Firms using AI-powered accounting tools on Jordan's Amman Stock Exchange obtain stronger market valuations, and those gains are partly explained by better disclosure and more comparable financial information; the association holds across fixed-effects and GMM specifications.

Main Finding

AI-enabled accounting intelligence (AIAI) is positively and significantly associated with better capital market performance (CMP) for listed firms on the Amman Stock Exchange. The study documents a direct effect (β = 0.412, p < 0.01) and shows that this effect is partially transmitted through improved disclosure quality (DQ) and greater financial information comparability (FIC). Significant mediation is confirmed by Sobel tests (DQ Sobel z = 6.142; FIC Sobel z = 5.873). Approximate indirect effects: AIAI→DQ→CMP ≈ 0.487 × 0.318 ≈ 0.155; AIAI→FIC→CMP ≈ 0.453 × 0.276 ≈ 0.125 (total indirect ≈ 0.28), indicating substantial partial mediation.

Key Points

• Sample: Balanced panel of 215 publicly listed firms on the Amman Stock Exchange across six sectors (banking, financial services, technology, industrial manufacturing, telecommunications, energy), 2017–2024 (1,720 firm‑year observations).
• Main measures:
  • AIAI: composite index (AI-accounting adoption, digital reporting integration, automated audit deployment, intelligent analytics) scaled 0–1.
  • DQ: combined voluntary disclosure index, accrual quality, and timeliness (Dechow–Dichev).
  • FIC: accounting comparability index (industry peer averages).
  • CMP: Tobin’s Q.
• Core empirical results:
  • AIAI → CMP: β = 0.412, p < 0.01.
  • AIAI → DQ: β = 0.487, p < 0.01.
  • AIAI → FIC: β = 0.453, p < 0.01.
  • DQ → CMP: β = 0.318, p < 0.01.
  • FIC → CMP: β = 0.276, p < 0.01.
  • Mediation: DQ and FIC partially mediate AIAI→CMP; Sobel z-statistics significant.
• Robustness: findings hold to lagged specifications, alternative dependent variables, trimming (1st/99th percentiles), and two-step system GMM (Arellano–Bover / Blundell–Bond) to address dynamics and endogeneity.
• Theoretical framing: integrates information asymmetry, agency, signaling, efficient market, and resource-based view arguments to explain how AIAI creates valuation benefits.

Data & Methods

• Data sources: annual reports, Bloomberg, Refinitiv Eikon / DataStream, Compustat-derived indices, stock exchange records.
• Panel design: balanced panel (215 firms × 8 years = 1,720 observations); sectors selected for AI intensity.
• Controls: SIZE, LEV, ROA, AGE, AUDQ (Big‑4 dummy), MTB, LIQ, INST.
• Estimation strategy:
  • Baseline: pooled OLS, fixed- and random-effects models; Hausman and Breusch‑Pagan LM tests used for specification.
  • Mediation: Baron & Kenny stepwise procedure plus Sobel z-tests for significance of indirect effects.
  • Endogeneity/dynamic specification: two-step system GMM (lagged dependent variable included); heteroskedasticity- and autocorrelation-robust inference; panel unit‑root diagnostics.
  • Robustness checks: lag structures, alternative outcome measures, outlier trimming.
• Construction/operationalization notes:
  • AIAI is a composite adoption/integration index (0–1) based on disclosures and Bloomberg ESG items.
  • DQ blends voluntary disclosure scores with accrual‑quality and timeliness metrics.
  • FIC computed as averaged comparability measures across industry peer pairs.

Implications for AI Economics

• Valuation of AI investments: The study provides empirical evidence that AI investments in accounting/reporting functions act as strategic intangible resources that raise firm valuations (higher Tobin’s Q). For economic models valuing digital/intangible capital, AIAI should be treated as a productive, value‑relevant asset that reduces information frictions.
• Mechanisms matter: The value effect is not only direct; much of it operates through improved disclosure quality and cross‑firm comparability. AI’s economic returns therefore include informational externalities (lower information acquisition costs, higher forecast accuracy, improved price discovery) which economists should model explicitly.
• Corporate strategy and allocation: Firms considering AI spending should account for its signaling and governance benefits (reduced agency costs, stronger audit trails) beyond efficiency gains—these benefits can translate into lower cost of capital and superior market performance.
• Policy and standard-setting: Regulators and standard-setters should consider promoting interoperable digital reporting standards (e.g., machine‑readable tagging, XBRL, harmonized data protocols) that amplify comparability gains from firm-level AI adoption and aggregate market efficiency.
• Measurement and investment appraisal: For cost‑benefit analysis and productivity accounting, empirical work should incorporate information-channel outcomes (DQ, FIC) as intermediate variables when estimating returns to AI capital.
• Research directions: Cross-country replications and causal identification (e.g., quasi‑experimental adoption events, difference‑in‑differences, randomized pilots) are needed to test external validity beyond Amman and the selected sectors. Micro‑level studies on how specific AI modules (NLP for disclosure drafting, anomaly detection in audits) contribute to DQ and FIC would refine policy and managerial guidance.
• Caveats for economists: Although robust techniques (system GMM, Sobel mediation, multiple checks) are used, measurement of AIAI as a composite index and the single‑country sample (Jordan) limit broad generalization. Future work should address measurement heterogeneity, heterogenous investor responses, and possible distributional effects (e.g., on analysts, auditors, labor markets).

If you want, I can: - Extract the precise regression tables and compute share of total effect mediated using the paper’s reported direct effect in the full mediation model; - Draft a short policy brief for regulators summarizing actionable recommendations; or - Propose a follow‑up empirical design (e.g., diff‑in‑diff or instrument) to strengthen causal claims.