Resampling, frequencies and Indian financial years

The canteen’s 313 recorded days become four quarterly totals.

Resamplingresampling: Grouping time-indexed observations into new time bins, or introducing new timestamps when changing frequency; the labels and aggregation define the result. changes the frequency of a time series. Downsampling turns many short intervals into fewer long ones, such as days into weeks, months or quarters, and aggregates each bin with a sum, a mean or a count; upsampling goes the other way and creates empty slots that something must fill. In pandas, sales.resample("ME").sum() works like a group-by on time. Its rules are frequencies written as short strings, and pandas 3 removed the old spellings "M", "Q", "Y" and "H", which now raise a ValueError. Indian financial years run from April to March. "QE-MAR" groups quarter-end bins; "Q-MAR" Periods supply April-based quarter numbers and the financial year’s ending year. A bin covers a stretch of time whether or not any rows fell into it, which is why resampling also exposes missing days.

Choose bins before adding canteen sales

A time bintime bin: A time interval used to place observations in a reporting group. Its endpoints, included boundaries and label are part of the metric definition. has a boundary rule and a label, both chosen before the sum.

The examples use one financial year of synthetic sales from a college canteen that is closed on Sundays, so 52 days of the year have no row at all.

"""Daily canteen sales for one financial year, April to March, closed on Sundays (shared by the examples)."""
import numpy as np
import pandas as pd


def canteen_sales():
    days = pd.date_range("2025-04-01", "2026-03-31", freq="D")
    days = days[days.dayofweek != 6]  # closed on Sundays: those days have no row at all
    rng = np.random.default_rng(26)
    return pd.Series(rng.integers(2800, 5200, len(days)), index=days, name="sales")

"""One financial year of daily sales, resampled to weeks, months and financial-year quarters."""
from canteen import canteen_sales

sales = canteen_sales()
print(f"{len(sales)} days with sales, from {sales.index[0]:%a %d %b %Y} to {sales.index[-1]:%a %d %b %Y}")

weeks = sales.resample("W").agg(["sum", "count"])
print("\nFirst weeks ('W': weeks end on Sunday and are labelled with it):")
print(weeks.head(3).to_string())

months = sales.resample("ME").sum()
print("\nMonths ('ME': labelled with the last day of the month):")
print(months.head(4).to_string())

quarters = sales.resample("QE-MAR").sum()
fy = quarters.index.to_period("Q-MAR")
print("\nQuarters of the financial year ('QE-MAR'):")
for end, period, total in zip(quarters.index, fy, quarters):
    print(f"{end:%Y-%m-%d}  {period}  FY{period.qyear - 1}-{period.qyear % 100:02d} Q{period.quarter}  {total:,}")
print("\nYear total:", f"{sales.sum():,}", "| quarters add up:", quarters.sum() == sales.sum())

313 days with sales, from Tue 01 Apr 2025 to Tue 31 Mar 2026

First weeks ('W': weeks end on Sunday and are labelled with it):
              sum  count
2025-04-06  20079      5
2025-04-13  24486      6
2025-04-20  26152      6

Months ('ME': labelled with the last day of the month):
2025-04-30    106435
2025-05-31    110839
2025-06-30    104841
2025-07-31    108984
Freq: ME

Quarters of the financial year ('QE-MAR'):
2025-06-30  2026Q1  FY2025-26 Q1  322,115
2025-09-30  2026Q2  FY2025-26 Q2  317,095
2025-12-31  2026Q3  FY2025-26 Q3  312,439
2026-03-31  2026Q4  FY2025-26 Q4  317,522

Year total: 1,269,171 | quarters add up: True

"W" makes weeks that end on Sunday and labels each with that Sunday. The first week holds 5 days, not 6, because the financial year starts on a Tuesday. Calendar rules choose the bin boundaries; count measures the usable observations supplied within them. "ME" labels each month with its last day, and "MS" would label the same months with their first day. The four quarters of "QE-MAR" end in June, September, December and March, and their totals add up to the year, a check worth keeping in every report. resample needs a datetime index, or a datetime column named with on="date"; on a plain numbered index it raises a TypeError. After the rule, any aggregation that a group-by accepts works, including agg with a list of functions.

Look at the first weekly count and the four quarter totals in this walk-through:

Use offset aliases for time-point bins

pandas 3 enforced a rename of several frequencies that pandas 2 had deprecated. Old code fails loudly, which is better than failing quietly:

"""The frequency strings pandas 3 removed, what it says, and what to write instead."""
import re

import pandas as pd

from canteen import canteen_sales

sales = canteen_sales()


def reason(error):
    """The useful part of pandas' message."""
    found = re.search(r"'[\w-]+' is no longer supported for offsets\. Please use '[\w-]+' instead\.|Did you mean [\w-]+\?|for Period, please use '[\w-]+' instead of '[\w-]+'", str(error))
    return found.group(0) if found else str(error)


print("resample() takes offset aliases:")
for old, new in [("M", "ME"), ("Q", "QE"), ("Y", "YE"), ("H", "h"), ("Q-MAR", "QE-MAR")]:
    try:
        sales.resample(old).sum()
    except ValueError as error:
        print(f"  {old!r:9} ValueError: {reason(error)}")
    print(f"  {new!r:9} {len(sales.resample(new).sum())} bins")

print("\nPeriods take period aliases:")
day = sales.index[200]
for freq in ["QE-MAR", "Q-MAR"]:
    try:
        print(f"  {freq!r:9} {pd.Period(day, freq=freq)}")
    except ValueError as error:
        print(f"  {freq!r:9} ValueError: {reason(error)}")

resample() takes offset aliases:
  'M'       ValueError: 'M' is no longer supported for offsets. Please use 'ME' instead.
  'ME'      12 bins
  'Q'       ValueError: 'Q' is no longer supported for offsets. Please use 'QE' instead.
  'QE'      4 bins
  'Y'       ValueError: 'Y' is no longer supported for offsets. Please use 'YE' instead.
  'YE'      2 bins
  'H'       ValueError: Did you mean h?
  'h'       8737 bins
  'Q-MAR'   ValueError: 'Q-MAR' is no longer supported for offsets. Please use 'QE-MAR' instead.
  'QE-MAR'  4 bins

Periods take period aliases:
  'QE-MAR'  ValueError: for Period, please use 'Q-MAR' instead of 'QE-MAR'
  'Q-MAR'   2026Q3

The two columns are the trap. Offsets mark points on the time line, so they say whether a bin is labelled by its end (E) or its start (S). Periods are spans of time with no such choice, so they keep the short names. A financial-year quarter is therefore "QE-MAR" in resample and "Q-MAR" in to_period, and each function refuses the other’s spelling with a message that names the right one.

Walk through the documented change and the current native behavior; keep the span notation separate:

Number fiscal quarters from April

Reports by Indian financial years name the year by both calendar years it touches, such as FY2025-26, and number its quarters from April: Q1 is April to June, and Q4 is January to March. The Reserve Bank of India labels quarters the same way in its policy statements, for example Q1:2027-28 for the first quarter of the 2027-28 year.

pandas has the building blocks but its own names. to_period("Q-MAR") turns each day into a quarter period, and the period’s qyear is the calendar year in which its financial year ends, so every quarter of FY2025-26 is labelled 2026, from 2026Q1 to 2026Q4. Its quarter is already the financial-year quarter. The label a finance team reads is one f-string away: f"FY{p.qyear - 1}-{p.qyear % 100:02d} Q{p.quarter}". Build labels from qyear, never from the calendar year of the date, which gives January to March the wrong year. Labels in this form also sort correctly as text, because the starting year comes first. Timestamps stored in UTC need converting to IST before the period is taken: an order placed at 20:00 UTC on the last day of March is already in April on the IST clock, and so in the next financial year. The quarter-end dates of plain QE and QE-MAR coincide, because December and March anchors share the same three-month boundaries. Their fiscal numbering differs: use Q-MAR periods for the ending year and quarter, rather than treating an end-date label as a financial-year label. Compare both representations in this run:

"""The same canteen bin boundaries, with explicit frequency and Period labels."""
import pandas as pd

from canteen import canteen_sales

sales = canteen_sales()
calendar = sales.resample("QE").sum()
fiscal = sales.resample("QE-MAR").sum()
print("Values and quarter-end timestamps match:", calendar.equals(fiscal))
print("Frequency metadata:", calendar.index.freqstr, "and", fiscal.index.freqstr)
print("\nThe same quarter ends labelled as calendar and fiscal periods:")
for end, total in fiscal.items():
    print(f"{end:%Y-%m-%d}: {total:,} | calendar {end.to_period('Q')} | fiscal {end.to_period('Q-MAR')}")
print("\nPeriod aliases are a separate vocabulary:")
for alias in ["M", "ME", "Q-MAR", "QE-MAR", "Y-MAR", "YE-MAR"]:
    try:
        print(alias, "->", pd.Period(sales.index[0], freq=alias))
    except ValueError as error:
        print(alias, "->", type(error).__name__, str(error).split(".")[0])

Values and quarter-end timestamps match: True
Frequency metadata: QE-DEC and QE-MAR

The same quarter ends labelled as calendar and fiscal periods:
2025-06-30: 322,115 | calendar 2025Q2 | fiscal 2026Q1
2025-09-30: 317,095 | calendar 2025Q3 | fiscal 2026Q2
2025-12-31: 312,439 | calendar 2025Q4 | fiscal 2026Q3
2026-03-31: 317,522 | calendar 2026Q1 | fiscal 2026Q4

Period aliases are a separate vocabulary:
M -> 2025-04
ME -> ValueError Invalid frequency: ME
Q-MAR -> 2026Q1
QE-MAR -> ValueError Invalid frequency: QE-MAR
Y-MAR -> 2026
YE-MAR -> ValueError Invalid frequency: YE-MAR

Follow the same reporting dates through fiscal numbering, then compare the Period labels.

Bin edges in depth: weeks, business days and clock changes Pro: opens with a Pro, Premium or Ultimate pass

Financial-year reports: quarters, halves and year to date, with numbers Pro: opens with a Pro, Premium or Ultimate pass

Gaps that resampling exposes: closed, missing or zero Pro: opens with a Pro, Premium or Ultimate pass

Practise financial-year totals

The exercise builds the same labels with the standard library, which also runs in the browser.

from datetime import date


def fy_quarter(day):
    """The financial-year quarter of a date, such as "FY2026-27 Q3" for date(2026, 10, 4)."""
    # Replace this line with your code.
    return ""


def revenue_by_fy_quarter(rows):
    """{"FY2025-26 Q4": 2500, "FY2026-27 Q1": 1200, ...} in calendar order, with 0 for quarters without sales.

    rows is a list of (day, amount) pairs, such as ("2026-04-02", 1200).
    """
    # Replace this line with your code.
    return {}

from datetime import date

from fy_quarters import fy_quarter, revenue_by_fy_quarter


def test_quarter_labels():
    """labels the quarters of a financial year that starts in April"""
    assert fy_quarter(date(2026, 4, 1)) == "FY2026-27 Q1"
    assert fy_quarter(date(2026, 10, 4)) == "FY2026-27 Q3"
    assert fy_quarter(date(2027, 3, 31)) == "FY2026-27 Q4"


def test_january_to_march():
    """puts January to March in the financial year that started the April before"""
    assert fy_quarter(date(2026, 3, 31)) == "FY2025-26 Q4"
    assert fy_quarter(date(2026, 1, 1)) == "FY2025-26 Q4"


def test_two_digit_end_year():
    """writes the end year with two digits, also at the turn of a century"""
    assert fy_quarter(date(2099, 12, 1)) == "FY2099-00 Q3"


def test_totals_across_a_year_boundary():
    """adds up each quarter, across the start of a new financial year, in calendar order"""
    rows = [("2026-04-02", 1200), ("2026-02-27", 900), ("2026-03-31", 1600), ("2026-06-30", 300)]
    result = revenue_by_fy_quarter(rows)
    assert result == {"FY2025-26 Q4": 2500, "FY2026-27 Q1": 1500}
    assert list(result) == ["FY2025-26 Q4", "FY2026-27 Q1"]


def test_empty_quarters_are_zero():
    """gives a quarter without sales a total of 0, as resample does"""
    rows = [("2026-05-10", 400), ("2027-01-15", 250)]
    assert list(revenue_by_fy_quarter(rows).items()) == [
        ("FY2026-27 Q1", 400),
        ("FY2026-27 Q2", 0),
        ("FY2026-27 Q3", 0),
        ("FY2026-27 Q4", 250),
    ]

Interview questions

Key takeaways

• resample(rule) groups rows into bins of time and aggregates each bin; the bins cover the calendar, so count the rows in each.
• pandas 3 refuses M, Q, Y and H: write ME or MS, QE or QS, YE or YS, and h.
• Offsets end in E or S (QE-MAR for resample); periods do not (Q-MAR for to_period).
• An Indian financial year runs from April to March; build its labels from qyear, which names the year in which it ends.