Dear guys,   I have a time series of hourly deviations of the electricity market saturation from zero, in other words I have data on whether the total electric system in my country was in black or red quantitites for each hour of the last months. The intention is to predict the likelihood that in two hours this deviation will be above zero. The data favor an AR(2) model with some GARCH component, but that's good only for the point forecast. To obtain the likelihood, I was going to use quantile regressions - by finding the specification for tau, which predicts a 2-step ahead forecast of 0 deviation, I would find my probability. So I wanted to ask whether QR in gretl is OK with autocorrelated series of 'y'. I know there are no assumptions on the error term distribution, but I found some extensions to QR to handle autocorrelated observations (but they were pretty scarce).   Many thanks, Daniel