Measurement Error in Google Ticker Search

Abstract

We quantify and illustrate the effects of measurement error in the Google ticker search volume index (“SVI”)—a commonly used proxy for investor attention. Based on a dataset of roughly 2.7 billion website visits following S&P 500 ticker searches, we estimate that 69% of searches are not by investors searching for information, and find that this measurement error is highly correlated with firm characteristics. We then show that measurement error in SVI can cause erroneous inferences in three common types of tests. First, in tests of investor attention around information events, measurement error biases coefficients towards zero and can generate false-negative results (type 2 errors). Second, because SVI measurement error is correlated with firm characteristics, it can easily generate false-positive results in cross-sectional tests (type 1 errors). Third, tests that compare SVI to other attention proxies can produce erroneous inferences due to differences
in measurement error between variables. We conclude by introducing a refined SVI specification and by providing guidance for future researchers using SVI as a proxy for attention.

Keywords: Google ticker search; SVI; investor attention; measurement error; noise search

JEL Classification: C13, C15, M41