Optimal Pricing under Mixed Logit Choice

Ruben van de Geer, Sandjai BhulaiarXiv 2016 · DOI · arXiv

TL;DR

This paper tackles the complex problem of setting optimal prices for multiple products by incorporating a flexible customer choice model (mixed logit). By approximating the resulting revenue maximization problem, they demonstrate that accounting for customer heterogeneity and varied substitution patterns can significantly boost revenue compared to simpler models.

What Problem Does This Paper Address?

When a company sells multiple, similar products (differentiated products), changing the price of one product doesn't just affect its own sales. It also influences the demand for the other products due to customers switching between them (substitution effects). This makes finding the optimal prices to maximize revenue a significant challenge. Traditional models like the Multinomial Logit (MNL) model, while popular, have limitations in capturing the nuances of customer behavior, such as individual differences in how much they value product features (customer heterogeneity) and how they substitute between products. This paper aims to address this by using the more flexible Mixed Logit model to better represent customer choices and then finding the revenue-maximizing prices.

Methods

The core of this paper's methodology lies in tackling the computational complexity of the Mixed Logit model within a pricing framework. The customer choice probabilities in the Mixed Logit model involve a multidimensional integral that lacks a closed-form solution, making direct optimization difficult. To overcome this, the authors employ Laplace's method of integration to derive a closed-form approximation of these probabilities. This approximation allows them to then formulate and analyze the revenue maximization problem. For the single-product case, they investigate the log-concavity of the problem, a property that guarantees a unique optimal solution. In the multi-product setting, they rely on numerical experiments to explore the impact of the Mixed Logit model's flexibility on optimal pricing and revenue.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and numerical experiments with simulated data based on the proposed model.

Key Insights and Ideas

  • Mixed Logit is a Game Changer for Pricing: The paper strongly suggests that moving beyond simpler choice models like MNL to the Mixed Logit model can unlock significant revenue gains by better reflecting real-world customer behavior.
  • Customer Heterogeneity Matters: Ignoring that different customers value product attributes differently leads to suboptimal pricing. The Mixed Logit model, by allowing for random coefficients, captures this, and the numerical results show that higher heterogeneity can lead to higher optimal prices and revenue.
  • Flexible Substitution is Key: The MNL model's restrictive "Independence of Irrelevant Alternatives" (IIA) property is a major limitation. The Mixed Logit model allows for more realistic substitution patterns, which is crucial for accurate multi-product pricing.
  • Laplace Approximation is a Practical Tool: The paper demonstrates that Laplace integration can provide a tractable way to approximate the complex integrals in the Mixed Logit model, making it amenable to optimization.
  • Single-Product Log-Concavity is Conditional: While the single-product optimization problem isn't always log-concave, it can be shown to be strictly log-concave under certain conditions on the variance of customer preferences, offering theoretical guarantees for optimality.

Key Results

  • Single-Product Case: For a single product, the paper shows that customer heterogeneity (modeled by a normal distribution for quality valuation) can lead to higher optimal prices and revenue compared to the MNL model, especially as variance increases. In their example, the Mixed Logit-based price was higher and yielded more revenue.
  • Multi-Product Case: Numerical experiments with two products indicate that increased customer heterogeneity and negative covariance between product quality valuations (leading to more distinct preferences) generally result in higher attainable revenue. The optimal prices under the Mixed Logit model are also shown to be more differentiated than under MNL, which is more realistic for distinct products.
  • Approximation Accuracy: The Laplace approximation is shown to be reasonably accurate in a sub-asymptotic setting, providing a practical alternative to computationally intensive methods like Monte Carlo integration for optimization.

Limitations and Future Work

  • Approximation Limitations: The Laplace approximation, while useful, has a relative error that is not necessarily small in all cases, and its practical utility is observed in a sub-asymptotic setting.
  • Log-Concavity in Multi-Product Case: Proving log-concavity for the revenue function in the multi-product case with the Mixed Logit model is analytically challenging and remains an open question.
  • Product-Specific Price Sensitivities: The current model assumes a common price sensitivity across products. Future work could explore the impact of product-specific price sensitivities.
  • Assortment Optimization: The authors suggest that their framework could be extended to assortment optimization, where the decision is not just about price but also about which subset of products to offer.
  • Real-World Data: The numerical results are based on simulated data. Applying the model to real-world purchase data would be a valuable next step.

Summary updated

Abstract

In this paper we consider the problem of pricing multiple differentiated products. This is challenging as a price change in one product, not only changes the demand of that particular product, but also the demand for the other products. To address this problem, customer choice models have recently been introduced as these are capable of describing customer choice behavior across differentiated products. In the present paper the objective is to obtain the revenue-maximizing prices when the customer's decision making process is modelled according to a particular customer choice model, namely the mixed logit model. The main advantage of using the mixed logit model, also known as the random coefficients logit model, for this purpose is its flexibility. In the single-product case we establish log-concavity of the optimization problem under certain regularity conditions. In addition, in the multi-product case, we present the results of our extensive numerical experiments. These suggest that the mixed logit model, by taking unobserved customer heterogeneity and flexible substitution patterns into account, can significantly improve the attainable revenue.
BibTeX
@misc{geer2016optimal,
  author = {Ruben van de Geer and Sandjai Bhulai},
  title = {{Optimal Pricing under Mixed Logit Choice}},
  year = {2016},
  eprint = {1605.08340},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1605.08340},
}
Figure