spb/uqo-research-db

1\relax 2\providecommand\hyper@newdestlabel[2]{}3\providecommand\HyField@AuxAddToFields[1]{}4\providecommand\HyField@AuxAddToCoFields[2]{}5\@writefile{toc}{\contentsline {section}{\numberline {1}Introduction}{4}{section.1}\protected@file@percent }6\@writefile{toc}{\contentsline {paragraph}{Contributions.}{5}{paragraph*.1}\protected@file@percent }7\@writefile{toc}{\contentsline {paragraph}{Caveats up front.}{5}{paragraph*.2}\protected@file@percent }8\citation{loughran2011liability,li2008annual}9\citation{hoberg2016text,hassan2019firm}10\citation{gentzkow2019text}11\citation{korinek2023language,bybee2023ghost}12\citation{hayashi1982tobin,fazzari1988financing}13\citation{peters2017intangible,crouzet2019understanding}14\citation{griliches1957hybrid,bass1969new}15\citation{babina2024artificial,acemoglu2022artificial}16\citation{hidalgo2007product}17\@writefile{toc}{\contentsline {section}{\numberline {2}Related Literature}{6}{section.2}\protected@file@percent }18\newlabel{sec:related}{{2}{6}{Related Literature}{section.2}{}}19\@writefile{toc}{\contentsline {paragraph}{Text as data.}{6}{paragraph*.3}\protected@file@percent }20\@writefile{toc}{\contentsline {paragraph}{LLMs for extraction and measurement.}{6}{paragraph*.4}\protected@file@percent }21\@writefile{toc}{\contentsline {paragraph}{Corporate investment.}{6}{paragraph*.5}\protected@file@percent }22\@writefile{toc}{\contentsline {paragraph}{Innovation and diffusion.}{6}{paragraph*.6}\protected@file@percent }23\@writefile{toc}{\contentsline {paragraph}{Networks and representations.}{7}{paragraph*.7}\protected@file@percent }24\@writefile{toc}{\contentsline {section}{\numberline {3}Data and Extraction Pipeline}{7}{section.3}\protected@file@percent }25\newlabel{sec:data}{{3}{7}{Data and Extraction Pipeline}{section.3}{}}26\@writefile{toc}{\contentsline {subsection}{\numberline {3.1}Source corpus}{7}{subsection.3.1}\protected@file@percent }27\@writefile{toc}{\contentsline {subsection}{\numberline {3.2}LLM extraction}{7}{subsection.3.2}\protected@file@percent }28\@writefile{toc}{\contentsline {subsection}{\numberline {3.3}Resolution, graph, and embeddings}{7}{subsection.3.3}\protected@file@percent }29\@writefile{toc}{\contentsline {subsection}{\numberline {3.4}The economics of extraction}{8}{subsection.3.4}\protected@file@percent }30\@writefile{toc}{\contentsline {subsection}{\numberline {3.5}Analysis datasets}{8}{subsection.3.5}\protected@file@percent }31\@writefile{lot}{\contentsline {table}{\numberline {1}{\ignorespaces Summary statistics. Amounts are in millions of USD; duration is the span between a project's first and last appearance in filings.}}{9}{table.caption.8}\protected@file@percent }32\providecommand*\caption@xref[2]{\@setref\relax\@undefined{#1}}33\newlabel{tab:summary}{{1}{9}{Summary statistics. Amounts are in millions of USD; duration is the span between a project's first and last appearance in filings}{table.caption.8}{}}34\@writefile{toc}{\contentsline {section}{\numberline {4}Descriptive Evidence and Stylized Facts}{9}{section.4}\protected@file@percent }35\newlabel{sec:descriptives}{{4}{9}{Descriptive Evidence and Stylized Facts}{section.4}{}}36\@writefile{toc}{\contentsline {paragraph}{Volume and trend.}{9}{paragraph*.9}\protected@file@percent }37\@writefile{lof}{\contentsline {figure}{\numberline {1}{\ignorespaces Project mentions extracted per year, by filing form. The corpus is roughly balanced across years, with 8-K and 10-K filings supplying the bulk of project signals.}}{9}{figure.caption.10}\protected@file@percent }38\newlabel{fig:mentions}{{1}{9}{Project mentions extracted per year, by filing form. The corpus is roughly balanced across years, with 8-K and 10-K filings supplying the bulk of project signals}{figure.caption.10}{}}39\@writefile{toc}{\contentsline {paragraph}{Composition.}{9}{paragraph*.11}\protected@file@percent }40\@writefile{lof}{\contentsline {figure}{\numberline {2}{\ignorespaces Distribution of resolved projects across the 21-category taxonomy.}}{10}{figure.caption.12}\protected@file@percent }41\newlabel{fig:typedist}{{2}{10}{Distribution of resolved projects across the 21-category taxonomy}{figure.caption.12}{}}42\@writefile{lot}{\contentsline {table}{\numberline {2}{\ignorespaces Projects by type: counts, aggregate announced capital, AI share, and mean extraction confidence. Capital is the sum of disclosed amounts and is an upper bound (see Section~\ref {sec:limitations}).}}{11}{table.caption.13}\protected@file@percent }43\newlabel{tab:bytype}{{2}{11}{Projects by type: counts, aggregate announced capital, AI share, and mean extraction confidence. Capital is the sum of disclosed amounts and is an upper bound (see Section~\ref {sec:limitations})}{table.caption.13}{}}44\@writefile{toc}{\contentsline {paragraph}{Sectoral structure.}{11}{paragraph*.14}\protected@file@percent }45\@writefile{lof}{\contentsline {figure}{\numberline {3}{\ignorespaces Project-type composition by GICS sector (row-normalized within-sector shares). Darker cells indicate a larger share of a sector's projects in that category.}}{12}{figure.caption.15}\protected@file@percent }46\newlabel{fig:heatmap}{{3}{12}{Project-type composition by GICS sector (row-normalized within-sector shares). Darker cells indicate a larger share of a sector's projects in that category}{figure.caption.15}{}}47\@writefile{lot}{\contentsline {table}{\numberline {3}{\ignorespaces Sector breakdown: number of firms, projects, aggregate announced capital, and AI share.}}{12}{table.caption.16}\protected@file@percent }48\newlabel{tab:bysector}{{3}{12}{Sector breakdown: number of firms, projects, aggregate announced capital, and AI share}{table.caption.16}{}}49\@writefile{toc}{\contentsline {paragraph}{Capital and geography.}{12}{paragraph*.17}\protected@file@percent }50\newlabel{fig:capex}{{4a}{13}{Announced capital and project count by year}{figure.caption.18}{}}51\newlabel{sub@fig:capex}{{a}{13}{Announced capital and project count by year}{figure.caption.18}{}}52\newlabel{fig:locations}{{4b}{13}{Top 15 project locations}{figure.caption.18}{}}53\newlabel{sub@fig:locations}{{b}{13}{Top 15 project locations}{figure.caption.18}{}}54\@writefile{lof}{\contentsline {figure}{\numberline {4}{\ignorespaces Capital and geography of the corporate project economy.}}{13}{figure.caption.18}\protected@file@percent }55\@writefile{toc}{\contentsline {subsection}{\numberline {4.1}Six stylized facts}{13}{subsection.4.1}\protected@file@percent }56\@writefile{toc}{\contentsline {section}{\numberline {5}Econometric Framework}{13}{section.5}\protected@file@percent }57\newlabel{sec:framework}{{5}{13}{Econometric Framework}{section.5}{}}58\@writefile{toc}{\contentsline {paragraph}{(i) Count models.}{14}{paragraph*.19}\protected@file@percent }59\newlabel{eq:poisson}{{1}{14}{(i) Count models}{equation.1}{}}60\@writefile{toc}{\contentsline {paragraph}{(ii) Hedonic capital regressions.}{14}{paragraph*.20}\protected@file@percent }61\newlabel{eq:hedonic}{{2}{14}{(ii) Hedonic capital regressions}{equation.2}{}}62\@writefile{toc}{\contentsline {paragraph}{(iii) Diffusion.}{14}{paragraph*.21}\protected@file@percent }63\@writefile{toc}{\contentsline {paragraph}{(iv) Difference-in-differences.}{14}{paragraph*.22}\protected@file@percent }64\newlabel{eq:did}{{3}{14}{(iv) Difference-in-differences}{equation.3}{}}65\@writefile{toc}{\contentsline {paragraph}{(v) Survival.}{14}{paragraph*.23}\protected@file@percent }66\@writefile{toc}{\contentsline {paragraph}{(vi) Networks and representations.}{14}{paragraph*.24}\protected@file@percent }67\@writefile{toc}{\contentsline {paragraph}{(vii) Prediction.}{15}{paragraph*.25}\protected@file@percent }68\@writefile{toc}{\contentsline {section}{\numberline {6}The Extensive Margin: Count Models}{15}{section.6}\protected@file@percent }69\newlabel{sec:counts}{{6}{15}{The Extensive Margin: Count Models}{section.6}{}}70\@writefile{lot}{\contentsline {table}{\numberline {4}{\ignorespaces Panel count models of project-announcement intensity. Dependent variable: number of projects announced by firm $i$ in year $t$. Cluster-robust standard errors (by firm) in parentheses. $^{*}p<0.1$, $^{**}p<0.05$, $^{***}p<0.01$.}}{15}{table.caption.26}\protected@file@percent }71\newlabel{tab:counts}{{4}{15}{Panel count models of project-announcement intensity. Dependent variable: number of projects announced by firm $i$ in year $t$. Cluster-robust standard errors (by firm) in parentheses. $^{*}p<0.1$, $^{**}p<0.05$, $^{***}p<0.01$}{table.caption.26}{}}72\@writefile{lof}{\contentsline {figure}{\numberline {5}{\ignorespaces Within-firm drivers of project-announcement intensity (Poisson with firm and year effects, 95\% confidence intervals).}}{16}{figure.caption.27}\protected@file@percent }73\newlabel{fig:countcoef}{{5}{16}{Within-firm drivers of project-announcement intensity (Poisson with firm and year effects, 95\% confidence intervals)}{figure.caption.27}{}}74\@writefile{toc}{\contentsline {section}{\numberline {7}The Intensive Margin: What Drives Announced Capital?}{16}{section.7}\protected@file@percent }75\newlabel{sec:amount}{{7}{16}{The Intensive Margin: What Drives Announced Capital?}{section.7}{}}76\@writefile{lot}{\contentsline {table}{\numberline {5}{\ignorespaces Hedonic regressions for log announced capital (projects with a disclosed amount, $N=9{,}321$). Baseline project type: M\&A integration. Cluster-robust SE (by firm). $^{*}p<0.1$, $^{**}p<0.05$, $^{***}p<0.01$.}}{17}{table.caption.28}\protected@file@percent }77\newlabel{tab:amount}{{5}{17}{Hedonic regressions for log announced capital (projects with a disclosed amount, $N=9{,}321$). Baseline project type: M\&A integration. Cluster-robust SE (by firm). $^{*}p<0.1$, $^{**}p<0.05$, $^{***}p<0.01$}{table.caption.28}{}}78\@writefile{lof}{\contentsline {figure}{\numberline {6}{\ignorespaces Announced-capital premia by project type relative to M\&A integration (firm and year effects, 95\% CI). Capital-intensive physical projects command the largest premia.}}{18}{figure.caption.29}\protected@file@percent }79\newlabel{fig:amountcoef}{{6}{18}{Announced-capital premia by project type relative to M\&A integration (firm and year effects, 95\% CI). Capital-intensive physical projects command the largest premia}{figure.caption.29}{}}80\@writefile{toc}{\contentsline {section}{\numberline {8}Technology Diffusion: The Rise of Corporate AI}{18}{section.8}\protected@file@percent }81\newlabel{sec:diffusion}{{8}{18}{Technology Diffusion: The Rise of Corporate AI}{section.8}{}}82\@writefile{lof}{\contentsline {figure}{\numberline {7}{\ignorespaces AI-project diffusion. Left: cumulative share of firms ever announcing an AI project, with a fitted logistic (S-curve) diffusion model; the vertical line marks the estimated inflection year ($\approx $2023). Right: AI share of all new project announcements by year.}}{19}{figure.caption.30}\protected@file@percent }83\newlabel{fig:diffusion}{{7}{19}{AI-project diffusion. Left: cumulative share of firms ever announcing an AI project, with a fitted logistic (S-curve) diffusion model; the vertical line marks the estimated inflection year ($\approx $2023). Right: AI share of all new project announcements by year}{figure.caption.30}{}}84\@writefile{lot}{\contentsline {table}{\numberline {6}{\ignorespaces Firm-year logit for the probability of announcing $\geq 1$ AI project. Average marginal effects in the right column. Cluster-robust SE (by firm). $^{*}p<0.1$, $^{**}p<0.05$, $^{***}p<0.01$.}}{19}{table.caption.31}\protected@file@percent }85\newlabel{tab:ailogit}{{6}{19}{Firm-year logit for the probability of announcing $\geq 1$ AI project. Average marginal effects in the right column. Cluster-robust SE (by firm). $^{*}p<0.1$, $^{**}p<0.05$, $^{***}p<0.01$}{table.caption.31}{}}86\@writefile{toc}{\contentsline {section}{\numberline {9}COVID-19 and Digitalization: A Difference-in-Differences Test}{19}{section.9}\protected@file@percent }87\newlabel{sec:did}{{9}{19}{COVID-19 and Digitalization: A Difference-in-Differences Test}{section.9}{}}88\@writefile{lof}{\contentsline {figure}{\numberline {8}{\ignorespaces Event study of COVID-19 and digitalization. Points are treatment$\times $year coefficients on the digital share of announcements relative to 2019 (95\% CI). Flat pre-trends and post-period coefficients near zero indicate no differential digital acceleration of physically intensive incumbents.}}{20}{figure.caption.32}\protected@file@percent }89\newlabel{fig:did}{{8}{20}{Event study of COVID-19 and digitalization. Points are treatment$\times $year coefficients on the digital share of announcements relative to 2019 (95\% CI). Flat pre-trends and post-period coefficients near zero indicate no differential digital acceleration of physically intensive incumbents}{figure.caption.32}{}}90\@writefile{lot}{\contentsline {table}{\numberline {7}{\ignorespaces Difference-in-differences: COVID-19 and digital project share. Cluster-robust SE (by firm).}}{20}{table.caption.33}\protected@file@percent }91\newlabel{tab:did}{{7}{20}{Difference-in-differences: COVID-19 and digital project share. Cluster-robust SE (by firm)}{table.caption.33}{}}92\@writefile{toc}{\contentsline {section}{\numberline {10}Project Lifecycles: Survival Analysis}{20}{section.10}\protected@file@percent }93\newlabel{sec:survival}{{10}{20}{Project Lifecycles: Survival Analysis}{section.10}{}}94\@writefile{lof}{\contentsline {figure}{\numberline {9}{\ignorespaces Kaplan--Meier survival curves by project family. M\&A integrations complete fastest; digital and AI programs persist longest.}}{21}{figure.caption.34}\protected@file@percent }95\newlabel{fig:km}{{9}{21}{Kaplan--Meier survival curves by project family. M\&A integrations complete fastest; digital and AI programs persist longest}{figure.caption.34}{}}96\@writefile{lot}{\contentsline {table}{\numberline {8}{\ignorespaces Cox proportional-hazards model of time to project completion (hazard ratios). Project-type controls included. $^{*}p<0.1$, $^{**}p<0.05$, $^{***}p<0.01$.}}{21}{table.caption.35}\protected@file@percent }97\newlabel{tab:cox}{{8}{21}{Cox proportional-hazards model of time to project completion (hazard ratios). Project-type controls included. $^{*}p<0.1$, $^{**}p<0.05$, $^{***}p<0.01$}{table.caption.35}{}}98\@writefile{toc}{\contentsline {section}{\numberline {11}The Innovation Network}{22}{section.11}\protected@file@percent }99\newlabel{sec:network}{{11}{22}{The Innovation Network}{section.11}{}}100\@writefile{lof}{\contentsline {figure}{\numberline {10}{\ignorespaces Degree distribution of the 32{,}226-node project knowledge graph (log-log). The heavy tail is consistent with a scale-free-like organization.}}{22}{figure.caption.36}\protected@file@percent }101\newlabel{fig:degree}{{10}{22}{Degree distribution of the 32{,}226-node project knowledge graph (log-log). The heavy tail is consistent with a scale-free-like organization}{figure.caption.36}{}}102\@writefile{lof}{\contentsline {figure}{\numberline {11}{\ignorespaces Technology co-occurrence network (top 60 technologies, edges for co-use in $\geq 2$ projects). Node color denotes community; node size scales with project frequency. Four ecosystems emerge (modularity 0.52), with energy-transition technologies most central.}}{23}{figure.caption.37}\protected@file@percent }103\newlabel{fig:technet}{{11}{23}{Technology co-occurrence network (top 60 technologies, edges for co-use in $\geq 2$ projects). Node color denotes community; node size scales with project frequency. Four ecosystems emerge (modularity 0.52), with energy-transition technologies most central}{figure.caption.37}{}}104\@writefile{lot}{\contentsline {table}{\numberline {9}{\ignorespaces Most central technologies in the co-occurrence network (eigenvector centrality) with project frequency and community membership.}}{23}{table.caption.38}\protected@file@percent }105\newlabel{tab:network}{{9}{23}{Most central technologies in the co-occurrence network (eigenvector centrality) with project frequency and community membership}{table.caption.38}{}}106\@writefile{toc}{\contentsline {section}{\numberline {12}Latent Themes: Embedding the Project Corpus}{24}{section.12}\protected@file@percent }107\newlabel{sec:embeddings}{{12}{24}{Latent Themes: Embedding the Project Corpus}{section.12}{}}108\@writefile{lof}{\contentsline {figure}{\numberline {12}{\ignorespaces Two-dimensional t-SNE projection of project embeddings (stratified sample of 5{,}000 projects), colored by $K$-means cluster. Themes separate while overlapping at the boundaries, reflecting the continuous semantic space of corporate initiatives.}}{24}{figure.caption.39}\protected@file@percent }109\newlabel{fig:tsne}{{12}{24}{Two-dimensional t-SNE projection of project embeddings (stratified sample of 5{,}000 projects), colored by $K$-means cluster. Themes separate while overlapping at the boundaries, reflecting the continuous semantic space of corporate initiatives}{figure.caption.39}{}}110\@writefile{lof}{\contentsline {figure}{\numberline {13}{\ignorespaces Silhouette score versus number of clusters; $K=6$ is selected.}}{25}{figure.caption.40}\protected@file@percent }111\newlabel{fig:silhouette}{{13}{25}{Silhouette score versus number of clusters; $K=6$ is selected}{figure.caption.40}{}}112\@writefile{lot}{\contentsline {table}{\numberline {10}{\ignorespaces Embedding clusters: size, dominant project type, top technologies, and aggregate announced capital. Clusters cross sector boundaries, capturing themes the taxonomy alone does not.}}{25}{table.caption.41}\protected@file@percent }113\newlabel{tab:clusters}{{10}{25}{Embedding clusters: size, dominant project type, top technologies, and aggregate announced capital. Clusters cross sector boundaries, capturing themes the taxonomy alone does not}{table.caption.41}{}}114\@writefile{toc}{\contentsline {section}{\numberline {13}Predicting Mega-Projects}{25}{section.13}\protected@file@percent }115\newlabel{sec:ml}{{13}{25}{Predicting Mega-Projects}{section.13}{}}116\@writefile{lof}{\contentsline {figure}{\numberline {14}{\ignorespaces Mega-project classifier. Left: 5-fold cross-validated ROC curve (AUC = 0.74). Right: permutation feature importance (drop in AUC).}}{26}{figure.caption.42}\protected@file@percent }117\newlabel{fig:roc}{{14}{26}{Mega-project classifier. Left: 5-fold cross-validated ROC curve (AUC = 0.74). Right: permutation feature importance (drop in AUC)}{figure.caption.42}{}}118\@writefile{lot}{\contentsline {table}{\numberline {11}{\ignorespaces Gradient-boosted mega-project classifier: feature-group importance and performance ($N=9{,}321$ projects with disclosed amounts).}}{26}{table.caption.43}\protected@file@percent }119\newlabel{tab:ml}{{11}{26}{Gradient-boosted mega-project classifier: feature-group importance and performance ($N=9{,}321$ projects with disclosed amounts)}{table.caption.43}{}}120\@writefile{toc}{\contentsline {section}{\numberline {14}Robustness and Validation}{26}{section.14}\protected@file@percent }121\newlabel{sec:robustness}{{14}{26}{Robustness and Validation}{section.14}{}}122\@writefile{toc}{\contentsline {paragraph}{Extraction quality.}{26}{paragraph*.44}\protected@file@percent }123\@writefile{toc}{\contentsline {paragraph}{Estimator robustness.}{27}{paragraph*.45}\protected@file@percent }124\@writefile{toc}{\contentsline {paragraph}{Parallel trends.}{27}{paragraph*.46}\protected@file@percent }125\@writefile{toc}{\contentsline {paragraph}{Level vs.\ composition.}{27}{paragraph*.47}\protected@file@percent }126\@writefile{toc}{\contentsline {paragraph}{Sample.}{27}{paragraph*.48}\protected@file@percent }127\@writefile{toc}{\contentsline {section}{\numberline {15}Limitations}{27}{section.15}\protected@file@percent }128\newlabel{sec:limitations}{{15}{27}{Limitations}{section.15}{}}129\@writefile{toc}{\contentsline {section}{\numberline {16}Discussion and Implications}{28}{section.16}\protected@file@percent }130\newlabel{sec:discussion}{{16}{28}{Discussion and Implications}{section.16}{}}131\@writefile{toc}{\contentsline {paragraph}{For methodology.}{28}{paragraph*.49}\protected@file@percent }132\@writefile{toc}{\contentsline {paragraph}{For the study of investment.}{28}{paragraph*.50}\protected@file@percent }133\@writefile{toc}{\contentsline {paragraph}{For the AI-diffusion debate.}{28}{paragraph*.51}\protected@file@percent }134\@writefile{toc}{\contentsline {paragraph}{For practice.}{28}{paragraph*.52}\protected@file@percent }135\@writefile{toc}{\contentsline {section}{\numberline {17}Conclusion}{29}{section.17}\protected@file@percent }136\newlabel{sec:conclusion}{{17}{29}{Conclusion}{section.17}{}}137\bibcite{acemoglu2022artificial}{{1}{2022}{{Acemoglu et al.}}{{}}}138\bibcite{babina2024artificial}{{2}{2024}{{Babina et al.}}{{}}}139\bibcite{bass1969new}{{3}{1969}{{Bass}}{{}}}140\bibcite{bybee2023ghost}{{4}{2023}{{Bybee et al.}}{{}}}141\bibcite{crouzet2019understanding}{{5}{2019}{{Crouzet \& Eberly}}{{}}}142\@writefile{toc}{\contentsline {section}{\numberline {A}Project Taxonomy and Variable Definitions}{30}{appendix.A}\protected@file@percent }143\newlabel{app:vars}{{A}{30}{Project Taxonomy and Variable Definitions}{appendix.A}{}}144\@writefile{toc}{\contentsline {section}{\numberline {B}Pipeline and Reproducibility}{30}{appendix.B}\protected@file@percent }145\newlabel{app:repro}{{B}{30}{Pipeline and Reproducibility}{appendix.B}{}}146\bibcite{fazzari1988financing}{{6}{1988}{{Fazzari et al.}}{{}}}147\bibcite{gentzkow2019text}{{7}{2019}{{Gentzkow et al.}}{{}}}148\bibcite{griliches1957hybrid}{{8}{1957}{{Griliches}}{{}}}149\bibcite{hassan2019firm}{{9}{2019}{{Hassan et al.}}{{}}}150\bibcite{hayashi1982tobin}{{10}{1982}{{Hayashi}}{{}}}151\bibcite{hidalgo2007product}{{11}{2007}{{Hidalgo et al.}}{{}}}152\bibcite{hoberg2016text}{{12}{2016}{{Hoberg \& Phillips}}{{}}}153\bibcite{korinek2023language}{{13}{2023}{{Korinek}}{{}}}154\bibcite{li2008annual}{{14}{2008}{{Li}}{{}}}155\bibcite{loughran2011liability}{{15}{2011}{{Loughran \& McDonald}}{{}}}156\bibcite{peters2017intangible}{{16}{2017}{{Peters \& Taylor}}{{}}}157\gdef \@abspage@last{31}158