Paper: arXiv 2512.23078

Authors: Jianping Mei, Michael Moses, Jan Waelty, Yucheng Yang

Abstract

We study how deep learning can improve valuation in the art market by incorporating the visual content of artworks into predictive models. Using a large repeated-sales dataset from major auction houses, we benchmark classical hedonic regressions and tree-based methods against modern deep architectures, including multi-modal models that fuse tabular and image data. We find that while artist identity and prior transaction history dominate overall predictive power, visual embeddings provide a distinct and economically meaningful contribution for fresh-to-market works where historical anchors are absent. Interpretability analyses using Grad-CAM and embedding visualizations show that models attend to compositional and stylistic cues. Our findings demonstrate that multi-modal deep learning delivers significant value precisely when valuation is hardest, namely first-time sales, and thus offers new insights for both academic research and practice in art market valuation.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper relies on standard deep learning architectures (ResNet, XGBoost, Neural Nets) rather than complex mathematical derivations. However, it excels in empirical rigor by utilizing a massive real-world dataset (1970–2024), strict out-of-sample forward-looking testing, and nuanced state-dependent analysis comparing fresh-to-market versus repeated sales.

Research Flowchart

  flowchart TD
    A[Research Goal: Improve Art Valuation via Multi-Modal Deep Learning] --> B[Data Inputs: Large Repeated-Sales Dataset<br/>Tabular Metadata & Artwork Images]
    B --> C[Methodology: Benchmark Hedonic Regressions & Tree Methods vs Multi-Modal Deep Models]
    C --> D[Computational Process: Fuse Tabular Features & Visual Embeddings + Grad-CAM Analysis]
    D --> E[Outcome 1: Artist Identity & History Dominate Overall Predictive Power]
    D --> F[Outcome 2: Visual Embeddings Significantly Enhance Valuation for First-Time Sales]

Cited References (70)

  1. Deep learning with long short-term memory networks for financial market predictions (2018)
  2. Stacked regressions (1996)
  3. Stacked generalization (1992)
  4. Tabular Data Models for Predicting Art Auction Results (2024)
  5. Unbiased Gradient Boosting Decision Tree with Unbiased Feature Importance (2023) — link
  6. Predicting property prices with machine learning algorithms (2021)
  7. A Review on Methods and Applications in Multimodal Deep Learning (2022) — link
  8. Empirical evidence of anchoring and loss aversion from art auctions (2023)
and 62 more
  1. Comparing Explanation Methods for Traditional Machine Learning Models Part 2: Quantifying Model Explainability Faithfulness and Improvements with Dimensionality Reduction (2022) — link
  2. Sale Rates and Price Movements in Art Auctions (2011)
  3. The “Death-Effect” in Art Prices: A Demand-Side Exploration (2000)
  4. Dropout: a simple way to prevent neural networks from overfitting (2014)
  5. Biased Auctioneers (2023)
  6. Vested Interest and Biased Price Estimates: Evidence from an Auction Market (2005)
  7. From local explanations to global understanding with explainable AI for trees (2020)
  8. A Unified Approach to Interpreting Model Predictions (2017) — link
  9. Prediction and Analysis of Artwork Price Based on Deep Neural Network (2022)
  10. The right price for art collectibles. A quantile hedonic regression investigation of Picasso paintings (2011)
  11. TorchVision: PyTorch’s Computer Vision library (2016) — link
  12. Art as an Asset: Evidence from Keynes the Collector (2020)
  13. PyTorch Image Models (2019) — link
  14. The art market, global economy and information transparency (2003)
  15. The computation of prices indices (2006)
  16. The Art Basel & UBS Art Market Report 2025 (2025) — link
  17. Principled approach to the selection of the embedding dimension of networks (2021)
  18. Can machine learning predict the price of art at auction? (2020)
  19. Reading between the lines in the art market: Lack of transparency and price heterogeneity as an indicator of multiple equilibria (2021)
  20. Machine Learning Algorithms and Fine Art Pricing: Comparative performance to regression results for the South African market, 2009-22 (2025)
  21. A Unified Approach to Interpreting Model Predictions (2017) — link
  22. Least squares after model selection in high-dimensional sparse models (2013)
  23. Hedonic Pricing Models: A Selective and Applied Review (2003)
  24. Art as an Investment: The Market for Modern Prints (1993)
  25. Principal component analysis: a review and recent developments (2016)
  26. Declining Values and the Afternoon Effect: Evidence from Art Auctions (1997)
  27. Adam: A Method for Stochastic Optimization (2017) — link
  28. The Missing Indicator Method: From Low to High Dimensions (2023)
  29. PyTorch library for CAM methods (2021) — link
  30. A neural network boosting regression model based on XGBoost (2022)
  31. Research and application of XGBoost in imbalanced data (2022)
  32. Accounting for Taste: Art and the Financial Markets Over Three Centuries (1993)
  33. Hedonic prices and implicit markets: product differentiation in pure competition (1974)
  34. Who performs better? AVMs vs hedonic models (2020)
  35. A CNN-LSTM model for gold price time-series forecasting (2020)
  36. Adam: A Method for Stochastic Optimization (2017) — link
  37. Buying Beauty: On Prices and Returns in the Art Market (2013)
  38. THE PRICE OF ART: Uncertainty and reputation in the art field (2013)
  39. Asset Pricing Theory and the Valuation of Canadian Paintings (2004)
  40. Investing in Emotional Assets (2014)
  41. Anchoring or loss aversion? Empirical evidence from art auctions (2014)
  42. Residual variance and asset pricing in the art market (2023)
  43. Deep residual learning for image recognition (2016)
  44. Convolutional neural networks are not invariant to translation, but they can learn to be (2021)
  45. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (2021) — link
  46. XGBoost : A scalable tree boosting system (2016)
  47. Why do tree-based models still outperform deep learning on tabular data? (2022) — link
  48. Greedy function approximation: A gradient boosting machine (2001)
  49. The Elements of Statistical Learning: Data Mining, Inference, and Prediction (2009)
  50. Tabular data: Deep learning is not all you need (2022)
  51. The Art Basel & UBS Art Market Report 2024 (2024) — link
  52. Grad-CAM : Visual explanations from deep networks via gradient-based localization (2017)
  53. Empirical asset pricing via machine learning (2020)
  54. (Re-) Imag (in) ing price trends (2023)
  55. Deep learning for search and matching models (2025)
  56. DeepHAM: A Global Solution Method For Heterogeneous Agent Models With Aggregate Shocks (2021) — link
  57. The knowledge graph for macroeconomic analysis with alternative big data (2020) — link
  58. Imagenet: A large-scale hierarchical image database (2009)
  59. Imagenet classification with deep convolutional neural networks (2012)
  60. Interpretable factors of firm characteristics (2025) — link
  61. Textual factors: A scalable, interpretable, and data-driven approach to analyzing unstructured information (2025)
  62. Machine learning in the Chinese stock market (2022)

0 of 70 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.