose to delay their purchases and wait for lower prices.
- This behavior leads to bias in estimating the demand gradient: during the "high" - price period, consumers' delayed purchases will underestimate the actual demand; during the "low" - price period, consumers' early purchases will overestimate the actual demand.

3. **Research Objectives**:
- Quantify the impact of forward - looking behavior on the demand gradient estimation.
- Prove that in the standard two - price switching experiment, this bias cannot be completely eliminated.
- Propose an improved three - price switching experiment design and its corresponding simple estimator to eliminate the bias caused by forward - looking behavior.

### Main Contributions

1. **Model Construction and Bias Quantification**:
- A simple and general model is introduced to quantify how consumers' patience level (discount factor γ) affects the estimation bias.

2. **Limitations of the Two - Price Switching Experiment**:
- It is proved that even in an infinitely long experiment, the two - price switching experiment cannot accurately identify the true demand gradient.

3. **Three - Price Switching Experiment Design and Estimator**:
- A simple three - price switching experiment design is proposed, and a corresponding estimator is given, which can completely eliminate the bias brought by forward - looking behavior.
- Specifically, by introducing three price points \( p, p-\epsilon, p - 2\epsilon \) and constructing an estimator by the difference method:
\[
\hat{d}=\frac{\Delta_1-(\Delta_2-\Delta_1)}{\epsilon}
\]
where \(\Delta_1\) is the average same - day sales difference between prices \(p-\epsilon\) and \(p - 2\epsilon\), and \(\Delta_2\) is the average same - day sales difference between prices \(p\) and \(p-\epsilon\).

### Conclusion

The main purpose of this paper is to understand and solve the problem of bias in estimating the demand gradient due to consumers' forward - looking behavior, and propose an improved experiment design and estimation method, enabling retailers to more accurately estimate the demand gradient in price experiments and thus better optimize the pricing strategy.