Counting number of non-trading days/days without price changes - sql

Question

I have a table of closing prices for bonds over time, with the essential structure:
bond_id | tdate | price
---------+------------+-------
EIX1923 | 2014-01-01 | 100.12
EIX1923 | 2014-01-02 | 100.10
EIX1923 | 2014-01-05 | 100.10
EIX1923 | 2014-01-10 | 100.15
As you can see, I don't have prices for every day -- because the bond does not trade every day. I would like to count how often this occurs in a given year and, if the bond price hasn't changed between consecutive days, I take that as the same result.
That is, for a year with N trading days (excluding weekends, ignoring holidays), I would essentially want to generate a series of dates and count how many days the price is (1) unchanged from the previous day or (2) is not recored for that day and divide it over N.
I'm using PostgreSQL, so I started out with generate_series('2014-01-01'::timestamp, '2015-01-01'::timestamp, '1 day'::interval); I can SELECT from this series and do a WHERE to exclude weekends:
SELECT dd
FROM generate_series(
'2014-01-01'::timestamp,
'2015-01-01'::timestamp,
'1 day'::timestamp
) dd
WHERE EXTRACT(dow FROM dd) NOT IN (0, 6);
Now, I figure I would like to generate a "column" of bond_id to JOIN against the trade table with, but I'm not sure how. Essentially, I figured the simplest structure would be a LEFT JOIN so that I get something like:
EIX1923 | 2014-01-01 | 100.12
EIX1923 | 2014-01-02 | 100.10
EIX1923 | 2014-01-03 |
EIX1923 | 2014-01-04 |
EIX1923 | 2014-01-05 | 100.10
EIX1923 | 2014-01-06 |
EIX1923 | 2014-01-07 |
EIX1923 | 2014-01-08 |
EIX1923 | 2014-01-09 |
EIX1923 | 2014-01-10 | 100.15
Then I could just fill in the gaps with the most recently available price and count the number of ABS(∆P) == 0 in application code. But if there are solutions to do this entirely in SQL that would be nice too! I have no idea if the approach above is the right one to go with.
(I didn't bother to check if the first days of January 2014 are weekends or not, since it's just for illustration here; but they would be excluded from the results, obviously).
EDIT: Seems there might be a number of similar questions already. Hope it's not too much of a duplicate!
EDIT: So, I played a bit more with this and this solution "works" (and I feel silly for not realizing sooner) in the above sense:
SELECT
'EI653670', dd, t.price
FROM
generate_series('2014-01-01'::timestamp, '2015-01-01'::timestamp, '1 day'::interval) dd
LEFT JOIN
trade t ON dd = t.tdate AND t.id = 'EI653670'
WHERE
EXTRACT(dow FROM dd) NOT IN (0, 6) ORDER BY dd;
Is there a better way?

Answer 1

I think you can do this logic with lag(). The following show the general idea -- get the previous date and price and do some logic:
select bond_id,
sum(case when prev_price = price
then date - prev_date + 1
when prev_date = date + interval '1 day'
then 0
else date - prev_date
end)
from (select t.*,
lag(t.date) over (partition by t.bond_id order by t.date) as prev_date,
lag(t.price) over (partition by t.bond_id order by t.date) as prev_price
trade t
) t
group by bond_id;
One caveat is that this probably won't handle boundary conditions the way that you want.