Get rowset based on distinct combination of columns - sql

Question

Given this dataset, each stock has a yearly snapshot of value.
+----+------+------+-------+-------+
| ID | Name | Year | Stock | Value |
+----+------+------+-------+-------+
| 1 | John | 2019 | ABC | 123 |
| 1 | John | 2020 | ABC | 123 |
| 1 | John | 2021 | ABC | 123 |
| 1 | John | 2021 | XYZ | 200 |
| 1 | John | 2022 | ABC | 123 |
| 1 | John | 2022 | XYZ | 200 |
| 1 | John | 2023 | ABC | 630 |
| 1 | John | 2023 | XYZ | 200 |
+----+------+------+-------+-------+
In 2019, John only holds ABC with a value of 123
In 2020, John also only holds ABC, with a value of 123 (has not changed)
In 2021, John holds ABC but has also acquired XYZ, with a value of 200
in 2022, John holds ABC and XYZ, both of which values haven't changed.
In 2023, John holds ABC and XYZ, with ABC's value increasing to 630 and XYZ's value remaining at 200.
I would like to return rows so that
Per year, if nothing of John's portfolio has changed SINCE THE LAST YEAR, no rows are returned
If anything in John's portfolio has changed SINCE THE LAST YEAR, all his current holdings are listed
For example,
+----+------+------+-------+-------+
| ID | Name | Year | Stock | Value |
+----+------+------+-------+-------+
| 1 | John | 2019 | ABC | 123 |
| 1 | John | 2021 | ABC | 123 |
| 1 | John | 2021 | XYZ | 200 |
| 1 | John | 2023 | ABC | 630 |
| 1 | John | 2023 | XYZ | 200 |
+----+------+------+-------+-------+
How would I do this, whether it be through functions in PL/SQL or in pure SQL?

Answer 1

If there are not too many rows per user, then listagg() provides a convenient solution:
select ny.*
from (select name, year,
listagg(stock || ':' || value, ',') within group (order by stock) as stocks,
lag(listagg(stock || ':' || value, ',') within group (order by stock)) as prev_stocks,
lag(year) over (partition by name order by year) as prev_year
from t
group by name, year
) ny
where prev_year is null or prev_year <> year - 1 or prev_stocks <> stocks;
Alternatively, you can check each row individually and use an analytic function to project the information over all rows in a name/year:
select t.*
from (select t.*,
sum(case when prev_nsv_year = year then 0 else 1 end) over (partition by name, year) as num_diff,
lag(cnt) over (partition by name order by year) as prev_cnt
from (select t.*,
lag(year) over (partition by name, stock, value over order by year) as prev_nsv_year,
count(*) over (partition by name, year) as cnt
from t
) t
) t
where cnt <> prev_cnt or prev_cnt is null or
num_diff > 0;