Commit
FIRC0002 Stage E exhibits (Agent 9): every table and figure rebuilt to the Part X standard — EDIT-11/16, ADVE-12/13/14, ECON-12/14/16/17, DATA-05/09, LITE-11/12, AI_C-15, THEO-17, engine audit (notes blocks, unreferenced tables)
- scripts/r3k_20_tables_v2.py: 34 self-contained table floats (\ExhibitTable wrapper: booktabs + threeparttable, caption, label, notes naming the inference, N/clusters/R2 rows, economic-magnitude line, generated-by comment); robust statistics in every headline table (winsorized sorts with NW/plain t and CIs, two-way t, wild p, MDE, CDA by benchmark); BEFORE values only in tab_hygiene Panel B; new tab_mht summary + member-level parts A/B; appendix_exhibits.tex with \ExhibitOnce guards; EXHIBIT_MAP.json/.md; macros.tex keeps the 442 legacy names and carries the wrapper macros + 43 new macros. - scripts/r3k_19_figures_v2.py: 17 figures redrawn (Okabe-Ito, Times 8-9 pt, width = textwidth, markers + dashes + hatches for grayscale, 95% intervals from stated procedures, CDA band on the event-study path, event-time reference year at zero, spec curve with NW SE bars) + float wrappers paper/figures_r3k/<name>.tex with scientific captions; old PNG copies in paper/figures_r3k removed (PDFs only). - Render-and-inspect audit done on every exhibit page (see quality_control/_FIRC-0002_*_audit.csv). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
added
figures/r3k_v2/FIGURE_MAP.json
+138 −0
@@ -0,0 +1,138 @@ | ||
| 1 | +[ | |
| 2 | + { | |
| 3 | + "kind": "figure", | |
| 4 | + "file": "figures/r3k_v2/fig1_quintiles.pdf", | |
| 5 | + "wrapper": "paper/figures_r3k/fig1_quintiles.tex", | |
| 6 | + "label": "fig:quintiles", | |
| 7 | + "title": "Mean forward return by AI-composite quintile before and after the ChatGPT launch.", | |
| 8 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 9 | + }, | |
| 10 | + { | |
| 11 | + "kind": "figure", | |
| 12 | + "file": "figures/r3k_v2/fig2_cumulative_ls.pdf", | |
| 13 | + "wrapper": "paper/figures_r3k/fig2_cumulative_ls.tex", | |
| 14 | + "label": "fig:cumls", | |
| 15 | + "title": "Cumulative return of the AI-composite long--short portfolio by size group.", | |
| 16 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 17 | + }, | |
| 18 | + { | |
| 19 | + "kind": "figure", | |
| 20 | + "file": "figures/r3k_v2/fig3_cohort_spreads.pdf", | |
| 21 | + "wrapper": "paper/figures_r3k/fig3_cohort_spreads.tex", | |
| 22 | + "label": "fig:cohorts", | |
| 23 | + "title": "Annual AI-composite spreads by formation cohort, Russell~3000.", | |
| 24 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 25 | + }, | |
| 26 | + { | |
| 27 | + "kind": "figure", | |
| 28 | + "file": "figures/r3k_v2/fig4_trend.pdf", | |
| 29 | + "wrapper": "paper/figures_r3k/fig4_trend.tex", | |
| 30 | + "label": "fig:trend", | |
| 31 | + "title": "The secular rise of AI language in SEC filings.", | |
| 32 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 33 | + }, | |
| 34 | + { | |
| 35 | + "kind": "figure", | |
| 36 | + "file": "figures/r3k_v2/fig5_fm_byyear.pdf", | |
| 37 | + "wrapper": "paper/figures_r3k/fig5_fm_byyear.tex", | |
| 38 | + "label": "fig:fmyear", | |
| 39 | + "title": "Year-by-year Fama--MacBeth slopes on the AI composite, Russell~3000.", | |
| 40 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 41 | + }, | |
| 42 | + { | |
| 43 | + "kind": "figure", | |
| 44 | + "file": "figures/r3k_v2/fig6_placebo.pdf", | |
| 45 | + "wrapper": "paper/figures_r3k/fig6_placebo.tex", | |
| 46 | + "label": "fig:placebo", | |
| 47 | + "title": "Placebo across eight semantic factors scored with the same pipeline.", | |
| 48 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 49 | + }, | |
| 50 | + { | |
| 51 | + "kind": "figure", | |
| 52 | + "file": "figures/r3k_v2/fig7_event_car.pdf", | |
| 53 | + "wrapper": "paper/figures_r3k/fig7_event_car.tex", | |
| 54 | + "label": "fig:event", | |
| 55 | + "title": "Cumulative abnormal returns around the ChatGPT launch by pre-launch AI tercile.", | |
| 56 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 57 | + }, | |
| 58 | + { | |
| 59 | + "kind": "figure", | |
| 60 | + "file": "figures/r3k_v2/fig8_double_sort.pdf", | |
| 61 | + "wrapper": "paper/figures_r3k/fig8_double_sort.tex", | |
| 62 | + "label": "fig:double", | |
| 63 | + "title": "AI language $\\times$ size double sort before and after the launch, Russell~3000.", | |
| 64 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 65 | + }, | |
| 66 | + { | |
| 67 | + "kind": "figure", | |
| 68 | + "file": "figures/r3k_v2/fig9_eventtime.pdf", | |
| 69 | + "wrapper": "paper/figures_r3k/fig9_eventtime.tex", | |
| 70 | + "label": "fig:eventtime", | |
| 71 | + "title": "Event-time profile of the premium on pre-determined AI exposure, Russell~3000.", | |
| 72 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 73 | + }, | |
| 74 | + { | |
| 75 | + "kind": "figure", | |
| 76 | + "file": "figures/r3k_v2/fig10_filing_type.pdf", | |
| 77 | + "wrapper": "paper/figures_r3k/fig10_filing_type.tex", | |
| 78 | + "label": "fig:filingtype", | |
| 79 | + "title": "Pre- and post-launch spreads of the form-specific AI composites.", | |
| 80 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 81 | + }, | |
| 82 | + { | |
| 83 | + "kind": "figure", | |
| 84 | + "file": "figures/r3k_v2/fig11_placebo_breaks.pdf", | |
| 85 | + "wrapper": "paper/figures_r3k/fig11_placebo_breaks.tex", | |
| 86 | + "label": "fig:placebobreaks", | |
| 87 | + "title": "Placebo break dates and contiguous-block comparison, AI composite, Russell~3000.", | |
| 88 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 89 | + }, | |
| 90 | + { | |
| 91 | + "kind": "figure", | |
| 92 | + "file": "figures/r3k_v2/fig12_size_groups.pdf", | |
| 93 | + "wrapper": "paper/figures_r3k/fig12_size_groups.tex", | |
| 94 | + "label": "fig:sizegroups", | |
| 95 | + "title": "Pre- and post-launch AI-composite spreads by size group and index membership.", | |
| 96 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 97 | + }, | |
| 98 | + { | |
| 99 | + "kind": "figure", | |
| 100 | + "file": "figures/r3k_v2/fig13_cohorts_both.pdf", | |
| 101 | + "wrapper": "paper/figures_r3k/fig13_cohorts_both.tex", | |
| 102 | + "label": "fig:cohortsboth", | |
| 103 | + "title": "Annual AI-composite spreads by cohort, Russell~3000 universe versus S\\&P~500 members.", | |
| 104 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 105 | + }, | |
| 106 | + { | |
| 107 | + "kind": "figure", | |
| 108 | + "file": "figures/r3k_v2/fig14_cumulative_both.pdf", | |
| 109 | + "wrapper": "paper/figures_r3k/fig14_cumulative_both.tex", | |
| 110 | + "label": "fig:cumboth", | |
| 111 | + "title": "Cumulative long--short return on the AI composite in the two universes.", | |
| 112 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 113 | + }, | |
| 114 | + { | |
| 115 | + "kind": "figure", | |
| 116 | + "file": "figures/r3k_v2/fig15_spec_curve.pdf", | |
| 117 | + "wrapper": "paper/figures_r3k/fig15_spec_curve.tex", | |
| 118 | + "label": "fig:speccurve", | |
| 119 | + "title": "Specification curve of the AI-composite spread, Russell~3000.", | |
| 120 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 121 | + }, | |
| 122 | + { | |
| 123 | + "kind": "figure", | |
| 124 | + "file": "figures/r3k_v2/fig16_placebo_launches.pdf", | |
| 125 | + "wrapper": "paper/figures_r3k/fig16_placebo_launches.tex", | |
| 126 | + "label": "fig:placebolaunches", | |
| 127 | + "title": "Placebo launch dates: the event-study design applied to 30 November of 2015--2021.", | |
| 128 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 129 | + }, | |
| 130 | + { | |
| 131 | + "kind": "figure", | |
| 132 | + "file": "figures/r3k_v2/fig17_attrition.pdf", | |
| 133 | + "wrapper": "paper/figures_r3k/fig17_attrition.tex", | |
| 134 | + "label": "fig:attrition", | |
| 135 | + "title": "Attrition of IWV constituents by market-value tercile, from dated holdings captures.", | |
| 136 | + "script": "scripts/r3k_19_figures_v2.py" | |
| 137 | + } | |
| 138 | +] | |
| \ No newline at end of file | ||
modified
figures/r3k_v2/fig10_filing_type.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig10_filing_type.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig11_placebo_breaks.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig11_placebo_breaks.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig12_size_groups.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig12_size_groups.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig13_cohorts_both.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig13_cohorts_both.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig14_cumulative_both.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig14_cumulative_both.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig15_spec_curve.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig15_spec_curve.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig16_placebo_launches.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig16_placebo_launches.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig17_attrition.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig17_attrition.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig1_quintiles.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig1_quintiles.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig2_cumulative_ls.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig2_cumulative_ls.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig3_cohort_spreads.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig3_cohort_spreads.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig4_trend.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig4_trend.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig5_fm_byyear.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig5_fm_byyear.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig6_placebo.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig6_placebo.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig7_event_car.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig7_event_car.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig8_double_sort.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig8_double_sort.png
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig9_eventtime.pdf
+0 −0
Binary file not shown.
modified
figures/r3k_v2/fig9_eventtime.png
+0 −0
Binary file not shown.
modified
paper/figures_r3k/fig10_filing_type.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig10_filing_type.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig10_filing_type.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig10_filing_type.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig10_filing_type} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:filingtype}{\textbf{Pre- and post-launch spreads of the form-specific AI composites.} Equal-weighted Q5$-$Q1 spreads (forward return winsorized within the formation year) of the composite rebuilt from one filing form at a time and of the equal-form composite, averaged over the pre- and post-launch cohorts with 95\% $t$-intervals over cohorts. The HAC Wald test of equal Fama--MacBeth slopes across the three forms has $p$ = 0.016 on winsorized and 0.234 on raw returns (Table~\ref{tab:filingtype}), so the ranking of the forms is not statistically established.}{\includegraphics[width=\textwidth]{fig10_filing_type.pdf}} | |
modified
paper/figures_r3k/fig11_placebo_breaks.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig11_placebo_breaks.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig11_placebo_breaks.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig11_placebo_breaks.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig11_placebo_breaks} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:placebobreaks}{\textbf{Placebo break dates and contiguous-block comparison, AI composite, Russell~3000.} Left: difference between the mean winsorized Q5$-$Q1 spread of the cohorts from the candidate break year onward and the mean of the earlier cohorts, for every candidate between 2013 and 2022; candidates with few post cohorts have larger sampling variance, so the ranking is descriptive (bootstrap and Romano--Wolf $p$-values in Table~\ref{tab:mhtfullb}). Right: mean spread of each contiguous three-cohort block minus the mean of the remaining cohorts; the 2022--24 block is exceeded by 2 of the 12 blocks and by a share 0.31 of all 455 three-cohort subsets (exact permutation; Table~\ref{tab:cohorts}).}{\includegraphics[width=\textwidth]{fig11_placebo_breaks.pdf}} | |
modified
paper/figures_r3k/fig12_size_groups.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig12_size_groups.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig12_size_groups.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig12_size_groups.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig12_size_groups} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:sizegroups}{\textbf{Pre- and post-launch AI-composite spreads by size group and index membership.} Equal-weighted Q5$-$Q1 spreads with quintiles formed within each sub-sample and cohort, averaged over the pre-launch (2010--21, twelve cohorts) and post-launch (2022--24, three cohorts) formation cohorts, on the raw and the within-year winsorized return, with 95\% $t$-intervals over cohorts. The post-launch intervals are wide because they rest on three cohorts; the point estimates rise for the largest 1,000 firms and the S\&P~500 members only, but the formal large-versus-small contrast (+10.8 pp, firm $t$ 1.83, two-way $t$ 1.55, wild $p$ 0.22, MDE 16.5 pp; Table~\ref{tab:contrasts}) is not significant and the large-cap post mean without the 2024 cohort equals its pre mean (Table~\ref{tab:groups}).}{\includegraphics[width=\textwidth]{fig12_size_groups.pdf}} | |
modified
paper/figures_r3k/fig13_cohorts_both.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig13_cohorts_both.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig13_cohorts_both.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig13_cohorts_both.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig13_cohorts_both} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:cohortsboth}{\textbf{Annual AI-composite spreads by cohort, Russell~3000 universe versus S\&P~500 members.} Equal-weighted Q5$-$Q1 spread of the AI composite (forward return winsorized within the formation year) for each formation cohort, with quintiles formed within each universe and cohort. In both universes the 2022 cohort is negative and the post-launch average is driven by the 2024 cohort (Russell~3000 post cohorts: 2022: -5.0, 2023: +3.1, 2024: +19.9 pp; S\&P~500: 2022: +14.8, 2023: -7.1, 2024: +32.0 pp). Cohort-level confidence intervals and the pooled member-versus-non-member test are in Tables~\ref{tab:prepostboth} and \ref{tab:contrasts}.}{\includegraphics[width=\textwidth]{fig13_cohorts_both.pdf}} | |
modified
paper/figures_r3k/fig14_cumulative_both.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig14_cumulative_both.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig14_cumulative_both.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig14_cumulative_both.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig14_cumulative_both} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:cumboth}{\textbf{Cumulative long--short return on the AI composite in the two universes.} Cumulative return of the equal-weighted, annually selected (June) Q5$-$Q1 portfolio on the AI composite, held July--June with monthly rebalancing within the year, in the Russell~3000 universe and among S\&P~500 members (180 months). Mean returns, Sharpe ratios and factor alphas with HAC inference are in Table~\ref{tab:alphaboth}; a cumulative path carries no interval.}{\includegraphics[width=\textwidth]{fig14_cumulative_both.pdf}} | |
added
paper/figures_r3k/fig15_spec_curve.pdf
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig15_spec_curve.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig15_spec_curve.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig15_spec_curve} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:speccurve}{\textbf{Specification curve of the AI-composite spread, Russell~3000.} Top: the equal- or value-weighted Q5$-$Q1 spread of the AI composite in each of the 1,152 specifications obtained by crossing the universe (all firms, excluding the 500 largest, largest 1,000, remaining), the return treatment (raw, 1/99 or 5/95 winsorized within year, median), the weighting, the horizon, the filing form and the chunk aggregation, ranked by size; thin bars are $\pm 1.96$ Newey--West standard errors over cohorts; 84\% of specifications are positive, 39\% positive with $|$NW $t| > 1.96$ (35\% with the plain $t$), median +3.9 pp, interquartile range [+0.7, +8.5]. Bottom: the specification choices of each ranked estimate. Specifications are not independent (they share cohorts and firms), so the share significant is descriptive, not a test.}{\includegraphics[width=\textwidth]{fig15_spec_curve.pdf}} | |
added
paper/figures_r3k/fig16_placebo_launches.pdf
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig16_placebo_launches.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig16_placebo_launches.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig16_placebo_launches} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:placebolaunches}{\textbf{Placebo launch dates: the event-study design applied to 30 November of 2015--2021.} The ChatGPT event study (IWV market model, terciles of the prior cohort's AI composite, one-year horizon) re-run with a fictitious launch on 30 November of each year 2015--2021, beside the actual 2022 launch. Left: one-year High$-$Low cumulative abnormal return for the full sample and the largest 1,000 firms; right: the dependence-robust CDA $t$ of the full-sample High$-$Low portfolio with the $\pm 1.96$ reference lines. The 2022 launch ranks 1 of 8 by High$-$Low and 1 by $t$; the placebo $t$-statistics range over -1.11 to 1.61 (Table~\ref{tab:eventcda}).}{\includegraphics[width=\textwidth]{fig16_placebo_launches.pdf}} | |
added
paper/figures_r3k/fig17_attrition.pdf
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig17_attrition.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig17_attrition.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig17_attrition} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:attrition}{\textbf{Attrition of IWV constituents by market-value tercile, from dated holdings captures.} Share of the constituents of each dated iShares Russell~3000 ETF holdings file (17 Internet Archive captures, 2006 and May 2019--August 2025) that are no longer held on 28 August 2025, by market-value tercile within the capture. The small tercile's annualised exit rate is two to four times the large tercile's (Table~\ref{tab:attrition}). No capture exists between 2007 and May 2019, and delisted firms have no AI scores, so attrition by AI quintile cannot be computed; the figure documents the size gradient of the survivorship that conditions the whole panel.}{\includegraphics[width=\textwidth]{fig17_attrition.pdf}} | |
modified
paper/figures_r3k/fig1_quintiles.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig1_quintiles.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig1_quintiles.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig1_quintiles.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig1_quintiles} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:quintiles}{\textbf{Mean forward return by AI-composite quintile before and after the ChatGPT launch.} Equal-weighted mean forward twelve-month return of the within-cohort quintiles of the AI composite, averaged over the pre-launch cohorts (2010--2021) and the post-launch cohorts (2022--2024), for the raw return (left) and the return winsorized at 1/99 within the formation year (right). Error bars are 95\% $t$-intervals over cohorts; the post-launch intervals rest on three cohorts. Russell~3000 universe, $N$ = 27,354 firm-years. The winsorized full-sample Q5$-$Q1 spread is +3.9 pp (Table~\ref{tab:sortsboth}).}{\includegraphics[width=\textwidth]{fig1_quintiles.pdf}} | |
modified
paper/figures_r3k/fig2_cumulative_ls.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig2_cumulative_ls.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig2_cumulative_ls.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig2_cumulative_ls.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig2_cumulative_ls} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:cumls}{\textbf{Cumulative return of the AI-composite long--short portfolio by size group.} Cumulative return of the equal-weighted Q5$-$Q1 portfolio on the AI composite, selected every June from the cohort quintiles and held July--June, for all firms and within the two size groups (largest 1,000 firms by corrected capitalization versus the rest; quintiles formed within group). Monthly returns, splice months set to missing; 180 months. No uncertainty band is drawn on a cumulative path; the monthly mean, HAC $t$ and alphas are in Table~\ref{tab:alphaboth}. The lines use distinct dashes and markers so that they remain distinguishable in grayscale.}{\includegraphics[width=\textwidth]{fig2_cumulative_ls.pdf}} | |
modified
paper/figures_r3k/fig3_cohort_spreads.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig3_cohort_spreads.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig3_cohort_spreads.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig3_cohort_spreads.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig3_cohort_spreads} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:cohorts}{\textbf{Annual AI-composite spreads by formation cohort, Russell~3000.} Equal-weighted Q5$-$Q1 forward-return spread of the AI composite for each formation cohort, on the raw return, the return winsorized at 1/99 within the formation year, and as the median difference. The dashed line and the shaded band are the mean and the 95\% $t$-interval of the twelve pre-launch winsorized spreads (+3.4 pp, [-1.6, +8.4]); the dotted vertical line separates the cohorts held before and after the ChatGPT launch. The post-launch cohorts are 2022: -5.0, 2023: +3.1, 2024: +19.9 pp (wins.); tests of a 2022 break are in Table~\ref{tab:cohorts}.}{\includegraphics[width=\textwidth]{fig3_cohort_spreads.pdf}} | |
modified
paper/figures_r3k/fig4_trend.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig4_trend.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig4_trend.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig4_trend.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig4_trend} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:trend}{\textbf{The secular rise of AI language in SEC filings.} Cross-sectional mean of the section-weighted cosine similarity of the firm-year's filing chunks to the AI-adoption anchor, for all firms in the estimation sample, the largest 1,000 firms by corrected capitalization and the remaining firms, and of the AI-infrastructure score for all firms, by filing year (2010--2024; $N$ = 27,354 firm-years). The series are unstandardized levels (the regressions use within-year $z$-scores), so the rise is the secular drift that the within-year standardization removes. No confidence band is drawn: with more than a thousand firms per year the standard error of the mean is smaller than the marker.}{\includegraphics[width=\textwidth]{fig4_trend.pdf}} | |
modified
paper/figures_r3k/fig5_fm_byyear.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig5_fm_byyear.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig5_fm_byyear.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig5_fm_byyear.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig5_fm_byyear} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:fmyear}{\textbf{Year-by-year Fama--MacBeth slopes on the AI composite, Russell~3000.} Slope of the forward twelve-month return on the within-year standardized AI composite in each annual cross-sectional regression with size, momentum and volatility controls, for the raw and the within-year winsorized return, with 95\% confidence intervals from heteroskedasticity-robust standard errors within the year. The dashed line is the time-series mean of the winsorized slopes (+1.88 pp per SD, NW $t$ 2.41; Table~\ref{tab:fmboth}). Within-year intervals do not account for the cross-cohort dependence that the Newey--West and fixed-$b$ inference of the tables addresses.}{\includegraphics[width=\textwidth]{fig5_fm_byyear.pdf}} | |
modified
paper/figures_r3k/fig6_placebo.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig6_placebo.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig6_placebo.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig6_placebo.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig6_placebo} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:placebo}{\textbf{Placebo across eight semantic factors scored with the same pipeline.} Equal-weighted Q5$-$Q1 spreads of the forward return (winsorized at 1/99 within the formation year) for the three AI anchors and five non-AI anchors scored identically, averaged over the pre- and post-launch cohorts, with 95\% $t$-intervals over cohorts. The post-launch intervals rest on three cohorts and are correspondingly wide; the pooled interaction tests with cohort-level inference and the multiple-testing adjustment over the eight factors are in Table~\ref{tab:placebo}.}{\includegraphics[width=\textwidth]{fig6_placebo.pdf}} | |
modified
paper/figures_r3k/fig7_event_car.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig7_event_car.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig7_event_car.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig7_event_car.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig7_event_car} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:event}{\textbf{Cumulative abnormal returns around the ChatGPT launch by pre-launch AI tercile.} Equal-weighted cumulative market-model abnormal returns of the three terciles of 2021-cohort AI exposure (2,215 firms; IWV market model estimated 2022-01-03..2022-10-31 (209 days)) and of the High$-$Low portfolio, from twenty trading days before the launch to 395 days after. The shaded band is $\pm 1.96\,\hat\sigma\sqrt{k}$ around the High$-$Low level on day 0, where $\hat\sigma$ is the estimation-window daily standard deviation of the High$-$Low portfolio implied by the CDA statistic (crude dependence adjustment; the band widens with the horizon $k$). All terciles drift negative because the model was estimated in the 2022 bear market; only differences between terciles are interpretable, and the one-year High$-$Low difference of +14.2 pp (CDA $t$ 2.23) depends on the benchmark (Table~\ref{tab:eventcda}).}{\includegraphics[width=\textwidth]{fig7_event_car.pdf}} | |
modified
paper/figures_r3k/fig8_double_sort.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig8_double_sort.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig8_double_sort.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig8_double_sort.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig8_double_sort} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:double}{\textbf{AI language $\times$ size double sort before and after the launch, Russell~3000.} Mean forward twelve-month return (winsorized at 1/99 within the formation year, percent per year) by tercile of the AI composite within tercile of the corrected market capitalization, both formed within cohort on ranks, averaged over the pre-launch and the post-launch cohorts (viridis scale, common across panels; cell values printed; High$-$Low differences in the panel titles). The differences are descriptive; the formal large-versus-small test with clustered inference and its minimum detectable effect are in Table~\ref{tab:contrasts}, and the post-launch cells average three cohorts.}{\includegraphics[width=\textwidth]{fig8_double_sort.pdf}} | |
modified
paper/figures_r3k/fig9_eventtime.pdf
+0 −0
Binary file not shown.
deleted
paper/figures_r3k/fig9_eventtime.png
+0 −0
Binary file not shown.
added
paper/figures_r3k/fig9_eventtime.tex
+3 −0
@@ -0,0 +1,3 @@ | ||
| 1 | +% Generated by scripts/r3k_19_figures_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). Figure file: figures/r3k_v2/fig9_eventtime.pdf (copy in paper/figures_r3k/). | |
| 2 | +% \input{figures_r3k/fig9_eventtime} on its own places the float with its caption; inside an existing figure wrapper only the graphic is printed. | |
| 3 | +\ExhibitFigure{fig:eventtime}{\textbf{Event-time profile of the premium on pre-determined AI exposure, Russell~3000.} Coefficient on pre-determined AI exposure $\times$ outcome-cohort dummies in a regression of the forward return (winsorized within the formation year) with the main effect of exposure included, so that each point is the difference from the 2021 reference cohort (zero by construction; the 2021 slope itself is +5.9 pp per SD, $t$ 5.95). Bars are 95\% intervals from firm-clustered standard errors ($N$ = 21,111). The committed v2.0 figure plotted absolute slopes from a model without the main effect, which is why it appeared flat. Shaded cohorts interact outcomes with language written afterwards (look-back selection, not a pre-trend test); the trailing-exposure variant avoids this. The pre-launch differences are not zero, so parallel trends do not hold (Table~\ref{tab:eventtime}).}{\includegraphics[width=\textwidth]{fig9_eventtime.pdf}} | |
added
paper/tables/EXHIBIT_MAP.json
+312 −0
@@ -0,0 +1,312 @@ | ||
| 1 | +{ | |
| 2 | + "generated_by": "scripts/r3k_20_tables_v2.py", | |
| 3 | + "baseline": "3cf7560", | |
| 4 | + "exhibits": [ | |
| 5 | + { | |
| 6 | + "kind": "table", | |
| 7 | + "file": "paper/tables/tab_summary.tex", | |
| 8 | + "label": "tab:summ", | |
| 9 | + "title": "Summary statistics of the estimation sample.", | |
| 10 | + "placement": "main", | |
| 11 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 12 | + "sources": "results/r3k_v2/t1_summary_stats.csv, panel_meta_v2.json, mktcap_diag.json, hygiene_summary.json" | |
| 13 | + }, | |
| 14 | + { | |
| 15 | + "kind": "table", | |
| 16 | + "file": "paper/tables/tab_sorts_both.tex", | |
| 17 | + "label": "tab:sortsboth", | |
| 18 | + "title": "Quintile sorts on AI language in the Russell~3000 universe and among S\\&P~500 members.", | |
| 19 | + "placement": "main", | |
| 20 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 21 | + "sources": "results/r3k_v2/results.json (samples.*.sorts, vw_diag), results2.json (mht)" | |
| 22 | + }, | |
| 23 | + { | |
| 24 | + "kind": "table", | |
| 25 | + "file": "paper/tables/tab_sorts.tex", | |
| 26 | + "label": "tab:sorts", | |
| 27 | + "title": "Quintile sorts on every signal and weighting, Russell~3000 universe.", | |
| 28 | + "placement": "appendix", | |
| 29 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 30 | + "sources": "results/r3k_v2/results.json (sorts, vw_diag)" | |
| 31 | + }, | |
| 32 | + { | |
| 33 | + "kind": "table", | |
| 34 | + "file": "paper/tables/tab_alpha_both.tex", | |
| 35 | + "label": "tab:alphaboth", | |
| 36 | + "title": "Monthly long--short portfolios on AI language: mean return, Sharpe ratio and factor alphas.", | |
| 37 | + "placement": "main", | |
| 38 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 39 | + "sources": "results/r3k_v2/results.json (samples.*.ls), results2.json (mht.deflated_sharpe)" | |
| 40 | + }, | |
| 41 | + { | |
| 42 | + "kind": "table", | |
| 43 | + "file": "paper/tables/tab_alpha.tex", | |
| 44 | + "label": "tab:alpha", | |
| 45 | + "title": "Monthly long--short portfolios by signal and sub-period, Russell~3000 universe.", | |
| 46 | + "placement": "appendix", | |
| 47 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 48 | + "sources": "results/r3k_v2/results.json (ls_alpha)" | |
| 49 | + }, | |
| 50 | + { | |
| 51 | + "kind": "table", | |
| 52 | + "file": "paper/tables/tab_panel.tex", | |
| 53 | + "label": "tab:panel", | |
| 54 | + "title": "Two-way fixed-effects panel regressions of the forward return on AI language, Russell~3000.", | |
| 55 | + "placement": "main", | |
| 56 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 57 | + "sources": "results/r3k_v2/results.json (panel_fe)" | |
| 58 | + }, | |
| 59 | + { | |
| 60 | + "kind": "table", | |
| 61 | + "file": "paper/tables/tab_fm_both.tex", | |
| 62 | + "label": "tab:fmboth", | |
| 63 | + "title": "Fama--MacBeth regressions of the forward return on AI language, Russell~3000 and S\\&P~500.", | |
| 64 | + "placement": "main", | |
| 65 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 66 | + "sources": "results/r3k_v2/results.json (samples.*.fm, fama_macbeth)" | |
| 67 | + }, | |
| 68 | + { | |
| 69 | + "kind": "table", | |
| 70 | + "file": "paper/tables/tab_fm.tex", | |
| 71 | + "label": "tab:fm", | |
| 72 | + "title": "Fama--MacBeth regressions, all specifications, Russell~3000 universe.", | |
| 73 | + "placement": "appendix", | |
| 74 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 75 | + "sources": "results/r3k_v2/results.json (fama_macbeth)" | |
| 76 | + }, | |
| 77 | + { | |
| 78 | + "kind": "table", | |
| 79 | + "file": "paper/tables/tab_prepost_both.tex", | |
| 80 | + "label": "tab:prepostboth", | |
| 81 | + "title": "AI-language spreads before and after the ChatGPT launch, Russell~3000 and S\\&P~500.", | |
| 82 | + "placement": "main", | |
| 83 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 84 | + "sources": "results/r3k_v2/results.json (samples.*.sub, pooled, samples.sp500.pooled_post_w)" | |
| 85 | + }, | |
| 86 | + { | |
| 87 | + "kind": "table", | |
| 88 | + "file": "paper/tables/tab_cohorts.tex", | |
| 89 | + "label": "tab:cohorts", | |
| 90 | + "title": "Cohort-by-cohort spreads of the AI composite and tests of a 2022 break, Russell~3000.", | |
| 91 | + "placement": "main", | |
| 92 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 93 | + "sources": "results/r3k_v2/results2.json (cohorts*, placebo_breaks), results_surv.json (then_members_present)" | |
| 94 | + }, | |
| 95 | + { | |
| 96 | + "kind": "table", | |
| 97 | + "file": "paper/tables/tab_groups.tex", | |
| 98 | + "label": "tab:groups", | |
| 99 | + "title": "AI-composite spreads by size group and index membership, Russell~3000.", | |
| 100 | + "placement": "main", | |
| 101 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 102 | + "sources": "results/r3k_v2/results.json (groups, size_neutral, contrasts)" | |
| 103 | + }, | |
| 104 | + { | |
| 105 | + "kind": "table", | |
| 106 | + "file": "paper/tables/tab_contrasts.tex", | |
| 107 | + "label": "tab:contrasts", | |
| 108 | + "title": "Formal tests of the large-versus-small and member-versus-non-member post-launch differences.", | |
| 109 | + "placement": "main", | |
| 110 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 111 | + "sources": "results/r3k_v2/results.json (contrasts)" | |
| 112 | + }, | |
| 113 | + { | |
| 114 | + "kind": "table", | |
| 115 | + "file": "paper/tables/tab_doublesort.tex", | |
| 116 | + "label": "tab:doublesort", | |
| 117 | + "title": "AI language $\\times$ size double sort, Russell~3000.", | |
| 118 | + "placement": "appendix", | |
| 119 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 120 | + "sources": "results/r3k_v2/results2.json (double_sort_*)" | |
| 121 | + }, | |
| 122 | + { | |
| 123 | + "kind": "table", | |
| 124 | + "file": "paper/tables/tab_sector_tech.tex", | |
| 125 | + "label": "tab:sectortech", | |
| 126 | + "title": "Within-sector estimates of the AI-language premium and technology-factor spanning, Russell~3000.", | |
| 127 | + "placement": "main", | |
| 128 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 129 | + "sources": "results/r3k_v2/results2.json (sector_controls, sector_neutral, residual_sort, monthly_ls)" | |
| 130 | + }, | |
| 131 | + { | |
| 132 | + "kind": "table", | |
| 133 | + "file": "paper/tables/tab_placebo.tex", | |
| 134 | + "label": "tab:placebo", | |
| 135 | + "title": "Placebo across eight semantic factors scored with the same pipeline, Russell~3000.", | |
| 136 | + "placement": "appendix", | |
| 137 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 138 | + "sources": "results/r3k_v2/results2.json (placebo, mht)" | |
| 139 | + }, | |
| 140 | + { | |
| 141 | + "kind": "table", | |
| 142 | + "file": "paper/tables/tab_robust.tex", | |
| 143 | + "label": "tab:robust", | |
| 144 | + "title": "Construction, horizon, benchmark, universe and survivorship checks on the composite spread, Russell~3000.", | |
| 145 | + "placement": "main", | |
| 146 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 147 | + "sources": "results/r3k_v2/results2.json (subperiod_spread, sector_neutral, residual_sort, mag7, alt_*, entrants, breaks, cluster_k), results_surv.json (conditional_spreads), results.json" | |
| 148 | + }, | |
| 149 | + { | |
| 150 | + "kind": "table", | |
| 151 | + "file": "paper/tables/tab_filing_type.tex", | |
| 152 | + "label": "tab:filingtype", | |
| 153 | + "title": "Form-specific AI composites, Russell~3000.", | |
| 154 | + "placement": "appendix", | |
| 155 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 156 | + "sources": "results/r3k_v2/results2.json (filing_type, mht.forms_fm_w)" | |
| 157 | + }, | |
| 158 | + { | |
| 159 | + "kind": "table", | |
| 160 | + "file": "paper/tables/tab_event.tex", | |
| 161 | + "label": "tab:event", | |
| 162 | + "title": "ChatGPT event study: cumulative abnormal returns by pre-launch AI tercile under the IWV market model, Russell~3000.", | |
| 163 | + "placement": "main", | |
| 164 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 165 | + "sources": "results/r3k_v2/results_event.json (by_benchmark.mm_2022, welch_reference, tercile_characteristics)" | |
| 166 | + }, | |
| 167 | + { | |
| 168 | + "kind": "table", | |
| 169 | + "file": "paper/tables/tab_event_cda.tex", | |
| 170 | + "label": "tab:eventcda", | |
| 171 | + "title": "Benchmark dependence of the one-year event-study gradient and placebo launch dates.", | |
| 172 | + "placement": "main", | |
| 173 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 174 | + "sources": "results/r3k_v2/results_event.json (by_benchmark, ff6_calendar_time, placebo_launches, placebo_summary, mht_event)" | |
| 175 | + }, | |
| 176 | + { | |
| 177 | + "kind": "table", | |
| 178 | + "file": "paper/tables/tab_did_spid.tex", | |
| 179 | + "label": "tab:didspid", | |
| 180 | + "title": "Difference-in-differences on pre-determined exposure and the SPID project signal, Russell~3000.", | |
| 181 | + "placement": "main", | |
| 182 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 183 | + "sources": "results/r3k_v2/results2.json (did, spid), t17_refit.json" | |
| 184 | + }, | |
| 185 | + { | |
| 186 | + "kind": "table", | |
| 187 | + "file": "paper/tables/tab_spid_binary.tex", | |
| 188 | + "label": "tab:spidbinary", | |
| 189 | + "title": "Binary contrasts on the SPID project signal with minimum detectable effects, Russell~3000.", | |
| 190 | + "placement": "main", | |
| 191 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 192 | + "sources": "results/r3k_v2/results2.json (spid.binary_contrasts, spid.tercile_nonzero_w), results.json (spid)" | |
| 193 | + }, | |
| 194 | + { | |
| 195 | + "kind": "table", | |
| 196 | + "file": "paper/tables/tab_reversal.tex", | |
| 197 | + "label": "tab:reversal", | |
| 198 | + "title": "Reversal tests and the cybersecurity-vocabulary check, Russell~3000.", | |
| 199 | + "placement": "appendix", | |
| 200 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 201 | + "sources": "results/r3k_v2/results2.json (cluster_k)" | |
| 202 | + }, | |
| 203 | + { | |
| 204 | + "kind": "table", | |
| 205 | + "file": "paper/tables/tab_mht.tex", | |
| 206 | + "label": "tab:mht", | |
| 207 | + "title": "Multiple-testing adjustments by family of tests.", | |
| 208 | + "placement": "main", | |
| 209 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 210 | + "sources": "results/r3k_v2/results2.json (mht, spec_curve), results_event.json (mht_event)" | |
| 211 | + }, | |
| 212 | + { | |
| 213 | + "kind": "table", | |
| 214 | + "file": "paper/tables/tab_mht_full_a.tex", | |
| 215 | + "label": "tab:mhtfull", | |
| 216 | + "title": "Member-level multiple-testing adjustments, part A: portfolio-sort and form families.", | |
| 217 | + "placement": "appendix", | |
| 218 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 219 | + "sources": "results/r3k_v2/results2.json (mht)" | |
| 220 | + }, | |
| 221 | + { | |
| 222 | + "kind": "table", | |
| 223 | + "file": "paper/tables/tab_mht_full_b.tex", | |
| 224 | + "label": "tab:mhtfullb", | |
| 225 | + "title": "Member-level multiple-testing adjustments, part B: robustness rows, placebo interactions and break years.", | |
| 226 | + "placement": "appendix", | |
| 227 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 228 | + "sources": "results/r3k_v2/results2.json (mht)" | |
| 229 | + }, | |
| 230 | + { | |
| 231 | + "kind": "table", | |
| 232 | + "file": "paper/tables/tab_eventtime.tex", | |
| 233 | + "label": "tab:eventtime", | |
| 234 | + "title": "Event-time profile of the premium on pre-determined AI exposure, Russell~3000.", | |
| 235 | + "placement": "appendix", | |
| 236 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 237 | + "sources": "results/r3k_v2/results2.json (did_eventtime*)" | |
| 238 | + }, | |
| 239 | + { | |
| 240 | + "kind": "table", | |
| 241 | + "file": "paper/tables/tab_attrition.tex", | |
| 242 | + "label": "tab:attrition", | |
| 243 | + "title": "Survivorship diagnostics on the current-membership universe, Russell~3000.", | |
| 244 | + "placement": "main", | |
| 245 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 246 | + "sources": "results/r3k_v2/results_surv.json (attrition, then_members_present, ew_vs_index, bounding), iwv_holdings_manifest.json" | |
| 247 | + }, | |
| 248 | + { | |
| 249 | + "kind": "table", | |
| 250 | + "file": "paper/tables/tab_construct_validity.tex", | |
| 251 | + "label": "tab:constructvalidity", | |
| 252 | + "title": "Construct validity of the semantic AI score: keyword baseline, stripped anchors, sector composition and encoder truncation.", | |
| 253 | + "placement": "main", | |
| 254 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 255 | + "sources": "results/r3k_v2/results_construct.json, results/n1/tokenizer_audit.json, results/n1/keyword_baseline.json" | |
| 256 | + }, | |
| 257 | + { | |
| 258 | + "kind": "table", | |
| 259 | + "file": "paper/tables/tab_corr.tex", | |
| 260 | + "label": "tab:corr", | |
| 261 | + "title": "Correlations of the within-year standardized variables.", | |
| 262 | + "placement": "appendix", | |
| 263 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 264 | + "sources": "results/r3k_v2/t1b_correlations.csv" | |
| 265 | + }, | |
| 266 | + { | |
| 267 | + "kind": "table", | |
| 268 | + "file": "paper/tables/tab_topfirms.tex", | |
| 269 | + "label": "tab:topfirms", | |
| 270 | + "title": "Firms with the highest average AI composite.", | |
| 271 | + "placement": "appendix", | |
| 272 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 273 | + "sources": "data/processed_r3k_v2/analysis_panel_v2.parquet" | |
| 274 | + }, | |
| 275 | + { | |
| 276 | + "kind": "table", | |
| 277 | + "file": "paper/tables/tab_coverage.tex", | |
| 278 | + "label": "tab:coverage", | |
| 279 | + "title": "Estimation sample by formation cohort: firms, size groups and text coverage.", | |
| 280 | + "placement": "appendix", | |
| 281 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 282 | + "sources": "results/r3k_v2/results2.json (coverage)" | |
| 283 | + }, | |
| 284 | + { | |
| 285 | + "kind": "table", | |
| 286 | + "file": "paper/tables/tab_corpus.tex", | |
| 287 | + "label": "tab:corpus", | |
| 288 | + "title": "Corpus composition by form.", | |
| 289 | + "placement": "main", | |
| 290 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 291 | + "sources": "data/raw_russell3000/filings, results/n1/corpus_counts.json" | |
| 292 | + }, | |
| 293 | + { | |
| 294 | + "kind": "table", | |
| 295 | + "file": "paper/tables/tab_hygiene.tex", | |
| 296 | + "label": "tab:hygiene", | |
| 297 | + "title": "Return hygiene and the effect of the data corrections on the headline statistics.", | |
| 298 | + "placement": "appendix", | |
| 299 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 300 | + "sources": "results/r3k_v2/hygiene_log.csv, hygiene_summary.json, results/r3k/*.json (HEAD 3cf7560), results/r3k_v2/results*.json" | |
| 301 | + }, | |
| 302 | + { | |
| 303 | + "kind": "table", | |
| 304 | + "file": "paper/tables/tab_break.tex", | |
| 305 | + "label": "tab:break", | |
| 306 | + "title": "Pooled test of a post-launch change in the signal slope, by signal, Russell~3000.", | |
| 307 | + "placement": "appendix", | |
| 308 | + "script": "scripts/r3k_20_tables_v2.py", | |
| 309 | + "sources": "results/r3k_v2/results.json (structural_break, subperiod_spread)" | |
| 310 | + } | |
| 311 | + ] | |
| 312 | +} | |
| \ No newline at end of file | ||
added
paper/tables/EXHIBIT_MAP.md
+57 −0
@@ -0,0 +1,57 @@ | ||
| 1 | +# Exhibit map — FIRC-0002 v2.1 (generated by scripts/r3k_20_tables_v2.py; do not edit) | |
| 2 | + | |
| 3 | +Every table file in `paper/tables/` and every figure wrapper in `paper/figures_r3k/` is a self-contained float: `\input{tables/<name>}` or `\input{figures_r3k/<name>}` prints the float with its caption, label and notes. `paper/tables/appendix_exhibits.tex` (input from `main.tex` after the appendix) prints, behind `\ExhibitOnce` guards, every table not already placed and the appendix-bound figures. Placement = the Stage E decision (main text vs appendix). | |
| 4 | + | |
| 5 | +| Label | File | Placement | Title | Sources | | |
| 6 | +|---|---|---|---|---| | |
| 7 | +| `tab:summ` | `paper/tables/tab_summary.tex` | main | Summary statistics of the estimation sample. | results/r3k_v2/t1_summary_stats.csv, panel_meta_v2.json, mktcap_diag.json, hygiene_summary.json | | |
| 8 | +| `tab:sortsboth` | `paper/tables/tab_sorts_both.tex` | main | Quintile sorts on AI language in the Russell~3000 universe and among S\&P~500 members. | results/r3k_v2/results.json (samples.*.sorts, vw_diag), results2.json (mht) | | |
| 9 | +| `tab:sorts` | `paper/tables/tab_sorts.tex` | appendix | Quintile sorts on every signal and weighting, Russell~3000 universe. | results/r3k_v2/results.json (sorts, vw_diag) | | |
| 10 | +| `tab:alphaboth` | `paper/tables/tab_alpha_both.tex` | main | Monthly long--short portfolios on AI language: mean return, Sharpe ratio and factor alphas. | results/r3k_v2/results.json (samples.*.ls), results2.json (mht.deflated_sharpe) | | |
| 11 | +| `tab:alpha` | `paper/tables/tab_alpha.tex` | appendix | Monthly long--short portfolios by signal and sub-period, Russell~3000 universe. | results/r3k_v2/results.json (ls_alpha) | | |
| 12 | +| `tab:panel` | `paper/tables/tab_panel.tex` | main | Two-way fixed-effects panel regressions of the forward return on AI language, Russell~3000. | results/r3k_v2/results.json (panel_fe) | | |
| 13 | +| `tab:fmboth` | `paper/tables/tab_fm_both.tex` | main | Fama--MacBeth regressions of the forward return on AI language, Russell~3000 and S\&P~500. | results/r3k_v2/results.json (samples.*.fm, fama_macbeth) | | |
| 14 | +| `tab:fm` | `paper/tables/tab_fm.tex` | appendix | Fama--MacBeth regressions, all specifications, Russell~3000 universe. | results/r3k_v2/results.json (fama_macbeth) | | |
| 15 | +| `tab:prepostboth` | `paper/tables/tab_prepost_both.tex` | main | AI-language spreads before and after the ChatGPT launch, Russell~3000 and S\&P~500. | results/r3k_v2/results.json (samples.*.sub, pooled, samples.sp500.pooled_post_w) | | |
| 16 | +| `tab:cohorts` | `paper/tables/tab_cohorts.tex` | main | Cohort-by-cohort spreads of the AI composite and tests of a 2022 break, Russell~3000. | results/r3k_v2/results2.json (cohorts*, placebo_breaks), results_surv.json (then_members_present) | | |
| 17 | +| `tab:groups` | `paper/tables/tab_groups.tex` | main | AI-composite spreads by size group and index membership, Russell~3000. | results/r3k_v2/results.json (groups, size_neutral, contrasts) | | |
| 18 | +| `tab:contrasts` | `paper/tables/tab_contrasts.tex` | main | Formal tests of the large-versus-small and member-versus-non-member post-launch differences. | results/r3k_v2/results.json (contrasts) | | |
| 19 | +| `tab:doublesort` | `paper/tables/tab_doublesort.tex` | appendix | AI language $\times$ size double sort, Russell~3000. | results/r3k_v2/results2.json (double_sort_*) | | |
| 20 | +| `tab:sectortech` | `paper/tables/tab_sector_tech.tex` | main | Within-sector estimates of the AI-language premium and technology-factor spanning, Russell~3000. | results/r3k_v2/results2.json (sector_controls, sector_neutral, residual_sort, monthly_ls) | | |
| 21 | +| `tab:placebo` | `paper/tables/tab_placebo.tex` | appendix | Placebo across eight semantic factors scored with the same pipeline, Russell~3000. | results/r3k_v2/results2.json (placebo, mht) | | |
| 22 | +| `tab:robust` | `paper/tables/tab_robust.tex` | main | Construction, horizon, benchmark, universe and survivorship checks on the composite spread, Russell~3000. | results/r3k_v2/results2.json (subperiod_spread, sector_neutral, residual_sort, mag7, alt_*, entrants, breaks, cluster_k), results_surv.json (conditional_spreads), results.json | | |
| 23 | +| `tab:filingtype` | `paper/tables/tab_filing_type.tex` | appendix | Form-specific AI composites, Russell~3000. | results/r3k_v2/results2.json (filing_type, mht.forms_fm_w) | | |
| 24 | +| `tab:event` | `paper/tables/tab_event.tex` | main | ChatGPT event study: cumulative abnormal returns by pre-launch AI tercile under the IWV market model, Russell~3000. | results/r3k_v2/results_event.json (by_benchmark.mm_2022, welch_reference, tercile_characteristics) | | |
| 25 | +| `tab:eventcda` | `paper/tables/tab_event_cda.tex` | main | Benchmark dependence of the one-year event-study gradient and placebo launch dates. | results/r3k_v2/results_event.json (by_benchmark, ff6_calendar_time, placebo_launches, placebo_summary, mht_event) | | |
| 26 | +| `tab:didspid` | `paper/tables/tab_did_spid.tex` | main | Difference-in-differences on pre-determined exposure and the SPID project signal, Russell~3000. | results/r3k_v2/results2.json (did, spid), t17_refit.json | | |
| 27 | +| `tab:spidbinary` | `paper/tables/tab_spid_binary.tex` | main | Binary contrasts on the SPID project signal with minimum detectable effects, Russell~3000. | results/r3k_v2/results2.json (spid.binary_contrasts, spid.tercile_nonzero_w), results.json (spid) | | |
| 28 | +| `tab:reversal` | `paper/tables/tab_reversal.tex` | appendix | Reversal tests and the cybersecurity-vocabulary check, Russell~3000. | results/r3k_v2/results2.json (cluster_k) | | |
| 29 | +| `tab:mht` | `paper/tables/tab_mht.tex` | main | Multiple-testing adjustments by family of tests. | results/r3k_v2/results2.json (mht, spec_curve), results_event.json (mht_event) | | |
| 30 | +| `tab:mhtfull` | `paper/tables/tab_mht_full_a.tex` | appendix | Member-level multiple-testing adjustments, part A: portfolio-sort and form families. | results/r3k_v2/results2.json (mht) | | |
| 31 | +| `tab:mhtfullb` | `paper/tables/tab_mht_full_b.tex` | appendix | Member-level multiple-testing adjustments, part B: robustness rows, placebo interactions and break years. | results/r3k_v2/results2.json (mht) | | |
| 32 | +| `tab:eventtime` | `paper/tables/tab_eventtime.tex` | appendix | Event-time profile of the premium on pre-determined AI exposure, Russell~3000. | results/r3k_v2/results2.json (did_eventtime*) | | |
| 33 | +| `tab:attrition` | `paper/tables/tab_attrition.tex` | main | Survivorship diagnostics on the current-membership universe, Russell~3000. | results/r3k_v2/results_surv.json (attrition, then_members_present, ew_vs_index, bounding), iwv_holdings_manifest.json | | |
| 34 | +| `tab:constructvalidity` | `paper/tables/tab_construct_validity.tex` | main | Construct validity of the semantic AI score: keyword baseline, stripped anchors, sector composition and encoder truncation. | results/r3k_v2/results_construct.json, results/n1/tokenizer_audit.json, results/n1/keyword_baseline.json | | |
| 35 | +| `tab:corr` | `paper/tables/tab_corr.tex` | appendix | Correlations of the within-year standardized variables. | results/r3k_v2/t1b_correlations.csv | | |
| 36 | +| `tab:topfirms` | `paper/tables/tab_topfirms.tex` | appendix | Firms with the highest average AI composite. | data/processed_r3k_v2/analysis_panel_v2.parquet | | |
| 37 | +| `tab:coverage` | `paper/tables/tab_coverage.tex` | appendix | Estimation sample by formation cohort: firms, size groups and text coverage. | results/r3k_v2/results2.json (coverage) | | |
| 38 | +| `tab:corpus` | `paper/tables/tab_corpus.tex` | main | Corpus composition by form. | data/raw_russell3000/filings, results/n1/corpus_counts.json | | |
| 39 | +| `tab:hygiene` | `paper/tables/tab_hygiene.tex` | appendix | Return hygiene and the effect of the data corrections on the headline statistics. | results/r3k_v2/hygiene_log.csv, hygiene_summary.json, results/r3k/*.json (HEAD 3cf7560), results/r3k_v2/results*.json | | |
| 40 | +| `tab:break` | `paper/tables/tab_break.tex` | appendix | Pooled test of a post-launch change in the signal slope, by signal, Russell~3000. | results/r3k_v2/results.json (structural_break, subperiod_spread) | | |
| 41 | +| `fig:quintiles` | `paper/figures_r3k/fig1_quintiles.tex` (PDF `figures/r3k_v2/fig1_quintiles.pdf`) | appendix | Mean forward return by AI-composite quintile before and after the ChatGPT launch. | scripts/r3k_19_figures_v2.py | | |
| 42 | +| `fig:cumls` | `paper/figures_r3k/fig2_cumulative_ls.tex` (PDF `figures/r3k_v2/fig2_cumulative_ls.pdf`) | appendix | Cumulative return of the AI-composite long--short portfolio by size group. | scripts/r3k_19_figures_v2.py | | |
| 43 | +| `fig:cohorts` | `paper/figures_r3k/fig3_cohort_spreads.tex` (PDF `figures/r3k_v2/fig3_cohort_spreads.pdf`) | main | Annual AI-composite spreads by formation cohort, Russell~3000. | scripts/r3k_19_figures_v2.py | | |
| 44 | +| `fig:trend` | `paper/figures_r3k/fig4_trend.tex` (PDF `figures/r3k_v2/fig4_trend.pdf`) | main | The secular rise of AI language in SEC filings. | scripts/r3k_19_figures_v2.py | | |
| 45 | +| `fig:fmyear` | `paper/figures_r3k/fig5_fm_byyear.tex` (PDF `figures/r3k_v2/fig5_fm_byyear.pdf`) | main | Year-by-year Fama--MacBeth slopes on the AI composite, Russell~3000. | scripts/r3k_19_figures_v2.py | | |
| 46 | +| `fig:placebo` | `paper/figures_r3k/fig6_placebo.tex` (PDF `figures/r3k_v2/fig6_placebo.pdf`) | appendix | Placebo across eight semantic factors scored with the same pipeline. | scripts/r3k_19_figures_v2.py | | |
| 47 | +| `fig:event` | `paper/figures_r3k/fig7_event_car.tex` (PDF `figures/r3k_v2/fig7_event_car.pdf`) | main | Cumulative abnormal returns around the ChatGPT launch by pre-launch AI tercile. | scripts/r3k_19_figures_v2.py | | |
| 48 | +| `fig:double` | `paper/figures_r3k/fig8_double_sort.tex` (PDF `figures/r3k_v2/fig8_double_sort.pdf`) | appendix | AI language $\times$ size double sort before and after the launch, Russell~3000. | scripts/r3k_19_figures_v2.py | | |
| 49 | +| `fig:eventtime` | `paper/figures_r3k/fig9_eventtime.tex` (PDF `figures/r3k_v2/fig9_eventtime.pdf`) | main | Event-time profile of the premium on pre-determined AI exposure, Russell~3000. | scripts/r3k_19_figures_v2.py | | |
| 50 | +| `fig:filingtype` | `paper/figures_r3k/fig10_filing_type.tex` (PDF `figures/r3k_v2/fig10_filing_type.pdf`) | appendix | Pre- and post-launch spreads of the form-specific AI composites. | scripts/r3k_19_figures_v2.py | | |
| 51 | +| `fig:placebobreaks` | `paper/figures_r3k/fig11_placebo_breaks.tex` (PDF `figures/r3k_v2/fig11_placebo_breaks.pdf`) | main | Placebo break dates and contiguous-block comparison, AI composite, Russell~3000. | scripts/r3k_19_figures_v2.py | | |
| 52 | +| `fig:sizegroups` | `paper/figures_r3k/fig12_size_groups.tex` (PDF `figures/r3k_v2/fig12_size_groups.pdf`) | main | Pre- and post-launch AI-composite spreads by size group and index membership. | scripts/r3k_19_figures_v2.py | | |
| 53 | +| `fig:cohortsboth` | `paper/figures_r3k/fig13_cohorts_both.tex` (PDF `figures/r3k_v2/fig13_cohorts_both.pdf`) | main | Annual AI-composite spreads by cohort, Russell~3000 universe versus S\&P~500 members. | scripts/r3k_19_figures_v2.py | | |
| 54 | +| `fig:cumboth` | `paper/figures_r3k/fig14_cumulative_both.tex` (PDF `figures/r3k_v2/fig14_cumulative_both.pdf`) | main | Cumulative long--short return on the AI composite in the two universes. | scripts/r3k_19_figures_v2.py | | |
| 55 | +| `fig:speccurve` | `paper/figures_r3k/fig15_spec_curve.tex` (PDF `figures/r3k_v2/fig15_spec_curve.pdf`) | main | Specification curve of the AI-composite spread, Russell~3000. | scripts/r3k_19_figures_v2.py | | |
| 56 | +| `fig:placebolaunches` | `paper/figures_r3k/fig16_placebo_launches.tex` (PDF `figures/r3k_v2/fig16_placebo_launches.pdf`) | main | Placebo launch dates: the event-study design applied to 30 November of 2015--2021. | scripts/r3k_19_figures_v2.py | | |
| 57 | +| `fig:attrition` | `paper/figures_r3k/fig17_attrition.tex` (PDF `figures/r3k_v2/fig17_attrition.pdf`) | appendix | Attrition of IWV constituents by market-value tercile, from dated holdings captures. | scripts/r3k_19_figures_v2.py | | |
added
paper/tables/appendix_exhibits.tex
+48 −0
@@ -0,0 +1,48 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Appendix exhibits. Every \input below is wrapped in \ExhibitOnce{label}{...}: an exhibit that was already \input earlier in the | |
| 3 | +% document (main text or appendix prose) is skipped here, so this block is a safety net as well as the appendix placement. | |
| 4 | +% Every main-text table is listed first (guarded), so a table the sections do not \input yet prints here with an appendix number until | |
| 5 | +% the writing agent places it; the appendix-bound tables follow. Figures placed through the writing agent's own | |
| 6 | +% \begin{figure} wrappers do not set the flag, so no figure fallback is listed). | |
| 7 | +\clearpage | |
| 8 | +\renewcommand{\thesection}{\Alph{section}}\renewcommand{\thetable}{A\arabic{table}}\renewcommand{\thefigure}{A\arabic{figure}} | |
| 9 | +\section{Additional Exhibits}\label{app:exhibits} | |
| 10 | +This appendix collects the secondary exhibits referenced in the text: the full-signal versions of the main tables, the member-level multiple-testing adjustments, the return-hygiene and before/after record, and the diagnostic figures. Every exhibit is generated by \texttt{scripts/r3k\_19\_figures\_v2.py} or \texttt{scripts/r3k\_20\_tables\_v2.py} from the result files named in its notes. | |
| 11 | + | |
| 12 | +\ExhibitOnce{tab:summ}{\input{tables/tab_summary}} | |
| 13 | +\ExhibitOnce{tab:sortsboth}{\input{tables/tab_sorts_both}} | |
| 14 | +\ExhibitOnce{tab:alphaboth}{\input{tables/tab_alpha_both}} | |
| 15 | +\ExhibitOnce{tab:panel}{\input{tables/tab_panel}} | |
| 16 | +\ExhibitOnce{tab:fmboth}{\input{tables/tab_fm_both}} | |
| 17 | +\ExhibitOnce{tab:prepostboth}{\input{tables/tab_prepost_both}} | |
| 18 | +\ExhibitOnce{tab:cohorts}{\input{tables/tab_cohorts}} | |
| 19 | +\ExhibitOnce{tab:groups}{\input{tables/tab_groups}} | |
| 20 | +\ExhibitOnce{tab:contrasts}{\input{tables/tab_contrasts}} | |
| 21 | +\ExhibitOnce{tab:sectortech}{\input{tables/tab_sector_tech}} | |
| 22 | +\ExhibitOnce{tab:robust}{\input{tables/tab_robust}} | |
| 23 | +\ExhibitOnce{tab:event}{\input{tables/tab_event}} | |
| 24 | +\ExhibitOnce{tab:eventcda}{\input{tables/tab_event_cda}} | |
| 25 | +\ExhibitOnce{tab:didspid}{\input{tables/tab_did_spid}} | |
| 26 | +\ExhibitOnce{tab:spidbinary}{\input{tables/tab_spid_binary}} | |
| 27 | +\ExhibitOnce{tab:mht}{\input{tables/tab_mht}} | |
| 28 | +\ExhibitOnce{tab:attrition}{\input{tables/tab_attrition}} | |
| 29 | +\ExhibitOnce{tab:constructvalidity}{\input{tables/tab_construct_validity}} | |
| 30 | +\ExhibitOnce{tab:corpus}{\input{tables/tab_corpus}} | |
| 31 | +\ExhibitOnce{tab:corr}{\input{tables/tab_corr}} | |
| 32 | +\ExhibitOnce{tab:topfirms}{\input{tables/tab_topfirms}} | |
| 33 | +\ExhibitOnce{tab:coverage}{\input{tables/tab_coverage}} | |
| 34 | +\ExhibitOnce{tab:hygiene}{\input{tables/tab_hygiene}} | |
| 35 | +\ExhibitOnce{tab:sorts}{\input{tables/tab_sorts}} | |
| 36 | +\ExhibitOnce{tab:alpha}{\input{tables/tab_alpha}} | |
| 37 | +\ExhibitOnce{tab:fm}{\input{tables/tab_fm}} | |
| 38 | +\ExhibitOnce{tab:break}{\input{tables/tab_break}} | |
| 39 | +\ExhibitOnce{tab:doublesort}{\input{tables/tab_doublesort}} | |
| 40 | +\ExhibitOnce{tab:placebo}{\input{tables/tab_placebo}} | |
| 41 | +\ExhibitOnce{tab:filingtype}{\input{tables/tab_filing_type}} | |
| 42 | +\ExhibitOnce{tab:eventtime}{\input{tables/tab_eventtime}} | |
| 43 | +\ExhibitOnce{tab:reversal}{\input{tables/tab_reversal}} | |
| 44 | +\ExhibitOnce{tab:mhtfull}{\input{tables/tab_mht_full_a}} | |
| 45 | +\ExhibitOnce{tab:mhtfullb}{\input{tables/tab_mht_full_b}} | |
| 46 | +\ExhibitOnce{fig:quintiles}{\input{figures_r3k/fig1_quintiles}} | |
| 47 | +\ExhibitOnce{fig:cumls}{\input{figures_r3k/fig2_cumulative_ls}} | |
| 48 | +\ExhibitOnce{fig:attrition}{\input{figures_r3k/fig17_attrition}} | |
modified
paper/tables/macros.tex
+56 −0
@@ -1,3 +1,16 @@ | ||
| 1 | +% ---- Exhibit wrappers (Stage E, Agent 9). Each generated table/figure file is a self-contained float. ---- | |
| 2 | +% \ExhibitTable{label}{caption}{size}{body}: prints a table float when \input on its own; inside an existing \begin{table} | |
| 3 | +% wrapper (\@captype defined) it prints only the body, so the old wrappers in the sections keep compiling until they are replaced | |
| 4 | +% by a bare \input{tables/<name>}. \ExhibitFigure{label}{caption}{content} does the same for figures. | |
| 5 | +% Both wrappers record the label as placed, and \ExhibitOnce{label}{...} prints its argument only if the label has not been placed yet | |
| 6 | +% (used by tables/appendix_exhibits.tex as a safety net: an exhibit placed in the sections, in either convention, is not printed twice). | |
| 7 | +\makeatletter | |
| 8 | +\providecommand{\ExhibitTable}{}\renewcommand{\ExhibitTable}[4]{\global\@namedef{exh@done@#1}{}\ifx\@captype\@undefined\begin{table}[htbp]\centering #3\caption{#2}\label{#1}#4\end{table}\else #4\fi} | |
| 9 | +\providecommand{\ExhibitFigure}{}\renewcommand{\ExhibitFigure}[3]{\global\@namedef{exh@done@#1}{}\ifx\@captype\@undefined\begin{figure}[htbp]\centering #3\caption{#2}\label{#1}\end{figure}\else #3\fi} | |
| 10 | +\providecommand{\ExhibitOnce}{}\renewcommand{\ExhibitOnce}[2]{\@ifundefined{exh@done@#1}{\global\@namedef{exh@done@#1}{}#2}{}} | |
| 11 | +\makeatother | |
| 12 | +% ---- Number macros: every value below is read from a result file by scripts/r3k_20_tables_v2.py; legacy names keep the legacy | |
| 13 | +% statistic (e.g. \meventT is the Welch t), the robust versions carry suffixes W / CDA / Ttw / CI / MDE / Pwild. ---- | |
| 1 | 14 | \newcommand{\mnObs}{27{,}354} |
| 2 | 15 | \newcommand{\mnFirms}{2{,}375} |
| 3 | 16 | \newcommand{\mnFirmYearsPanel}{32{,}818} |
@@ -440,3 +453,46 @@ | ||
| 440 | 453 | \newcommand{\maaplRankTwentyEleven}{2} |
| 441 | 454 | \newcommand{\mlogCapMin}{15.5} |
| 442 | 455 | \newcommand{\mlogCapMax}{29.0} |
| 456 | +\newcommand{\mnExhibitTables}{34} | |
| 457 | +\newcommand{\mnIWVcaptures}{17} | |
| 458 | +\newcommand{\mattritionSmallExitMax}{16.1} | |
| 459 | +\newcommand{\mattritionLargeExitMax}{3.3} | |
| 460 | +\newcommand{\mboundReqExitHalf}{5.7} | |
| 461 | +\newcommand{\mboundReqExitFull}{3.3} | |
| 462 | +\newcommand{\mevLargeWelchRef}{5.29} | |
| 463 | +\newcommand{\mevFullWelchRef}{3.92} | |
| 464 | +\newcommand{\mplaceboTmin}{-1.11} | |
| 465 | +\newcommand{\mplaceboTmax}{1.61} | |
| 466 | +\newcommand{\mevHolmMin}{0.088} | |
| 467 | +\newcommand{\mevBHMin}{0.036} | |
| 468 | +\newcommand{\mmhtRobustPreRWmin}{0.38} | |
| 469 | +\newcommand{\mmhtRobustPostRWmin}{0.50} | |
| 470 | +\newcommand{\mmhtSortsFullRWmin}{0.15} | |
| 471 | +\newcommand{\mcyberIntRWp}{0.05} | |
| 472 | +\newcommand{\mlargePostExTwentyFourWCI}{[-2.2, +8.8]} | |
| 473 | +\newcommand{\mcontrastLargeMinusSmallBootCI}{[-4.0, +22.0]} | |
| 474 | +\newcommand{\mcontrastLargeMinusSmallBoot}{+10.0} | |
| 475 | +\newcommand{\msectorFEslopeFull}{+0.75} | |
| 476 | +\newcommand{\msectorFEslopeFullTtw}{2.78} | |
| 477 | +\newcommand{\msectorFEslopeFullPwild}{0.024} | |
| 478 | +\newcommand{\mspecIQRlo}{+0.7} | |
| 479 | +\newcommand{\mspecIQRhi}{+8.5} | |
| 480 | +\newcommand{\mspecNegSig}{1.5} | |
| 481 | +\newcommand{\mformWaldStat}{8.25} | |
| 482 | +\newcommand{\mcohortTwentyTwoPostFilingShare}{4.6} | |
| 483 | +\newcommand{\mthenMembersCohorts}{2018: all +19.0 vs members +20.4, 2019: all +1.4 vs members +4.2, 2020: all -6.6 vs members -6.7, 2021: all +12.1 vs members +14.4, 2022: all -5.0 vs members -2.9, 2023: all +3.1 vs members +4.2, 2024: all +19.9 vs members +16.4} | |
| 484 | +\newcommand{\mkwSpearman}{0.13} | |
| 485 | +\newcommand{\mkwFullW}{+4.7} | |
| 486 | +\newcommand{\mkwFullWT}{2.30} | |
| 487 | +\newcommand{\mkwFMW}{+0.61} | |
| 488 | +\newcommand{\mkwFMWT}{1.25} | |
| 489 | +\newcommand{\maltSpearman}{0.98} | |
| 490 | +\newcommand{\maltFullW}{+4.1} | |
| 491 | +\newcommand{\maltFullWT}{2.33} | |
| 492 | +\newcommand{\maltFMW}{+1.79} | |
| 493 | +\newcommand{\maltFMWT}{2.70} | |
| 494 | +\newcommand{\mexUtilFullW}{+4.6} | |
| 495 | +\newcommand{\mexUtilFullWT}{2.16} | |
| 496 | +\newcommand{\mutilShareQfive}{6.5} | |
| 497 | +\newcommand{\mutilShareAll}{3.2} | |
| 498 | +\newcommand{\mkwShareZero}{84} | |
modified
paper/tables/tab_alpha.tex
+38 −23
@@ -1,35 +1,50 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results.json (ls_alpha) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: appendix. \input{tables/tab_alpha} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:alpha}{\textbf{Monthly long--short portfolios by signal and sub-period, Russell~3000 universe.} Companion to Table~\ref{tab:alphaboth}: equal-weighted Q5$-$Q1 monthly portfolios (annually selected in June, held July--June) for every signal, over the full sample, the pre-launch months (to June 2023) and the post-launch months (from July 2023). Mean and alphas in percent per year.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 1 | 5 | \begin{tabular}{llrrrrrrrr} |
| 2 | 6 | \toprule |
| 3 | −Signal & Period & Months & Mean (\%/yr) & HAC $t$ & plain $t$ & Sharpe & CAPM $\alpha$ (\%/yr) & FF6 $\alpha$ (\%/yr) & FF6 $t$ \\ | |
| 7 | +Signal & Period & Months & Mean (\%/yr) & [HAC $t$] & [plain $t$] & Sharpe & CAPM $\alpha$ (\%/yr) & FF6 $\alpha$ (\%/yr) & [$t$] \\ | |
| 4 | 8 | \midrule |
| 5 | −AI composite & full & 180 & +3.43 & 1.67 & 1.76 & 0.45 & +1.73 & +2.20 & 1.44 \\ | |
| 6 | − & pre & 144 & +3.76 & 1.62 & 1.82 & 0.52 & +2.54 & +2.54 & 1.47 \\ | |
| 7 | − & post & 36 & +2.10 & 0.45 & 0.40 & 0.23 & -2.54 & -0.27 & -0.08 \\ | |
| 9 | +AI composite & full & 180 & +3.43 & [1.67] & [1.76] & 0.45 & +1.73 & +2.20 & [1.44] \\ | |
| 10 | + & pre & 144 & +3.76 & [1.62] & [1.82] & 0.52 & +2.54 & +2.54 & [1.47] \\ | |
| 11 | + & post & 36 & +2.10 & [0.45] & [0.40] & 0.23 & -2.54 & -0.27 & [-0.08] \\ | |
| 8 | 12 | \addlinespace |
| 9 | −AI adoption & full & 180 & +3.72 & 1.80 & 1.88 & 0.49 & +1.71 & +2.50 & 1.74 \\ | |
| 10 | − & pre & 144 & +4.50 & 1.95 & 2.14 & 0.62 & +2.92 & +3.15 & 1.96 \\ | |
| 11 | − & post & 36 & +0.59 & 0.12 & 0.11 & 0.07 & -3.98 & -0.17 & -0.05 \\ | |
| 13 | +AI adoption & full & 180 & +3.72 & [1.80] & [1.88] & 0.49 & +1.71 & +2.50 & [1.74] \\ | |
| 14 | + & pre & 144 & +4.50 & [1.95] & [2.14] & 0.62 & +2.92 & +3.15 & [1.96] \\ | |
| 15 | + & post & 36 & +0.59 & [0.12] & [0.11] & 0.07 & -3.98 & -0.17 & [-0.05] \\ | |
| 12 | 16 | \addlinespace |
| 13 | −AI infrastructure & full & 180 & +2.74 & 1.49 & 1.53 & 0.39 & +1.57 & +1.90 & 1.28 \\ | |
| 14 | − & pre & 144 & +2.60 & 1.25 & 1.40 & 0.41 & +1.88 & +1.87 & 1.17 \\ | |
| 15 | − & post & 36 & +3.30 & 0.79 & 0.64 & 0.37 & -0.62 & +0.27 & 0.07 \\ | |
| 17 | +AI infrastructure & full & 180 & +2.74 & [1.49] & [1.53] & 0.39 & +1.57 & +1.90 & [1.28] \\ | |
| 18 | + & pre & 144 & +2.60 & [1.25] & [1.40] & 0.41 & +1.88 & +1.87 & [1.17] \\ | |
| 19 | + & post & 36 & +3.30 & [0.79] & [0.64] & 0.37 & -0.62 & +0.27 & [0.07] \\ | |
| 16 | 20 | \addlinespace |
| 17 | −AI disruption risk & full & 180 & +7.76 & 2.53 & 2.84 & 0.73 & +5.36 & +7.50 & 4.91 \\ | |
| 18 | − & pre & 144 & +6.35 & 1.95 & 2.20 & 0.64 & +4.97 & +6.42 & 4.40 \\ | |
| 19 | − & post & 36 & +13.37 & 1.65 & 1.82 & 1.05 & +4.89 & +12.50 & 2.79 \\ | |
| 21 | +AI disruption risk & full & 180 & +7.76 & [2.53] & [2.84] & 0.73 & +5.36 & +7.50 & [4.91] \\ | |
| 22 | + & pre & 144 & +6.35 & [1.95] & [2.20] & 0.64 & +4.97 & +6.42 & [4.40] \\ | |
| 23 | + & post & 36 & +13.37 & [1.65] & [1.82] & 1.05 & +4.89 & +12.50 & [2.79] \\ | |
| 20 | 24 | \addlinespace |
| 21 | −PC1 & full & 180 & +4.77 & 2.00 & 2.13 & 0.55 & +2.78 & +3.96 & 2.76 \\ | |
| 22 | − & pre & 144 & +4.77 & 1.81 & 2.05 & 0.59 & +3.56 & +3.99 & 2.52 \\ | |
| 23 | − & post & 36 & +4.77 & 0.81 & 0.76 & 0.44 & -2.01 & +2.05 & 0.58 \\ | |
| 25 | +PC1 & full & 180 & +4.77 & [2.00] & [2.13] & 0.55 & +2.78 & +3.96 & [2.76] \\ | |
| 26 | + & pre & 144 & +4.77 & [1.81] & [2.05] & 0.59 & +3.56 & +3.99 & [2.52] \\ | |
| 27 | + & post & 36 & +4.77 & [0.81] & [0.76] & 0.44 & -2.01 & +2.05 & [0.58] \\ | |
| 24 | 28 | \addlinespace |
| 25 | −Equal-form composite & full & 180 & +5.20 & 1.86 & 1.93 & 0.50 & +1.76 & +2.44 & 1.46 \\ | |
| 26 | − & pre & 144 & +4.60 & 1.54 & 1.70 & 0.49 & +2.03 & +1.96 & 1.15 \\ | |
| 27 | − & post & 36 & +7.61 & 1.02 & 0.94 & 0.54 & -0.98 & +5.12 & 0.84 \\ | |
| 29 | +Equal-form composite & full & 180 & +5.20 & [1.86] & [1.93] & 0.50 & +1.76 & +2.44 & [1.46] \\ | |
| 30 | + & pre & 144 & +4.60 & [1.54] & [1.70] & 0.49 & +2.03 & +1.96 & [1.15] \\ | |
| 31 | + & post & 36 & +7.61 & [1.02] & [0.94] & 0.54 & -0.98 & +5.12 & [0.84] \\ | |
| 28 | 32 | \addlinespace |
| 29 | −SPID tech projects & full & 180 & +2.20 & 1.14 & 1.22 & 0.31 & -0.33 & +1.15 & 0.78 \\ | |
| 30 | − & pre & 144 & +1.84 & 0.94 & 1.00 & 0.29 & -0.03 & +0.60 & 0.46 \\ | |
| 31 | − & post & 36 & +3.65 & 0.64 & 0.68 & 0.39 & -2.84 & +4.85 & 0.91 \\ | |
| 33 | +10-K composite & full & 180 & +5.44 & [1.80] & [1.88] & 0.49 & +1.86 & +2.50 & [1.46] \\ | |
| 34 | + & pre & 144 & +4.90 & [1.53] & [1.71] & 0.49 & +2.31 & +2.10 & [1.27] \\ | |
| 35 | + & post & 36 & +7.60 & [0.93] & [0.85] & 0.49 & -1.89 & +6.42 & [0.94] \\ | |
| 36 | +\addlinespace | |
| 37 | +SPID tech projects & full & 180 & +2.20 & [1.14] & [1.22] & 0.31 & -0.33 & +1.15 & [0.78] \\ | |
| 38 | + & pre & 144 & +1.84 & [0.94] & [1.00] & 0.29 & -0.03 & +0.60 & [0.46] \\ | |
| 39 | + & post & 36 & +3.65 & [0.64] & [0.68] & 0.39 & -2.84 & +4.85 & [0.91] \\ | |
| 40 | +\addlinespace | |
| 41 | +Cybersecurity (placebo) & full & 180 & +1.33 & [0.84] & [0.82] & 0.21 & +0.10 & +0.73 & [0.60] \\ | |
| 42 | + & pre & 144 & -0.14 & [-0.09] & [-0.09] & -0.03 & -0.74 & +0.91 & [0.73] \\ | |
| 43 | + & post & 36 & +7.20 & [1.46] & [1.36] & 0.79 & +2.27 & +2.54 & [0.46] \\ | |
| 32 | 44 | \addlinespace |
| 33 | 45 | \bottomrule |
| 34 | 46 | \end{tabular} |
| 35 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Monthly equal-weighted Q5$-$Q1 portfolios formed every June from the cohort quintiles and held July--June; splice months set to missing. HAC = Newey--West (3 lags). Post = months from July 2023. Deflated Sharpe ratio of the composite (Bailey and L\'opez de Prado 2014, trials = 93 long--short series tried): full 0.35, post 0.29 (probability that the true Sharpe exceeds the multiple-testing hurdle).} | |
| \ No newline at end of file | ||
| 47 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 48 | +\item \emph{Notes:} HAC = Newey--West (3 lags); plain = i.i.d. The post period has 36 months, so its alphas are imprecise; the AI disruption-risk facet is the only signal whose post FF6 alpha has $|t| > 2$ (Table~\ref{tab:robust}). Factor data: Ken French daily FF5 + momentum, compounded to months. Splice months set to missing. | |
| 49 | +\end{tablenotes} | |
| 50 | +\end{threeparttable}} | |
modified
paper/tables/tab_alpha_both.tex
+25 −15
@@ -1,19 +1,29 @@ | ||
| 1 | −\begin{tabular}{llrrrrrrr} | |
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results.json (samples.*.ls), results2.json (mht.deflated_sharpe) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_alpha_both} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:alphaboth}{\textbf{Monthly long--short portfolios on AI language: mean return, Sharpe ratio and factor alphas.} Equal-weighted Q5$-$Q1 portfolios selected every June from the cohort quintiles and held July--June (monthly returns, annualized). Mean and alphas in percent per year; $t$-statistics in brackets use Newey--West (HAC, 3 lags) standard errors. FF6 = Fama--French five factors plus momentum (Ken French daily factors aggregated to months). Post = months from July 2023, the first holding year that follows the ChatGPT launch cohort.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 5 | +\begin{tabular}{llrrrrrrrrr} | |
| 2 | 6 | \toprule |
| 3 | −Signal & Sample & Months & Mean (\%/yr) & Sharpe & FF6 $\alpha$ full (\%/yr) & ($t$) & FF6 $\alpha$ post (\%/yr) & ($t$) \\ | |
| 7 | +Signal & Universe & Months & Mean (\%/yr) & [HAC $t$] & Sharpe & FF6 $\alpha$ (\%/yr) & [$t$] & Post months & Post FF6 $\alpha$ & [$t$] \\ | |
| 4 | 8 | \midrule |
| 5 | −AI adoption & Russell 3000 & 180 & +3.72 & 0.49 & +2.50 & (1.74) & -0.17 & (-0.05) \\ | |
| 6 | − & S\&P 500 & 180 & +4.91 & 0.65 & +3.57 & (2.08) & +2.84 & (0.58) \\ | |
| 7 | −AI infrastructure & Russell 3000 & 180 & +2.74 & 0.39 & +1.90 & (1.28) & +0.27 & (0.07) \\ | |
| 8 | − & S\&P 500 & 180 & +3.97 & 0.54 & +2.97 & (1.88) & +3.04 & (0.87) \\ | |
| 9 | −AI disruption risk & Russell 3000 & 180 & +7.76 & 0.73 & +7.50 & (4.91) & +12.50 & (2.79) \\ | |
| 10 | − & S\&P 500 & 180 & +6.04 & 0.68 & +5.15 & (3.07) & +3.54 & (0.72) \\ | |
| 11 | −AI composite & Russell 3000 & 180 & +3.43 & 0.45 & +2.20 & (1.44) & -0.27 & (-0.08) \\ | |
| 12 | − & S\&P 500 & 180 & +4.33 & 0.57 & +3.21 & (1.95) & +1.15 & (0.26) \\ | |
| 13 | −PC1 of three facets & Russell 3000 & 180 & +4.77 & 0.55 & +3.96 & (2.76) & +2.05 & (0.58) \\ | |
| 14 | − & S\&P 500 & 180 & +4.57 & 0.55 & +3.29 & (2.02) & +0.72 & (0.16) \\ | |
| 15 | −SPID tech projects (legacy quintiles) & Russell 3000 & 180 & +2.20 & 0.31 & +1.15 & (0.78) & +4.85 & (0.91) \\ | |
| 16 | − & S\&P 500 & 180 & +1.01 & 0.17 & +0.44 & (0.34) & -0.71 & (-0.19) \\ | |
| 9 | +AI adoption & Russell 3000 & 180 & +3.72 & [1.80] & 0.49 & +2.50 & [1.74] & 36 & -0.17 & [-0.05] \\ | |
| 10 | + & S\&P 500 & 180 & +4.91 & [2.40] & 0.65 & +3.57 & [2.08] & 36 & +2.84 & [0.58] \\ | |
| 11 | +\addlinespace | |
| 12 | +AI infrastructure & Russell 3000 & 180 & +2.74 & [1.49] & 0.39 & +1.90 & [1.28] & 36 & +0.27 & [0.07] \\ | |
| 13 | + & S\&P 500 & 180 & +3.97 & [2.09] & 0.54 & +2.97 & [1.88] & 36 & +3.04 & [0.87] \\ | |
| 14 | +\addlinespace | |
| 15 | +AI disruption risk & Russell 3000 & 180 & +7.76 & [2.53] & 0.73 & +7.50 & [4.91] & 36 & +12.50 & [2.79] \\ | |
| 16 | + & S\&P 500 & 180 & +6.04 & [2.69] & 0.68 & +5.15 & [3.07] & 36 & +3.54 & [0.72] \\ | |
| 17 | +\addlinespace | |
| 18 | +AI composite & Russell 3000 & 180 & +3.43 & [1.67] & 0.45 & +2.20 & [1.44] & 36 & -0.27 & [-0.08] \\ | |
| 19 | + & S\&P 500 & 180 & +4.33 & [2.20] & 0.57 & +3.21 & [1.95] & 36 & +1.15 & [0.26] \\ | |
| 20 | +\addlinespace | |
| 21 | +PC1 of three facets & Russell 3000 & 180 & +4.77 & [2.00] & 0.55 & +3.96 & [2.76] & 36 & +2.05 & [0.58] \\ | |
| 22 | + & S\&P 500 & 180 & +4.57 & [2.13] & 0.55 & +3.29 & [2.02] & 36 & +0.72 & [0.16] \\ | |
| 23 | +\addlinespace | |
| 17 | 24 | \bottomrule |
| 18 | 25 | \end{tabular} |
| 19 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Annually selected (June), monthly equal-weighted long--short portfolios; HAC $t$ (3 lags).} | |
| \ No newline at end of file | ||
| 26 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 27 | +\item \emph{Notes:} Deflated Sharpe ratio of the Russell~3000 composite long--short (Bailey and L\'opez de Prado 2014; trials = 93 long--short series tried in this study, benchmark SR 0.55): full sample 0.35, pre 0.46, post 0.29 (probability that the true Sharpe ratio exceeds the multiple-testing hurdle; values below 0.95 do not clear it). Economic magnitude: the composite earns +3.43\% per year over 180 months in the Russell~3000, a Sharpe ratio of 0.45. Splice months (Table~\ref{tab:hygiene}) are set to missing. All signals and sub-periods: Table~\ref{tab:alpha}. | |
| 28 | +\end{tablenotes} | |
| 29 | +\end{threeparttable}} | |
modified
paper/tables/tab_attrition.tex
+44 −29
@@ -1,36 +1,51 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results_surv.json (attrition, then_members_present, ew_vs_index, bounding), iwv_holdings_manifest.json -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_attrition} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:attrition}{\textbf{Survivorship diagnostics on the current-membership universe, Russell~3000.} Panel A: for each of the 17 dated IWV holdings files captured by the Internet Archive, the share of then-constituents no longer held on 28 August 2025 (absent = ISIN/CUSIP not in the 2025 holdings and ticker not in the 2026 universe), by market-value tercile within the snapshot, the implied annualised exit rates of the small and large terciles, and the share of then-constituents present in the estimation panel. Panel B: the composite spread of each cohort computed on all panel firms and on the firms that were IWV members at formation. Panel C: the equal-weighted panel return against the index total return, and the exit-rate difference between the bottom and top AI quintiles that survivorship would need to generate the entire observed spread.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable}\setlength{\tabcolsep}{4pt} | |
| 1 | 5 | \begin{tabular}{lrrrrrr} |
| 2 | 6 | \toprule |
| 3 | −IWV holdings as of & Small tercile & Mid & Large & All & Annualised exit, small / large & Then-members in panel \\ | |
| 7 | +\multicolumn{7}{@{}l}{\textit{Panel A: share of then-IWV constituents absent from the 28 August 2025 holdings, by market-value tercile within the snapshot}} \\ | |
| 8 | +IWV holdings as of & Small & Mid & Large & All & Annualised exit, small / large & Then-members in the panel \\ | |
| 4 | 9 | \midrule |
| 5 | −2006-09-29 & 77.7\% & 65.3\% & 47.2\% & 63.4\% & 7.6\% / 3.3\% & 32.3\% \\ | |
| 6 | −2019-05-31 & 61.2\% & 31.1\% & 16.3\% & 36.2\% & 14.1\% / 2.8\% & 55.5\% \\ | |
| 7 | −2019-11-29 & 57.8\% & 27.0\% & 14.8\% & 33.2\% & 14.0\% / 2.8\% & 57.8\% \\ | |
| 8 | −2019-12-20 & 58.2\% & 26.9\% & 14.2\% & 33.1\% & 14.2\% / 2.7\% & 58.0\% \\ | |
| 9 | −2020-12-31 & 51.6\% & 25.4\% & 10.9\% & 29.3\% & 14.4\% / 2.5\% & 61.4\% \\ | |
| 10 | −2021-12-31 & 39.8\% & 19.6\% & 8.2\% & 22.5\% & 13.0\% / 2.3\% & 67.0\% \\ | |
| 11 | −2022-12-30 & 26.7\% & 13.5\% & 5.2\% & 15.1\% & 11.0\% / 2.0\% & 73.4\% \\ | |
| 12 | −2023-06-30 & 22.8\% & 10.9\% & 4.6\% & 12.8\% & 11.3\% / 2.1\% & 75.2\% \\ | |
| 13 | −2023-07-18 & 22.9\% & 10.7\% & 4.2\% & 12.6\% & 11.6\% / 2.0\% & 75.5\% \\ | |
| 14 | −2023-09-11 & 21.9\% & 9.9\% & 4.2\% & 12.0\% & 11.8\% / 2.2\% & 75.9\% \\ | |
| 15 | −2023-09-21 & 21.4\% & 10.3\% & 4.0\% & 11.9\% & 11.7\% / 2.1\% & 76.0\% \\ | |
| 16 | −2023-12-29 & 21.4\% & 8.8\% & 3.2\% & 11.1\% & 13.5\% / 1.9\% & 76.7\% \\ | |
| 17 | −2024-02-02 & 21.4\% & 7.9\% & 3.3\% & 10.9\% & 14.3\% / 2.1\% & 76.8\% \\ | |
| 18 | −2024-04-04 & 21.2\% & 6.8\% & 2.7\% & 10.2\% & 15.6\% / 1.9\% & 77.5\% \\ | |
| 19 | −2024-04-30 & 20.8\% & 6.7\% & 2.8\% & 10.1\% & 16.1\% / 2.1\% & 77.7\% \\ | |
| 20 | −2024-12-06 & 12.0\% & 3.5\% & 1.9\% & 5.8\% & 16.1\% / 2.6\% & 81.2\% \\ | |
| 21 | −2025-08-28 & 0.1\% & 0.0\% & 0.0\% & 0.0\% & -- / -- & 85.8\% \\ | |
| 10 | +2006-09-29 & 77.7\% & 65.3\% & 47.2\% & 63.4\% & 7.6 / 3.3\% & 32.3\% \\ | |
| 11 | +2019-05-31 & 61.2\% & 31.1\% & 16.3\% & 36.2\% & 14.1 / 2.8\% & 55.5\% \\ | |
| 12 | +2019-11-29 & 57.8\% & 27.0\% & 14.8\% & 33.2\% & 14.0 / 2.8\% & 57.8\% \\ | |
| 13 | +2019-12-20 & 58.2\% & 26.9\% & 14.2\% & 33.1\% & 14.2 / 2.7\% & 58.0\% \\ | |
| 14 | +2020-12-31 & 51.6\% & 25.4\% & 10.9\% & 29.3\% & 14.4 / 2.5\% & 61.4\% \\ | |
| 15 | +2021-12-31 & 39.8\% & 19.6\% & 8.2\% & 22.5\% & 13.0 / 2.3\% & 67.0\% \\ | |
| 16 | +2022-12-30 & 26.7\% & 13.5\% & 5.2\% & 15.1\% & 11.0 / 2.0\% & 73.4\% \\ | |
| 17 | +2023-06-30 & 22.8\% & 10.9\% & 4.6\% & 12.8\% & 11.3 / 2.1\% & 75.2\% \\ | |
| 18 | +2023-07-18 & 22.9\% & 10.7\% & 4.2\% & 12.6\% & 11.6 / 2.0\% & 75.5\% \\ | |
| 19 | +2023-09-11 & 21.9\% & 9.9\% & 4.2\% & 12.0\% & 11.8 / 2.2\% & 75.9\% \\ | |
| 20 | +2023-09-21 & 21.4\% & 10.3\% & 4.0\% & 11.9\% & 11.7 / 2.1\% & 76.0\% \\ | |
| 21 | +2023-12-29 & 21.4\% & 8.8\% & 3.2\% & 11.1\% & 13.5 / 1.9\% & 76.7\% \\ | |
| 22 | +2024-02-02 & 21.4\% & 7.9\% & 3.3\% & 10.9\% & 14.3 / 2.1\% & 76.8\% \\ | |
| 23 | +2024-04-04 & 21.2\% & 6.8\% & 2.7\% & 10.2\% & 15.6 / 1.9\% & 77.5\% \\ | |
| 24 | +2024-04-30 & 20.8\% & 6.7\% & 2.8\% & 10.1\% & 16.1 / 2.1\% & 77.7\% \\ | |
| 25 | +2024-12-06 & 12.0\% & 3.5\% & 1.9\% & 5.8\% & 16.1 / 2.6\% & 81.2\% \\ | |
| 26 | +2025-08-28 & 0.1\% & 0.0\% & 0.0\% & 0.0\% & -- / --\% & 85.8\% \\ | |
| 22 | 27 | \midrule |
| 23 | −\multicolumn{7}{l}{\textit{Then-members-present cohort spreads (wins.), all panel firms vs firms that were IWV members at formation}} \\ | |
| 24 | −Cohort 2018 (snapshot 2019-05) & \multicolumn{6}{l}{86\% members then; spread all +19.0 vs members +20.4 pp; mean return members -2.9 vs non-members +8.5} \\ | |
| 25 | −Cohort 2019 (snapshot 2020-12) & \multicolumn{6}{l}{88\% members then; spread all +1.4 vs members +4.2 pp; mean return members +70.7 vs non-members +169.6} \\ | |
| 26 | −Cohort 2020 (snapshot 2020-12) & \multicolumn{6}{l}{84\% members then; spread all -6.6 vs members -6.7 pp; mean return members -14.6 vs non-members -26.8} \\ | |
| 27 | −Cohort 2021 (snapshot 2021-12) & \multicolumn{6}{l}{82\% members then; spread all +12.1 vs members +14.4 pp; mean return members +15.9 vs non-members +28.5} \\ | |
| 28 | −Cohort 2022 (snapshot 2023-06) & \multicolumn{6}{l}{88\% members then; spread all -5.0 vs members -2.9 pp; mean return members +13.1 vs non-members +35.2} \\ | |
| 29 | −Cohort 2023 (snapshot 2024-04) & \multicolumn{6}{l}{90\% members then; spread all +3.1 vs members +4.2 pp; mean return members +12.3 vs non-members +107.6} \\ | |
| 30 | −Cohort 2024 (snapshot 2025-08) & \multicolumn{6}{l}{94\% members then; spread all +19.9 vs members +16.4 pp; mean return members +39.1 vs non-members +188.4} \\ | |
| 28 | +\multicolumn{7}{@{}l}{\textit{Panel B: composite Q5$-$Q1 spread (wins., pp) for all panel firms vs the firms that were IWV members at formation}} \\ | |
| 29 | +Cohort & Snapshot & Members then & Spread, all & Spread, members & Mean return, members & Mean return, non-members \\ | |
| 31 | 30 | \midrule |
| 32 | −\multicolumn{7}{l}{EW panel return minus IWV total return, mean over cohorts: all +5.7 pp (wins. +3.6; median -3.1); largest 1,000 -0.1; remaining vs IWM +14.5; S\&P members +4.3; pre +5.0 / post +8.4.} \\ | |
| 33 | −\multicolumn{7}{l}{Bounding: Q1 $-$ Q5 exit-rate difference needed to generate the whole observed spread (+3.9 pp, wins.) for exit returns of $-$30 / $-$50 / $-$70 / $-$100\%: 8.1 pp / 5.7 pp / 4.4 pp / 3.3 pp; measured annualised exit rate, small tercile 16.1\% at most; forward-volatility gradient Q5 $-$ Q1 +4.6 pp.} \\ | |
| 31 | +2018 & 2019-05 & 86\% & +19.0 & +20.4 & -2.9\% & +8.5\% \\ | |
| 32 | +2019 & 2020-12 & 88\% & +1.4 & +4.2 & +70.7\% & +169.6\% \\ | |
| 33 | +2020 & 2020-12 & 84\% & -6.6 & -6.7 & -14.6\% & -26.8\% \\ | |
| 34 | +2021 & 2021-12 & 82\% & +12.1 & +14.4 & +15.9\% & +28.5\% \\ | |
| 35 | +2022 & 2023-06 & 88\% & -5.0 & -2.9 & +13.1\% & +35.2\% \\ | |
| 36 | +2023 & 2024-04 & 90\% & +3.1 & +4.2 & +12.3\% & +107.6\% \\ | |
| 37 | +2024 & 2025-08 & 94\% & +19.9 & +16.4 & +39.1\% & +188.4\% \\ | |
| 38 | +\midrule | |
| 39 | +\multicolumn{7}{@{}l}{\textit{Panel C: survivor signature at the level and bounding of the spread (pp per year)}} \\ | |
| 40 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.52\textwidth}}{EW panel return minus IWV total return: all firms, raw / wins.\ / median} & \multicolumn{6}{l}{+5.7 / +3.6 / -3.1} \\ | |
| 41 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.52\textwidth}}{\quad largest 1,000 minus IWV / remaining minus IWM / S\&P members minus IWV} & \multicolumn{6}{l}{-0.1 / +14.5 / +4.3} \\ | |
| 42 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.52\textwidth}}{\quad pre-launch cohorts / post-launch cohorts} & \multicolumn{6}{l}{+5.0 / +8.4} \\ | |
| 43 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.52\textwidth}}{Q1 $-$ Q5 exit-rate difference needed to generate the whole spread (+3.9 pp, wins.), exit return $-$30 / $-$50 / $-$70 / $-$100\%} & \multicolumn{6}{l}{8.1 / 5.7 / 4.4 / 3.3 pp/yr} \\ | |
| 44 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.52\textwidth}}{Measured small-minus-large annualised exit rate, largest gap across snapshots} & \multicolumn{6}{l}{14.0 pp/yr} \\ | |
| 45 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.52\textwidth}}{Forward-volatility gradient of the composite sort, Q5 $-$ Q1 (annualized SD)} & \multicolumn{6}{l}{+4.6 pp} \\ | |
| 34 | 46 | \bottomrule |
| 35 | 47 | \end{tabular} |
| 36 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Point-in-time IWV constituents from the dated iShares holdings files captured by the Internet Archive (17 captures; no capture exists for June 2010--2018). Absent = ISIN/CUSIP not in the 28 Aug 2025 holdings and ticker not in the 2026 universe. Size terciles by market value within the snapshot. Delisted firms have no AI scores, so attrition by AI quintile cannot be computed on this corpus (author-side package X1).} | |
| \ No newline at end of file | ||
| 48 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 49 | +\item \emph{Notes:} No capture exists between 2007 and May 2019, so the then-member comparison covers cohorts 2018--2024 only; for 2010--2017 the entry-side checks of Table~\ref{tab:robust} are the only executable diagnostics. Delisted firms have no AI scores in any artefact, so attrition by AI quintile (the test the editor asked for) cannot be computed on this corpus; the small-tercile exit rate is two to four times the large-tercile rate, and the forward-volatility gradient shows that Q5 firms are riskier, so the small-firm pre-launch premium is reported as conditional on survival to October 2026. Bounding logic: if exiting firms earn $r_x$ and are dropped, the survivor spread overstates the true spread by $(e_1 - e_5)(\bar r_{\text{surv}} - r_x)$. Source: results/r3k\_v2/results\_surv.json; captures listed in iwv\_holdings\_manifest.json. | |
| 50 | +\end{tablenotes} | |
| 51 | +\end{threeparttable}} | |
modified
paper/tables/tab_break.tex
+16 −9
@@ -1,14 +1,21 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results.json (structural_break, subperiod_spread) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: appendix. \input{tables/tab_break} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:break}{\textbf{Pooled test of a post-launch change in the signal slope, by signal, Russell~3000.} Pooled regression of the within-year winsorized forward return on the signal, a post-launch cohort dummy, their interaction, size and momentum (pp per SD). Pre slope with its two-way clustered $t$; the interaction (post-launch change of the slope) with firm-clustered and two-way $t$, the firm-clustered 95\% CI and the Webb wild-cluster bootstrap $p$ by cohort; the last two columns repeat the cohort-level pre and post winsorized spreads of Table~\ref{tab:prepostboth}.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable}\setlength{\tabcolsep}{4pt} | |
| 1 | 5 | \begin{tabular}{lrrrrrrrrr} |
| 2 | 6 | \toprule |
| 3 | −Signal & Pre slope & Post extra & $t$ (firm) & $t$ (firm$\times$year) & wild $p$ (Webb) & Pre Q5$-$Q1 (wins.) & ($n$, NW $t$) & Post Q5$-$Q1 (wins.) & ($n$, NW $t$) \\ | |
| 7 | +Signal & Pre slope & [$t$ two-way] & Slope $\times$ post & [$t$ firm] & [$t$ two-way] & 95\% CI & wild $p$ & Pre Q5$-$Q1 & Post Q5$-$Q1 \\ | |
| 4 | 8 | \midrule |
| 5 | −AI adoption & +2.03 & +0.93 & 0.93 & 0.47 & 0.759 & +4.1 & (12, 2.13) & +2.8 & (3, 0.81) \\ | |
| 6 | −AI infrastructure & +1.59 & +1.98 & 1.93 & 1.16 & 0.409 & +2.4 & (12, 1.11) & +6.5 & (3, 1.68) \\ | |
| 7 | −AI composite & +2.02 & +1.63 & 1.51 & 0.80 & 0.608 & +3.4 & (12, 1.65) & +6.0 & (3, 1.18) \\ | |
| 8 | −AI disruption risk & +1.66 & +4.18 & 3.96 & 0.96 & 0.604 & +5.0 & (12, 2.01) & +16.2 & (3, 1.74) \\ | |
| 9 | −PC1 & +1.92 & +2.58 & 2.51 & 0.90 & 0.613 & +3.8 & (12, 1.61) & +8.6 & (3, 1.24) \\ | |
| 10 | −Equal-form & +2.84 & +2.67 & 2.62 & 0.96 & 0.487 & +4.3 & (12, 1.40) & +12.2 & (3, 1.85) \\ | |
| 11 | −SPID tech projects & +0.49 & -1.79 & -1.80 & -0.95 & 0.545 & +1.8 & (12, 1.13) & -0.5 & (3, -0.23) \\ | |
| 9 | +AI composite & +2.02 & [2.26] & +1.63 & [1.51] & [0.80] & [-0.49, +3.76] & 0.608 & +3.4 & +6.0 \\ | |
| 10 | +AI adoption & +2.03 & [2.26] & +0.93 & [0.93] & [0.47] & [-1.02, +2.88] & 0.759 & +4.1 & +2.8 \\ | |
| 11 | +AI infrastructure & +1.59 & [2.12] & +1.98 & [1.93] & [1.16] & [-0.03, +3.99] & 0.409 & +2.4 & +6.5 \\ | |
| 12 | +AI disruption risk & +1.66 & [1.27] & +4.18 & [3.96] & [0.96] & [+2.11, +6.26] & 0.604 & +5.0 & +16.2 \\ | |
| 13 | +PC1 & +1.92 & [1.88] & +2.58 & [2.51] & [0.90] & [+0.57, +4.59] & 0.613 & +3.8 & +8.6 \\ | |
| 14 | +Equal-form composite & +2.84 & [2.27] & +2.67 & [2.62] & [0.96] & [+0.67, +4.67] & 0.487 & +4.3 & +12.2 \\ | |
| 15 | +SPID tech projects & +0.49 & [0.74] & -1.79 & [-1.80] & [-0.95] & [-3.74, +0.16] & 0.545 & +1.8 & -0.5 \\ | |
| 12 | 16 | \bottomrule |
| 13 | 17 | \end{tabular} |
| 14 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Pooled regression of the within-year winsorized forward return on the signal, a post-2022 cohort dummy, their interaction, size and momentum; wild-cluster bootstrap by cohort (Webb six-point weights, 9,999 draws, 15 clusters). NW $t$ on three post cohorts has no useful sampling distribution and is printed with $n$ only.} | |
| \ No newline at end of file | ||
| 18 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 19 | +\item \emph{Notes:} $N$ = 26{,}933 firm-years, 15 cohort clusters, 9,999 bootstrap draws. None of the interactions is significant once clustering by cohort or the wild bootstrap is used; the firm-clustered $t$ overstates precision for a regressor that varies at the cohort level. Source: results/r3k\_v2/results.json (structural\_break). | |
| 20 | +\end{tablenotes} | |
| 21 | +\end{threeparttable}} | |
modified
paper/tables/tab_cohorts.tex
+38 −18
@@ -1,22 +1,42 @@ | ||
| 1 | −\begin{tabular}{lrrr@{\hspace{2em}}lr} | |
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results2.json (cohorts*, placebo_breaks), results_surv.json (then_members_present) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_cohorts} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:cohorts}{\textbf{Cohort-by-cohort spreads of the AI composite and tests of a 2022 break, Russell~3000.} Panel A: equal-weighted Q5$-$Q1 forward-return spread of the composite for each formation cohort (raw, within-year winsorized, median difference), the same spread restricted to firms that were IWV constituents at formation (Wayback captures exist from 2019 only), and the pre- and post-launch means with 95\% $t$-intervals and leave-one-out post means. Panel B: an exact permutation test of the 2022--24 block against every three-cohort subset, the contiguous-block comparison, the rank of 2022 among the 10 candidate break years, and the Andrews sup-Wald test for a mean shift in the cohort series (simulated finite-sample $p$).}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 5 | +\begin{tabular}{lrrrl} | |
| 2 | 6 | \toprule |
| 3 | −Cohort & Q5$-$Q1 raw (pp) & wins. & median & Pooled / permutation test & Estimate \\ | |
| 7 | +\multicolumn{5}{@{}l}{\textit{Panel A: equal-weighted Q5$-$Q1 spread of the composite by formation cohort (pp)}} \\ | |
| 8 | +Cohort & raw & wins. & median & Then-IWV-members only (wins.) \\ | |
| 4 | 9 | \midrule |
| 5 | −2010 & -6.5 & -7.6 & -9.9 & Post cohorts, Q5 vs Q1, raw: firm $t$ / two-way $t$ / wild $p$ & +8.7 (1.95 / 1.10 / 0.366), $N$=2{,}784 \\ | |
| 6 | −2011 & +0.3 & +0.1 & -3.2 & Post, winsorized & +6.2 (2.14 / 0.84 / 0.631) \\ | |
| 7 | −2012 & +9.4 & +7.9 & +5.1 & Post 2022--2023 only (wins.) & -0.9 (-0.36) \\ | |
| 8 | −2013 & -3.4 & -3.5 & -2.4 & Pre cohorts (wins.) & +3.8 (4.15) \\ | |
| 9 | −2014 & +1.7 & +0.9 & -4.3 & Difference post $-$ pre (wins.): firm / two-way / wild $p$; MDE & +2.4 (0.78 / 0.36 / 0.761); 8.5 pp \\ | |
| 10 | −2015 & +1.5 & +1.2 & -3.8 & Post mean / leave-one-out (raw) & +8.5 / +16.0, +6.8, +2.6 \\ | |
| 11 | −2016 & +6.7 & +7.1 & +8.2 & Post mean / leave-one-out (wins.) & +6.0 / +11.5, +7.5, -0.9 \\ | |
| 12 | −2017 & +9.2 & +8.7 & +7.9 & Pre mean, 95\% CI (raw / wins.) & +4.8 [-0.7, 10.2] / +3.4 [-1.6, 8.4] \\ | |
| 13 | −2018 & +21.4 & +19.0 & +18.0 & Permutation $p$, 2022--24 vs all 455 three-cohort blocks (raw / wins.) & 0.28 (rank 126) / 0.31 (rank 141) \\ | |
| 14 | −2019 & +11.6 & +1.4 & -1.7 & Contiguous three-cohort blocks above 2022--24 (raw / wins.) & 3 / 2 of 12 \\ | |
| 15 | −2020 & -7.4 & -6.6 & -10.2 & sup-Wald mean-shift test (Andrews 1993, 15\% trim), raw / wins.: stat ($p$ simulated) & 3.43 (0.43) / 2.57 (0.59) \\ | |
| 16 | −2021 & +12.6 & +12.1 & +12.4 & & \\ | |
| 17 | −2022 & -6.7 & -5.0 & -5.6 & & \\ | |
| 18 | −2023 & +11.8 & +3.1 & -0.4 & & \\ | |
| 19 | −2024 & +20.2 & +19.9 & -2.2 & & \\ | |
| 10 | +2010 & -6.5 & -7.6 & -9.9 & -- \\ | |
| 11 | +2011 & +0.3 & +0.1 & -3.2 & -- \\ | |
| 12 | +2012 & +9.4 & +7.9 & +5.1 & -- \\ | |
| 13 | +2013 & -3.4 & -3.5 & -2.4 & -- \\ | |
| 14 | +2014 & +1.7 & +0.9 & -4.3 & -- \\ | |
| 15 | +2015 & +1.5 & +1.2 & -3.8 & -- \\ | |
| 16 | +2016 & +6.7 & +7.1 & +8.2 & -- \\ | |
| 17 | +2017 & +9.2 & +8.7 & +7.9 & -- \\ | |
| 18 | +2018 & +21.4 & +19.0 & +18.0 & +20.4 (86\% members) \\ | |
| 19 | +2019 & +11.6 & +1.4 & -1.7 & +4.2 (88\% members) \\ | |
| 20 | +2020 & -7.4 & -6.6 & -10.2 & -6.7 (84\% members) \\ | |
| 21 | +2021 & +12.6 & +12.1 & +12.4 & +14.4 (82\% members) \\ | |
| 22 | +2022 & -6.7 & -5.0 & -5.6 & -2.9 (88\% members) \\ | |
| 23 | +2023 & +11.8 & +3.1 & -0.4 & +4.2 (90\% members) \\ | |
| 24 | +2024 & +20.2 & +19.9 & -2.2 & +16.4 (94\% members) \\ | |
| 25 | +\midrule | |
| 26 | +Pre mean (2010--21), 95\% CI & +4.8 [-0.7, +10.2] & +3.4 [-1.6, +8.4] & +1.4 & \\ | |
| 27 | +Post mean (2022--24), 95\% CI & +8.5 [-25.8, +42.7] & +6.0 [-25.5, +37.6] & -2.7 & \\ | |
| 28 | +Post mean leaving out 2022 / 2023 / 2024 & +16.0 / +6.8 / +2.6 & +11.5 / +7.5 / -0.9 & -1.3 / -3.9 / -3.0 & \\ | |
| 29 | +\midrule | |
| 30 | +\multicolumn{5}{@{}l}{\textit{Panel B: is the 2022--24 block unusual? (composite)}} \\ | |
| 31 | +Test & \multicolumn{2}{l}{raw} & \multicolumn{2}{l}{wins.} \\ | |
| 32 | +\midrule | |
| 33 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.4\textwidth}}{Permutation $p$ (one-sided), 2022--24 vs all 455 three-cohort subsets (rank)} & \multicolumn{2}{l}{0.28 (rank 126)} & \multicolumn{2}{l}{0.31 (rank 141)} \\ | |
| 34 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.4\textwidth}}{Contiguous three-cohort blocks with a higher mean than 2022--24} & \multicolumn{2}{l}{3 of 12} & \multicolumn{2}{l}{2 of 12} \\ | |
| 35 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.4\textwidth}}{Rank of 2022 among 10 candidate break years (post $-$ pre mean)} & \multicolumn{2}{l}{8 of 10} & \multicolumn{2}{l}{8 of 10} \\ | |
| 36 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.4\textwidth}}{sup-Wald mean-shift statistic, Andrews (1993), 15\% trim (simulated $p$)} & \multicolumn{2}{l}{3.43 (0.43)} & \multicolumn{2}{l}{2.57 (0.59)} \\ | |
| 20 | 37 | \bottomrule |
| 21 | 38 | \end{tabular} |
| 22 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} NW = Newey--West (Bartlett, 3 lags) $t$ of the mean of the cohort series; plain = i.i.d. $t$; CI = 95\% $t$-interval over cohorts; $n$ = number of cohorts. Returns are the look-ahead-free 12-month forward returns (July $Y{+}1$--June $Y{+}2$); `wins.' = 1/99 winsorized within the formation year; `median' = median spread within cohort. Rebuilt panel ($N$ = 27{,}354 firm-years, 2{,}375 firms, keyed on CIK $\times$ year, cleaned returns, corrected market capitalization).} | |
| \ No newline at end of file | ||
| 39 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 40 | +\item \emph{Notes:} Inference: NW $t$ = Newey--West (Bartlett kernel, 3 lags) $t$-statistic of the mean of the annual cohort series; plain $t$ = i.i.d. $t$ over cohorts; 95\% CI = $t$-interval over cohorts (the post period has 3 cohorts, so its CIs are wide by construction). Returns are the look-ahead-free twelve-month forward returns (July $Y{+}1$--June $Y{+}2$); `wins.' = 1/99 winsorized within the formation year; `median' = median spread within cohort; `raw' = unwinsorized. Rebuilt panel: $N$ = 27{,}354 firm-years, 2{,}375 firms, keyed on CIK $\times$ year, cleaned returns, corrected market capitalization. No significance stars are used. Then-members-only spreads use the IWV holdings capture nearest the formation month (results/r3k\_v2/results\_surv.json). The permutation $p$ is the share of the 455 three-cohort subsets whose mean spread is at least the 2022--24 mean; the sup-Wald $p$ is from 20,000 simulated series of length 15. Economic reading: the post mean leaving out 2024 (wins.\ -0.9 pp) is at the pre mean (+3.4); none of the break tests rejects at conventional levels. | |
| 41 | +\end{tablenotes} | |
| 42 | +\end{threeparttable}} | |
modified
paper/tables/tab_construct_validity.tex
+27 −11
@@ -1,17 +1,33 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results_construct.json, results/n1/tokenizer_audit.json, results/n1/keyword_baseline.json -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_construct_validity} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:constructvalidity}{\textbf{Construct validity of the semantic AI score: keyword baseline, stripped anchors, sector composition and encoder truncation.} Panel A: the composite and five alternative measures scored on the same firm-years: Spearman correlation with the composite, equal-weighted Q5$-$Q1 spreads (within-year winsorized, pp per year; [NW $t$] [plain $t$] and 95\% CI over cohorts; pre and post means) and Fama--MacBeth slopes with controls (pp per SD, [NW $t$]) on winsorized and raw returns. The stripped anchors re-score the stored chunk embeddings against anchor texts without the cybersecurity and automation vocabulary (no corpus re-embedding). Panel B: horse races. Panel C: what the encoder actually reads and how much of the score is lexical.}{\scriptsize}{% | |
| 4 | +\begin{threeparttable}\setlength{\tabcolsep}{3.5pt} | |
| 1 | 5 | \begin{tabular}{lrrrrrrrr} |
| 2 | 6 | \toprule |
| 3 | −Signal & $N$ & $\rho$ with composite & Full Q5$-$Q1 raw & Full Q5$-$Q1 wins. (NW $t$ / plain $t$) & Pre wins. & Post wins. & FM slope raw (NW $t$) & FM slope wins. (NW $t$) \\ | |
| 7 | +Signal & $\rho$ with composite & Full Q5$-$Q1 & [NW $t$] [plain $t$] & 95\% CI & Pre & Post & FM slope, wins. & FM slope, raw \\ | |
| 4 | 8 | \midrule |
| 5 | −AI composite (baseline) & 27399 & 1.00 & +5.7 & +4.1 (2.08 / 1.84) & +3.6 & +6.0 & +2.75 (2.45) & +1.87 (2.41) \\ | |
| 6 | −Keyword baseline (core AI terms per 1,000 words) & 27399 & 0.13 & +7.2 & +4.7 (2.30 / 1.82) & +3.1 & +11.1 & +0.51 (1.03) & +0.61 (1.25) \\ | |
| 7 | −Keyword baseline, log(1 + x) & 27399 & 0.14 & +7.2 & +4.7 (2.30 / 1.82) & +3.1 & +11.1 & +0.77 (2.08) & +0.81 (1.95) \\ | |
| 8 | −Stripped-anchor composite (AR-2) & 27399 & 0.98 & +5.8 & +4.1 (2.33 / 2.13) & +4.1 & +4.0 & +2.62 (2.65) & +1.79 (2.70) \\ | |
| 9 | −Core-AI adoption anchor (AR-2) & 27399 & 0.80 & +6.1 & +4.8 (2.69 / 1.99) & +4.7 & +5.2 & +2.26 (3.15) & +1.74 (3.02) \\ | |
| 10 | −Cybersecurity anchor stripped of AI terms (AR-2) & 27399 & 0.35 & +2.8 & +1.9 (0.97 / 1.12) & +0.1 & +8.8 & +0.90 (1.07) & +0.92 (1.49) \\ | |
| 11 | −AI composite, excluding Utilities and Energy & 25420 & & +6.4 & +4.6 (2.16 / 1.84) & +4.1 & +6.6 & +3.02 (2.47) & +2.06 (2.46) \\ | |
| 9 | +\multicolumn{9}{@{}l}{\textit{Panel A: alternative measures of AI language (same panel, same return)}} \\ | |
| 10 | +AI composite (baseline) & 1.00 & +4.1 & [2.08] [1.84] & [-0.7, +8.8] & +3.6 & +6.0 & +1.87 [2.41] & +2.75 [2.45] \\ | |
| 11 | +Keyword baseline (core AI terms per 1,000 words) & 0.13 & +4.7 & [2.30] [1.82] & [-0.9, +10.3] & +3.1 & +11.1 & +0.61 [1.25] & +0.51 [1.03] \\ | |
| 12 | +Keyword baseline, $\log(1 + x)$ & 0.14 & +4.7 & [2.30] [1.82] & [-0.9, +10.3] & +3.1 & +11.1 & +0.81 [1.95] & +0.77 [2.08] \\ | |
| 13 | +Stripped-anchor composite & 0.98 & +4.1 & [2.33] [2.13] & [-0.0, +8.3] & +4.1 & +4.0 & +1.79 [2.70] & +2.62 [2.65] \\ | |
| 14 | +Core-AI adoption anchor & 0.80 & +4.8 & [2.69] [1.99] & [-0.4, +10.0] & +4.7 & +5.2 & +1.74 [3.02] & +2.26 [3.15] \\ | |
| 15 | +Cyber anchor stripped of AI terms & 0.35 & +1.9 & [0.97] [1.12] & [-1.7, +5.5] & +0.1 & +8.8 & +0.92 [1.49] & +0.90 [1.07] \\ | |
| 16 | +Composite ex Utilities and Energy ($N$ = 25{,}420) & -- & +4.6 & [2.16] [1.84] & [-0.8, +9.9] & +4.1 & +6.6 & +2.06 [2.46] & +3.02 [2.47] \\ | |
| 12 | 17 | \midrule |
| 13 | −Horse race (FM, wins.): composite vs keyword & \multicolumn{8}{l}{z_AI_COMPOS: +1.85 (2.37); z_KW: +0.36 (0.68)} \\ | |
| 14 | −Horse race (FM, wins.): composite vs stripped & \multicolumn{8}{l}{z_AI_COMPOS: +0.15 (0.06); z_ALT: +1.66 (0.79)} \\ | |
| 18 | +\multicolumn{9}{@{}l}{\textit{Panel B: Fama--MacBeth horse races (wins., pp per SD [NW $t$])}} \\ | |
| 19 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.3\textwidth}}{Composite vs keyword baseline} & \multicolumn{8}{l}{composite +1.85 [2.37]; keyword +0.36 [0.68]} \\ | |
| 20 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.3\textwidth}}{Composite vs stripped-anchor composite} & \multicolumn{8}{l}{composite +0.15 [0.06]; stripped +1.66 [0.79]} \\ | |
| 21 | +\midrule | |
| 22 | +\multicolumn{9}{@{}l}{\textit{Panel C: encoder truncation and lexical content of the score (node audit of the stored chunks and embeddings)}} \\ | |
| 23 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.5\textwidth}}{Encoder window (tokens) / chunk length (words) / median tokens per chunk} & \multicolumn{8}{l}{256 / 512 / 634} \\ | |
| 24 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.5\textwidth}}{Chunks truncated (frame K, $n$ = 10{,}000) / median words read / share of token mass dropped} & \multicolumn{8}{l}{99.8\% / 205 / 60\%} \\ | |
| 25 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.5\textwidth}}{AI keyword occurrences beyond the encoder window} & \multicolumn{8}{l}{61\% of 40{,}522} \\ | |
| 26 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.5\textwidth}}{Spearman correlation of the keyword baseline with the composite / firm-years with zero core AI hits} & \multicolumn{8}{l}{0.13 / 84\%} \\ | |
| 27 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.5\textwidth}}{Utilities: share of top-quintile firm-years vs share of the sample} & \multicolumn{8}{l}{6.5\% vs 3.2\%} \\ | |
| 15 | 28 | \bottomrule |
| 16 | 29 | \end{tabular} |
| 17 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Keyword baseline from the node audit (code/n1, published AI lexicon over the section texts; share of firm-years with zero core hits 0.839); stripped-anchor composite = anchors without the cybersecurity / automation vocabulary, scored against the stored chunk embeddings (AR-FIRC0002-2, no corpus re-embedding). Utilities are 6.5\% of top-quintile firm-years vs 3.2\% of the sample. NW $t$ over cohorts (3 lags); FM controls size, momentum, volatility.} | |
| \ No newline at end of file | ||
| 30 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 31 | +\item \emph{Notes:} Keyword baseline = core AI terms (published AI lexicons) per 1,000 words of the section texts, computed on the node audit (code/n1); the keyword and semantic measures agree weakly, and the keyword measure carries a similar sort spread but a weaker regression slope. The encoder reads roughly the first 200 words of each 512-word chunk, so the score is a function of chunk openings; re-chunking to the encoder window (X4) was not executed and the consequences for the premium are unknown. No human- or LLM-labelled validation sample exists: the 1,700-chunk coding frames and codebook were delivered without labels (results/n1/validation\_*), so precision and recall of the score are not measured. $N$ = 27{,}399 firm-years in Panel A. Source: results/r3k\_v2/results\_construct.json, results/n1/tokenizer\_audit.json, keyword\_baseline.json. | |
| 32 | +\end{tablenotes} | |
| 33 | +\end{threeparttable}} | |
modified
paper/tables/tab_contrasts.tex
+27 −17
@@ -1,30 +1,40 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results.json (contrasts) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_contrasts} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:contrasts}{\textbf{Formal tests of the large-versus-small and member-versus-non-member post-launch differences.} Panel A: pooled regressions of the forward return of Q1 and Q5 firm-years on a top-quintile indicator (hi), a post-launch dummy, the group indicator and all interactions, with cohort fixed effects. The triple interaction is the post-launch change of the spread in the group relative to the reference group. $t$-statistics in brackets from firm-clustered and two-way (firm $\times$ cohort) standard errors; 95\% CI from the firm-clustered SE; wild $p$ from the restricted wild-cluster bootstrap by cohort (Rademacher / Webb weights, 9,999 draws, 15 clusters); MDE = minimum detectable effect at 80\% power (2.8 $\times$ firm-clustered SE). Panel B: the difference of the post $-$ pre changes of the cohort-level spreads with a 95\% cohort-bootstrap interval (12 pre and 3 post cohorts resampled, 9,999 draws).}{\footnotesize}{% | |
| 4 | +\begin{threeparttable}\setlength{\tabcolsep}{4pt} | |
| 1 | 5 | \begin{tabular}{llrrrrrr} |
| 2 | 6 | \toprule |
| 3 | −Contrast & Returns & Coefficient (pp) & $t$ (firm) & $t$ (firm$\times$year) & 95\% CI (firm) & wild $p$ (Rademacher / Webb) & MDE (80\% power) \\ | |
| 7 | +Term & Returns & Coefficient (pp) & [$t$ firm] & [$t$ firm$\times$cohort] & 95\% CI & wild $p$ (Rad.\ / Webb) & MDE (pp) \\ | |
| 4 | 8 | \midrule |
| 5 | −hi $\times$ post (reference group) & wins. & -1.23 & -0.25 & -0.12 & [-11.0, +8.5] & -- / -- & 13.9 \\ | |
| 6 | −hi $\times$ group (pre) & wins. & -4.06 & -1.84 & -0.99 & [-8.4, +0.3] & -- / -- & 6.2 \\ | |
| 7 | −hi $\times$ post $\times$ largest-1,000 & wins. & +10.79 & 1.83 & 1.55 & [-0.8, +22.4] & 0.226 / 0.225 & 16.5 \\ | |
| 9 | +\multicolumn{8}{@{}l}{\textit{Panel A: pooled firm-level interactions (Q1/Q5 firm-years, cohort fixed effects)}} \\ | |
| 10 | +hi $\times$ post (reference group) & wins. & -1.23 & [-0.25] & [-0.12] & [-11.0, +8.5] & -- / -- & 13.9 \\ | |
| 11 | +hi $\times$ group (pre-launch) & wins. & -4.06 & [-1.84] & [-0.99] & [-8.4, +0.3] & -- / -- & 6.2 \\ | |
| 12 | +hi $\times$ post $\times$ largest-1,000 & wins. & +10.79 & [1.83] & [1.55] & [-0.8, +22.4] & 0.226 / 0.225 & 16.5 \\ | |
| 8 | 13 | \quad $N$ = 10{,}826 & & & & & & & \\ |
| 9 | 14 | \addlinespace |
| 10 | −hi $\times$ post (reference group) & raw & -0.28 & -0.03 & -0.02 & [-17.6, +17.0] & -- / -- & 24.7 \\ | |
| 11 | −hi $\times$ group (pre) & raw & -8.99 & -2.44 & -1.40 & [-16.2, -1.8] & -- / -- & 10.3 \\ | |
| 12 | −hi $\times$ post $\times$ largest-1,000 & raw & +10.81 & 1.14 & 0.82 & [-7.8, +29.4] & 0.492 / 0.495 & 26.6 \\ | |
| 15 | +hi $\times$ post (reference group) & raw & -0.28 & [-0.03] & [-0.02] & [-17.6, +17.0] & -- / -- & 24.7 \\ | |
| 16 | +hi $\times$ group (pre-launch) & raw & -8.99 & [-2.44] & [-1.40] & [-16.2, -1.8] & -- / -- & 10.3 \\ | |
| 17 | +hi $\times$ post $\times$ largest-1,000 & raw & +10.81 & [1.14] & [0.82] & [-7.8, +29.4] & 0.492 / 0.495 & 26.6 \\ | |
| 13 | 18 | \quad $N$ = 10{,}826 & & & & & & & \\ |
| 14 | 19 | \addlinespace |
| 15 | −hi $\times$ post (reference group) & wins. & +1.57 & 0.43 & 0.20 & [-5.5, +8.7] & -- / -- & 10.1 \\ | |
| 16 | −hi $\times$ group (pre) & wins. & +0.87 & 0.49 & 0.41 & [-2.6, +4.3] & -- / -- & 4.9 \\ | |
| 17 | −hi $\times$ post $\times$ S\&P member & wins. & +8.57 & 1.39 & 1.07 & [-3.5, +20.7] & 0.319 / 0.345 & 17.3 \\ | |
| 20 | +hi $\times$ post (reference group) & wins. & +1.57 & [0.43] & [0.20] & [-5.5, +8.7] & -- / -- & 10.1 \\ | |
| 21 | +hi $\times$ group (pre-launch) & wins. & +0.87 & [0.49] & [0.41] & [-2.6, +4.3] & -- / -- & 4.9 \\ | |
| 22 | +hi $\times$ post $\times$ S\&P member & wins. & +8.57 & [1.39] & [1.07] & [-3.5, +20.7] & 0.319 / 0.345 & 17.3 \\ | |
| 18 | 23 | \quad $N$ = 10{,}949 & & & & & & & \\ |
| 19 | 24 | \addlinespace |
| 20 | −hi $\times$ post (reference group) & raw & +2.77 & 0.46 & 0.29 & [-9.0, +14.6] & -- / -- & 16.9 \\ | |
| 21 | −hi $\times$ group (pre) & raw & -1.21 & -0.50 & -0.42 & [-5.9, +3.5] & -- / -- & 6.7 \\ | |
| 22 | −hi $\times$ post $\times$ S\&P member & raw & +8.53 & 1.03 & 0.67 & [-7.7, +24.8] & 0.508 / 0.530 & 23.2 \\ | |
| 25 | +hi $\times$ post (reference group) & raw & +2.77 & [0.46] & [0.29] & [-9.0, +14.6] & -- / -- & 16.9 \\ | |
| 26 | +hi $\times$ group (pre-launch) & raw & -1.21 & [-0.50] & [-0.42] & [-5.9, +3.5] & -- / -- & 6.7 \\ | |
| 27 | +hi $\times$ post $\times$ S\&P member & raw & +8.53 & [1.03] & [0.67] & [-7.7, +24.8] & 0.508 / 0.530 & 23.2 \\ | |
| 23 | 28 | \quad $N$ = 10{,}949 & & & & & & & \\ |
| 24 | 29 | \addlinespace |
| 25 | −(post $-$ pre)$_{\text{large}}$ $-$ (post $-$ pre)$_{\text{small}}$, cohort spreads, wins. & & +9.96 & & & cohort-bootstrap CI [-4.0, +22.0] & & \\ | |
| 26 | −same, raw & & +11.40 & & & cohort-bootstrap CI [-14.9, +31.7] & & \\ | |
| 27 | −(post $-$ pre)$_{\text{S\&P}}$ $-$ (post $-$ pre)$_{\text{non}}$, wins. & & +9.08 & & & cohort-bootstrap CI [-13.2, +24.1] & & \\ | |
| 30 | +\midrule | |
| 31 | +\multicolumn{8}{@{}l}{\textit{Panel B: difference of cohort-level post $-$ pre changes, cohort bootstrap}} \\ | |
| 32 | +(post $-$ pre)$_{\text{largest 1,000}}$ $-$ (post $-$ pre)$_{\text{remaining}}$ & wins. & +9.96 & & & [-4.0, +22.0] & & \\ | |
| 33 | +same, raw returns & raw & +11.40 & & & [-14.9, +31.7] & & \\ | |
| 34 | +(post $-$ pre)$_{\text{S\&P}}$ $-$ (post $-$ pre)$_{\text{non-members}}$ & wins. & +9.08 & & & [-13.2, +24.1] & & \\ | |
| 28 | 35 | \bottomrule |
| 29 | 36 | \end{tabular} |
| 30 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Pooled regressions of Q1/Q5 forward returns on hi, post, group and all interactions with cohort fixed effects; wild-cluster bootstrap by cohort (15 clusters, 9,999 draws). Cohort-bootstrap CIs resample the 12 pre and 3 post cohort spreads (9,999 draws). MDE = 2.8 $\times$ firm-clustered SE.} | |
| \ No newline at end of file | ||
| 37 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 38 | +\item \emph{Notes:} Economic reading: the point estimate of the large-cap post-launch increase relative to small caps is +10.8 pp per year, but its MDE is 16.5 pp and the wild-bootstrap $p$ is 0.22, so the difference between the two size groups is not statistically established; the cohort-bootstrap interval in Panel B covers zero. The S\&P-member contrast is weaker still. Returns winsorized within the formation year unless marked raw. Source: results/r3k\_v2/results.json (contrasts). | |
| 39 | +\end{tablenotes} | |
| 40 | +\end{threeparttable}} | |
modified
paper/tables/tab_corpus.tex
+9 −2
@@ -1,6 +1,10 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from data/raw_russell3000/filings, results/n1/corpus_counts.json -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_corpus} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:corpus}{\textbf{Corpus composition by form.} Filings of the 2{,}499 issuers of the universe (S\&P~500 constituents and iShares Russell~3000 holdings) indexed in the EDGAR Research Database for 2010--2026, the number parsed into sections and 512-word chunks, and the share of the 27{,}354 estimation firm-years whose signal includes at least one scored filing of that form.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 1 | 5 | \begin{tabular}{lrrrr} |
| 2 | 6 | \toprule |
| 3 | −Form & Filings 2010--2026 & Issuers & Parsed & Firm-years with a scored filing (estimation sample) \\ | |
| 7 | +Form & Filings 2010--2026 & Issuers & Parsed & Firm-years with a scored filing \\ | |
| 4 | 8 | \midrule |
| 5 | 9 | 10-K & 31{,}775 & 2{,}468 & 31{,}772 & 96.8\% \\ |
| 6 | 10 | 10-Q & 95{,}167 & 2{,}494 & 95{,}155 & 99.6\% \\ |
@@ -9,4 +13,7 @@ Form & Filings 2010--2026 & Issuers & Parsed & Firm-years with a scored filing ( | ||
| 9 | 13 | Total & 527{,}548 & 2{,}499 & 527{,}532 & \\ |
| 10 | 14 | \bottomrule |
| 11 | 15 | \end{tabular} |
| 12 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Filings index of the EDGAR Research Database (data/raw\_russell3000/filings). Node audit of the parquet copy (results/n1): 10{,}720{,}890 chunks, 1{,}342{,}283 sections, 527{,}548 filings; median chunk length 512 words, 84.2\% of chunks at the 512-word cap.} | |
| \ No newline at end of file | ||
| 16 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 17 | +\item \emph{Notes:} Filings index: data/raw\_russell3000/filings. Node audit of the parquet copy (results/n1): 10{,}720{,}890 chunks, 1{,}342{,}283 sections, 527{,}548 filings; median chunk length 512 words, 84.2\% of chunks at the 512-word cap. The encoder reads only the first 256 tokens of each chunk (Table~\ref{tab:constructvalidity}). Unparsed filings are parser errors or files missing on disk. | |
| 18 | +\end{tablenotes} | |
| 19 | +\end{threeparttable}} | |
modified
paper/tables/tab_corr.tex
+8 −1
@@ -1,3 +1,7 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/t1b_correlations.csv -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: appendix. \input{tables/tab_corr} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:corr}{\textbf{Correlations of the within-year standardized variables.} Pearson correlations over the 27{,}354 firm-years of the estimation sample; signals and controls standardized within formation year; the forward return is the raw twelve-month return.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 1 | 5 | \begin{tabular}{lrrrrrrrrr} |
| 2 | 6 | \toprule |
| 3 | 7 | & AI adoption & AI infrastructure & AI disruption risk & PC1 & SPID tech projects & Cybersecurity & Size & Momentum & Forward return \\ |
@@ -13,4 +17,7 @@ Momentum & 0.03 & 0.02 & 0.02 & 0.02 & -0.01 & 0.01 & 0.14 & 1.00 & \\ | ||
| 13 | 17 | Forward return & 0.02 & 0.03 & 0.05 & 0.04 & 0.00 & 0.01 & -0.11 & -0.04 & 1.00 \\ |
| 14 | 18 | \bottomrule |
| 15 | 19 | \end{tabular} |
| 16 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Pearson correlations over the 27{,}354 firm-years of the estimation sample (signals standardized within year).} | |
| \ No newline at end of file | ||
| 20 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 21 | +\item \emph{Notes:} The three AI facets are highly correlated with each other (and with PC1) and weakly with the SPID project count, which is a count variable with a zero median. Source: results/r3k\_v2/t1b\_correlations.csv. | |
| 22 | +\end{tablenotes} | |
| 23 | +\end{threeparttable}} | |
modified
paper/tables/tab_coverage.tex
+25 −18
@@ -1,22 +1,29 @@ | ||
| 1 | −\begin{tabular}{lrrrrrrr} | |
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results2.json (coverage) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: appendix. \input{tables/tab_coverage} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:coverage}{\textbf{Estimation sample by formation cohort: firms, size groups and text coverage.} Number of firms with a valid forward return in each formation cohort, the size-group counts under the two partitions (largest 1,000 at formation on the corrected capitalization; largest 42\% within cohort), the firm-years without a capitalization, the S\&P~500 members, and the median number of scored filings and text chunks per firm-year.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 5 | +\begin{tabular}{lrrrrrrrr} | |
| 2 | 6 | \toprule |
| 3 | −Cohort & Firms & Largest 1,000 & Remaining & No cap & Largest 42\% & Median filings & Median chunks \\ | |
| 7 | +Cohort & Firms & Largest 1,000 & Remaining & No cap & Largest 42\% & S\&P 500 & Median filings & Median chunks \\ | |
| 4 | 8 | \midrule |
| 5 | −2010 & 1{,}342 & 1{,}000 & 328 & 14 & 557 & 15 & 258 \\ | |
| 6 | −2011 & 1{,}376 & 1{,}000 & 361 & 15 & 571 & 16 & 268 \\ | |
| 7 | −2012 & 1{,}425 & 1{,}000 & 410 & 15 & 592 & 16 & 275 \\ | |
| 8 | −2013 & 1{,}491 & 1{,}000 & 478 & 13 & 620 & 15 & 281 \\ | |
| 9 | −2014 & 1{,}574 & 1{,}000 & 560 & 14 & 655 & 16 & 288 \\ | |
| 10 | −2015 & 1{,}645 & 1{,}000 & 628 & 17 & 684 & 16 & 293 \\ | |
| 11 | −2016 & 1{,}705 & 1{,}000 & 686 & 19 & 708 & 15 & 297 \\ | |
| 12 | −2017 & 1{,}774 & 1{,}000 & 752 & 22 & 736 & 15 & 308 \\ | |
| 13 | −2018 & 1{,}853 & 1{,}000 & 832 & 21 & 770 & 15 & 318 \\ | |
| 14 | −2019 & 1{,}935 & 1{,}000 & 912 & 23 & 804 & 15 & 323 \\ | |
| 15 | −2020 & 2{,}060 & 1{,}000 & 1{,}035 & 25 & 855 & 16 & 339 \\ | |
| 16 | −2021 & 2{,}221 & 999 & 1{,}196 & 26 & 921 & 15 & 334 \\ | |
| 17 | −2022 & 2{,}267 & 999 & 1{,}241 & 27 & 940 & 15 & 327 \\ | |
| 18 | −2023 & 2{,}312 & 1{,}000 & 1{,}284 & 28 & 959 & 14 & 320 \\ | |
| 19 | −2024 & 2{,}374 & 1{,}000 & 1{,}345 & 29 & 984 & 14 & 303 \\ | |
| 9 | +2010 & 1{,}342 & 1{,}000 & 328 & 14 & 557 & 402 & 15 & 258 \\ | |
| 10 | +2011 & 1{,}376 & 1{,}000 & 361 & 15 & 571 & 407 & 16 & 268 \\ | |
| 11 | +2012 & 1{,}425 & 1{,}000 & 410 & 15 & 592 & 418 & 16 & 275 \\ | |
| 12 | +2013 & 1{,}491 & 1{,}000 & 478 & 13 & 620 & 425 & 15 & 281 \\ | |
| 13 | +2014 & 1{,}574 & 1{,}000 & 560 & 14 & 655 & 430 & 16 & 288 \\ | |
| 14 | +2015 & 1{,}645 & 1{,}000 & 628 & 17 & 684 & 437 & 16 & 293 \\ | |
| 15 | +2016 & 1{,}705 & 1{,}000 & 686 & 19 & 708 & 441 & 15 & 297 \\ | |
| 16 | +2017 & 1{,}774 & 1{,}000 & 752 & 22 & 736 & 450 & 15 & 308 \\ | |
| 17 | +2018 & 1{,}853 & 1{,}000 & 832 & 21 & 770 & 454 & 15 & 318 \\ | |
| 18 | +2019 & 1{,}935 & 1{,}000 & 912 & 23 & 804 & 463 & 15 & 323 \\ | |
| 19 | +2020 & 2{,}060 & 1{,}000 & 1{,}035 & 25 & 855 & 470 & 16 & 339 \\ | |
| 20 | +2021 & 2{,}221 & 999 & 1{,}196 & 26 & 921 & 475 & 15 & 334 \\ | |
| 21 | +2022 & 2{,}267 & 999 & 1{,}241 & 27 & 940 & 478 & 15 & 327 \\ | |
| 22 | +2023 & 2{,}312 & 1{,}000 & 1{,}284 & 28 & 959 & 482 & 14 & 320 \\ | |
| 23 | +2024 & 2{,}374 & 1{,}000 & 1{,}345 & 29 & 984 & 487 & 14 & 303 \\ | |
| 20 | 24 | \bottomrule |
| 21 | 25 | \end{tabular} |
| 22 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Estimation sample by formation cohort; size groups at formation on the corrected capitalization.} | |
| \ No newline at end of file | ||
| 26 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 27 | +\item \emph{Notes:} The panel grows from 1{,}342 firms in 2010 to 2{,}374 in 2024 because the universe is the October 2026 membership: early cohorts contain only the firms that survived to 2026 (Table~\ref{tab:attrition}). Under the top-1,000 partition the `remaining' group has 328 firms in 2010 (quintiles of about 65 firms); the 42\% partition keeps the group shares constant. Rank autocorrelation of the composite: lag 1 0.87, lag 5 0.72. Source: results/r3k\_v2/results2.json (coverage). | |
| 28 | +\end{tablenotes} | |
| 29 | +\end{threeparttable}} | |
modified
paper/tables/tab_did_spid.tex
+32 −22
@@ -1,31 +1,41 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results2.json (did, spid), t17_refit.json -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_did_spid} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:didspid}{\textbf{Difference-in-differences on pre-determined exposure and the SPID project signal, Russell~3000.} Panel A: forward return (within-year winsorized, pp) on the interaction of pre-determined AI exposure (firm mean of the composite over filing years 2019--2021, binary above-median or continuous) with the post-launch dummy, with firm and year fixed effects and two-way clustered $t$ (brackets); the event-time version with the reference year at zero is in Figure~\ref{fig:eventtime} and Table~\ref{tab:eventtime}. Panel B: pooled regressions with the explicit terms signal, post, signal $\times$ post, composite, composite $\times$ post, size and momentum (pp per SD, or pp per unit of the indicator); $t$-statistics clustered by firm and two-way by firm and cohort; 95\% CI from the firm-clustered SE.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 1 | 5 | \begin{tabular}{lrrrr} |
| 2 | 6 | \toprule |
| 3 | −Specification & Coefficient (pp) & $t$ (firm) & $t$ (firm$\times$year) & 95\% CI (firm) \\ | |
| 7 | +Specification & Coefficient (pp) & [$t$ firm] & [$t$ firm$\times$cohort] & 95\% CI (firm) \\ | |
| 4 | 8 | \midrule |
| 5 | −DiD: treated (above-median 2019--21 exposure) $\times$ post, firm + year FE & +4.71 & & 1.28 & \\ | |
| 6 | −DiD: continuous exposure $\times$ post, firm + year FE & +2.79 & & 1.28 & \\ | |
| 9 | +\multicolumn{5}{@{}l}{\textit{Panel A: difference-in-differences with firm and year fixed effects (treatment = pre-determined 2019--21 exposure)}} \\ | |
| 10 | +Treated (above-median exposure) $\times$ post & +4.71 & -- & [1.28] & -- \\ | |
| 11 | +Continuous exposure ($z$) $\times$ post & +2.79 & -- & [1.28] & -- \\ | |
| 12 | +\quad $N$ = 26{,}664; firm + year FE; two-way clustered & & & & \\ | |
| 7 | 13 | \midrule |
| 8 | −SPID technology projects ($z$) & +0.40 & 1.21 & 0.60 & [-0.25, +1.04] \\ | |
| 9 | −SPID technology projects ($z$) $\times$ post & -1.93 & -1.93 & -0.96 & [-3.88, +0.03] \\ | |
| 10 | −AI composite ($z$) & +2.00 & 5.78 & 2.24 & [+1.32, +2.68] \\ | |
| 11 | −AI composite $\times$ post & +1.77 & 1.63 & 0.82 & [-0.36, +3.91] \\ | |
| 12 | −Post-launch cohorts & +9.54 & 9.91 & 0.85 & [+7.65, +11.43] \\ | |
| 13 | −\quad $N$ = 26{,}933; design rank 8, condition number 2.8 \\ | |
| 14 | +\multicolumn{5}{@{}l}{\textit{Panel B: pooled regressions of the forward return on the SPID project signal, the composite and their post-launch interactions}} \\ | |
| 15 | +SPID technology projects ($z$) & +0.40 & [1.21] & [0.60] & [-0.25, +1.04] \\ | |
| 16 | +SPID technology projects ($z$) $\times$ post & -1.93 & [-1.93] & [-0.96] & [-3.88, +0.03] \\ | |
| 17 | +AI composite ($z$) & +2.00 & [5.78] & [2.24] & [+1.32, +2.68] \\ | |
| 18 | +AI composite $\times$ post & +1.77 & [1.63] & [0.82] & [-0.36, +3.91] \\ | |
| 19 | +Post-launch cohorts & +9.54 & [9.91] & [0.85] & [+7.65, +11.43] \\ | |
| 20 | +\quad $N$ = 26{,}933; design rank 8 of 8; condition number 2.8 & & & & \\ | |
| 14 | 21 | \addlinespace |
| 15 | −Any AI-initiative project (0/1) & +0.73 & 0.18 & 0.07 & [-7.01, +8.46] \\ | |
| 16 | −Any AI-initiative project (0/1) $\times$ post & +4.20 & 0.62 & 0.32 & [-9.00, +17.39] \\ | |
| 17 | −AI composite ($z$) & +2.02 & 5.89 & 2.31 & [+1.34, +2.69] \\ | |
| 18 | −AI composite $\times$ post & +1.49 & 1.36 & 0.69 & [-0.65, +3.63] \\ | |
| 19 | −Post-launch cohorts & +9.34 & 9.59 & 0.83 & [+7.43, +11.25] \\ | |
| 20 | −\quad $N$ = 26{,}933; design rank 8, condition number 18.5 \\ | |
| 22 | +Any AI-initiative project (0/1) & +0.73 & [0.18] & [0.07] & [-7.01, +8.46] \\ | |
| 23 | +Any AI-initiative project (0/1) $\times$ post & +4.20 & [0.62] & [0.32] & [-9.00, +17.39] \\ | |
| 24 | +AI composite ($z$) & +2.02 & [5.89] & [2.31] & [+1.34, +2.69] \\ | |
| 25 | +AI composite $\times$ post & +1.49 & [1.36] & [0.69] & [-0.65, +3.63] \\ | |
| 26 | +Post-launch cohorts & +9.34 & [9.59] & [0.83] & [+7.43, +11.25] \\ | |
| 27 | +\quad $N$ = 26{,}933; design rank 8 of 8; condition number 18.5 & & & & \\ | |
| 21 | 28 | \addlinespace |
| 22 | −Any technology project (0/1) & +1.41 & 1.76 & 0.62 & [-0.16, +2.98] \\ | |
| 23 | −Any technology project (0/1) $\times$ post & -4.52 & -2.02 & -0.89 & [-8.90, -0.14] \\ | |
| 24 | −AI composite ($z$) & +1.99 & 5.74 & 2.22 & [+1.31, +2.67] \\ | |
| 25 | −AI composite $\times$ post & +1.75 & 1.61 & 0.82 & [-0.38, +3.88] \\ | |
| 26 | −Post-launch cohorts & +10.70 & 9.49 & 0.91 & [+8.49, +12.91] \\ | |
| 27 | −\quad $N$ = 26{,}933; design rank 8, condition number 6.2 \\ | |
| 29 | +Any technology project (0/1) & +1.41 & [1.76] & [0.62] & [-0.16, +2.98] \\ | |
| 30 | +Any technology project (0/1) $\times$ post & -4.52 & [-2.02] & [-0.89] & [-8.90, -0.14] \\ | |
| 31 | +AI composite ($z$) & +1.99 & [5.74] & [2.22] & [+1.31, +2.67] \\ | |
| 32 | +AI composite $\times$ post & +1.75 & [1.61] & [0.82] & [-0.38, +3.88] \\ | |
| 33 | +Post-launch cohorts & +10.70 & [9.49] & [0.91] & [+8.49, +12.91] \\ | |
| 34 | +\quad $N$ = 26{,}933; design rank 8 of 8; condition number 6.2 & & & & \\ | |
| 28 | 35 | \addlinespace |
| 29 | 36 | \bottomrule |
| 30 | 37 | \end{tabular} |
| 31 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Explicit-terms formula (signal, post, signal $\times$ post, composite, composite $\times$ post, size, momentum), rebuilt panel, dependent variable winsorized within year. The committed exhibit (HEAD 3cf7560) printed $t$ = 0.00 throughout from a degenerate fit; the refit of the committed panel (L1) gave SPID $\times$ post $-1.57$ pp ($t$ $-1.72$ firm / $-0.88$ two-way).} | |
| \ No newline at end of file | ||
| 38 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 39 | +\item \emph{Notes:} The committed v2.0 exhibit printed $t$ = 0.00 in every row of Panel B from a rank-deficient fit; the refit of the committed panel (design rank 8 of 8, condition number 2.82, $N$ = 25{,}135) gives SPID $\times$ post -1.57 pp [$t$ -1.72 firm / -0.88 two-way], and the rebuilt panel above gives -1.93 [-1.93 / -0.96]. The parallel-trends condition of Panel A is not satisfied (Table~\ref{tab:eventtime}), so the DiD coefficients are descriptive, not causal. The SPID measure is sparse (any-AI prevalence 2.0\% of firm-years) and unvalidated; its minimum detectable effects are in Table~\ref{tab:spidbinary}. Source: results/r3k\_v2/results2.json (did, spid), t17\_refit.json. | |
| 40 | +\end{tablenotes} | |
| 41 | +\end{threeparttable}} | |
modified
paper/tables/tab_doublesort.tex
+21 −13
@@ -1,22 +1,30 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results2.json (double_sort_*) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: appendix. \input{tables/tab_doublesort} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:doublesort}{\textbf{AI language $\times$ size double sort, Russell~3000.} Mean forward twelve-month return (percent) by tercile of the AI composite within tercile of the corrected market capitalization (both formed within cohort on ranks), for the pre- and post-launch cohorts and for the within-year winsorized and raw returns, with the High$-$Low difference and the number of firm-years per size tercile.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 1 | 5 | \begin{tabular}{lllrrrrr} |
| 2 | 6 | \toprule |
| 3 | −Period & Returns & Size tercile & Low AI & Mid & High AI & High$-$Low & $N$ \\ | |
| 7 | +Period & Returns & Size tercile & Low AI & Mid & High AI & High$-$Low & Firm-years \\ | |
| 4 | 8 | \midrule |
| 5 | −Pre-ChatGPT (2010--2021) & raw & Small & 20.1 & 29.4 & 29.6 & +9.5 & 6{,}727 \\ | |
| 6 | − & & Mid & 13.4 & 12.5 & 16.6 & +3.2 & 6{,}721 \\ | |
| 7 | − & & Large & 11.6 & 13.4 & 14.7 & +3.1 & 6{,}729 \\ | |
| 8 | −\midrule | |
| 9 | − & wins. & Small & 18.2 & 22.3 & 22.4 & +4.2 & 6{,}727 \\ | |
| 9 | +Pre-ChatGPT (2010--2021) & wins. & Small & 18.2 & 22.3 & 22.4 & +4.2 & 6{,}727 \\ | |
| 10 | 10 | & & Mid & 13.2 & 12.0 & 16.3 & +3.2 & 6{,}721 \\ |
| 11 | 11 | & & Large & 11.6 & 13.3 & 14.7 & +3.1 & 6{,}729 \\ |
| 12 | −\midrule | |
| 13 | −Post-ChatGPT (2022--2024) & raw & Small & 48.3 & 37.8 & 63.4 & +15.1 & 2{,}290 \\ | |
| 14 | − & & Mid & 18.6 & 14.2 & 23.6 & +5.0 & 2{,}288 \\ | |
| 15 | − & & Large & 12.6 & 14.0 & 26.1 & +13.5 & 2{,}291 \\ | |
| 16 | −\midrule | |
| 17 | − & wins. & Small & 40.8 & 36.3 & 46.8 & +6.1 & 2{,}290 \\ | |
| 12 | +\addlinespace | |
| 13 | + & raw & Small & 20.1 & 29.4 & 29.6 & +9.5 & 6{,}727 \\ | |
| 14 | + & & Mid & 13.4 & 12.5 & 16.6 & +3.2 & 6{,}721 \\ | |
| 15 | + & & Large & 11.6 & 13.4 & 14.7 & +3.1 & 6{,}729 \\ | |
| 16 | +\addlinespace | |
| 17 | +Post-ChatGPT (2022--2024) & wins. & Small & 40.8 & 36.3 & 46.8 & +6.1 & 2{,}290 \\ | |
| 18 | 18 | & & Mid & 18.3 & 14.2 & 22.4 & +4.1 & 2{,}288 \\ |
| 19 | 19 | & & Large & 12.6 & 14.1 & 25.1 & +12.6 & 2{,}291 \\ |
| 20 | +\addlinespace | |
| 21 | + & raw & Small & 48.3 & 37.8 & 63.4 & +15.1 & 2{,}290 \\ | |
| 22 | + & & Mid & 18.6 & 14.2 & 23.6 & +5.0 & 2{,}288 \\ | |
| 23 | + & & Large & 12.6 & 14.0 & 26.1 & +13.5 & 2{,}291 \\ | |
| 24 | +\addlinespace | |
| 20 | 25 | \bottomrule |
| 21 | 26 | \end{tabular} |
| 22 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Size terciles on the corrected capitalization within cohort (rank-based), AI terciles within size tercile; mean forward return in \%.} | |
| \ No newline at end of file | ||
| 27 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 28 | +\item \emph{Notes:} Descriptive cell means without standard errors; the formal large-versus-small test with clustered inference and MDE is in Table~\ref{tab:contrasts}, and the post-launch cells average only three cohorts. Firm-years without a capitalization are excluded. Source: results/r3k\_v2/results2.json (double\_sort\_*). | |
| 29 | +\end{tablenotes} | |
| 30 | +\end{threeparttable}} | |
modified
paper/tables/tab_event.tex
+20 −14
@@ -1,25 +1,31 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results_event.json (by_benchmark.mm_2022, welch_reference, tercile_characteristics) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_event} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:event}{\textbf{ChatGPT event study: cumulative abnormal returns by pre-launch AI tercile under the IWV market model, Russell~3000.} Market-model cumulative abnormal returns (percent) of the high and low terciles of 2021-cohort AI exposure (2{,}215 firms; terciles on ranks), the High$-$Low difference and three $t$-statistics: CDA = the dependence-robust crude-dependence-adjustment statistic of the daily High$-$Low portfolio (CAR divided by the estimation-window SD of the portfolio $\times \sqrt{\text{days}}$), event-SD = the same with the event-window SD, BMP = Boehmer--Musumeci--Poulsen standardized-CAR difference, which treats firms as independent. Market model estimated on the iShares Russell~3000 ETF (IWV) over 2022-01-03..2022-10-31 (209 days); the pre-launch block is reported separately from the event window.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 1 | 5 | \begin{tabular}{lrrrrrrr} |
| 2 | 6 | \toprule |
| 3 | −Horizon (day 0 = 1 Dec 2022) & Days & High & Low & High$-$Low & CDA $t$ & event-SD $t$ & BMP $t$ (H$-$L) \\ | |
| 7 | +Horizon & Days & High (\%) & Low (\%) & High$-$Low (pp) & CDA $t$ & event-SD $t$ & BMP $t$ \\ | |
| 4 | 8 | \midrule |
| 9 | +\multicolumn{8}{@{}l}{\textit{Panel A: full sample by horizon (day 0 = 1 December 2022)}} \\ | |
| 5 | 10 | Pre-launch block [$-$20, $-$1] (1--29 Nov 2022) & 20 & -2.34 & -2.06 & -0.28 & -0.16 & -0.17 & -0.71 \\ |
| 6 | −Days {}[$-$1, $+$1] & 3 & +0.37 & -0.27 & +0.65 & 0.93 & 0.98 & 3.91 \\ | |
| 7 | −Days {}[0, $+$5] & 6 & +0.41 & +0.01 & +0.41 & 0.41 & 0.44 & 2.66 \\ | |
| 8 | −Days {}[0, $+$20] & 21 & +1.50 & +0.54 & +0.96 & 0.52 & 0.55 & 2.53 \\ | |
| 11 | +Days [$-$1, $+$1] & 3 & +0.37 & -0.27 & +0.65 & 0.93 & 0.98 & 3.91 \\ | |
| 12 | +Days [0, $+$5] & 6 & +0.41 & +0.01 & +0.41 & 0.41 & 0.44 & 2.66 \\ | |
| 13 | +Days [0, $+$20] & 21 & +1.50 & +0.54 & +0.96 & 0.52 & 0.55 & 2.53 \\ | |
| 9 | 14 | 60 trading days & 60 & +7.08 & +3.65 & +3.43 & 1.10 & 1.16 & 1.78 \\ |
| 10 | 15 | 125 days (6 months) & 125 & -0.45 & -13.36 & +12.91 & 2.87 & 3.02 & 5.00 \\ |
| 11 | 16 | 250 days (1 year) & 250 & -3.91 & -18.10 & +14.20 & 2.23 & 2.35 & 5.17 \\ |
| 12 | 17 | 395 days (18 months) & 395 & -12.04 & -21.52 & +9.49 & 1.19 & 1.25 & 3.16 \\ |
| 13 | 18 | \midrule |
| 14 | −\multicolumn{8}{l}{\textit{Sub-samples, 1-year H$-$L CAR (CDA $t$; BMP $t$)}} \\ | |
| 15 | −Largest 1,000 (corrected cap) ($n$ = 393+290) & 250 & -5.54 & -25.57 & +20.03 & 3.02 & 3.53 & 5.34 \\ | |
| 16 | −Remaining firms ($n$ = 341+440) & 250 & -1.53 & -13.76 & +12.23 & 1.51 & 1.79 & 3.09 \\ | |
| 17 | −Largest 1,000, terciles within group ($n$ = 333+331) & 250 & -4.52 & -25.21 & +20.70 & 2.97 & 3.49 & 5.33 \\ | |
| 18 | −Remaining, terciles within group ($n$ = 398+399) & 250 & -2.82 & -14.47 & +11.66 & 1.48 & 1.72 & 3.36 \\ | |
| 19 | −Largest 1,000 (author's cap) ($n$ = 390+281) & 250 & -8.24 & -27.47 & +19.24 & 3.30 & 3.57 & 4.94 \\ | |
| 20 | −Remaining (author's cap) ($n$ = 275+385) & 250 & -2.51 & -12.99 & +10.48 & 1.33 & 1.34 & 2.84 \\ | |
| 21 | −S\&P 500 members ($n$ = 212+111) & 250 & -3.50 & -15.56 & +12.06 & 1.96 & 2.24 & 2.51 \\ | |
| 22 | −Non-members ($n$ = 529+627) & 250 & -4.07 & -18.55 & +14.48 & 1.85 & 2.25 & 4.42 \\ | |
| 19 | +\multicolumn{8}{@{}l}{\textit{Panel B: sub-samples, one-year (250-day) horizon}} \\ | |
| 20 | +Largest 1,000 (corrected cap) ($n$ = 393 + 290) & 250 & -5.54 & -25.57 & +20.03 & 3.02 & 3.53 & 5.34 \\ | |
| 21 | +Remaining firms ($n$ = 341 + 440) & 250 & -1.53 & -13.76 & +12.23 & 1.51 & 1.79 & 3.09 \\ | |
| 22 | +Largest 1,000, terciles within group ($n$ = 333 + 331) & 250 & -4.52 & -25.21 & +20.70 & 2.97 & 3.49 & 5.33 \\ | |
| 23 | +Remaining, terciles within group ($n$ = 398 + 399) & 250 & -2.82 & -14.47 & +11.66 & 1.48 & 1.72 & 3.36 \\ | |
| 24 | +S\&P 500 members ($n$ = 212 + 111) & 250 & -3.50 & -15.56 & +12.06 & 1.96 & 2.24 & 2.51 \\ | |
| 25 | +Non-members ($n$ = 529 + 627) & 250 & -4.07 & -18.55 & +14.48 & 1.85 & 2.25 & 4.42 \\ | |
| 23 | 26 | \bottomrule |
| 24 | 27 | \end{tabular} |
| 25 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} IWV market model estimated 3 Jan--31 Oct 2022; 2{,}215 firms with 2021-cohort exposure; terciles on rank. CDA $t$ = H$-$L daily-portfolio CAR divided by the estimation-window SD of the H$-$L portfolio $\times \sqrt{\text{days}}$ (dependence-robust); event-SD $t$ uses the event-window SD; BMP = Boehmer--Musumeci--Poulsen standardized-CAR difference (independence-based; the Kolari--Pynn\"onen factor 0.131 for $\bar\rho$ = 0.0374 would deflate it further, but the common component nets out of the H$-$L portfolio, which is why the portfolio CDA and the KP-deflated firm-level $t$ differ). Welch $t$ (independence-based, author's statistic) for reference: full 3.92, largest 1,000 5.29, remaining 2.04, S\&P 1.96, non-S\&P 3.29. Tercile characteristics: Low: $\beta$ 0.99, log cap 21.47, share large 39\%; Mid: $\beta$ 1.08, log cap 21.76, share large 43\%; High: $\beta$ 1.17, log cap 22.09, share large 53\%.} | |
| \ No newline at end of file | ||
| 28 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 29 | +\item \emph{Notes:} The CDA $t$ is the statistic the text relies on; the BMP $t$ is independence-based and overstates significance under a common event date (the Kolari--Pynn\"onen factor for $\bar\rho$ = 0.0374 is 0.131; it deflates the firm-level $t$, whereas the common component nets out of the High$-$Low portfolio, which is why the two differ). The Welch $t$ reported in v2.0 (independence-based) was: full 3.92, largest 1,000 5.29, remaining 2.04, S\&P 1.96, non-S\&P 3.29. Tercile characteristics (estimation window): Low $\beta$ 0.99, log cap 21.47, share large 39\%; High $\beta$ 1.17, log cap 22.09, share large 53\% --- the terciles differ in beta and size, so Table~\ref{tab:eventcda} re-estimates the one-year gradient under five benchmarks. Levels drift negative because the model was estimated in the 2022 bear market; only differences between terciles are interpretable. Source: results/r3k\_v2/results\_event.json. | |
| 30 | +\end{tablenotes} | |
| 31 | +\end{threeparttable}} | |
modified
paper/tables/tab_event_cda.tex
+26 −8
@@ -1,26 +1,44 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results_event.json (by_benchmark, ff6_calendar_time, placebo_launches, placebo_summary, mht_event) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: main. \input{tables/tab_event_cda} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:eventcda}{\textbf{Benchmark dependence of the one-year event-study gradient and placebo launch dates.} Panel A: the one-year High$-$Low abnormal return of the pre-launch AI terciles and its dependence-robust CDA $t$ under six benchmarks: the IWV market model of Table~\ref{tab:event}, a beta-only adjustment, the market-adjusted return, a market model estimated over calendar 2021, a size $\times$ momentum matched benchmark (5 $\times$ 5 cells of the 2021-cohort panel) and a Fama--French five-factor-plus-momentum calendar-time regression of the daily High$-$Low portfolio (alpha $\times$ 250 days, HAC $t$). Panel B: the same design re-run with a fictitious launch on 30 November of 2015--2021. Every cell is dependence-robust (daily High$-$Low portfolio).}{\footnotesize}{% | |
| 4 | +\begin{threeparttable}\setlength{\tabcolsep}{4pt} | |
| 1 | 5 | \begin{tabular}{llrrrrrr} |
| 2 | 6 | \toprule |
| 3 | 7 | Benchmark & Statistic & Full & Largest 1,000 & Remaining & S\&P 500 & Non-S\&P & Within-size large / small \\ |
| 4 | 8 | \midrule |
| 5 | −(a) IWV market model, est. Jan--Oct 2022 & 1-y H$-$L (pp) & +14.20 & +20.03 & +12.23 & +12.06 & +14.48 & 20.70 / 11.66 \\ | |
| 9 | +\multicolumn{8}{@{}l}{\textit{Panel A: one-year High$-$Low abnormal return (pp) and CDA $t$ by benchmark and sub-sample}} \\ | |
| 10 | +(a) IWV market model, est.\ Jan--Oct 2022 & H$-$L (pp) & +14.20 & +20.03 & +12.23 & +12.06 & +14.48 & 20.70 / 11.66 \\ | |
| 6 | 11 | & CDA $t$ & 2.23 & 3.02 & 1.51 & 1.96 & 1.85 & 2.97 / 1.48 \\ |
| 7 | 12 | \addlinespace |
| 8 | −(b) Beta-only, AR $= r - \hat\beta m$ & 1-y H$-$L (pp) & +8.63 & +7.40 & +11.61 & +7.39 & +8.53 & 8.62 / 11.05 \\ | |
| 13 | +(b) Beta-only, $r - \hat\beta m$ & H$-$L (pp) & +8.63 & +7.40 & +11.61 & +7.39 & +8.53 & 8.62 / 11.05 \\ | |
| 9 | 14 | & CDA $t$ & 1.36 & 1.12 & 1.43 & 1.20 & 1.09 & 1.24 / 1.41 \\ |
| 10 | 15 | \addlinespace |
| 11 | −(c) Market-adjusted, $r - m$ & 1-y H$-$L (pp) & +10.84 & +9.33 & +14.29 & +8.58 & +11.32 & 10.82 / 13.87 \\ | |
| 16 | +(c) Market-adjusted, $r - m$ & H$-$L (pp) & +10.84 & +9.33 & +14.29 & +8.58 & +11.32 & 10.82 / 13.87 \\ | |
| 12 | 17 | & CDA $t$ & 1.40 & 1.22 & 1.47 & 1.30 & 1.18 & 1.31 / 1.43 \\ |
| 13 | 18 | \addlinespace |
| 14 | −(d) Market model estimated over 2021 & 1-y H$-$L (pp) & +20.12 & +15.69 & +30.21 & +9.21 & +24.16 & 16.32 / 28.98 \\ | |
| 19 | +(d) Market model estimated over 2021 & H$-$L (pp) & +20.12 & +15.69 & +30.21 & +9.21 & +24.16 & 16.32 / 28.98 \\ | |
| 15 | 20 | & CDA $t$ & 2.90 & 2.09 & 4.10 & 1.54 & 3.23 & 2.14 / 4.05 \\ |
| 16 | 21 | \addlinespace |
| 17 | −(f) Size $\times$ momentum matched & 1-y H$-$L (pp) & +10.72 & +7.93 & +13.60 & +8.02 & +10.58 & 9.31 / 13.19 \\ | |
| 22 | +(e) Size $\times$ momentum matched (5 $\times$ 5) & H$-$L (pp) & +10.72 & +7.93 & +13.60 & +8.02 & +10.58 & 9.31 / 13.19 \\ | |
| 18 | 23 | & CDA $t$ & 1.81 & 1.38 & 1.89 & 1.30 & 1.59 & 1.56 / 1.89 \\ |
| 19 | 24 | \addlinespace |
| 20 | −(e) FF5+MOM calendar-time, 250 days & $\alpha \times$ days (pp) & +4.89 & +2.21 & +10.94 & +2.03 & +6.57 & \\ | |
| 25 | +(f) FF5 + MOM calendar-time, 250 days & $\alpha \times$ days (pp) & +4.89 & +2.21 & +10.94 & +2.03 & +6.57 & \\ | |
| 21 | 26 | & HAC $t$ & 1.20 & 0.52 & 1.86 & 0.50 & 1.37 & \\ |
| 22 | 27 | \midrule |
| 23 | −\multicolumn{8}{l}{Placebo launches (30 Nov of each year, same design): 1-year H$-$L (pp) / CDA $t$: 2015: +5.8 / 1.61; 2016: -2.4 / -0.70; 2017: -5.1 / -1.11; 2018: -3.2 / -0.65; 2019: +7.3 / 1.59; 2020: -13.2 / -1.06; 2021: +5.5 / 0.78; 2022: +14.2 / 2.23. 2022 ranks 1 of 8 by H$-$L and 1 by $t$.} \\ | |
| 28 | +\multicolumn{8}{@{}l}{\textit{Panel B: placebo launch dates (30 November of each year, same design, benchmark (a))}} \\ | |
| 29 | +Launch year & Firms & \multicolumn{2}{l}{Full: H$-$L (pp) / CDA $t$} & \multicolumn{2}{l}{Largest 1,000: H$-$L / CDA $t$} & \multicolumn{2}{l}{Remaining: H$-$L / CDA $t$} \\ | |
| 30 | +\midrule | |
| 31 | +2015 & 1{,}571 & \multicolumn{2}{l}{+5.8 / 1.61} & \multicolumn{2}{l}{+2.3 / 0.64} & \multicolumn{2}{l}{+12.7 / 2.16} \\ | |
| 32 | +2016 & 1{,}641 & \multicolumn{2}{l}{-2.4 / -0.70} & \multicolumn{2}{l}{-2.3 / -0.62} & \multicolumn{2}{l}{-1.8 / -0.33} \\ | |
| 33 | +2017 & 1{,}701 & \multicolumn{2}{l}{-5.1 / -1.11} & \multicolumn{2}{l}{-7.8 / -1.77} & \multicolumn{2}{l}{+0.7 / 0.11} \\ | |
| 34 | +2018 & 1{,}770 & \multicolumn{2}{l}{-3.2 / -0.65} & \multicolumn{2}{l}{-9.1 / -1.76} & \multicolumn{2}{l}{+3.9 / 0.62} \\ | |
| 35 | +2019 & 1{,}849 & \multicolumn{2}{l}{+7.3 / 1.59} & \multicolumn{2}{l}{+5.8 / 1.34} & \multicolumn{2}{l}{+9.9 / 1.58} \\ | |
| 36 | +2020 & 1{,}931 & \multicolumn{2}{l}{-13.2 / -1.06} & \multicolumn{2}{l}{-14.1 / -1.21} & \multicolumn{2}{l}{-9.8 / -0.74} \\ | |
| 37 | +2021 & 2{,}056 & \multicolumn{2}{l}{+5.5 / 0.78} & \multicolumn{2}{l}{-5.5 / -0.74} & \multicolumn{2}{l}{+18.0 / 2.64} \\ | |
| 38 | +2022 (ChatGPT) & 2{,}215 & \multicolumn{2}{l}{+14.2 / 2.23} & \multicolumn{2}{l}{+20.0 / 3.02} & \multicolumn{2}{l}{+12.2 / 1.51} \\ | |
| 24 | 39 | \bottomrule |
| 25 | 40 | \end{tabular} |
| 26 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Every cell is dependence-robust (H$-$L daily portfolio). Holm-adjusted $p$ over the 35 sub-sample $\times$ horizon CDA statistics of benchmark (a): smallest 0.088; BH smallest 0.036. Size $\times$ momentum matching uses 5 $\times$ 5 cells of the 2021-cohort panel (B/M not available).} | |
| \ No newline at end of file | ||
| 41 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 42 | +\item \emph{Notes:} The largest-1,000 gradient is significant only under the market models (a) and (d); under (b), (c), (e) and (f) its CDA $t$ is below 2, so the event-study evidence for a large-cap re-rating is benchmark-dependent and is reported as such. Holm-adjusted $p$ over the 35 sub-sample $\times$ horizon CDA statistics of benchmark (a): smallest 0.088 (BH: 0.036). The 2022 launch ranks 1 of 8 by full-sample H$-$L and 1 by CDA $t$ among the eight launch dates; the placebo $t$'s range over -1.11 to 1.61. A book-to-market match was not run (book equity not in the panel). Source: results/r3k\_v2/results\_event.json. | |
| 43 | +\end{tablenotes} | |
| 44 | +\end{threeparttable}} | |
modified
paper/tables/tab_eventtime.tex
+11 −3
@@ -1,6 +1,11 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results2.json (did_eventtime*) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: appendix. \input{tables/tab_eventtime} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:eventtime}{\textbf{Event-time profile of the premium on pre-determined AI exposure, Russell~3000.} Slope of the forward return (within-year winsorized, pp per SD) on pre-determined AI exposure by outcome cohort, with firm-clustered standard errors in parentheses. The first pair of columns reproduces the committed coding (interaction without a main effect, which estimates one absolute slope per cohort); the second pair restores the main effect so that the coefficients are differences from the 2021 reference cohort (zero by construction); the third pair replaces the 2019--21 exposure by the mean composite over the three years before each outcome year.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 1 | 5 | \begin{tabular}{lrrrrrr} |
| 2 | 6 | \toprule |
| 3 | −Outcome cohort & Author coding: absolute slope (pp/SD) & (SE) & Difference from 2021 (pp/SD) & (SE) & Trailing exposure ($Y{-}3$..$Y{-}1$): difference from 2021 & (SE) \\ | |
| 7 | +Outcome cohort & \multicolumn{2}{c}{Committed coding: absolute slope} & \multicolumn{2}{c}{Difference from reference} & \multicolumn{2}{c}{Trailing exposure: difference} \\ | |
| 8 | +\cmidrule(lr){2-3}\cmidrule(lr){4-5}\cmidrule(lr){6-7} & (pp/SD) & (SE) & (pp/SD) & (SE) & (pp/SD) & (SE) \\ | |
| 4 | 9 | \midrule |
| 5 | 10 | 2014 & +3.1 & (0.8) & -2.8 & (1.3) & -5.4 & (1.3) \\ |
| 6 | 11 | 2015 & +3.4 & (0.9) & -2.5 & (1.3) & -4.1 & (1.4) \\ |
@@ -9,10 +14,13 @@ Outcome cohort & Author coding: absolute slope (pp/SD) & (SE) & Difference from | ||
| 9 | 14 | 2018 & +7.7 & (1.0) & +1.8 & (1.3) & +1.5 & (1.3) \\ |
| 10 | 15 | 2019 & +0.2 & (2.0) & -5.8 & (2.1) & -4.0 & (2.2) \\ |
| 11 | 16 | 2020 & -0.9 & (0.7) & -6.8 & (1.2) & -7.5 & (1.2) \\ |
| 12 | −2021 & +5.9 & (1.0) & +0.0 & (0.0) & +0.0 & (0.0) \\ | |
| 17 | +2021 (reference) & +5.9 & (1.0) & +0.0 & (0.0) & +0.0 & (0.0) \\ | |
| 13 | 18 | 2022 & +0.4 & (1.0) & -5.6 & (1.2) & -5.5 & (1.3) \\ |
| 14 | 19 | 2023 & +4.0 & (1.2) & -1.9 & (1.6) & -2.2 & (1.6) \\ |
| 15 | 20 | 2024 & +7.6 & (2.4) & +1.7 & (2.7) & +1.7 & (2.7) \\ |
| 16 | 21 | \bottomrule |
| 17 | 22 | \end{tabular} |
| 18 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Firm-clustered SEs. Author's coding (no main effect) estimates one absolute slope per cohort; with the main effect restored (2021 slope +5.9 pp/SD, $t$ 5.95) the coefficients are differences from 2021. With the 2019--21 exposure, rows before 2019 interact future exposure with past outcomes (look-back, not pre-trends); the trailing-exposure column uses the mean composite over the three years before each outcome year. $N$ = 21{,}111.} | |
| \ No newline at end of file | ||
| 23 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 24 | +\item \emph{Notes:} Reference-cohort slope with the main effect restored: +5.9 pp/SD ($t$ 5.95); $N$ = 21{,}111 (trailing: 20{,}600). With the 2019--21 exposure, cohorts before 2019 interact future language with past outcomes (look-back selection, not a pre-trend test). The pre-launch differences are not zero, so the parallel-trends condition behind Table~\ref{tab:didspid} Panel A fails and the post-launch coefficients are relative to a reference year that itself carried the largest slope. Plotted with 95\% CIs in Figure~\ref{fig:eventtime}. Source: results/r3k\_v2/results2.json (did\_eventtime, did\_eventtime\_trailing). | |
| 25 | +\end{tablenotes} | |
| 26 | +\end{threeparttable}} | |
modified
paper/tables/tab_filing_type.tex
+23 −9
@@ -1,15 +1,29 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results2.json (filing_type, mht.forms_fm_w) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: appendix. \input{tables/tab_filing_type} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:filingtype}{\textbf{Form-specific AI composites, Russell~3000.} Panel A: the composite rebuilt from the filing-level scores of one form at a time: share of estimation firm-years with at least one scored filing of that form, correlation with the baseline composite, equal-weighted Q5$-$Q1 spreads (within-year winsorized, pp per year; full sample with 95\% CI over cohorts, pre and post), and the Fama--MacBeth slope with controls (pp per SD; [NW $t$; plain $t$]). Panel B: the three form composites entered jointly, with a HAC Wald test of equal slopes. Panel C: how much of the 2022 cohort's text post-dates the launch and whether it changes the cohort spread.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable}\setlength{\tabcolsep}{4pt} | |
| 1 | 5 | \begin{tabular}{lrrrrrrr} |
| 2 | 6 | \toprule |
| 3 | −Signal built from & Firm-years covered & Corr. w/ baseline & Full Q5$-$Q1 (wins.) & Pre (wins.) & Post (wins.) & ($n$, NW $t$) & FM slope wins. (NW $t$ / plain $t$) \\ | |
| 7 | +Signal built from & Firm-years covered & Corr.\ with baseline & Full Q5$-$Q1 & 95\% CI & Pre & Post & FM slope (pp/SD) \\ | |
| 4 | 8 | \midrule |
| 5 | −Form 10-K only & 96.8\% & 0.62 & +6.2 & +4.9 & +11.6 & (3, 2.06) & +2.67 (2.09 / 2.36) \\ | |
| 6 | −Form 10-Q only & 99.6\% & 0.61 & +5.6 & +4.6 & +9.4 & (3, 1.58) & +2.14 (1.95 / 2.01) \\ | |
| 7 | −Form 8-K only & 99.9\% & 0.87 & +2.7 & +2.1 & +4.9 & (3, 2.43) & +1.34 (2.37 / 2.48) \\ | |
| 8 | −Equal-form composite (mean of three $z$) & 96.7\% & 0.86 & +5.9 & +4.3 & +12.2 & (3, 1.85) & +2.42 (2.18 / 2.33) \\ | |
| 9 | +\multicolumn{8}{@{}l}{\textit{Panel A: one form at a time}} \\ | |
| 10 | +Form 10-K only & 96.8\% & 0.62 & +6.2 & [-0.8, +13.2] & +4.9 & +11.6 & +2.67 [2.09; 2.36] \\ | |
| 11 | +Form 10-Q only & 99.6\% & 0.61 & +5.6 & [-0.9, +12.0] & +4.6 & +9.4 & +2.14 [1.95; 2.01] \\ | |
| 12 | +Form 8-K only & 99.9\% & 0.87 & +2.7 & [-0.1, +5.5] & +2.1 & +4.9 & +1.34 [2.37; 2.48] \\ | |
| 13 | +Equal-form composite (mean of three $z$) & 96.7\% & 0.86 & +5.9 & [-0.8, +12.6] & +4.3 & +12.2 & +2.42 [2.18; 2.33] \\ | |
| 9 | 14 | \midrule |
| 10 | −Horse race (FM, all three, wins.): & \multicolumn{7}{l}{10-K: +2.78 (2.23); 10-Q: -0.32 (-0.87); 8-K: +0.55 (1.29); HAC Wald test of equal slopes $\chi^2_2$ = 8.25, $p$ = 0.02} \\ | |
| 11 | −Horse race (raw): & \multicolumn{7}{l}{10-K: +3.29 (1.95); 10-Q: -0.07 (-0.11); 8-K: +1.25 (2.26); Wald $\chi^2_2$ = 2.90, $p$ = 0.23} \\ | |
| 12 | −2022 cohort text filed after 30 Nov 2022: & \multicolumn{7}{l}{4.6\% of filings, 1.8\% of chunks; 2022 spread on pre-launch filings only: -5.2 pp (wins.) vs -5.0 on the full-year text, same 2{,}265 firms; rank correlation of the two signals 0.99} \\ | |
| 15 | +\multicolumn{8}{@{}l}{\textit{Panel B: horse race, the three form composites entered jointly (Fama--MacBeth, pp per SD [NW $t$])}} \\ | |
| 16 | +Returns & 10-K & 10-Q & 8-K & \multicolumn{4}{l}{Wald $\chi^2_2$ ($p$), equal slopes} \\ | |
| 17 | +\midrule | |
| 18 | +wins. & +2.78 [2.23] & -0.32 [-0.87] & +0.55 [1.29] & \multicolumn{4}{l}{8.25 (0.016)} \\ | |
| 19 | +raw & +3.29 [1.95] & -0.07 [-0.11] & +1.25 [2.26] & \multicolumn{4}{l}{2.90 (0.234)} \\ | |
| 20 | +\midrule | |
| 21 | +\multicolumn{8}{@{}l}{\textit{Panel C: text of the 2022 cohort filed after the launch (30 November 2022)}} \\ | |
| 22 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.3\textwidth}}{Share of the cohort's filings / chunks dated after the launch} & \multicolumn{7}{l}{4.6\% / 1.8\%} \\ | |
| 23 | +\multicolumn{1}{>{\raggedright\arraybackslash}p{0.3\textwidth}}{Cohort spread, pre-launch filings only vs full-year text (2{,}265 firms, wins., pp)} & \multicolumn{7}{l}{-5.2 vs -5.0; rank correlation of the two signals 0.99} \\ | |
| 13 | 24 | \bottomrule |
| 14 | 25 | \end{tabular} |
| 15 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Form-specific composites = $\frac12[z(\text{adoption}) + z(\text{infrastructure})]$ from the filings of that form only; the equal-form composite averages the three form $z$-scores. Romano--Wolf $p$ over the three form slopes: 10k: 0.05, 10q: 0.07, 8k: 0.05. Equation (3) aggregation reproduced from filing-level scores for 100.00\% of firm-years.} | |
| \ No newline at end of file | ||
| 26 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 27 | +\item \emph{Notes:} Form composites = $\tfrac12[z(\text{adoption}) + z(\text{infrastructure})]$ from the filings of that form only; the equal-form composite averages the three form $z$-scores. The Wald test rejects equality at 5\% on winsorized returns but not on raw returns (wins.\ $p$ = 0.016, raw 0.234); Romano--Wolf $p$ over the three single-form slopes: 10-K: 0.05, 10-Q: 0.07, 8-K: 0.05. The claim that annual reports carry the strongest premium is therefore not statistically distinguishable from the other forms. Equation (3) aggregation reproduced from filing-level scores for 100.00\% of firm-years. Source: results/r3k\_v2/results2.json (filing\_type). | |
| 28 | +\end{tablenotes} | |
| 29 | +\end{threeparttable}} | |
modified
paper/tables/tab_fm.tex
+38 −31
@@ -1,51 +1,58 @@ | ||
| 1 | +% Generated by scripts/r3k_20_tables_v2.py from results/r3k_v2/results.json (fama_macbeth) -- do not edit by hand (FIRC-0002 v2.1, Stage E exhibits). | |
| 2 | +% Placement decided at Stage E: appendix. \input{tables/tab_fm} on its own prints the whole float; inside an existing float wrapper only the body is printed. | |
| 3 | +\ExhibitTable{tab:fm}{\textbf{Fama--MacBeth regressions, all specifications, Russell~3000 universe.} Companion to Table~\ref{tab:fmboth}: time-series mean of the 15 annual cross-sectional slopes (pp per SD, or pp per unit for the 0/1 indicator) of the forward return on each signal with size, momentum and volatility controls, for the within-year winsorized and the raw return.}{\footnotesize}{% | |
| 4 | +\begin{threeparttable} | |
| 1 | 5 | \begin{tabular}{llrrrr} |
| 2 | 6 | \toprule |
| 3 | −Regressor & Returns & Slope (pp/SD) & (NW $t$) & (plain $t$) & 95\% CI \\ | |
| 7 | +Regressor & Returns & Slope (pp/SD) & [NW $t$] & [plain $t$] & 95\% CI \\ | |
| 4 | 8 | \midrule |
| 5 | −AI composite & raw & +2.77 & (2.44) & (2.75) & [+0.61, +4.92] \\ | |
| 6 | − & wins. & +1.88 & (2.41) & (2.51) & [+0.27, +3.49] \\ | |
| 9 | +AI composite & wins. & +1.88 & [2.41] & [2.51] & [+0.27, +3.49] \\ | |
| 10 | + & raw & +2.77 & [2.44] & [2.75] & [+0.61, +4.92] \\ | |
| 7 | 11 | \addlinespace |
| 8 | −AI adoption (joint) & raw & -1.28 & (-1.13) & (-1.08) & [-3.83, +1.26] \\ | |
| 9 | − & wins. & +0.13 & (0.11) & (0.11) & [-2.36, +2.61] \\ | |
| 12 | +AI adoption (joint) & wins. & +0.13 & [0.11] & [0.11] & [-2.36, +2.61] \\ | |
| 13 | + & raw & -1.28 & [-1.13] & [-1.08] & [-3.83, +1.26] \\ | |
| 10 | 14 | \addlinespace |
| 11 | −AI infrastructure (joint) & raw & +1.95 & (1.37) & (1.54) & [-0.76, +4.67] \\ | |
| 12 | − & wins. & +0.99 & (1.09) & (1.32) & [-0.62, +2.61] \\ | |
| 15 | +AI infrastructure (joint) & wins. & +0.99 & [1.09] & [1.32] & [-0.62, +2.61] \\ | |
| 16 | + & raw & +1.95 & [1.37] & [1.54] & [-0.76, +4.67] \\ | |
| 13 | 17 | \addlinespace |
| 14 | −AI disruption risk (joint) & raw & +2.51 & (1.46) & (1.27) & [-1.71, +6.73] \\ | |
| 15 | − & wins. & +0.90 & (0.57) & (0.52) & [-2.81, +4.61] \\ | |
| 18 | +AI disruption risk (joint) & wins. & +0.90 & [0.57] & [0.52] & [-2.81, +4.61] \\ | |
| 19 | + & raw & +2.51 & [1.46] & [1.27] & [-1.71, +6.73] \\ | |
| 16 | 20 | \addlinespace |
| 17 | −AI disruption risk (alone) & raw & +2.79 & (2.48) & (2.39) & [+0.28, +5.30] \\ | |
| 18 | − & wins. & +1.69 & (1.93) & (1.69) & [-0.45, +3.83] \\ | |
| 21 | +AI disruption risk (alone) & wins. & +1.69 & [1.93] & [1.69] & [-0.45, +3.83] \\ | |
| 22 | + & raw & +2.79 & [2.48] & [2.39] & [+0.28, +5.30] \\ | |
| 19 | 23 | \addlinespace |
| 20 | −PC1 of three facets & raw & +2.80 & (2.54) & (2.72) & [+0.59, +5.00] \\ | |
| 21 | − & wins. & +1.83 & (2.33) & (2.24) & [+0.08, +3.57] \\ | |
| 24 | +PC1 of three facets & wins. & +1.83 & [2.33] & [2.24] & [+0.08, +3.57] \\ | |
| 25 | + & raw & +2.80 & [2.54] & [2.72] & [+0.59, +5.00] \\ | |
| 22 | 26 | \addlinespace |
| 23 | −Equal-form composite & raw & +3.50 & (2.38) & (2.58) & [+0.59, +6.41] \\ | |
| 24 | − & wins. & +2.42 & (2.18) & (2.33) & [+0.20, +4.64] \\ | |
| 27 | +Equal-form composite & wins. & +2.42 & [2.18] & [2.33] & [+0.20, +4.64] \\ | |
| 28 | + & raw & +3.50 & [2.38] & [2.58] & [+0.59, +6.41] \\ | |
| 25 | 29 | \addlinespace |
| 26 | −10-K composite & raw & +3.63 & (2.28) & (2.54) & [+0.57, +6.70] \\ | |
| 27 | − & wins. & +2.69 & (2.13) & (2.39) & [+0.27, +5.11] \\ | |
| 30 | +10-K composite & wins. & +2.69 & [2.13] & [2.39] & [+0.27, +5.11] \\ | |
| 31 | + & raw & +3.63 & [2.28] & [2.54] & [+0.57, +6.70] \\ | |
| 28 | 32 | \addlinespace |
| 29 | −SPID tech projects ($z$) & raw & +0.04 & (0.04) & (0.03) & [-2.58, +2.66] \\ | |
| 30 | − & wins. & -0.23 & (-0.46) & (-0.36) & [-1.63, +1.16] \\ | |
| 33 | +SPID tech projects ($z$) & wins. & -0.23 & [-0.46] & [-0.36] & [-1.63, +1.16] \\ | |
| 34 | + & raw & +0.04 & [0.04] & [0.03] & [-2.58, +2.66] \\ | |
| 31 | 35 | \addlinespace |
| 32 | −Any AI project (0/1, pp) & raw & +1.59 & (0.33) & (0.29) & [-10.35, +13.54] \\ | |
| 33 | − & wins. & +1.68 & (0.45) & (0.35) & [-8.50, +11.86] \\ | |
| 36 | +Any AI project (0/1, pp) & wins. & +1.68 & [0.45] & [0.35] & [-8.50, +11.86] \\ | |
| 37 | + & raw & +1.59 & [0.33] & [0.29] & [-10.35, +13.54] \\ | |
| 34 | 38 | \addlinespace |
| 35 | −Cybersecurity ($z$, with composite) & raw & -0.05 & (-0.06) & (-0.07) & [-1.61, +1.51] \\ | |
| 36 | − & wins. & +0.32 & (0.52) & (0.59) & [-0.84, +1.48] \\ | |
| 39 | +Cybersecurity ($z$, with composite) & wins. & +0.32 & [0.52] & [0.59] & [-0.84, +1.48] \\ | |
| 40 | + & raw & -0.05 & [-0.06] & [-0.07] & [-1.61, +1.51] \\ | |
| 37 | 41 | \addlinespace |
| 38 | −Size & raw & -5.55 & (-4.02) & (-3.26) & [-9.19, -1.90] \\ | |
| 39 | − & wins. & -3.45 & (-3.88) & (-2.90) & [-6.00, -0.90] \\ | |
| 42 | +Size & wins. & -3.45 & [-3.88] & [-2.90] & [-6.00, -0.90] \\ | |
| 43 | + & raw & -5.55 & [-4.02] & [-3.26] & [-9.19, -1.90] \\ | |
| 40 | 44 | \addlinespace |
| 41 | −Momentum & raw & -2.90 & (-1.94) & (-1.39) & [-7.39, +1.59] \\ | |
| 42 | − & wins. & -1.05 & (-0.89) & (-0.63) & [-4.65, +2.55] \\ | |
| 45 | +Momentum & wins. & -1.05 & [-0.89] & [-0.63] & [-4.65, +2.55] \\ | |
| 46 | + & raw & -2.90 & [-1.94] & [-1.39] & [-7.39, +1.59] \\ | |
| 43 | 47 | \addlinespace |
| 44 | −Volatility & raw & +8.06 & (2.62) & (2.22) & [+0.28, +15.85] \\ | |
| 45 | − & wins. & +4.60 & (1.79) & (1.58) & [-1.63, +10.83] \\ | |
| 48 | +Volatility & wins. & +4.60 & [1.79] & [1.58] & [-1.63, +10.83] \\ | |
| 49 | + & raw & +8.06 & [2.62] & [2.22] & [+0.28, +15.85] \\ | |
| 46 | 50 | \addlinespace |
| 47 | 51 | \midrule |
| 48 | −Years & & 15 & \multicolumn{3}{l}{mean cross-section 1{,}795 firms} \\ | |
| 52 | +Cohorts & & 15 & \multicolumn{3}{l}{mean cross-section 1{,}795 firms} \\ | |
| 49 | 53 | \bottomrule |
| 50 | 54 | \end{tabular} |
| 51 | −\par\vspace{2pt}{\footnotesize\emph{Notes:} Annual cross-sectional regressions of the forward return on the signal and size, momentum and volatility (all standardized within year; size from within-year winsorized log cap); slope = time-series mean of the 15 annual slopes. NW $t$ uses 3 Bartlett lags; the Kiefer--Vogelsang fixed-$b$ 5\% critical value for this bandwidth and $T$ = 15 is 2.71 (simulated), so NW $t$-statistics between 1.96 and 2.71 are not significant at 5\% under fixed-$b$ asymptotics. `raw' / `wins.' = raw / within-year 1/99-winsorized forward returns.} | |
| \ No newline at end of file | ||
| 55 | +\begin{tablenotes}[flushleft]\footnotesize | |
| 56 | +\item \emph{Notes:} NW $t$ uses 3 Bartlett lags; fixed-$b$ 5\% critical value 2.71 (simulated); plain = i.i.d.\ $t$ over the annual slopes; CI = 95\% $t$-interval over the annual slopes. The SPID $z$-score is $\log(1 + \text{mentions})$ standardized within year; `Any AI project' is the 0/1 indicator (Table~\ref{tab:spidbinary} for its minimum detectable effect). | |
| 57 | +\end{tablenotes} | |
| 58 | +\end{threeparttable}} | |
modified
paper/tables/tab_fm_both.tex
+44 −23
Large diff (72 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_groups.tex
+24 −31
Large diff (66 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_hygiene.tex
+28 −1
Large diff (40 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_mht.tex
+23 −75
Large diff (104 lines) is not loaded by default to keep the page fast.Load diff
added
paper/tables/tab_mht_full_a.tex
+43 −0
Large diff (43 lines) is not loaded by default to keep the page fast.Load diff
added
paper/tables/tab_mht_full_b.tex
+63 −0
Large diff (63 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_panel.tex
+20 −10
Large diff (47 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_placebo.tex
+17 −10
Large diff (33 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_prepost_both.tex
+38 −39
Large diff (88 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_reversal.tex
+23 −14
Large diff (44 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_robust.tex
+33 −27
Large diff (65 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_sector_tech.tex
+35 −26
Large diff (70 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_sorts.tex
+57 −58
Large diff (121 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_sorts_both.tex
+36 −55
Large diff (97 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_spid_binary.tex
+24 −15
Large diff (47 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_summary.tex
+22 −15
Large diff (42 lines) is not loaded by default to keep the page fast.Load diff
modified
paper/tables/tab_topfirms.tex
+12 −5
Large diff (32 lines) is not loaded by default to keep the page fast.Load diff
modified
results/r3k_v2/inline_numbers.json
+44 −1
Large diff (50 lines) is not loaded by default to keep the page fast.Load diff
modified
scripts/r3k_19_figures_v2.py
+179 −91
Large diff (273 lines) is not loaded by default to keep the page fast.Load diff
modified
scripts/r3k_20_tables_v2.py
+579 −235
Large diff (869 lines) is not loaded by default to keep the page fast.Load diff