Corporate AI Communication and Capital Market Response
This dashboard asks whether AI-related language in firms' 10-K annual reports is associated with abnormal stock-market reactions around the 10-K filing date. It studies communication in public filings, not direct observation of actual AI adoption.
1. Plain-Language Guide
For readers with no prior background
What is being tested?
When a firm talks more about AI in its annual report, does the market react differently around the day that report becomes public?
What did the evidence show?
10-K filings appear to attract investor attention, especially through abnormal trading volume, but AI-specific disclosure variation does not robustly explain short-window CAR in the current sample.
2. Underlying Mechanism
Why AI language might matter
Public 10-K disclosure
AI-related language becomes observable to investors through a formal annual report.
Investor attention
Investors process whether the AI language signals innovation opportunity, risk exposure, or routine disclosure.
Trading response
Information processing can appear as abnormal trading volume even when prices do not move strongly.
Price adjustment
CAR changes only if the AI signal is value-relevant and separable from the other bundled 10-K information.
Mechanism implication: the strongest expected pattern is not necessarily price movement. Because 10-K reports contain many types of information, trading attention can be clearer than AI-specific CAR.
3. IV: Independent Variables
AI-related 10-K communication measures
| Variable | Plain meaning | Interpretation boundary |
|---|---|---|
AI_Related_Disclosure_Intensity | How intensively the 10-K report discusses AI. | Keyword-based disclosure intensity; not direct actual AI adoption. |
AI_RiskRelated_Topic_Share | How much AI discussion appears in privacy, data, cybersecurity, legal, regulatory, or risk contexts. | Topic-derived risk-context share; not a manual sentiment score. |
AI_Risk_Orientation_Proxy | Whether AI communication is framed more toward risk orientation than infrastructure opportunity. | Proxy based on topic balance; not human-coded managerial intent. |
4. DV: Dependent Variables
Market reaction outcomes
Cumulative Abnormal Return (CAR)
CAR asks whether the firm's stock return moved more than expected after accounting for normal market movement. It is the price-reaction outcome.
CAR_m1_p1: [-1,+1]CAR_0_p1: [0,+1]CAR_0_p3: [0,+3]CAR_0_p5: [0,+5]
Abnormal Volume
Abnormal volume asks whether trading activity increased unusually around the filing date. It is interpreted as investor attention and information processing.
AbnormalVolume_0_p1: [0,+1]AbnormalVolume_0_p3: [0,+3]AbnormalVolume_0_p5: [0,+5]
5. Model and Formula
How the test is implemented
Abnormal return and CAR
Meaning: subtract normal market-driven returns from the firm's actual return, then add the abnormal returns over the event window.
Regression model
Meaning: Y is CAR or abnormal volume. β1 tests whether AI-related 10-K communication is associated with the market reaction. μi controls for firm fixed effects and λt controls for year fixed effects.
6. Results
Descriptive and regression evidence
| Evidence | Current output | Plain interpretation |
|---|---|---|
| Average CAR | CAR_m1_p1 0.0030; CAR_0_p1 0.0031; CAR_0_p3 0.0028; CAR_0_p5 0.0024 | Short-window average price reactions are small and slightly positive. |
| Average abnormal volume | AbnormalVolume_0_p1 0.6070; _0_p3 0.4614; _0_p5 0.3704 | Trading volume is visibly elevated, especially immediately after the filing. |
| Regression evidence | N=272; 21 coefficient rows; firm/year FE when feasible; firm-clustered SE or HC1 fallback | AI disclosure variables do not show a statistically significant short-window CAR association in the current sample. |
| Main conclusion | Attention evidence stronger than AI-specific price evidence | 10-K filings attract market attention, but robust AI-specific CAR evidence is not established. |
7. Placebo / Diagnostics
Checks before the event date
Placebo logic: if the same pattern appears before the filing, the event-window interpretation becomes weaker. The analysis uses pre-filing windows CAR_m10_m6, CAR_m5_m2, and AbnormalVolume_m5_m2. Current outputs contain 9 placebo coefficient rows, 10 model diagnostic rows, and merged_rows = 273.
8. Claim Boundaries
What this dashboard does not claim
Not actual AI adoption
The IVs are text-based disclosure proxies. They do not directly observe whether or how deeply a firm adopted AI.
Not standalone causality
The design provides event-window association evidence. It should not be read as standalone proof of an AI-specific causal effect.
Bundled 10-K information
10-K reports contain financial, risk, legal, and strategic information together. AI-specific effects are difficult to isolate.
9. Data Files
Generated repository outputs
| File | Status |
|---|---|
data/derived/market_extension/daily_market_data_10k_events.csv | Generated; 73,746 rows |
data/derived/market_extension/market_data_collection_report.csv | Generated; 273 rows |
data/derived/market_extension/post_filing_market_reaction_estimates.csv | Generated; 273 rows, 272 estimated |
data/derived/causal/ai_10k_event_study_analysis_dataset.csv | Generated; 273 rows |
data/derived/causal/causal_event_study_regression_summary.csv | Generated; 21 coefficient rows |
data/derived/causal/placebo_pre_filing_checks.csv | Generated; 9 placebo model rows |