Why Copula Models Use ARMA-GARCH Standardized Residuals

Summary

The document explains why a copula model for financial time series is often applied to standardized residuals after fitting a univariate ARMA or GARCH model, rather than directly to raw observations. Raw series may contain serial correlation and changing conditional variance, so treating their observations as independent can lead to incorrect inference and weaker forecasts.

The ARMA-GARCH step models each series’ time dependence and volatility; the standardized residuals are then used to represent the remaining variation for copula modeling. The copula is not a model detached from the original observations: the combined ARMA-GARCH-copula specification supports inference and forecasts for the original series. The answer gives a conceptual rationale rather than a worked example or empirical comparison, and frames the residual approach as a potentially better approximation in finance, not a universal guarantee.

Key ideas

• Raw financial time series can exhibit autocorrelation and conditional heteroskedasticity.
• An independent and identically distributed assumption may misstate inference and produce suboptimal forecasts when those effects are present.
• ARMA-GARCH models capture serial dependence and changing conditional variance before copula fitting.
• The full ARMA-GARCH-copula model still describes and supports forecasts for the original data.
• Using standardized residuals can improve the approximation, but the discussion does not establish universal superiority.

Tags

# In copula modeling for time series data, why do we need to fit ARIMA/GARCH and then work on standardized residulas.?


# In copula modeling for time series data, why do we need to fit ARIMA/GARCH and then work on standardized residulas.?












I have read that for standard copula modeling, you can get empirical cdf of data and use it for copulas. But for time series data, we must first fit ARIMA/GARCH, get standardized residuals, and only then start copula modeling. What is the reason for this using standardized residuals and not directly the time series data for the copula modeling?

## Answer by Richard Hardy (score 1, accepted)

https://quant.stackexchange.com/a/78291

If the univariate time series follow ARMA+GARCH, why do we not model them as i.i.d.? After all, it would make our job easier. The answer is, inference based on an i.i.d. assumption will be wrong and forecasts suboptimal when the series follow ARMA+GARCH (or contain autocorrelations and autoregressive conditional heteroskedasticity that are better approximated by ARMA+GARCH than by i.i.d.). In finance, using copulas on standardized residuals from ARMA+GARCH may often make a better approximation than using copulas on raw data.

And just in case you might wonder what use is a copula model on standardized residuals when you want a model for the original data: the entire model (ARMA+GARCH+copula) is actually for the original data, so you can make inference and forecasts for the original data.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.