Model description

The main purpose of the Activity Detector model is to identify structural changes in the price discovery of financial assets by analyzing volatility and trading volumes. The model detects sharp spikes in these indicators, evaluates their significance and classifies them as anomalous or not. When anomalies are detected, the system promptly notifies the user, helping him to react quickly to changes in market conditions.

Volatility and trading volume analysis methods

Activity Detector dynamically tracks changes in an asset's volatility and liquidity by analyzing time series of market data in real time. To do this, it:

Uses approximate statistical estimates of average and instantaneous volatility and volume values.
Applies smoothing filters and adaptive algorithms to minimize noise in the data and highlight significant changes.
Analyzes periods of accelerated changes in volatility and volume, comparing them to the historical context of the asset's market behavior.
Applies thresholds and dynamic adaptation to eliminate false signals.

Definition of abnormal activity

Abnormal activity is determined based on several factors at once:

The degree of deviation of current volatility and volume from their typical values over similar periods.
Rate of change of market parameters: sharp jumps in prices, volumes and spreads are analyzed.
Contextual parameters: general market conditions and macroeconomic factors are taken into account, excluding natural periods of increased activity (e.g., during the release of important economic news).

Using machine learning

Activity Detector uses machine learning algorithms to improve the accuracy of its analysis, but specific models are not disclosed. Basic principles of the system's use of AI include:

Training a model on historical data to recognize typical and atypical patterns of market behavior.
Automatic updating of statistical thresholds and logical dependencies based on dynamic market behavior.