Dealing with The Somewhat Naive Solution

[TomH488]

The classic NS is using the previous prediction as the current.
Predict weather tomorrow - say is same as today. Same thing with
forecasting stock price.

5 day into future prediction has a 5 day lag.

My current model is a 10 day future prediction that has a 3 day lag -
so its really a 7 day future prediction.

I really don't know how to get this bit of lag out of the net. I've
tried recurrent nets and they don't get rid of it at all.

So I'm thinking I'll just do a 14 day future prediction and get my 10
day prediction from a 4 day lag.

This is obviously an Engineering Solution.

Any comments?
Tom

[Greg Heath]

Why don't you look at peaks of the autocorrelation function and/or if there are inputs, the peaks of the corresponding cross-correlation functions?

Greg