# How to predict future reservations when data for the current day is incomplete?

I'm trying to build a model to predict reservations up to 15 days in advance.

So, if I want to predict how many reservations there will be tomorrow, I use historical data of how many total reservations there were on two days prior since when using the model to forecast for tomorrow, today won't be finished... Does that make sense?

I think it is an OK and unbiased model, however, it does not use all of the available data, namely, how many reservations for tomorrow have been made today. So, if I want my model to be more 'realtime' and account for current reervations as well, how do I do that? Do I look at reservation data over the course of the day, and just set up a proportion or something? For example, if I use my model at 13:00, find that I have 20 additional reservations, and I know that historically 40% of additional reservations are made by 13:00, do I just take 20+ 20*(60/40) = 50. So now I should 'expect' 50 more on top of what yesterdays total told me? This is all I can think of.

Edit:

" For example if it takes 1 hour to complete three innings at a baseball game you can rest assured that the next 6 innings is going to take a lot longer than 2 hours to complete." Well, what if you knew that historically the first 3 innings accounted for 25% of the time? Sticking with the ballgame example, that's good illustration of my problem which I think you understand Dave, but I'm not sure I've made it clear for the others...

If you wanted to predict the duration of a ball game, you might have some formula based on the current teams, the pitchers, etc.. But, consider a game where you predicted a 2.5 hour game, and you're in the 4th inning, and it's already taken 3 hours, now what do you do?

Dave, I've tried different techniques, involving the seasonal arima methods, etc. I'm not familiar with 'level shifts', although I've seen you mention them in many of your posts. I will do some further web surfing to understand this concept. I'd be very interested in a chat session. Please let me know when you're available, I'm +11 hours EST.

• Sorry, I'm a bit confused. When you say "reservation" you mean something like a restaurant reservation? It sounds like you're predicting when they will OCCUR, as opposed to when they are MADE, right? So reservations made today for tomorrow would enter into tomorrow's forecast along with a prediction based on reservations that occurred yesterday and today? Is this in an AR(2) sort of way? Sep 22, 2011 at 21:59
• Wayne, sorry I didn't answer this sooner. You are right, I'm not clear in my description. I want to predict how many total paying customers I will have on a particular day. (Call that variable: PAID.) I'm currently doing that as a function of how many reservations there are already for that particular day. (It's not deterministic as there are walk in customers, and cancellations.) (edit -- i hit the enter key or something) When I build my model, to predict tomorrows PAID, I use how many total reservations there are 2 days in advance.
Sep 23, 2011 at 4:05