Home Research Master's Thesis (NII)

The Narrative Inconsistency Index (NII)

Measuring the 'Optimism Gap' between MD&A and Risk Factors in the strategic disclosures of US technology firms using Generative AI (Gemini 2.5) to predict stock price crash risk.

PFORZHEIM UNIVERSITY • M.SC. DIGITAL BUSINESS MANAGEMENT

Academic Research Monograph

Author: Doğan Koç  |  First Examiner: Prof. Dr. Mario Boßlau  |  Date: June 11, 2026

Abstract

This master's thesis investigates executive strategic information-hoarding behaviors by engineering and validating the Narrative Inconsistency Index (NII). Analyzing 167 mandatory SEC 10-K annual filings (2016–2023) across 25 leading US technology firms (Tech25), the study quantifies the semantic asymmetry between optimistic management discourse (MD&A, Item 7) and statutory risk disclosures (Risk Factors, Item 1A).

Integrating Google Gemini 2.5 Pro and Flash Generative AI models via multi-step chain-of-thought prompting, the empirical framework demonstrates statistically significant superiority over legacy static word-count dictionaries (Loughran & McDonald, 2011) in capturing contextual nuances, obfuscation, and subtle tone divergences.

Generative Artificial Intelligence Google Gemini 2.5 Pro / Flash SEC 10-K Filings Audit Stock Crash Risk (NCSKEW) Agency Theory Voluntary Disclosure Theory

Key Empirical Contributions

  • The NII Metric: Formalized a normalized 0–200 scale capturing the executive 'Optimism Gap'.
  • 167 SEC Filings: Empirical evaluation across 8 years of audited filings for Apple, Microsoft, Nvidia, Alphabet, Amazon, Meta, Tesla, etc.
  • Crash Risk Forecasting: Confirmed predictive correlation with one-year-ahead negative skewness (NCSKEW) at p < 0.01.
  • LLM vs. Dictionaries: LLM contextual embedding outperformed static financial lexicon lists by detecting implicit risk suppression.
Interactive NII Simulator: Test Live Corporate Cases
Item 7: MD&A (Executive Optimism Language)

"Our user engagement metrics and multi-product ecosystem momentum continue to expand at an unprecedented trajectory, establishing a robust foundation for record-setting operating margins."

LLM Sentiment Score: +84% (Hyper-Optimistic)
Item 1A: Risk Factors (Disclosed Threat Severity)

"Severe component supply bottlenecks, geopolitical export constraints, and tightening anti-trust regulations may induce material delivery halts and margin contraction in upcoming quarters."

LLM Risk Severity: -78% (Severe Risk)
Computed Narrative Inconsistency Index (NII) Score: 154 / 200
0–60: Transparent & Low Crash Risk 61–120: Moderate Divergence 121–200: High Stock Price Crash Risk (NCSKEW Alert)
METHODOLOGICAL RIGOR

Research Pipeline & Theoretical Architecture

1. Theoretical Foundations

Grounding in Agency Theory (Jensen & Meckling, 1976) and Voluntary Disclosure Theory (Verrecchia, 1983; Dye, 1985), explaining executive incentives to conceal adverse information behind promotional language until an accumulation threshold triggers a crash.

2. GenAI Pipeline Architecture

Automated ingestion of raw 10-K text via SEC EDGAR APIs, parsing HTML into clean structured sections. Multi-turn zero-shot and few-shot Chain-of-Thought prompting using Gemini 2.5 to extract sentiment polarities and risk concealment vectors.

3. Econometric Panel Regressions

Formulating fixed-effects panel models controlling for firm size, leverage, ROA, stock return volatility, and share turnover. Statistically testing NII against future NCSKEW and DUVOL measures to establish causality.

Interested in the full academic manuscript or research data?

I am glad to discuss data extraction pipelines, LLM prompt templates, or econometric replication with researchers and industry leaders.

Get in Touch Regarding Research