Cumulative Calculation and Running Totals

Overview

Cumulative calculation (Running Total / Cumulative Aggregation) is a high-frequency requirement in data analysis: calculating cumulative sales up to a point in time, computing rolling N-day averages, tracking how much a metric has changed from its starting point. Singdata Lakehouse implements these calculations through window function ORDER BY clauses and frame specifications (ROWS BETWEEN) without requiring self-joins or subqueries.

SQL Syntax Covered

Syntax	Purpose
`SUM(...) OVER (ORDER BY ...)`	Cumulative sum (default frame: from first row to current row)
`ROWS BETWEEN N PRECEDING AND CURRENT ROW`	Rolling N-row window
`ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW`	From partition start to current row
`ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING`	Entire partition
`FIRST_VALUE / LAST_VALUE`	Get first/last row value in window
`LAG(col, n)`	Get value of n rows before, used for period-over-period calculation

Prerequisite Data

The following examples use two test tables:


-- Daily sales details table
CREATE TABLE IF NOT EXISTS doc_running_sales (
  sale_date DATE,
  region    VARCHAR(20),
  amount    DECIMAL(10,2)
);

INSERT INTO doc_running_sales VALUES
  (CAST('2024-01-01' AS DATE), 'East', 1200.00),
  (CAST('2024-01-02' AS DATE), 'East',  850.00),
  (CAST('2024-01-03' AS DATE), 'East', 1500.00),
  (CAST('2024-01-04' AS DATE), 'East',  600.00),
  (CAST('2024-01-05' AS DATE), 'East', 2100.00),
  (CAST('2024-01-01' AS DATE), 'West',  900.00),
  (CAST('2024-01-02' AS DATE), 'West', 1100.00),
  (CAST('2024-01-03' AS DATE), 'West',  750.00),
  (CAST('2024-01-04' AS DATE), 'West', 1300.00),
  (CAST('2024-01-05' AS DATE), 'West',  980.00);

-- Monthly revenue table
CREATE TABLE IF NOT EXISTS doc_running_monthly (
  month_date DATE,
  category   VARCHAR(20),
  revenue    DECIMAL(12,2)
);

INSERT INTO doc_running_monthly VALUES
  (CAST('2024-01-01' AS DATE), 'Apparel',   50000.00),
  (CAST('2024-02-01' AS DATE), 'Apparel',   45000.00),
  (CAST('2024-03-01' AS DATE), 'Apparel',   62000.00),
  (CAST('2024-04-01' AS DATE), 'Apparel',   58000.00),
  (CAST('2024-01-01' AS DATE), 'Electronics', 120000.00),
  (CAST('2024-02-01' AS DATE), 'Electronics', 135000.00),
  (CAST('2024-03-01' AS DATE), 'Electronics',  98000.00),
  (CAST('2024-04-01' AS DATE), 'Electronics', 145000.00);

Scenario 1: Cumulative Sum by Partition

Calculate the cumulative daily sales for each region.


SELECT
  sale_date,
  region,
  amount,
  SUM(amount) OVER (
    PARTITION BY region
    ORDER BY sale_date
  ) AS running_total
FROM doc_running_sales
ORDER BY region, sale_date;

Result:

sale_date	region	amount	running_total
2024-01-01	East	1200.00	1200.00
2024-01-02	East	850.00	2050.00
2024-01-03	East	1500.00	3550.00
2024-01-04	East	600.00	4150.00
2024-01-05	East	2100.00	6250.00
2024-01-01	West	900.00	900.00
2024-01-02	West	1100.00	2000.00
2024-01-03	West	750.00	2750.00
2024-01-04	West	1300.00	4050.00
2024-01-05	West	980.00	5030.00

Without ROWS BETWEEN, the default frame for ORDER BY is ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW, i.e., from the partition start to the current row.

Scenario 2: Cumulative Proportion of Total (YTD Progress)

Calculate each category's monthly YTD revenue and its proportion of total annual revenue.


SELECT
  month_date,
  category,
  revenue,
  SUM(revenue) OVER (
    PARTITION BY category
    ORDER BY month_date
  )                                                          AS ytd_revenue,
  SUM(revenue) OVER (PARTITION BY category)                 AS annual_total,
  ROUND(
    SUM(revenue) OVER (PARTITION BY category ORDER BY month_date)
    / SUM(revenue) OVER (PARTITION BY category) * 100,
    2
  )                                                          AS ytd_pct
FROM doc_running_monthly
ORDER BY category, month_date;

Result:

month_date	category	revenue	ytd_revenue	annual_total	ytd_pct
2024-01-01	Apparel	50000.00	50000.00	215000.00	23.26
2024-02-01	Apparel	45000.00	95000.00	215000.00	44.19
2024-03-01	Apparel	62000.00	157000.00	215000.00	73.02
2024-04-01	Apparel	58000.00	215000.00	215000.00	100.00
2024-01-01	Electronics	120000.00	120000.00	498000.00	24.10
2024-02-01	Electronics	135000.00	255000.00	498000.00	51.20
2024-03-01	Electronics	98000.00	353000.00	498000.00	70.88
2024-04-01	Electronics	145000.00	498000.00	498000.00	100.00

Scenario 3: Rolling N-Row Window (Moving Average)

Calculate a 3-day moving average of sales for each region.


SELECT
  sale_date,
  region,
  amount,
  AVG(amount) OVER (
    PARTITION BY region
    ORDER BY sale_date
    ROWS BETWEEN 2 PRECEDING AND CURRENT ROW
  ) AS moving_avg_3day
FROM doc_running_sales
ORDER BY region, sale_date;

Result:

sale_date	region	amount	moving_avg_3day
2024-01-01	East	1200.00	1200.000000
2024-01-02	East	850.00	1025.000000
2024-01-03	East	1500.00	1183.333333
2024-01-04	East	600.00	983.333333
2024-01-05	East	2100.00	1400.000000

ROWS BETWEEN 2 PRECEDING AND CURRENT ROW means the current row and the preceding 2 rows, for a total of 3 rows. At the partition boundary where there are fewer than 3 rows, the average uses only the available rows.

Scenario 4: Aggregate First, Then Roll (Two-Level Window)

First aggregate total daily sales across all regions, then calculate a 7-day rolling sum.


SELECT
  sale_date,
  SUM(amount)                                                    AS daily_total,
  SUM(SUM(amount)) OVER (
    ORDER BY sale_date
    ROWS BETWEEN 6 PRECEDING AND CURRENT ROW
  )                                                              AS rolling_7day_sum,
  AVG(SUM(amount)) OVER (
    ORDER BY sale_date
    ROWS BETWEEN 6 PRECEDING AND CURRENT ROW
  )                                                              AS rolling_7day_avg
FROM doc_running_sales
GROUP BY sale_date
ORDER BY sale_date;

Result:

sale_date	daily_total	rolling_7day_sum	rolling_7day_avg
2024-01-01	2100.00	2100.00	2100.000000
2024-01-02	1950.00	4050.00	2025.000000
2024-01-03	2250.00	6300.00	2100.000000
2024-01-04	1900.00	8200.00	2050.000000
2024-01-05	3080.00	11280.00	2256.000000

When applying window functions over aggregated results, the window function parameter must be an aggregate expression (SUM(SUM(amount))), not a plain column name.

Scenario 5: Cumulative Maximum / Minimum

Track the historical highest and lowest single-day sales for each region.


SELECT
  sale_date,
  region,
  amount,
  MAX(amount) OVER (
    PARTITION BY region
    ORDER BY sale_date
    ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
  ) AS running_max,
  MIN(amount) OVER (
    PARTITION BY region
    ORDER BY sale_date
    ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
  ) AS running_min
FROM doc_running_sales
ORDER BY region, sale_date;

Result (East region only):

sale_date	region	amount	running_max	running_min
2024-01-01	East	1200.00	1200.00	1200.00
2024-01-02	East	850.00	1200.00	850.00
2024-01-03	East	1500.00	1500.00	850.00
2024-01-04	East	600.00	1500.00	600.00
2024-01-05	East	2100.00	2100.00	600.00

Scenario 6: Period-over-Period Change (MoM / DoD)

Calculate the month-over-month change amount and change rate for each category.


SELECT
  month_date,
  category,
  revenue,
  LAG(revenue, 1) OVER (
    PARTITION BY category
    ORDER BY month_date
  )                                                              AS prev_month,
  revenue - LAG(revenue, 1) OVER (
    PARTITION BY category
    ORDER BY month_date
  )                                                              AS mom_change,
  ROUND(
    (revenue - LAG(revenue, 1) OVER (PARTITION BY category ORDER BY month_date))
    / LAG(revenue, 1) OVER (PARTITION BY category ORDER BY month_date) * 100,
    2
  )                                                              AS mom_pct
FROM doc_running_monthly
ORDER BY category, month_date;

Result (Apparel category):

month_date	category	revenue	prev_month	mom_change	mom_pct
2024-01-01	Apparel	50000.00	NULL	NULL	NULL
2024-02-01	Apparel	45000.00	50000.00	-5000.00	-10.00
2024-03-01	Apparel	62000.00	45000.00	17000.00	37.78
2024-04-01	Apparel	58000.00	62000.00	-4000.00	-6.45

The first row has no prior period data, so LAG returns NULL and the period-over-period calculation also yields NULL. Use COALESCE(LAG(...), 0) to replace with 0 if needed.

Scenario 7: Change from Starting Point (FIRST_VALUE)

Calculate each region's daily sales change relative to the first day, and the final day's sales.


SELECT
  sale_date,
  region,
  amount,
  FIRST_VALUE(amount) OVER (
    PARTITION BY region
    ORDER BY sale_date
  )                                                              AS first_day_amount,
  LAST_VALUE(amount) OVER (
    PARTITION BY region
    ORDER BY sale_date
    ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
  )                                                              AS last_day_amount,
  amount - FIRST_VALUE(amount) OVER (
    PARTITION BY region
    ORDER BY sale_date
  )                                                              AS change_from_start
FROM doc_running_sales
ORDER BY region, sale_date;

Result (East region only):

sale_date	region	amount	first_day_amount	last_day_amount	change_from_start
2024-01-01	East	1200.00	1200.00	2100.00	0.00
2024-01-02	East	850.00	1200.00	2100.00	-350.00
2024-01-03	East	1500.00	1200.00	2100.00	300.00
2024-01-04	East	600.00	1200.00	2100.00	-600.00
2024-01-05	East	2100.00	1200.00	2100.00	900.00

LAST_VALUE's default frame only goes up to the current row. You must explicitly specify ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING to get the value from the partition's last row.

Important Notes

Frame Specification Defaults: When ORDER BY is present, the default frame is RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW (value-range based). For date columns with duplicate dates, RANGE and ROWS may produce different results. Use explicit ROWS BETWEEN when precise control is needed.
Two-Level Windows: When applying window functions over GROUP BY aggregated results, the window function argument must wrap the aggregate function (e.g., SUM(SUM(col))), otherwise an error will occur.
LAST_VALUE Trap: Without ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING, LAST_VALUE only returns the current row's value (because the default frame stops at the current row).
NULL Handling: LAG / LEAD return NULL at boundaries. Use COALESCE(LAG(col, 1), 0) to substitute default values.