Topic models have been increasingly applied in economics and finance. Due to com-
putational limitations and issues in interpretability, the most popular topic model, the
Latent Dirichlet Allocation (LDA), exhibits mixed results in economic and financial
research. This paper introduces a topic modeling framework that extracts the hid-
den topics discussed in the Management Discussion and Analysis (MD&A) section
in 10-K filings in a data-driven and interpretable way. The results of our paper are
three-fold. Firstly, the hidden topics discovered by our model are conveniently in-
terpretable and distinguishable in the financial context. Secondly, the time series of
hidden topic loadings grant insights into time variation of topic prevalence. Finally,
regression analyses show variations of the relationships between firm characteristics
and the MD&A compilation behavior of firm managers. We find that firm managers
truthfully report the firm performance and accrual conditions in their MD&A.