Quant analytics: How many samples do you need to make a decent forecast of time series 1 step ahead ?
And how many for 3, 6, 12 step ahead? What does it depend on ? Any help appreciated
There is not, and cannot be, any universal answer because that would depend on the behavior of the process generating your data. If you are talking about a typical economic time series of the type people model as having seasonal and random components added to or multiplied by a long-term trend, then you would need at least a few years of data in order to have any hope of discerning any long-term trends underneath the seasonal and random components. But if your data come from a phenomenon without any seasonality, you would have a rather different sort of modeling problem. So it all depends.
NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!