Add study_014: Davis et al. (2011) — Optimal Reserve Prices in Auctions

studies/study_014/README.md

@@ -0,0 +1,58 @@
+# Do Auctioneers Pick Optimal Reserve Prices?
+
+**Authors:** Andrew M. Davis, Elena Katok, Anthony M. Kwasnica
+**Year:** 2011
+
+---
+
+## Description
+
+This study investigates how auctioneers set reserve prices in second-price sealed-bid auctions (described to sellers as English auctions). A well-established theoretical result, assuming risk neutrality of the seller, is that the optimal reserve price should not depend on the number of participating bidders (Roger B. Myerson 1981). In a set of controlled laboratory experiments, seller behavior often deviates from this theoretical benchmark.
+
+Specifically, this study tests the Cuberoot distribution treatment with bidder counts in {1, 4, 7, 10} under the NoInfo condition. Computerized bidders follow the weakly dominant strategy of bidding their private valuations, drawn from the Cuberoot distribution F(v) = (v/100)^(1/3) with mean 25 and support [0, 100]. The seller's personal valuation is zero, and the risk-neutral optimal reserve price is 42.
+
+The key finding: sellers systematically increase their reserve prices as the number of bidders grows (Pearson r = 0.42, p < .001), contrary to standard auction theory.
+
+LLM simulation baselines for this experiment are done and reported in [Feng et al., "Noise, Adaptation, and Strategy: Assessing LLM Fidelity in Decision-Making," EMNLP 2025](https://aclanthology.org/2025.emnlp-main.391.pdf). Baselines from GPT-4o and Claude Sonnet models are available in `benchmark/`.
+
+## Participants
+
+- **N:** 40
+- **Population:** University students
+
+## Replicated tests (human data)
+
+Due to the unavailability of the original human experiment data (Davis et al. 2011), we use the replicated experiment data (Davis et al. 2023) instead.
+
+- **Pearson correlation (reserve price vs. number of bidders)**
+  Reported: r = 0.42, p < .001
+  Significance level: 0.05
+
+## Effect sizes & auxiliary statistics
+
+- F1: Mean reserve prices by number of bidders (Cuberoot distribution, NoInfo):
+  1 bidder: 14.85 (SD = 19.15, n = 716),
+  4 bidders: 24.30 (SD = 17.90, n = 566),
+  7 bidders: 32.93 (SD = 21.25, n = 458),
+  10 bidders: 39.50 (SD = 25.12, n = 660).
+  Overall: mean = 27.31, SD = 23.23, sell-through rate = 0.74, mean profit = 32.83.
+  Risk-neutral optimal reserve price = 42.
+  Raw data present in `ground_truth.json`.
+
+## Files
+
+**source/**
+- ground_truth.json
+- materials/auction_instructions.json
+- metadata.json
+- specification.json
+
+**scripts/**
+- config.py
+- evaluator.py
+- stats_lib.py
+- study_utils.py
+
+**benchmark/**
+- human_data/auction_human_data.csv
+- llm_data