Distinct lists on dates where an ID is present (i.e. intersects) on consecutive dates - sql

Question

I'm trying to make an MSSQL query that produces lists of apartment prices. The ultimate goal of the query is to calculate the percentage change in average prices of apartments. However, this final calculation (namely taking averages) is something I can fix in code provided that the list(s) of prices that are retrieved are correct.
What makes this tricky is that apartments are sold and new ones added all the time, so when comparing prices from week to week (I have weekly data), I only want to compare prices for apartments that have a recorded price in weeks (t-1, t), (t, t+1), (t+1,t+2) etc. In other words, some apartments that had a recorded price in time (t-1) might not be there at time t, and some apartments may have been added at time t (and thus weren't there at time t-1). I only want to select prices in week t-1 and t where some ApartmentID exists in both week t-1 and t to calculate the average change in week t.
Example data
-------------------------------------------------------------
| RegistrationID | Date | Price | ApartmentID |
-------------------------------------------------------------
| 1 | 2014-04-04 | 5 | 1 |
| 2 | 2014-04-04 | 6 | 2 |
| 3 | 2014-04-04 | 4 | 3 |
| 4 | 2014-04-11 | 5.2 | 1 |
| 5 | 2014-04-11 | 4 | 3 |
| 6 | 2014-04-11 | 7 | 4 |
| 7 | 2014-04-19 | 5.1 | 1 |
| 8 | 2014-04-19 | 4.1 | 3 |
| 9 | 2014-04-19 | 7.1 | 4 |
| 10 | 2014-04-26 | 4.1 | 3 |
| 11 | 2014-04-26 | 7.2 | 4 |
-------------------------------------------------------------
Solutions thoughts
I think it makes sense to produce two different lists, one for odd-numbered weeks and one for even-numbered weeks. List 1 would then contain Date, Price and ApartmentID that are valid for the tuples (t-1,t), (t+1,t+2), (t+3,t+4) etc. while list 2 would contain the same for the tuples (t,t+1),(t+2,t+3),(t+4,t+5) etc. The reason I think two lists are needed is that for any given week t, there are two sets of apartments and corresponding prices that need to be produced - one that is "forward compatible" and one that is "backwards compatible".
If two such lists can be produced, then the rest is simply an exercise in taking averages over each distinct date.
I'm not really sure to begin here. I played a little around with Intersect, but I'm pretty sure I need to nest queries to get this to work.
Result
Using the methodology described above would yield two lists.
List 1:
Notice how RegistrationID 2 and 6 disappear because they don't exist in on both dates 2014-04-04 and 2014-04-11. The same goes for RegistrationID 7 as this apartment doesn't exist for both 2014-04-19 and 2014-04-26.
-------------------------------------------------------------
| RegistrationID | Date | Price | ApartmentID |
-------------------------------------------------------------
| 1 | 2014-04-04 | 5 | 1 |
| 3 | 2014-04-04 | 4 | 3 |
| 4 | 2014-04-11 | 5.2 | 1 |
| 5 | 2014-04-11 | 4 | 3 |
| 8 | 2014-04-19 | 4.1 | 3 |
| 9 | 2014-04-19 | 7.1 | 4 |
| 10 | 2014-04-26 | 4.1 | 3 |
| 11 | 2014-04-26 | 7.2 | 4 |
-------------------------------------------------------------
List 2:
Here, nothing disappears because every apartment is present in the tuples within the scope of this list.
-------------------------------------------------------------
| RegistrationID | Date | Price | ApartmentID |
-------------------------------------------------------------
| 4 | 2014-04-11 | 5.2 | 1 |
| 5 | 2014-04-11 | 4 | 3 |
| 6 | 2014-04-11 | 7 | 4 |
| 7 | 2014-04-19 | 5.1 | 1 |
| 8 | 2014-04-19 | 4.1 | 3 |
| 9 | 2014-04-19 | 7.1 | 4 |
-------------------------------------------------------------

Answer 1

Here's a solution. First, I get all the records from the table (I named it "ApartmentPrice"), computing the WeekOf (which is the Sunday of that week), PreviousWeek (the Sunday of the previous week), and NextWeek (the Sunday of the following week). I store that in a table variable (you could also put it in a CTE or a temp table).
declare #tempTable table(RegistrationId int, PriceDate date, Price decimal(8,2), ApartmentId int, WeekOf date, PreviousWeek date, NextWeek date)
Insert #tempTable
select ap.RegistrationId,
ap.PriceDate,
ap.Price,
ap.ApartmentId,
DATEADD(ww, DATEDIFF(ww,0,ap.PriceDate), 0) WeekOf,
DATEADD(ww, DATEDIFF(ww,0,dateadd(wk, -1, ap.PriceDate)), 0) PreviousWeek,
DATEADD(ww, DATEDIFF(ww,0,dateadd(wk, 1, ap.PriceDate)), 0) NextWeek
from ApartmentPrice ap
Then I join that table variable to itself where WeekOf equals either NextWeek or PreviousWeek. This gives the apartments that have a record in the adjoining week.
select distinct t.RegistrationId, t.PriceDate, t.Price, t.ApartmentId
from #tempTable t
join #tempTable t2 on t.ApartmentId = t2.ApartmentId and (t.WeekOf = t2.PreviousWeek or t.WeekOf = t2.NextWeek)
order by t.RegistrationId, t.ApartmentId, t.PriceDate
I'm using distinct because an apartment will appear more than once in the results if it does have an adjoining week record.
You can also find the average prices for each week like this:
select t.WeekOf, avg(distinct t.Price)
from #tempTable t
join #tempTable t2 on t.ApartmentId = t2.ApartmentId and (t.WeekOf = t2.PreviousWeek or t.WeekOf = t2.NextWeek)
group by t.WeekOf
order by t.WeekOf
Here's a SQL Fiddle. I added a few more rows to the test data to show that it handles dates that cross the end of the year boundary.