GARCH
Developed in 1982 by Robert F. Engle, the generalized autoregressive conditional heteroskedasticity (GARCH) process is an econometric term developed to describe an approach to estimate volatility in financial markets. It involves autoregression of prices (or any other variable) and assumes that the variability is heteroskedastic, meaning that the variability is not constant but changes with time.
There are three steps involved:
For a Generalized Autoregressive Conditional Heteroskedastic Model of Order (p, q),
1. First is to find autoregression model for the variable based on historic data
2. Second is to predict the error term ( since the variability is not constant and hence the error term changes over time and mean error is not zero)
3. The last step is to test the significance of the model using hypothesis testing.