XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
PRELIMINARIES: Please turn on "word wrap" when reading this document.

This document describes the Tax Holiday Instances database. The dataset identifies instances in which U.S. publicly traded firms disclose the presence of a foreign tax holiday in their annual Form 10-K filings. The purpose of this database is to provide a systematic and machine-readable record of tax holiday mentions across firms and years for use in empirical research.

Please review and reference the following paper when using these data:

Fox, Z. D., L. Krull, and S. G. Rane. 2025. Foreign tax holiday participation and U.S. job and investment loss. Contemporary Accounting Research, Forthcoming

* The original study did not use generative artificial intelligence. However, the present dataset extends the sample period and coverage using AI-assisted text identification applied to an expanded set of machine-readable 10-K filings. The data include observations spanning 1993 to 2024 following the approach described in Section 3 of the original study.

XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
DATA DESCRIPTION:

•	Number of observations (rows): 6,231
•	Unit of observation: A single mention of a foreign tax holiday within a firm’s Form 10-K filing.
•	Indexing variable: CIK (SEC Central Index Key)
•	Sample period: 1993–2024
•	Requirement for inclusion: The firm-year must have a machine-readable 10-K.

A single firm-year may contain multiple tax holiday instances if the filing references holidays in more than one country.

The dataset was constructed using the following procedure:
1.	Textual analysis was conducted on machine-readable 10-K filings obtained from SEC EDGAR for all available firm-years.
2.	Python-based text parsing procedures were used to scan filings for key terms related to tax holidays, including “tax holiday,” “tax incentive,” “tax exemption,” and closely related variants.
3.	Each relevant passage was extracted together with a context window of approximately 500 characters before and after the matched phrase, ensuring sufficient contextual information for accurate interpretation.
4.	Country names were identified when the filing explicitly stated where the tax holiday was received.
5.	Extracted passages were flagged as potential tax holiday mentions and passed through a secondary classification step using generative AI to determine whether each represented a true instance of tax holiday usage.
6.	A manual audit of 100 randomly selected passages was performed to benchmark accuracy; the review indicated an approximate 95% correct classification rate.
7.	Filings lacking machine-readable text were excluded because textual extraction could not be reliably performed.

XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
VARIBLE DEFINITIONS

The CSV file contains the following four variables:
(1) CIK
•	Type: Integer
•	Definition: SEC Central Index Key identifying the registrant firm.
•	Notes: The CIK uniquely identifies the filing entity. Users merging with Compustat should crosswalk CIKs to gvkeys.
(2) Filedate
•	Type: Date (YYYY-MM-DD)
•	Definition: The date on which the firm filed its Form 10-K with the SEC.
•	Notes: This corresponds to the official filing date recorded in EDGAR.
(3) Datadate
•	Type: Date (YYYY-MM-DD)
•	Definition: The financial statement date as reported within the 10-K (typically fiscal year-end).
•	Notes: This date reflects the period the filing pertains to, not when it was filed.
(4) Country
•	Type: String
•	Definition: The country in which the tax holiday is reported to have been granted.
•	Coding: Free-text country names standardized where possible; missing when the filing does not specify a country.
•	Notes: Some filings reference tax incentives without naming the specific country; these are recorded with a blank value.

XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
CONTACT INFORMATION

Questions, error reports, or suggestions for future updates may be directed to:
zack.fox@byu.edu