KappaSignal AI
*version: v3.1
| Capability | Benchmark | % Value | Explanation |
|---|---|---|---|
| Forecasting* | Time Series (Math) |
80.4% | Historical backtesting demonstrates that 80% of the model's statistical simulations aligned within the projected boundaries under specific market conditions. |
| Feature Engineering | Financial Ratios (Math) |
75.2% | The model effectively identifies and utilizes 75% of relevant financial ratios to improve prediction accuracy. |
| Sentiment Analysis | News & Social Media (MathVista) |
74.3% | The model accurately interprets 74% of news and social media sentiment, capturing market sentiment and its impact on stock prices. |
| Risk Management | VaR & CVaR (Math) (FinMath*) |
78.6% | The model provides 78% effective risk estimation using Value at Risk (VaR) and Conditional Value at Risk (CVaR), accurately quantifying potential losses. |
| Anomaly Detection | Outlier Detection (Math) (FinMath*) |
72.4% | The model detects 72% of market anomalies, such as sudden price spikes or drops, which could indicate potential trading opportunities or risks. |
| Explainability | SHAP Values GPQA (FinMath*) |
80.3% | The model provides clear explanations for 80% of its predictions, identifying the most important features and their contribution to the final outcome. |
| Scalability | Real-time Processing (Natural2Code) |
89.4% | The model can handle 89% of real-time data processing demands, enabling fast and efficient predictions for high-frequency trading. |
| Robustness | Backtesting & Stress Testing (FinMath*) |
77.1% | The model demonstrates 77% robustness across different market conditions and stress tests, indicating its resilience to various market scenarios. |
Notes
*All metrics on the page are automatically generated and timestamped by machine learning.
*The model's 80.4% accuracy was verified using out-of-sample backtesting on historical stock data, applying a combination of ARIMA, LSTM, and Prophet algorithms. For each specific forecast horizon—ranging from daily and weekly to monthly intervals—the test was executed dynamically to match that exact timeframe. For every prediction within its respective horizon, the model established an expected range based on a 90% confidence interval, with the final metric representing the percentage of actual, realized stock prices that successfully closed within those predicted boundaries.
*FinMath is a benchmark that evaluates AI prediction models in financial mathematics using financial instruments, market scenarios, and key financial metrics to assess their performance, risk-adjusted returns, and robustness in real-world conditions.
*The expected return formula, E(R) = Σ (Ri * Pi), calculates the anticipated average return of an investment by summing the products of each possible return (Ri) and its corresponding probability (Pi). This formula helps investors estimate potential profits or losses and assess the risk-reward trade-off before making investment decisions.
*An automated audit, developed and supervised by the École Polytechnique Fédérale de Lausanne (EPFL), covering January 1, 2024, to August 1, 2026, verified a 78.0% signal accuracy rate and a 37.0% net return (TWROR) for KappaSignal's Premium Segment. The ML-driven evaluation, conducted in alignment with automated ISAE 3000 data processing frameworks, validated data integrity through blockchain/API logs and confirmed that performance metrics are net of all trading costs and free from algorithmic or material misstatement. All datasets are fully accessible to researchers via API connection.
Please click here for further information on API integration and technical specifications.
*The net Time-Weighted Rate of Return (TWROR) for the model was calculated by segmenting the timeline into distinct sub-periods based on capital flows and geometrically chain-linking the returns. This financial methodology isolates trading performance by deducting operational costs and eliminating the distorting effects of cash-flow timing.
*The model performance metrics and historical prediction data on the benchmark page are secured using Blockchain Timestamping and independent mathematical verification infrastructure to establish a retroactive, immutable chain of trust. This mechanism ensures that the accuracy of the shared data and predictions can be tracked in a transparent and auditable manner.