Modeling Local Economies

Michael

Raseen

James

Introduction

What effect do music concert tours have on the local economy?

• Recent Taylor Swift and Beyoncé tours suggest that large concert tours have a significant impact on the local economy through ticket sales, hotel bookings, etc.

• We hypothesize that, because of this economic impact, we can use concert attendance to model local macroeconomic trends (e.g. GDP)

Project Goals

1. Use concert attendance and revenue data to predict GDP growth in the state where the concert occurred
2. Use Twitter data about concert tours to predict concert attendance in advance

Related Works

• Claveria and Torra (2021) predict GDP of European countries using business and consumer sentiment data
• Glaesar et al. (2017) use crowd-sourced Yelp data to measure economic output of local businesses
• Kreindler and Miyauchi (2021) leverage cell phone location data to map centers of high economic activity

Methodology

Data

Economic Data

• State-by-state quarterly GDP data from the Bureau of Economic Analysis
  • Manually calculate quarter-over-quarter GDP growth

Concert Data

• Concert attendance and revenue data for 8 high-grossing concert tours from the 2010s from Wikipedia (n = 453)
• Google Trends search data by state for each artist over a 5 year span

Google Trends

Problem

Problem: Google Treds data is relative, not absolute

Solution

Solution: Normalize relative values by a common search denominator

Preliminary Findings

GDP Growth Prediction

Attendance Revenue Prediction

Models

Bagging Regressor

• Using linear regression as base learners
• In terms of training error, performed about the same as the neural networks at a RMSE of 0.755 when predicting GDP growth from trend data

Neural Netowrks

• Feed forward design
• Numerous configurations tested, out of these a 4 layer network performed the best
• More sophisticated than the bagging regressor and with a root mean squared training error of 0.753 when predicting GDP growth from trend data

Limitations

• Data Granularity: We were limited to GDP data on a state level and on a quarterly time scale

  • It is difficult to determine just how much of an impact the concerts had on GDP growth versus general business cycle shifts

• Training Data Size: We limited our concert data to large-scale concert tours between 2015 and 2019 with available attendance and revenue figures (8 concerts)

• Confounding Factors: Our research problem lends itself to numerous confounding factors that are difficult to disentangle:

  • Do the concerts actually add economic impact beyond what would otherwise be expected?
  • e.g. most concerts happen in the summer (Q3)

• Google Trends vs. Twitter: We originally wanted to use NLP on tweets to extract more fine-grained data

Opportunities for Future Work

• Testing Hypothesis on Post-Pandemic Concerts: We hope to use the Taylor Swift and Beyoncé concert tours as a test set for our models. We postulate that worldwide concert tours after 2020 might have a greater impact on economic trends as economies recover, and concerts stimulate a more disproportionate level of consumer spending compared to before 2020.

• Additional/Different Trend Data Sources: Given that tweets provide hyper-local data (e.g. geotags) and a larger quantity of individual data points, better matches between tweet activity and concert attendance can be expected.