Querying historical balance data from a transactions table - postgresql

Question

In need of help to get the total balances of customers on a daily basis if I backtracked the data.
I have the following table structures in a Postgres database:
Table1: accounts (acc)
|id|acc_created|
|1 |2019-01-01 |
|2 |2019-01-01 |
|3 |2019-01-01 |
Table2: transactions
|transaction_id|acc_id|balance|txn_created |
|1 |1 |100 |2019-01-01 07:00:00|
|2 |1 |50 |2019-01-01 16:32:10|
|3 |1 |25 |2019-01-01 22:10:59|
|4 |2 |200 |2019-01-02 18:34:22|
|5 |3 |150 |2019-01-02 15:09:43|
|6 |1 |125 |2019-01-04 04:52:31|
|7 |1 |0 |2019-01-05 05:10:00|
|8 |2 |300 |2019-01-05 12:34:56|
|9 |3 |120 |2019-01-06 23:59:59|
The transactions table shows the balance after a transaction is made on the account.
To be honest, I am unsure how to write the query, or whether I am overthinking the situation. I know it would involve last_value() and coalesce(), and possibly lag() and lead(). Essentially the criterias I would like to fulfill are:
It takes the last balance value of that day, for that account.
(i.e. the balance for acc_id = '1' on 2019-01-01 would be $25, acc_id ='2' and '3' would be $0)
For days where there are no transaction made by an account, the balance would take from the previous balance of that account.
(i.e. the balance for acc_id = '1' on 2019-01-03 would be $25)
Lastly, I would like the total balance of all accounts aggregated by date.
(i.e. At end of 2019-01-02, the total balance should be $375 (=25+200+150)
I have tried the query below:
SELECT date_trunc('day',date), sum(balance_of_day) FROM (
SELECT txn.created as date,
acc_id,
row_number() over (partition BY acc_id ORDER BY txn_created ASC) as order_of_created,
last_value(balance) over (partition by acc_id ORDER BY txn_created RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) as balance_of_day
FROM transactions) X
where X.order_of_created = 1
GROUP BY 1
However, this only gives me the total balance if a transaction was made by any account on a certain day.
The expected end result (based on the example) should be:
|date |total_balance|
|2019-01-01 |25 |
|2019-01-02 |375 |
|2019-01-03 |375 |
|2019-01-04 |475 |
|2019-01-05 |450 |
|2019-01-06 |420 |
I won't need to present the different account numbers, just the total accumulated balance from all customers at the end of the day. Please let me know how I can solve this! Many thanks!

Answer 1

You can use a few cool postgres feature to accomplish this. First, to get the last balance per day, use DISTINCT ON:
SELECT DISTINCT on(acc_id, txn_created::date)
transaction_id, acc_id, balance, txn_created::date as day
FROM transactions
ORDER BY acc_id, txn_created::date, txn_created desc;
To figure out the balance on any given day, we'll use a daterange per row that includes the current row and excludes the next row, partitioned by acc_id:
SELECT transaction_id, acc_id, balance, daterange(day, lead(day, 1) OVER (partition by acc_id order by day), '[)')
FROM (
SELECT DISTINCT on(acc_id, txn_created::date)
transaction_id, acc_id, balance, txn_created::date as day
FROM transactions
ORDER BY acc_id, txn_created::date, txn_created desc
) sub;
Lastly, join to generate_series. We can join where the date in generate_series is contained by the daterange we created in the last step. The dateranges are intentionally not overlapping, so we can query on any date safely.
WITH balances as (
SELECT transaction_id, acc_id, balance, daterange(day, lead(day, 1) OVER (partition by acc_id order by day), '[)') as drange
FROM (
SELECT DISTINCT on(acc_id, txn_created::date)
transaction_id, acc_id, balance, txn_created::date as day
FROM transactions
ORDER BY acc_id, txn_created::date, txn_created desc
) sub
)
SELECT d::date, sum(balance)
FROM generate_series('2019-01-01'::date, '2019-01-06'::date, '1 day') as g(d)
JOIN balances ON d::date <# drange
GROUP BY d::date;
d | sum
------------+-----
2019-01-01 | 25
2019-01-02 | 375
2019-01-03 | 375
2019-01-04 | 475
2019-01-05 | 450
2019-01-06 | 420
(6 rows)
Here's a fiddle.