Music Ex Machina: How Spotify's Recommendations Shape Music Production 133 views

I examine how recommender systems have influenced the music industry and shaped music production. I provide a detailed analysis of the recorded music industry, including the structure of the industry, the characteristics of music, and how these characteristics have changed over time. Using data from Spotify, I document changes in song characteristics since the introduction of streaming services and recommender systems. I conduct reduced form analysis to show that the introduction of streaming services and recommender systems has led to a 40-second decrease in the average length of songs on Billboard's Hot 100 since 2010. Using a structural model of the recorded music industry, I analyze consumer behavior, platform recommendations, and rightsholder release decisions. I estimate a fixed cost of $170,000 for songs that enter Spotify's Top 200. Counterfactual analysis shows that with randomized recommendations, fewer songs would enter the market, reducing consumer welfare by 4%. The songs that do enter would be 33 seconds longer on average and vary more widely in length. Popularity-based recommendations that do not account for individual taste would generate a superstar effect, increasing gross profit margins for songs that enter the market to 40%, but reducing consumer welfare by 13%. Although recommender systems have reduced overall variety in music, they have also enabled additional entry and increased consumer welfare.

PHD (Doctor of Philosophy)

Economic of Music; Recommender Systems; Digital Economics

English

All rights reserved (no additional license for public reuse)

2025-04-25