Data Stationarity

Data stationarity is a property of a time series where its statistical properties, such as mean, variance, and autocorrelation, remain constant over time. Many financial models, including those used for VAR, require stationary data to produce reliable forecasts.

If the underlying process is non-stationary, such as a price series that exhibits a trend or changing volatility, the model may produce spurious results. In cryptocurrency markets, prices are notoriously non-stationary because they often exhibit strong trends and shifting volatility regimes.

To make the data suitable for modeling, practitioners typically transform the raw price data into log returns. This transformation often helps in stabilizing the mean and variance, making the series more amenable to statistical analysis.

However, even with transformations, structural breaks can occur, causing the series to become non-stationary again. Monitoring for stationarity is a critical part of the data preparation phase in quantitative finance.

If a model is trained on non-stationary data, its predictive power will be significantly compromised, leading to inaccurate risk assessments and potentially dangerous exposure levels.