DOES FINANCE BENEFIT SOCIETY?

A LANGUAGE EMBEDDING APPROACH

We ask a simple question ... what emotions do you feel when you hear the word "finance." We could see many finance professionals (most people reading this) have pleasant feelings about it. However, the sentiments differ across different regions and across time. In this paper, we look at Google's book data to see how the average denizens of eight countries view finance from 1870 to 2009 and how that affects economic outcomes.

Presentations

  • Greater China Area Finance Conference 2020
  • Norwegian School of Economics 2021
  • NYU Shanghai Volatility Institute 2020, Video Link
  • Michigan State, UI Chicago (Virtual Finance Seminar) 2020, Video Link
  • Washinton University 2020

Media Coverage

Abstract

We measure popular sentiment toward finance using a computational linguistics approach applied to millions of books published in eight countries over hundreds of years. We document persistent differences in finance sentiment across countries despite ample time-series variation. Finance sentiment declines after epidemics and earthquakes, but rises following droughts, floods, and landslides. These heterogeneous effects of natural disasters suggest finance sentiment responds differently to the realization of insured versus uninsured risks. Using local projections, we find that positive shocks to finance sentiment have positive and persistent effects on economic growth. Our estimates predict a contraction in finance sentiment due to the COVID-19 pandemic that will exacerbate its long-term economic damage.

Key Figures

Finance sentiment is based on the annual average projection of finance-mentioning sentences' embeddings onto the positive minus negative finance sentiment dimension. Sentences are from the Google Books Ngram corpus and embedded using BERT. Bands represent 95 percent confidence intervals produced by subsampling.
Circles indicate severe natural disasters, with size proportional to the logarithm of the number of death. We define severe disasters, as disasters with death above 20 per million population.