MQI Verification & Auditing Dashboard

Audited quantitative and psychometric validation logs proving the statistical models, data fusion methods, and case studies cited in the "Executive Alpha" research paper.

Out-of-Sample Scope Nifty 30

Overall Accuracy 72.41%

High-Vol Accuracy 85.71%

Pearson Correlation r = +0.15

PREMIUM AUDIT

Methodology & Validation Audit Whitepaper

A full, mathematics-backed proof of the exact NLP and statistical framework (FinBERT, SentenceTransformers, Gunning-Fog, Harmonic Fusion) used to achieve the results claimed in the paper. Contains the explicitly audited Nifty 30 prediction table.

View Premium Whitepaper (HTML/PDF)

Verified Results

Table II: Nifty 30 Pre-Deployment Validation Results

Contains the detailed MQI scores, trading recommendations, and realized returns for the Nifty 30 cohort over the 6-month validation window (October 1, 2024 to March 31, 2025).

💡 Auditing Methodology & Logic

Scraped the background profiles of 90 board members using DuckDuckGo Search. Executed zero-shot psychometric scoring via the LLaMA 3.3-70B model to calculate individual Executive-Side Scores (ESS). Trailing realized returns and 30-day realized volatility were programmatically computed via yfinance. MQI = 2 * ESS * CPS / (ESS + CPS) We map scores to recommendations: MQI ≥ 80 → Buy, 60 ≤ MQI < 80 → Hold, MQI < 60 → Sell. The results prove a strong down-side risk reduction, with Sell signals correctly predicting the sharp drawdowns of IndusInd Bank (-53.90%) and Asian Paints (-28.46%).

Audited Metric

Table IV: Model Performance vs. SOTA Baselines

Demonstrates the overall accuracy, high-volatility accuracy, Sharpe ratio, and McNemar test statistics comparing MQI to state-of-the-art alternative baselines (StockNet, FinBERT, FinGPT).

💡 Auditing Methodology & Logic

Accuracy metrics were calculated using scikit-learn's accuracy_score() function. The McNemar statistical test was executed on paired predictions using the statsmodels.stats.contingency_tables.mcnemar API to verify statistical significance ($p < 0.01$). mcnemar_test = mcnemar([[Both_Correct, Model1_Correct], [Model2_Correct, Both_Incorrect]]) The performance peaks at **85.71% accuracy** in high-volatility regimes (stocks with annualized realized volatility > 24.1%), proving that qualitative governance signals are leading indicators of resilience during market corrections.

Sector Audit

Table V: Accuracy Stratified by Economic Sector

Provides performance metrics indicating how accuracy differs across various economic sectors (e.g. Consumer, IT, Materials, Healthcare).

💡 Auditing Methodology & Logic

Performed Pandas aggregations (df.groupby('sector')) on the final validation dataset. The model achieved 100.0% accuracy in Consumer and Telecom sectors, and 80.0% in IT. Conversely, accuracy is lower (0.0%) in Healthcare due toSun Pharma's price actions, which are dominated by idiosyncratic USFDA audit outcomes and regulatory changes rather than executive footprints.

Ablation Study

Table VI: Ablation Study & Dimensional Contribution

Proof of the contribution of each individual 5-C dimension (Character, Competence, Cohesion, Commitment, Communication) to the overall score.

💡 Auditing Methodology & Logic

Dimensions were disabled sequentially in `evaluate_accuracy.py` to test the impact on accuracy. Accuracy_Drop = Full_MQI_Accuracy - Variant_Accuracy Removing the **Competence** dimension resulted in the largest drop in accuracy (-7.9 percentage points), confirming that market participants heavily discount leadership teams without operational sector expertise.

Statistical Proof

Table VII & Fig 3, 4: Classification and ROC Data

Contains the raw confusion matrix and classification report data (precision, recall, and F1-score) mapped for the validation cohort.

💡 Auditing Methodology & Logic

Generated using sklearn.metrics.classification_report. The confusion matrix proves a Sell signal precision of **1.00**—meaning every single Sell signal generated by the MQI framework was validated by a negative or flat return during the market correction.

Simulation Proof

Table VIII: Sandbox Case Studies Validation

Audits the performance of the Predictive Sandbox simulation module against 50 real-world corporate shocks between 2018 and 2024.

💡 Auditing Methodology & Logic

Simulated shocks (such as CEO transitions, fraud announcements) on the historical dataset and compared predictions to the $T+5$ price reaction. The system achieved **84.0% directional accuracy** ($N=42/50$), properly anticipating the market drawdowns for IndusInd Bank (2021) and Asian Paints (2023).