Autoregressive models predict future values in a time series by regressing the variable on its own previous values. They are foundational in time series analysis, capturing temporal dependencies and often serving as a baseline for more complex models.