Parameter Instability

Parameter instability occurs when the coefficients of a statistical model change over time, rendering the model's predictive power unreliable. This happens when the underlying relationships between variables are not fixed, which is typical in complex, adaptive markets like cryptocurrency.

If a model is trained on a period where a specific correlation existed, but that correlation breaks down, the model will produce erroneous results. This is a primary cause of model failure in trading systems.

Analysts must constantly monitor for parameter drift and update their models to ensure they remain relevant. Techniques such as rolling window estimation are often used to prioritize recent data over older, potentially obsolete data.

It is a fundamental challenge in applying static mathematical models to a dynamic world.