Stephen V. Brown

Visiting Assistant Professor

University of Florida

Biography

I’m a visiting professor in the Fisher School of Accounting at the University of Florida. My research has thus far primarily focused on the use of natural language in corporate disclosures and what we can learn from text beyond what’s obvious to a human reader. While my past papers have used various types of machine learning and artificial intelligence approaches, more current projects move even more in the direction of data science and how it can be used to better understand and evaluate companies. You can find more information on my CV and LinkedIn page.

In my downtime, I enjoy photography, playing acoustic guitar, and anything outdoors.

Interests

Data Science
Machine Learning
Artificial Intelligence
Computational Linguistics
Financial Accounting and Audit

Education

Ph.D. in Business Administration, 2012

University of Florida
Master of Business Administration, 2007

University of North Florida
B.S. in Computer Science, 1997

University of North Florida

Selected Publications

The Spillover Effect of SEC Comment Letters on Qualitative Corporate Disclosure: Evidence from the Risk Factor Disclosure

We focus on the indirect effect of SEC comment letters by analyzing how companies change their risk factor disclosures after other leaders and peers in the industry receive a comment letter, even though the original company did not receive a letter itself. Not only does this spillover effect influence how much companies change their disclosures, but it also increases the amount of company-specific information they disclose. We confirm that these actions by the company (that did not receive a letter) improve its disclosures such that they are less likely to receive a future comment letter.

Stephen V. Brown, Xiaoli Tian, Jennifer Wu Tucker

Contemporary Accounting Research (Conditionally Accepted), 2017.

Details PDF

Auditor-Client Compatibility and Audit Firm Selection

We examine the likelihood of a client switching auditors when that client does not fit well with other clients of the same auditor. The most dissimilar clients are around 10% more likely to switch auditors, and will tend to change to an auditor with which they are more compatible. Following an auditor change, there is a quick alignment in the footnotes of that client with existing clients of the new auditor, followed by slower changes in unaudited (but reviewed) sections of the 10-K. There is mixed evidence as to whether a better client-firm alignment improves audit quality.

Stephen V. Brown, W. Robert Knechel

Journal of Accounting Research, 2016.

Details PDF Dataset

Large-sample Evidence on Firms' Year-over-year MD&A Modifications

We measure year-over-year changes in companies’ 10-K MD&A disclosures, confirming these changes are more likely to appear following larger operational changes, but finding a decline in the magnitude of these changes and the stock market’s reaction to them. This latter result may indicate a decline in the usefulness of the MD&A for investors, coincident with the MD&A disclosures becoming longer and more boilerplate over time. We find no evidence that sell-side analysts use the MD&A in adjusting their forecasts, possibly due to having superior information sources.

Stephen V. Brown, Jennifer Wu Tucker

Journal of Accounting Research, 2011.

Details PDF Dataset

Working Papers

Specialization Through Client Commonality and Its Effect on Audit Production Costs

Augmenting earlier measures of between-client similarity based on narrative text, I introduce a new measure for quantifying the similarity using quantitative data on the face of the financial statements. Clients having greater overlap with other clients of the auditor have audit fees that are lower than otherwise expected, which I interpret as evidence of increased efficiencies and lower risks within the audit process as the auditor is able to apply common knowledge, technology, and skills across related clients. The lower-fee effect is strongest in industries where the auditor has greater economic incentives (i.e., can better profit from the overlap). When the narrative disclosures and financial statements portray inconsistent views of the client, I find higher audit fees in situations that likely increase risk for the auditor and lower fees when risks decrease.

Stephen V. Brown

Working Paper, 2017.

Details PDF

Work in Process

Disclosure Inconsistency and SEC Comment Letters

We focus on the risk of a company receiving an SEC comment letter when there are inconsistent messages between the company’s financial statements and associated 10-K narrative text. Companies can use the flexibility of the text to either clarify and expand on the financials, or to obscure them. We examine the relationship between this discretionary decision within the MD&A and the receipt of a comment letter from the SEC regarding that MD&A. So far, we have documented companies that look unusual relative to their peers have a higher risk of receiving a comment. We are currently performing a more in-depth, careful analysis of the data.

Stephen V. Brown, Lisa Hinson, Jennifer Wu Tucker

Work in Process, 2017.

Details

Automated Processing of 10-K Filings

Most public companies are required to file annual 10-K’s with the SEC, which may be found on its EDGAR system. Parsing these free-form, loosely structured documents is challenging due to inconsistent formatting, regulatory changes over time, and varying filing requirements across companies. In this paper, I will document the numerous assumptions and decisions made while developing version 3.0 of the DocNet system, an automated tool for parsing 10-K’s (and other narratives) into their component sections while also allowing for manual adjustments via a web-based editor. The core logic is written in Python and the web interface in Flask, built on a combination MongoDB and SQLite database backend.

Stephen V. Brown

Work in Process, 2017.

Details

Teaching

In Spring 2018, I will be teaching the following courses at the University of Florida:

ACG 5815: Accounting Regulation
ACG 4111: Financial Accounting and Reporting 2

Technology

I have around 20 years’ experience as a software developer, designing and building a wide variety of object-oriented libraries, databases, end-user applications, and back-end systems. Most of my work now is in direct support of my accounting research and data science activities.

Programming Languages


Python	Especially Pandas, SciPy, and NumPy; around 75% of my machine learning/AI work is in Python, the rest is in R
R	I use the tidyverse approach where possible; most of my R development is for statistical analysis and data visualization
SQL	Extensive database development experience; in recent years pretty equally divided between traditional SQL backends and noSQL data stores
JavaScript	Primarily to support web frontends
C	For high-performance operations, less often for systems development
Perl	Extensive use in earlier years, but less often now
XML, JSON	Data representation
HTML, CSS	Web development

General Data Science


Analytical Approach	Machine learning and artificial intelligence; traditional statistics and econometrics
Data Transformation	Processing, transforming, and merging large, semi-structured data; full extract-transform-load (ETL) process
Data Types	Accounting, financial, operational, and natural language (NLP)
Statistical Tools	Primarily R for traditional statistics; Python for machine learning; Stata occasionally
Data Visualization	ggplot2 when possible; matplotlib when in Python

Development Environment


Operating System	I feel most comfortable in a Linux environment, although I also use Windows regularly
IDE	I use RStudio when coding R, vim for everything else

Contact

stephen.brown@ufl.edu
University of Florida, Warrington College of Business, Fisher School of Accounting, PO Box 117166, Gainesville, Florida 32611-7166
Gerson 227B; Email for appointment