Date Ranges, Frequencies and Shifting

Why fixed frequencies matter

A real timestamp column (like order_time from the previous page) is often irregular — orders don't arrive exactly every hour. But for analysis you frequently want a fixed-frequency index: every day, every business day, every 15 minutes. Pandas gives you two tools for this:

• pd.date_range — generate a clean, evenly spaced DatetimeIndex.
• .shift() — move values (or timestamps) forward/backward along that index.

Think of date_range as drawing the ruler, and shift as sliding your data along it.

Generating date ranges with date_range

import pandas as pd
 
# Start + end (daily by default)
idx = pd.date_range("2024-01-01", "2024-01-06")
print(idx)
 
# Start + number of periods
idx2 = pd.date_range(start="2024-01-01", periods=5)
print(idx2)
 
# Business days only (skips weekends)
idx3 = pd.date_range("2024-01-01", periods=5, freq="B")
print(idx3)

date_range always needs at least two of: start, end, periods. Give it any two and it works out the rest.

Frequency strings and multiples

Frequencies combine a base alias with an optional multiplier:

import pandas as pd
 
print(pd.date_range("2024-01-01", periods=4, freq="4h"))    # every 4 hours
print(pd.date_range("2024-01-01", periods=4, freq="5min"))  # every 5 minutes
print(pd.date_range("2024-01-01", periods=4, freq="ME"))    # last day of each month

Shifting data: leading and lagging

shift moves the values, keeping the index fixed — this creates NaN at the start or end, which is exactly what you want for lag/lead features:

import pandas as pd
 
ts = pd.Series(
    [100, 105, 102, 108],
    index=pd.date_range("2024-01-31", periods=4, freq="ME"),
)
 
print(ts.shift(1))   # lag by 1 period — value moves down
print(ts.shift(-1))  # lead by 1 period — value moves up

A very common use of shift is computing period-over-period percent change:

pct_change = ts / ts.shift(1) - 1
print(pct_change)

If you pass a freq, shift moves the timestamps instead of the values, so no data is discarded:

print(ts.shift(2, freq="ME"))  # every date pushed forward by 2 months

Rolling dates forward with offsets

Pandas offset objects understand calendar rules. Adding MonthEnd() to any date "rolls forward" to the next month end — handy for snapping arbitrary dates onto a frequency grid:

import pandas as pd
from pandas.tseries.offsets import MonthEnd, Day
 
now = pd.Timestamp("2024-01-17")
 
print(now + 3 * Day())        # simple: 3 calendar days later
print(now + MonthEnd())       # rolls forward to 2024-01-31
print(now + MonthEnd(2))      # rolls forward 2 month-ends: 2024-02-29
 
offset = MonthEnd()
print(offset.rollforward(now))  # 2024-01-31
print(offset.rollback(now))     # 2023-12-31 (previous month end)

Common pitfalls

• Naive shift() (no freq) introduces NaN at the start or end — that's expected, not a bug.
• Legacy aliases "M", "T", "H", "A" are no longer accepted — use "ME", "min", "h", "YE".
• date_range needs exactly two of start / end / periods — passing all three (when inconsistent) or only one raises an error.

Visualize it

  flowchart LR
  A["pd.date_range(start, periods, freq)"] --> B["Fixed-frequency DatetimeIndex"]
  B --> C["Series/DataFrame on that index"]
  C --> D["ts.shift(1)
values slide down, NaN at top"]
  C --> E["ts.shift(1, freq='ME')
timestamps slide, no NaN"]

🧪 Try It Yourself

Exercise 1 – Build a Fixed-Frequency Range

Press Run to see what your code prints.

Exercise 2 – Lag a Series With shift

Press Run to see what your code prints.

Exercise 3 – Roll a Date Forward to Month End

Press Run to see what your code prints.

Check your understanding

Next

Fixed-frequency ranges become tricky once you cross regions — see Time Zone Handling for how to localize and convert timestamps safely.