Introduction

Extracting month and year from a pandas datetime column is a small transformation that shows up everywhere in reporting and feature engineering. The tricky part is not the syntax. It is handling parsing, missing values, and timezone boundaries in a way that keeps the calendar fields trustworthy. A good implementation separates those concerns instead of slicing strings and hoping the input stays clean forever.

Parse the Column Before Extracting Anything

If the source column is still a string, convert it to datetime first. That keeps the parsing logic centralized and lets pandas deal with invalid inputs consistently.

1import pandas as pd
2
3df = pd.DataFrame(
4    {
5        "created_at": [
6            "2026-01-10 09:15:00",
7            "2026-02-03 14:05:00",
8            "bad-value",
9            None,
10        ]
11    }
12)
13
14df["created_at"] = pd.to_datetime(df["created_at"], errors="coerce")
15print(df)

Using errors="coerce" turns bad values into missing timestamps instead of crashing the pipeline. That is often the right default for analytics workflows, because it lets you continue processing while still preserving evidence that the input was malformed.

Extract Typed Year and Month Columns

Once the dtype is datetime, use the .dt accessor rather than string formatting. It is clearer, and the result stays numeric.

1df["year"] = df["created_at"].dt.year.astype("Int64")
2df["month"] = df["created_at"].dt.month.astype("Int64")
3
4print(df[["created_at", "year", "month"]])

The nullable integer dtype matters here. Without it, missing values often force the result into floating-point columns, which is awkward for grouping and validation.

Use a Monthly Key for Grouping

If the real goal is monthly aggregation, a month-level key is often more convenient than carrying separate numeric columns through the whole workflow.

1df["month_key"] = df["created_at"].dt.to_period("M")
2
3monthly = (
4    df.groupby("month_key", dropna=True)
5      .size()
6      .reset_index(name="rows")
7)
8
9print(monthly)

Period values keep month and year tied together, sort naturally, and reduce the chance of mistakes such as grouping on month number alone and accidentally mixing January from different years.

Keep Display Labels Separate From Logic Fields

Developers often create a label such as 2026-01 and then use that string for both display and computation. It is better to keep the typed month key for logic and add a formatted label only when needed.

df["label"] = df["created_at"].dt.strftime("%Y-%m")
print(df[["month_key", "label"]])

This gives you the best of both worlds: stable typed fields for grouping and friendly strings for export or chart labels.

Convert Timezones Before You Extract Calendar Parts

Timezone-aware timestamps need special care. An event near midnight UTC can belong to a different day, month, or even year in the business timezone you actually report on.

1events = pd.DataFrame(
2    {"ts": ["2026-03-31T23:30:00Z", "2026-04-01T00:30:00Z"]}
3)
4
5events["ts"] = pd.to_datetime(events["ts"], utc=True)
6events["local_ts"] = events["ts"].dt.tz_convert("America/Toronto")
7events["year_local"] = events["local_ts"].dt.year
8events["month_local"] = events["local_ts"].dt.month
9
10print(events)

If reporting is defined in local business time, convert to that timezone first and extract the calendar fields second. Reversing that order creates subtle month-boundary bugs that are hard to notice until totals stop matching expectations.

Wrap the Pattern in a Helper

If this transformation appears in multiple notebooks or jobs, put it behind one helper function. That keeps the parsing and output rules consistent.

1def add_calendar_columns(frame: pd.DataFrame, column: str) -> pd.DataFrame:
2    if column not in frame.columns:
3        raise ValueError(f"Missing column: {column}")
4
5    out = frame.copy()
6    out[column] = pd.to_datetime(out[column], errors="coerce")
7    out["year"] = out[column].dt.year.astype("Int64")
8    out["month"] = out[column].dt.month.astype("Int64")
9    out["month_key"] = out[column].dt.to_period("M")
10    return out

The code is small, but centralizing it prevents each caller from handling nulls, parsing, and month keys slightly differently.

Common Pitfalls

The biggest mistake is extracting month and year from raw strings instead of parsed timestamps. Another is ignoring missing values and ending up with float columns or unexpected null behavior. Teams also often use formatted strings as the only month field, or they extract calendar values before converting to the reporting timezone.

Summary

• Parse the datetime column first with pd.to_datetime.
• Use .dt.year and .dt.month for typed calendar extraction.
• Prefer a monthly Period key when grouping is the real goal.
• Keep presentation labels separate from computation fields.
• Convert to the business timezone before extracting month and year.