SPB Git
spboucher.ai

spb/firc0001_uqo

Public

L4: references.bib — bloom2021 -> kalyani2025 (QJE 2025), 24 DOI-verified additions (Crossref/DataCite 2026-11-05), SEC and FTSE Russell URLs fixed, spid2026 relabelled, sibling WP 0002 and data citations

Items: FIRC0001-EDIT-10, FIRC0001-ADVE-17, FIRC0001-THEO-19, FIRC0001-LITE-01, FIRC0001-LITE-02, FIRC0001-LITE-03, FIRC0001-LITE-04, FIRC0001-LITE-05, FIRC0001-LITE-06, FIRC0001-LITE-11, FIRC0001-DATA-12.
Only the cite key bloom2021 was swapped in literature.tex:7 and applications.tex:11 so that the build stays clean; prose is Stage E (improvements/FIRC-0001/literature_additions.md).
Records: reproducibility/FIRC-0001/logs/L4_crossref_20261105-141223.json.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Simon-Pierre Boucher committed (Nov 5, 2026) parent 64ddafa Browse files

3 changed files +175 −19

3 changed files
Unified Split
modified paper/references.bib +173 −17
@@ -54,15 +54,6 @@
54 year = {2021}, volume = {3}, number = {3}, pages = {303--320}, 54 year = {2021}, volume = {3}, number = {3}, pages = {303--320},
55 doi = {10.1257/aeri.20190499} 55 doi = {10.1257/aeri.20190499}
56 } 56 }
57−@techreport{bloom2021,
58− author = {Bloom, Nicholas and Hassan, Tarek A. and Kalyani, Aakash and Lerner, Josh and Tahoun, Ahmed},
59− title = {The Diffusion of Disruptive Technologies},
60− institution = {National Bureau of Economic Research},
61− type = {NBER Working Paper},
62− number = {28999},
63− year = {2021},
64− doi = {10.3386/w28999}
65−}
66 @article{dyer2017, 57 @article{dyer2017,
67 author = {Dyer, Travis and Lang, Mark and Stice-Lawrence, Lorien}, 58 author = {Dyer, Travis and Lang, Mark and Stice-Lawrence, Lorien},
68 title = {The Evolution of 10-{K} Textual Disclosure: Evidence from {L}atent {D}irichlet {A}llocation}, 59 title = {The Evolution of 10-{K} Textual Disclosure: Evidence from {L}atent {D}irichlet {A}llocation},
@@ -116,22 +107,187 @@
116 @misc{secfairaccess, 107 @misc{secfairaccess,
117 author = {{U.S. Securities and Exchange Commission}}, 108 author = {{U.S. Securities and Exchange Commission}},
118 title = {Accessing {EDGAR} Data: Fair Access Policy}, 109 title = {Accessing {EDGAR} Data: Fair Access Policy},
119− howpublished = {\url{https://www.sec.gov/os/accessing-edgar-data}}, 110+ howpublished = {\url{https://www.sec.gov/search-filings/edgar-search-assistance/accessing-edgar-data}},
120− year = {2024}, 111+ year = {2026},
121− note = {Accessed October 2026} 112+ note = {Accessed October 2026; URL re-verified November 5, 2026 (the former \texttt{/os/accessing-edgar-data} address redirects here)}
122 } 113 }
123 @misc{ftserussell2026, 114 @misc{ftserussell2026,
124 author = {{FTSE Russell}}, 115 author = {{FTSE Russell}},
125− title = {Russell {US} Equity Indexes: Construction and Methodology}, 116+ title = {Russell {US} Equity Indexes: Construction and Methodology, v7.2},
126− howpublished = {\url{https://www.lseg.com/en/ftse-russell/indices/russell-us}}, 117+ howpublished = {\url{https://www.lseg.com/content/dam/ftse-russell/en_us/documents/ground-rules/russell-us-indexes-construction-and-methodology.pdf}},
118+ year = {2026},
119+ month = {August},
120+ note = {Document dated August 2026; downloaded November 5, 2026}
121+}
122+@misc{ftserussell3000factsheet2026,
123+ author = {{FTSE Russell}},
124+ title = {Russell 3000 Index Factsheet},
125+ howpublished = {\url{https://research.ftserussell.com/Analytics/FactSheets/Home/DownloadSingleIssue?issueName=US3000USD}},
127 year = {2026}, 126 year = {2026},
128− note = {Accessed October 2026} 127+ note = {Index characteristics as of September 30, 2026 (2,996 holdings); downloaded November 5, 2026}
129 } 128 }
130 @techreport{spid2026, 129 @techreport{spid2026,
131 author = {Tandjambianda, Djerrycharli and Boucher, Simon-Pierre and Kammoun, Manel}, 130 author = {Tandjambianda, Djerrycharli and Boucher, Simon-Pierre and Kammoun, Manel},
132 title = {{SPID} --- {SEC} Project Intelligence Database: A Methodological Report on the {S\&P}~500 Release}, 131 title = {{SPID} --- {SEC} Project Intelligence Database: A Methodological Report on the {S\&P}~500 Release},
133 institution = {Université du Québec en Outaouais, Département des sciences administratives}, 132 institution = {Université du Québec en Outaouais, Département des sciences administratives},
134− type = {Internal methodological report}, 133+ type = {Unpublished internal methodological report, available from the authors},
135 year = {2026}, 134 year = {2026},
136− note = {Revision df7905d; October 28, 2026} 135+ note = {June 2026 study of revision \texttt{df7905d}; the report file was regenerated on October 28, 2026}
136+}
137+@techreport{kammoun2026ai,
138+ author = {Kammoun, Manel and Tandjambianda, Djerrycharli and Boucher, Simon-Pierre},
139+ title = {Does the Language of Artificial Intelligence Move Markets? {A} Semantic Analysis of {R}ussell~3000 {SEC} Filings and the Cross-Section of Stock Returns, 2010--2026},
140+ institution = {Université du Québec en Outaouais, Département des sciences administratives},
141+ type = {UQO Working Paper (FIRC)},
142+ number = {0002},
143+ year = {2026},
144+ note = {Version 2.0, October 28, 2026}
145+}
146+@misc{iwvholdings2026,
147+ author = {{BlackRock, Inc.}},
148+ title = {{iShares} {R}ussell 3000 {ETF} ({IWV}): Fund holdings file (\texttt{latest-holdings.csv})},
149+ howpublished = {\url{https://www.ishares.com/us/products/239714/ishares-russell-3000-etf}},
150+ year = {2026},
151+ note = {Holdings as of October 23, 2026, downloaded October 27, 2026; subject to the iShares website terms of use; the downloaded file is not archived with the paper}
152+}
153+@misc{wikipediasp500,
154+ author = {{Wikipedia contributors}},
155+ title = {List of {S\&P} 500 companies},
156+ howpublished = {\url{https://en.wikipedia.org/wiki/List_of_S%26P_500_companies}},
157+ year = {2026},
158+ note = {Table scraped May 25, 2026; text available under the Creative Commons Attribution-ShareAlike 4.0 licence}
159+}
160+@article{kalyani2025,
161+ author = {Kalyani, Aakash and Bloom, Nicholas and Carvalho, Marcela and Hassan, Tarek and Lerner, Josh and Tahoun, Ahmed},
162+ title = {The Diffusion of New Technologies},
163+ journal = {Quarterly Journal of Economics},
164+ year = {2025}, volume = {140}, number = {2}, pages = {1299--1365},
165+ doi = {10.1093/qje/qjaf002},
166+ note = {Published version of NBER Working Paper 28999 (2021, DOI 10.3386/w28999)}
167+}
168+@techreport{jha2024,
169+ author = {Jha, Manish and Qian, Jialin and Weber, Michael and Yang, Baozhong},
170+ title = {{ChatGPT} and Corporate Policies},
171+ institution = {National Bureau of Economic Research}, type = {NBER Working Paper}, number = {32161},
172+ year = {2024}, doi = {10.3386/w32161}
173+}
174+@article{ludwig2025,
175+ author = {Ludwig, Jens and Mullainathan, Sendhil and Rambachan, Ashesh},
176+ title = {Large Language Models: An Applied Econometric Framework},
177+ journal = {Annual Review of Economics},
178+ year = {2026}, volume = {18}, number = {1}, pages = {283--316},
179+ doi = {10.1146/annurev-economics-120925-105620},
180+ note = {Also NBER Working Paper 33344 (2025, DOI 10.3386/w33344)}
181+}
182+@article{gilardi2023,
183+ author = {Gilardi, Fabrizio and Alizadeh, Meysam and Kubli, Maël},
184+ title = {{ChatGPT} outperforms crowd workers for text-annotation tasks},
185+ journal = {Proceedings of the National Academy of Sciences},
186+ year = {2023}, volume = {120}, number = {30}, pages = {e2305016120},
187+ doi = {10.1073/pnas.2305016120}
188+}
189+@article{dekok2025,
190+ author = {de Kok, Ties},
191+ title = {{ChatGPT} for Textual Analysis? How to Use Generative {LLMs} in Accounting Research},
192+ journal = {Management Science},
193+ year = {2025}, volume = {71}, number = {9}, pages = {7888--7906},
194+ doi = {10.1287/mnsc.2023.03253}
195+}
196+@techreport{kim2023,
197+ author = {Kim, Alex G. and Muhn, Maximilian and Nikolaev, Valeri V.},
198+ title = {From Transcripts to Insights: Uncovering Corporate Risks Using Generative AI},
199+ institution = {SSRN}, type = {Working paper}, number = {4593660},
200+ year = {2023}, doi = {10.2139/ssrn.4593660}
201+}
202+@misc{dolphin2026,
203+ author = {Dolphin, Rian and Dursun, Joe and Blankenship, Jarrett and Adams, Katie and Pike, Quinton},
204+ title = {Grounded Event Extraction from {SEC} {8-K} Filings with a Fine-Grained Taxonomy},
205+ howpublished = {arXiv:2607.08346; also SSRN 7087858 (DOI 10.2139/ssrn.7087858)},
206+ year = {2026}, doi = {10.48550/arXiv.2607.08346}
207+}
208+@techreport{crane2024,
209+ author = {Crane, Leland D. and Green, Emily and Harnish, Molly and McClennan, Will and Soto, Paul E. and Vrankovich, Betsy and Williams, Jacob},
210+ title = {Tracking Real Time Layoffs with {SEC} Filings: A Preliminary Investigation},
211+ institution = {Board of Governors of the Federal Reserve System}, type = {Finance and Economics Discussion Series}, number = {2024-020},
212+ year = {2024}, doi = {10.17016/feds.2024.020}
213+}
214+@inproceedings{loukas2021,
215+ author = {Loukas, Lefteris and Fergadiotis, Manos and Androutsopoulos, Ion and Malakasiotis, Prodromos},
216+ title = {{EDGAR-CORPUS}: Billions of Tokens Make The World Go Round},
217+ booktitle = {Proceedings of the Third Workshop on Economics and Natural Language Processing},
218+ year = {2021}, pages = {13--18},
219+ doi = {10.18653/v1/2021.econlp-1.2}
220+}
221+@misc{bettencourt2026,
222+ author = {Bettencourt, Nick and Ding, Xiaowei and Giesecke, Kay},
223+ title = {The {Stanford} {EDGAR} Filings Dataset: Reconstructing U.S. Corporate and Financial Disclosures into Layout-Faithful and Token-Efficient Pretraining Data},
224+ howpublished = {arXiv:2606.18192; also SSRN 7161699 (DOI 10.2139/ssrn.7161699)},
225+ year = {2026}, doi = {10.48550/arXiv.2606.18192}
226+}
227+@article{greenstone2010,
228+ author = {Greenstone, Michael and Hornbeck, Richard and Moretti, Enrico},
229+ title = {Identifying Agglomeration Spillovers: Evidence from Winners and Losers of Large Plant Openings},
230+ journal = {Journal of Political Economy},
231+ year = {2010}, volume = {118}, number = {3}, pages = {536--598},
232+ doi = {10.1086/653714}
233+}
234+@article{mcconnell1985,
235+ author = {McConnell, J and Muscarella, Chris J.},
236+ title = {Corporate capital expenditure decisions and the market value of the firm},
237+ journal = {Journal of Financial Economics},
238+ year = {1985}, volume = {14}, number = {3}, pages = {399--422},
239+ doi = {10.1016/0304-405X(85)90006-6}
240+}
241+@misc{yang2024,
242+ author = {Yang, Daniel and Tsai, Yao-Hung Hubert and Yamada, Makoto},
243+ title = {On Verbalized Confidence Scores for {LLMs}},
244+ howpublished = {arXiv:2412.14737},
245+ year = {2024}, doi = {10.48550/arXiv.2412.14737}
246+}
247+@inproceedings{ye2024,
248+ author = {Ye, Fanghua and Yang, Mingming and Pang, Jianhui and Wang, Longyue and Wong, Derek and Yilmaz, Emine and Shi, Shuming and Tu, Zhaopeng},
249+ title = {Benchmarking {LLMs} via Uncertainty Quantification},
250+ booktitle = {Advances in Neural Information Processing Systems 37 (NeurIPS 2024)},
251+ year = {2024}, pages = {15356--15385},
252+ doi = {10.52202/079017-0491}
253+}
254+@article{li2020,
255+ author = {Li, Yuliang and Li, Jinfeng and Suhara, Yoshihiko and Doan, AnHai and Tan, Wang-Chiew},
256+ title = {Deep entity matching with pre-trained language models},
257+ journal = {Proceedings of the VLDB Endowment},
258+ year = {2020}, volume = {14}, number = {1}, pages = {50--60},
259+ doi = {10.14778/3421424.3421431}
260+}
261+@misc{peeters2023,
262+ author = {Peeters, Ralph and Steiner, Aaron and Bizer, Christian},
263+ title = {Entity Matching using Large Language Models},
264+ howpublished = {arXiv:2310.11244},
265+ year = {2023}, doi = {10.48550/arXiv.2310.11244}
266+}
267+@article{bellstam2021,
268+ author = {Bellstam, Gustaf and Bhagat, Sanjai and Cookson, J. Anthony},
269+ title = {A Text-Based Analysis of Corporate Innovation},
270+ journal = {Management Science},
271+ year = {2021}, volume = {67}, number = {7}, pages = {4004--4031},
272+ doi = {10.1287/mnsc.2020.3682}
273+}
274+@article{sautner2023,
275+ author = {Sautner, Zacharias and van Lent, Laurence and Vilkov, Grigory and Zhang, Ruishen},
276+ title = {Firm-Level Climate Change Exposure},
277+ journal = {Journal of Finance},
278+ year = {2023}, volume = {78}, number = {3}, pages = {1449--1498},
279+ doi = {10.1111/jofi.13219}
280+}
281+@article{afzal2026,
282+ author = {Afzal, Muhammad Talha and Yan, Jinqiu (Claire) and Li, Huaxia and Vasarhelyi, Miklos},
283+ title = {Beyond the bottom line: using generative AI to extract narrative financial disclosures},
284+ journal = {International Journal of Accounting Information Systems},
285+ year = {2026}, volume = {57}, pages = {100777},
286+ doi = {10.1016/j.accinf.2026.100777}
287+}
288+@techreport{zolas2020,
289+ author = {Zolas, Nikolas and Kroff, Zachary and Brynjolfsson, Erik and McElheran, Kristina and Beede, David and Buffington, Cathy and Goldschlag, Nathan and Foster, Lucia and Dinlersoz, Emin},
290+ title = {Advanced Technologies Adoption and Use by U.S. Firms: Evidence from the Annual Business Survey},
291+ institution = {National Bureau of Economic Research}, type = {NBER Working Paper}, number = {28290},
292+ year = {2020}, doi = {10.3386/w28290}
137 } 293 }
modified paper/sections/applications.tex +1 −1
@@ -8,7 +8,7 @@ Four families of empirical designs are natural once the validation gates are pas
8 8
9 \paragraph{Project counts and composition as firm-year variables.} The issuer-year count of newly observed projects, by category, is a text-derived measure of disclosed strategic activity comparable in spirit to the thematic scores of the factor layer but with a discrete, interpretable unit. Because the panel is a snapshot universe, firm fixed effects and within-firm comparisons are safer than cross-sectional levels, and the later entry of smaller issuers (Figure~\ref{fig:coverage}) should be modeled explicitly. 9 \paragraph{Project counts and composition as firm-year variables.} The issuer-year count of newly observed projects, by category, is a text-derived measure of disclosed strategic activity comparable in spirit to the thematic scores of the factor layer but with a discrete, interpretable unit. Because the panel is a snapshot universe, firm fixed effects and within-firm comparisons are safer than cross-sectional levels, and the later entry of smaller issuers (Figure~\ref{fig:coverage}) should be modeled explicitly.
10 10
11−\paragraph{Technology adoption and diffusion.} The \texttt{technologies} field and the categories \texttt{ai\_initiative}, \texttt{data\_center}, \texttt{cloud\_migration}, \texttt{automation} and \texttt{erp\_implementation} allow the first-mention date of a technology to be located by issuer, sector and location, in the spirit of \citet{bloom2021} but from the issuer's own regulatory narrative. The small counts of these categories (a few hundred projects each) and the fact that technologies are free-text tags rather than a controlled vocabulary limit precision; a mapping of tags to a vocabulary is part of the proposed ontology work. 11+\paragraph{Technology adoption and diffusion.} The \texttt{technologies} field and the categories \texttt{ai\_initiative}, \texttt{data\_center}, \texttt{cloud\_migration}, \texttt{automation} and \texttt{erp\_implementation} allow the first-mention date of a technology to be located by issuer, sector and location, in the spirit of \citet{kalyani2025} but from the issuer's own regulatory narrative. The small counts of these categories (a few hundred projects each) and the fact that technologies are free-text tags rather than a controlled vocabulary limit precision; a mapping of tags to a vocabulary is part of the proposed ontology work.
12 12
13 \paragraph{Event studies.} Mentions in 8-K filings are dated to the day and are the closest the database comes to an announcement date (74{,}594 mentions). Because the primary document of an 8-K often refers to a press release filed as an exhibit that the pipeline does not read, the first \emph{public} announcement may precede the first mention in the corpus; event dates should be verified against the exhibit or the newswire before use. 13 \paragraph{Event studies.} Mentions in 8-K filings are dated to the day and are the closest the database comes to an announcement date (74{,}594 mentions). Because the primary document of an 8-K often refers to a press release filed as an exhibit that the pipeline does not read, the first \emph{public} announcement may precede the first mention in the corpus; event dates should be verified against the exhibit or the newswire before use.
14 14
modified paper/sections/literature.tex +1 −1
@@ -4,7 +4,7 @@
4 4
5 \paragraph{Text as data in accounting and finance.} The narrative content of SEC filings has become a standard empirical input. \citet{loughran2011} showed that general-purpose sentiment dictionaries misclassify financial text and introduced finance-specific word lists; \citet{loughran2016} survey the subsequent decade of textual analysis in accounting and finance, and \citet{gentzkow2019} provide the general econometric framework for treating text as data, including the measurement-error and validation issues that any text-derived variable inherits. \citet{dyer2017} document how 10-K disclosure has lengthened and become more boilerplate over time, a trend that matters for us because repeated boilerplate is the principal source of false-positive ``projects''. \citet{cohen2020} show that changes in otherwise repetitive filing language carry information, which motivates our emphasis on \emph{repeated} mentions of the same initiative across filings rather than on single occurrences. \citet{grimmer2013} offer the general methodological warning that automated text measures must be validated against human judgment on the specific task at hand, which is the stance adopted in Section~\ref{sec:validation}. 5 \paragraph{Text as data in accounting and finance.} The narrative content of SEC filings has become a standard empirical input. \citet{loughran2011} showed that general-purpose sentiment dictionaries misclassify financial text and introduced finance-specific word lists; \citet{loughran2016} survey the subsequent decade of textual analysis in accounting and finance, and \citet{gentzkow2019} provide the general econometric framework for treating text as data, including the measurement-error and validation issues that any text-derived variable inherits. \citet{dyer2017} document how 10-K disclosure has lengthened and become more boilerplate over time, a trend that matters for us because repeated boilerplate is the principal source of false-positive ``projects''. \citet{cohen2020} show that changes in otherwise repetitive filing language carry information, which motivates our emphasis on \emph{repeated} mentions of the same initiative across filings rather than on single occurrences. \citet{grimmer2013} offer the general methodological warning that automated text measures must be validated against human judgment on the specific task at hand, which is the stance adopted in Section~\ref{sec:validation}.
6 6
7−\paragraph{Measuring corporate activity from disclosures.} Closest to our construct are text-based measures of what firms do rather than how they feel. \citet{hoberg2016} build product-market classifications from 10-K business descriptions; \citet{hobergmaksimovic2015} identify financially constrained firms from the language of the liquidity discussion; \citet{hassan2019} measure firm-level political risk from the share of earnings-call text devoted to politics; \citet{li2021} measure corporate culture with word embeddings; and \citet{kelly2021} and \citet{bloom2021} track the emergence and diffusion of technologies through patents, calls and job postings. SPID differs from these approaches in its unit of observation. Rather than a firm-level score summarizing the intensity of a theme, it targets \emph{discrete initiatives}---objects with a name, a type, a status, cited years, a location, partners and an amount---that can be followed from filing to filing. This is the project-centric view that commercial intelligence products sell, but with open provenance and a documented extraction contract. 7+\paragraph{Measuring corporate activity from disclosures.} Closest to our construct are text-based measures of what firms do rather than how they feel. \citet{hoberg2016} build product-market classifications from 10-K business descriptions; \citet{hobergmaksimovic2015} identify financially constrained firms from the language of the liquidity discussion; \citet{hassan2019} measure firm-level political risk from the share of earnings-call text devoted to politics; \citet{li2021} measure corporate culture with word embeddings; and \citet{kelly2021} and \citet{kalyani2025} track the emergence and diffusion of technologies through patents, calls and job postings. SPID differs from these approaches in its unit of observation. Rather than a firm-level score summarizing the intensity of a theme, it targets \emph{discrete initiatives}---objects with a name, a type, a status, cited years, a location, partners and an amount---that can be followed from filing to filing. This is the project-centric view that commercial intelligence products sell, but with open provenance and a documented extraction contract.
8 8
9 \paragraph{Structured information extraction with language models.} The extraction step relies on an instruction-tuned LLM used in a constrained-output mode: a fixed system prompt, a strict JSON object schema, temperature zero, and an explicit instruction to extract only initiatives grounded in the supplied passage. This design follows the general lesson of the text-as-data literature that the measurement instrument must be fixed, versioned and auditable, and it is deliberately conservative: the model is asked to prefer fewer, concrete projects to many vague ones, to return an empty list when no concrete project is described, and to attach a self-reported confidence. Semantic search over projects uses sentence embeddings \citep{reimers2019}. We make no claim that self-reported confidence is calibrated; its calibration is one of the items of the validation protocol. 9 \paragraph{Structured information extraction with language models.} The extraction step relies on an instruction-tuned LLM used in a constrained-output mode: a fixed system prompt, a strict JSON object schema, temperature zero, and an explicit instruction to extract only initiatives grounded in the supplied passage. This design follows the general lesson of the text-as-data literature that the measurement instrument must be fixed, versioned and auditable, and it is deliberately conservative: the model is asked to prefer fewer, concrete projects to many vague ones, to return an empty list when no concrete project is described, and to attach a self-reported confidence. Semantic search over projects uses sentence embeddings \citep{reimers2019}. We make no claim that self-reported confidence is calibrated; its calibration is one of the items of the validation protocol.
10 10
11 11
Menu

Theme
↑↓ navigate ↵ open esc close / or ⌘K anywhere