regression serial correlation and ARMA

Hello,

Let's say that I have a daily return series of a stock of a company.

I would like to build a regression model. However instead of forecasting next day return I would like to forecast monthly return. This means dependent variable represents price(t+22)-price(t) (22 trading days) whereas independent variable is price(t)-price(t-1). Since there is such a lag, error terms become extremely correlated. I wanted to use an ARMA model, but I don't know how to ignore first 22 lags, as those inputs will be unknown when running the model out of sample. Basically I would like to use daily samples but monthly returns as output.

How do I go by doing this?

Thank you.