Logo image
Back
Applying deep learning to examine tax footnotes: a study of emotions and tax outcomes
Dissertation   Open access

Applying deep learning to examine tax footnotes: a study of emotions and tax outcomes

Tony Liang Ju Lin
Doctor of Philosophy (Ph.D.), Drexel University
Jun 2020
DOI:
https://doi.org/10.17918/00001042
pdf
Lin_Tony_20201.41 MBDownloadView
PDF Open Access Open Access (License Unspecified)

Abstract

Artificial intelligence Business enterprises--Taxation Business--Psychological aspects
This study applies deep learning algorithms provided and validated by the IBM Watson to examine the relation between qualitative textual characteristics (i.e., sentiment and emotion) detected in tax footnotes and tax outcomes. Extant studies provide evidence that a firm's reported unrecognized tax benefits (UTBs) consist of tax avoidance/complexity and financial reporting incentives. I use the emotion of joy conveyed in tax footnotes to measure the comfortability of a firm's tax position, employing the IBM Watson Natural Language Understanding (NLU) service. I provide evidence and show that, unlike emotion, sentiment (i.e., a positive, negative, or neutral tone) is easier to modify. Specifically, enabling targeted sentiment and emotion features on the service, I find that targeted sentiment, not targeted emotion, associated with and conveyed by the concept of uncertain tax positions explains financial reporting incentives of UTBs. I further find that detected emotion predicts UTBs related settlements with taxing authorities. In related analyses, I also find that these qualitative textual characteristics can explain a firm's demand for auditor provided tax services (APTS). Overall, my findings provide important initial evidence on the role of NLU detected qualitative textual characteristics and their relation to tax outcomes and APTS. Nevertheless, this study may help taxing authorities develop a more concrete audit strategy.

Metrics

61 File views/ downloads
85 Record Views

Details

Logo image