Please rotate your device to landscape mode to view the charts.

Biased Auctioneers

Journal: Journal of Finance

Date: 20230401

Author: Aubry, Mathieu; Kraussl, Roman; Manso, Gustavo; Spaenjers, Christophe

Abstract:
We construct a neural network algorithm that generates price predictions for art at auction, relying on both visual and nonvisual object characteristics. We find that higher automated valuations relative to auction house presale estimates are associated with substantially higher price-to-estimate ratios and lower buy-in rates, pointing to estimates' informational inefficiency. The relative contribution of machine learning is higher for artists with less dispersed and lower average prices. Furthermore, we show that auctioneers' prediction errors are persistent both at the artist and at the auction house level, and hence directly predictable themselves using information on past errors.

Link: Google Scholar

Background and Context

Research Setting

The study examines art auction houses' presale price estimates and uses machine learning to evaluate their accuracy and potential biases.

Data Coverage

Analysis uses data from 1.2 million painting auctions worldwide between 2008-2015, including detailed artwork information and images.

Methodology

Researchers developed a neural network algorithm to generate automated art valuations using both visual and non-visual artwork characteristics.

Machine Learning Predictions Outperform Traditional Methods

  • Neural network explains 74.2% of variation in art prices vs 67.7% for traditional hedonic model
  • Demonstrates superior predictive power of machine learning approach
  • Higher accuracy achieved by capturing complex relationships between artwork characteristics

Auction House Estimates Show Clear Bias Patterns

  • Systematic undervaluation of artists with poor recent performance
  • Shows persistence of auction house biases based on past returns
  • Demonstrates estimates are not purely objective valuations

Buy-in Rates Vary with Machine Learning Predictions

  • Buy-in rates drop from 45% to 25% as machine learning valuations increase relative to estimates
  • Indicates machine learning can predict likelihood of successful sales
  • Shows economic impact of auction house estimate biases

Prediction Accuracy Varies by Artist Characteristics

  • Machine learning performs better for high-volume artists with more historical data
  • Shows where automated valuations are most reliable
  • Helps identify when human expertise adds most value

Future Returns Affected by Estimate Biases

  • Artworks with high price-to-estimate ratios have lower future returns
  • Suggests market participants anchor on auction house estimates
  • Demonstrates real economic impact of estimate biases

Contribution and Implications

  • First study to use machine learning to systematically evaluate auction house estimate biases
  • Demonstrates potential for automated valuation tools to complement human expertise in art market
  • Shows economic consequences of estimate biases for market participants

Data Sources

  • Prediction accuracy comparison based on Table II and Figure 5
  • Bias patterns analysis based on Figure 1 and Figure 12
  • Buy-in rate analysis based on Figure 10
  • Artist characteristic analysis based on Figure 8
  • Returns analysis based on Table VII