I have daily stock return data (log returns). I want to forecast returns for the next two months. I am creating forecasts with both univariate ARIMA and GARCH models with regressors.
What are the dangers /assumptions I am making if I downsample the daily data to the month level of aggregation? I am concerned with the following:
- If I downsample to weekly data, is this better than monthly given my data is daily and the goal of forecast accuracy?
- Financial data does not include weekends (market's close). Does this affect how I should think about downsampling?
If these questions are too many for this post, are there time series books that have a focusing on the implications of down and upsampling on forecasting results?
