Calculating a rolling 30 day average with multiple values per day

Question

I have a table that looks something like this:

city	date	value
DC	2020-01-01	10
DC	2020-01-01	23
DC	2020-01-02	43
NYC	2020-01-01	43
NYC	2020-01-02	23
NYC	2020-01-03	10

There are multiple values per city, per date. I'm having a hard time calculating the forward looking 30 day average because the condition in the rolling average should be based on the date and not based on the number of rows. I want the output to be something like this:

city	date	value
DC	2020-01-01	25.33
DC	2020-01-02	43
NYC	2020-01-01	25.33
NYC	2020-01-02	16.5
NYC	2020-01-03	10

So I can't do something like this:

AVG(value) OVER (
           PARTITION BY city, date
           ORDER BY date DESC
           ROWS BETWEEN 31 PRECEDING AND 1 PRECEDING
           )

joining the table on itself like this:

SELECT t1.city, t1.date, avg(t2.values)
FROM table1 t1
JOIN table1 t2 ON t1.city=t2.city 
    AND t2.date < DATEADD('day',31,t1.date)
    AND t2.date >= t1.date

isn't an option because it's quite large and takes forever. How do I do this?

Answer 1

You want to use RANGE instead of ROWS in your window frame. But that requires ordering by a number. So you need to do:

over (... order by datediff(date,'1970-01-01') asc range between 31 preceding and 1 following)

(Though I think you want 0 following?)