|
Abstract:
|
Complaints in financial services in the U.S. are collected by the Consumer Financial Protection Bureau and made available to the public for review and analysis. We model complaint relief using different specifications, taking into account geography, products, and various attributes of the complaint database, including the complaint narratives. We estimate an ordinal regression of complaint outcomes measured by "relief." We uncover in the first phase how some complaints for some products and States achieve better outcomes relative to a baseline. One notable finding is that the presence of a complaint narrative may not necessarily result in better relief to the consumer. Certain special groups who might be expected to get relief (e.g., military/elderly) do not. Another phase of our modeling analyzes only complaints with narratives. In this phase, military/elderly do achieve better outcomes. A particular challenge we aim to address is how to analyze the linguistic facet of a complaint given that the writing style can be unconstrained, thus other dimensions beyond readability might exist. Additionally, we seek to compare narratives at more granular geographic levels in our data set.
|
