I am trying to do some time series analysis on the margin resulting from three specific commodity futures contracts and ultimately forecast the margin. The margin is calculated as M = F1 + F2 - F3. I have the daily close prices for F1, F2, and F3, with the associated date and margin calculation.
The point I would like to address is that the futures contracts I'm interested in only trade for 9 months of the year. I have been able to acquire the close price for each futures contract for each day they were traded on the market, so I don't have missing data in that sense. There are gaps in my data for 3 reasons:
- futures don't trade on weekends (no data Saturday's and Sunday's)
- futures don't trade on holidays (no data for days like New Year's)
- these futures don't trade year round (no data for 3 months of the year)
I know people analyze market data all the time, so the first 2 points don't seem to be an issue. The last point seems similar to the first two, but I haven't been able to find explicit examples of analysis with data like this. Rob Hyndman's book Forecasting: Principles and Practice briefly touches on missing data but I'm still not sure how this case should be handled (if at all). From Hyndman's book, I think this case is one where the missingness of the data will induce bias in the model, so maybe I should use dummy variables to represent the time when these contracts are not being traded.
Do I need to treat my data in any particular way to handle the missing trading days? How do methods like ARIMA and dynamic regression deal with market data that doesn't include weekends or holidays? Are there certain models I should lean towards or away from given the structure of my data? Should any trading day where these contracts were not traded be included in my data as NULL or dropped all together?
I am a little out of my element on this so I apologize for any ignorance and will appreciate any insights. This seems like it should have an easy solution, but I have not had a lot of luck answering my questions through research.