I’m working on a time series project of sales data for my work. My advisor guided me to use the Arch Garch method. How do I first make sure my data is suitable for this method? What tests to run and what do I look for?

First you must have at least 150 observations (sample size is mor than 150) and you run regression estimation for your time series and then you exmine the residuals of the estimated model if the residuals have a clustring period of high volatality followed by clustring period of low volatality and so then we can conclud that there is a arch effect so we can do a arch test. If the arch test is significance then we must estimate the Arch-Garch model. Eviews is considered a good for this situstion.

ARCH models were introduced in the literature of time series to model the heteroskedasticity (heterogeneity in variance). They are often considered as models for innovation processes. Indeed, when you have a time series which doesn't present any significant autocorrelation, you may notice a volatility clustering as in the attached figure. The square of this kind of series are sometimes exhibiting correlations and this shows that the conditional variance is not homogeneous.

In practice, you can check the homoskedasticity test to see if your series do present an ARCH effect. You'll find attached some links that may help you.

First you must have at least 150 observations (sample size is mor than 150) and you run regression estimation for your time series and then you exmine the residuals of the estimated model if the residuals have a clustring period of high volatality followed by clustring period of low volatality and so then we can conclud that there is a arch effect so we can do a arch test. If the arch test is significance then we must estimate the Arch-Garch model. Eviews is considered a good for this situstion.