S&P 500 AI Exposure and Cost of Equity Dataset: ICC–CAPM Valuation Diagnostics, 2020–2024
Description
This dataset provides a firm-year panel of S&P 500 companies for the period 2020–2024, designed to study the relation between artificial intelligence (AI) exposure, valuation diagnostics, and the cost of equity in U.S. capital markets. The dataset combines AI-related disclosure measures, governance/risk disclosure counts, standard firm controls, market-level equity risk premium inputs, and firm-level valuation outcomes. AI exposure is measured through keyword-based counts of AI-related terms in annual disclosure text, with log-transformed versions using log(1 + count). Governance/risk variables are based on counts of disclosure terms related to oversight, accountability, control, risk, and governance. Firm controls include standard accounting variables such as size, profitability, leverage, and sales growth. The valuation block includes firm-level implied cost of equity estimates based on the modified PEG (MPEG) model, firm implied ERP, CAPM-based cost of equity, and the ICC–CAPM gap. The main formulas are: ICC_MPEG = [DPS/P + sqrt((DPS/P)^2 + 4((EPS2 − EPS1)/P))]/2; firm_implied_erp = ICC_MPEG − risk_free_rate; capm_coe = risk_free_rate + beta × market_erp; and icc_capm_gap = ICC_MPEG − capm_coe. These variables allow comparison between analyst-growth-implied required returns and beta-based CAPM required returns. Most valuation inputs are based on Bloomberg point-in-time data as of each year-end valuation date. When Bloomberg data were unavailable, reliable archival financial sources were used and documented through source flags. The dataset is intended for research on AI exposure, cost of equity, equity risk premia, and the expectation–risk wedge in large U.S. listed firms during the post-2020 AI diffusion period.
Files
Steps to reproduce
To reproduce the dataset, begin with the S&P 500 firm-year sample for 2020–2024 and retain firms with available annual AI disclosure, firm controls, and valuation data. Construct AI exposure by counting AI-related keywords in each firm’s annual disclosure text and calculate log-transformed AI exposure as log(1 + AI count). Construct governance/risk disclosure measures by counting terms related to governance, oversight, accountability, control, and risk, again using log(1 + count) for transformed variables. Add firm controls from annual financial data, including size, profitability, leverage, and sales growth. For each firm-year, assign the year-end valuation date as the last trading day of the calendar year, collect price, forward EPS forecasts, expected dividend per share, beta, risk-free rate, and market ERP as of the valuation date. Most valuation inputs are obtained from Bloomberg point-in-time data; when unavailable, reliable archival financial sources are used and documented through source flags. Use the Bloomberg consensus EPS forecast for the next fiscal year as EPS1 and the second-forward fiscal year as EPS2. Use expected annual dividend per share, or four times the latest declared quarterly dividend when a forward dividend estimate is unavailable. Set dividends to zero for non-dividend-paying firms. Calculate the modified PEG implied cost of equity as: ICC_MPEG = [DPS/P + sqrt((DPS/P)^2 + 4((EPS2 − EPS1)/P))]/2, where P is year-end price, DPS is expected annual dividend per share, EPS1 is the next-year EPS forecast, and EPS2 is the second-forward EPS forecast. Convert the result into percentage points. Flag ICC observations as valid only when EPS2 > EPS1. Calculate firm_implied_erp as ICC_MPEG − risk_free_rate. Calculate CAPM cost of equity as risk_free_rate + beta × market_erp. Calculate the ICC–CAPM gap as ICC_MPEG − capm_coe. Merge the valuation variables with the AI disclosure, governance/risk disclosure, firm controls, and market variables by ticker and year. Check that each firm-year appears only once, that rates and premia are expressed in percentage points, and that source flags are complete. For regression analysis, winsorize valuation variables at the 1st and 99th percentiles to reduce the influence of extreme implied cost-of-equity values.
Institutions
- Ca' Foscari University of VeniceVeneto, Venice