Dear Kevin!
I'd much rather be lucky than good. I'm not going to share my work with you...I am going to try to help you by consulting with you, not by doing the job for you. I would be much more happier. Did not request the data to be lazy but to spare you time as I can by trying to understand on my own.
Zero is not missing! Egyptian accountants had a symbol for zero in 1740 BC. We should honor prehistory by using it in place of the missing data in your file. Yes true, statistically missing, financially zero.
But your graph of the 2014 data shows a monotonic increase that doesn't jive with the large numbers of zeroes (used to be missing) in the Revenue data. So I guess I am even more confused with this new information. Actually I could make a data table with the missing revenue data not displayed but each year the date of sales differ so I've thought I could make a calendar date column to sweep all. Under these circumstances I think making a "Week Of The Year" time variable will be much better.
The functional shapes on the graph all look similar year-over-year, and you may do better by just using the average of the functions at each day. That would certainly be more parsimonious than my approach, and it may even be more accurate. Remember, all you gave us was two partial years... I guess the exponential smoothing or weighted average would be the best as leisure & tourism in Turkey in 2016 is predicted to be more likely as 2015 rather than 2012 and I think I can handle it.
I know very little about the resort hotel business. But it makes some intuitive sense to me that, for instance, weekends would be more lucrative than weekdays, that summer vacationtime would be more lucrative than winter, that holidays like New Year's Eve would make more money than other days, etc. Actually what you were describing is the characteristics of "City Hotel" business. Weekend or weekdays do not play into Resort Hotel business since the average stay night is 12. Starts operating in May and ends with November (no christmas). So I made some variables like DayOfYear, DayOfWeek, WeekOfYear, Year, etc. I then fit "decent" Neural Net, Bootstrap Forest, and Boosted Tree models with the response of Revenue with those variables as Factors, and saved the prediction formulae. I made a variable that summarized the mean of those three model predictions. Then I incremented the dates through 2016, made the daily predictions, and summed the means by Year. It's a very naive approach that could probably be improved dramatically with some subject knowledge. Resort Hotels typically take very little direct reservations from their guests but make contracts with Tour Operators for the whole year in advance. 65 percent of the whole revenue is almost completed with the contracts months before the hotel starts operating (that's why the sales date for given operating year is consisting also the previous year). Predicting the remaining 35 percent is coming from "hot sales" which is the skills of Sales Management Team. They give a prediction about which market will produce what amount of sales and when, which is put in an excel file and called as "Revenue Budget" and "Cost Management Team" calculates the expected expenses depending on the data given by Revenue Budget and we call it "Expenses Budget". Total profit is the difference of the two. What I would like to monitor is the deviation of actual sales from the Revenue Budget as every day sales data flows in, which market is under performing or over performing and what would be the final actual sales amount compared to revenue budget.
I guess first I need to prepare the data table in such a format that I can use for Time Series and Fit Model and Fit Y by X but I don't know how to.. ( as michael.anderson mentioned "You can take out isolate seasonal effects from linear trends in revenue. This also gives the ability to do forecasting.")
I have added a data table " Predicting Future Values Years Weeks" I guess which is better for demonstration purposes.
Many thanks in advance,