Auction Catalogue Narratives, Moral-Historical Framing, and Auction Outcomes for Adrian Ghenie Lots: Data and Code
Description
This repository contains data, code, and replication materials for a qualitative-led mixed-method study of how Christie’s and Sotheby’s publicly narrate morally burdened historical references in Adrian Ghenie auction lots, and how such framing corresponds to observable auction outcomes. The dataset links public, author-archived lot materials to structured lot-level variables including auction house, sale date, lot number, title, dimensions, creation year, estimates, realized prices, currencies, catalogue text, and moral-language coding fields. The corpus covers sales from approximately 2013–2025 and was collected from public online catalogues in November 2025. No private client communications, internal notes, private negotiations, or other non-public documents are included. We coded moral-historical framing using a transparent, dictionary-assisted approach. Variables include moral_flag, moral_token_count, moral_terms_hit, and moral_intensity, an original ordinal 0–3 indicator of the most intense moral-historical reference identified in each lot text. It is descriptive and study-specific, not a psychological scale or measure of buyer response, trauma severity, ethical quality, or inherent artistic meaning. The analytic sample contains 106 lots. The revised econometric analysis accounts for catalogue-text length. catalogue_word_count measures the whitespace-delimited length of the archived sale-page text, while log_catalogue_word_count enters the principal specifications as a control. Because raw moral_token_count is mechanically related to text length, it is retained as a frequency/audit variable rather than interpreted as textual density. moral_tokens_per_1000_words provides a length-normalized robustness measure. The updated script, ghenie_econometrics_textlength_categorical.py, corrects the documented extraction error for Sotheby’s The Sunflowers in 1937, audits currency consistency, constructs the analytic sample, derives the revised variables, estimates the principal models, and produces diagnostic and robustness outputs. The quantitative analysis is exploratory and associational, not causal. moral_intensity is the focal textual predictor. Robustness checks include currency fixed effects, winsorization, moral_flag, influential-case analysis, GLM replication, auction-house interaction, length-normalized moral-language frequency, and categorical specifications of moral_intensity. The revised evidence does not provide robust support for an independent association between moral-historical intensity or length-normalized moral-language frequency and auction performance once catalogue-text length is accounted for. Apparent positive associations in selected unadjusted specifications attenuate after this adjustment. The results therefore do not support a stable independent “moral premium”; rather, the quantitative analysis bounds the qualitative argument about how traumatic history becomes part of the public narrative architecture of luxury valuation.
Files
Steps to reproduce
Download the repository files and place Ghenie_auctions_clean.csv and ghenie_econometrics_textlength_categorical.py in the same working directory, or manually specify their paths. Required Python packages: pandas, numpy, statsmodels, and matplotlib. Run: python ghenie_econometrics_textlength_categorical.py By default, the script reads Ghenie_auctions_clean.csv and creates the output folder ghenie_econometrics_results_textlength. Alternatively: python ghenie_econometrics_textlength_categorical.py --input path/to/Ghenie_auctions_clean.csv --output-dir path/to/ghenie_econometrics_results_textlength The input CSV contains lot-level auction, artwork, price/estimate, catalogue-text, and moral-language variables. The script applies the documented correction for Sotheby’s The Sunflowers in 1937, audits price/estimate currency consistency, restricts estimate-relative analyses to matched-currency observations, and constructs the 106-lot analytic sample. The script derives log_price, log_est_low, premium_ratio, log_premium_ratio, premium_dummy, log_area, catalogue_word_count, log_catalogue_word_count, and moral_tokens_per_1000_words. The latter is calculated as: 1000 × moral_token_count / catalogue_word_count Principal models use moral_intensity as the focal textual predictor and control for log_catalogue_word_count, logged artwork area, creation year, and auction house. The exploratory price model additionally includes logged low estimate and sale-currency indicators. OLS models use HC3 robust standard errors; logistic models use robust covariance. All results are associational, not causal. Robustness checks include currency fixed effects, winsorization, moral_flag, influential-case exclusion, binomial GLM replication, auction-house interaction, length-normalized moral-language frequency, categorical moral_intensity specifications, joint Wald tests, and legacy-versus-length-adjusted comparisons. Main outputs include the cleaned model dataset, descriptives, coefficient and fit tables, model summaries, currency and correction audits, influence diagnostics, odds ratios, text-length diagnostics, normalized-frequency and categorical-intensity robustness files, legacy-versus-adjusted comparisons, and diagnostic figures. Researchers who only want to reproduce the econometric analysis can run the script directly using the supplied Ghenie_auctions_clean.csv.
Institutions
- West University of TimişoaraTimiș County, Timișoara