Dynamic Pricing in the Linear Valuation Model using Shape Constraints
Authors: Daniele Bracale, Moulinath Banerjee, Yuekai Sun, Salam Turki, Kevin Stoll
TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Simulations and experiments with real-world data obtained by Welltower Inc (a major healthcare Real Estate Investment Trust) consistently demonstrate that our method attains lower empirical regret in comparison to several existing methods in the literature while offering the advantage of being tuning-parameter free. |
| Researcher Affiliation | Collaboration | Daniele Bracale EMAIL Department of Statistics University of Michigan Moulinath Banerjee EMAIL Department of Statistics University of Michigan Yuekai Sun EMAIL Department of Statistics University of Michigan Kevin Stoll EMAIL Welltower Inc. Salam Turki EMAIL Welltower Inc. |
| Pseudocode | Yes | Algorithm 1 Semi-Parametric Pricing Input: Time horizon T and length of the first epoch, τ1; the Hölder exponent α of S0 and the corresponding ν(α) defined in Equation (2); the minimum and maximum of price search range, pmin and pmax, H = pmax pmin; U defined in Equation (4). Set K = log (T/τ1) + 1 . for epoch k = 1, 2, . . . , K do |
| Open Source Code | Yes | 8 Code Availability The codes are available at https://github.com/dbracale/DP_via_Antitonic_TMLR_2025. |
| Open Datasets | No | This study applies our method to a real data set obtained by Welltower Inc to simulate the dynamic pricing process. The dataset consists of various characteristics and the transaction price for units in the United States (see Table 2 for more details). In our experiments, we present each rental unit to the dynamic pricing algorithm in a sequential fashion to simulate the dynamic pricing game. The unique aspect of the dataset is it includes the exact transaction price, which allows us to evaluate the regret of the algorithm directly. |
| Dataset Splits | No | We conducted 36 iterations, randomly shuffling the data before each run. |
| Hardware Specification | No | The paper does not explicitly mention any specific hardware specifications (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not explicitly mention any specific software dependencies with version numbers (e.g., Python version, specific library versions like PyTorch 1.9). |
| Experiment Setup | Yes | We set U = ( 1/2, 1/2) (known), the feature dimension d = 3 (known), the distribution of Xt Unif( p 2/3) d (unknown), and the coefficient θ 0 = (α0, β 0 ) (unknown), where α0 = 3, β0 = (2/3, 2/3, 2/3). We also choose pmin = 0 and pmax = 5 (known). The initial episode length was set to τ1 = 150, with subsequent episodes doubling in length according to τk = τ12k 1, for a total of K = 4 episodes. Each algorithm then chooses its exploration phase according to its rule. ... For our algorithm, we set α = 1. Prior to implementing the methods, we conducted cross-validation to tune the UCB algorithm s parameters λ and C2, as defined in Luo et al. (2022). We searched over a grid with (λ, C2) {0.1, 0.5, 1, 1.5, 2, 5} {5, 10, 15, 20, 30}. |