Celarvaux applies machine learning models to large volumes of market data to produce backtested recommendations, viewable from any workplace.
Recommendations are based on historical cycles analyzed continuously — no performance promises are made beyond documented backtesting results.
The Celarvaux engine carries out a multidimensional analysis of historical series – volatility, cross-correlations, seasonal cycles – before generating any recommendation. Each scenario is replayed over previous periods to measure its robustness before broadcast.
Raw data streams are normalized and removed from anomalies before any statistical processing.
The algorithms identify structural regularities and trend breaks across multiple horizons.
Each recommendation is weighted according to a calculated risk score before being transmitted to the user.
The following three functions make up the operational base of the platform and are designed for remote control, without continuous supervision.
Relevant positions and market indicators are continuously recalculated. Significant deviations from the backtested scenarios trigger a notification, without requiring physical presence at a fixed workstation.
Each identified opportunity receives a composite score incorporating historical volatility, liquidity and cross exposure. The score is recalculated each time the source data is updated, without manual intervention.
The model's outputs are formulated as arbitration avenues classified by relevance, the user is responsible for validating the final decision according to their own risk framework.
These scenarios illustrate how an operator working remotely can rely on the model's output to structure their decisions.
Detection of valuation differences between correlated instruments, with estimation of the historical convergence time observed on similar configurations.
Projection of likely trajectories from comparable past cycles, with an explicit confidence interval rather than a single value.
Weighting recommendations intended to reduce the overall correlation of a portfolio, recalculated at each significant change in market data.
The following answers specify the origin of the data and the accuracy limits of the model, without commercial formulation.
The data comes from public market feeds and suppliers specializing in price history. No personal portfolio data is used for training general models.
The precision varies depending on the asset class and the horizon considered. Backtesting results are documented per scenario and do not constitute a guarantee of future performance.
Models are re-evaluated at regular intervals as new market data becomes available, in order to limit performance drift over time.
Access is via a web interface that can be consulted remotely. No local installation is required, allowing workstation-independent use.
View backtested scenarios and current risk scores before formulating your next strategic decision.