Convert Parquet to Excel (XLSX) with DuckDB or Python

When CSV is not good enough

The quickest way into Excel is CSV — drop the file into the Parquet to CSV converter and open the result, as described in opening a Parquet file in Excel. That loses Parquet's types on the way, and Excel re-guesses them: 2026-03-04 may become April 3 depending on locale, leading zeros vanish, booleans turn into text. A real .xlsx keeps dates as date cells, numbers as numbers and booleans as booleans. Two tools produce one.

DuckDB: one COPY statement

Recent DuckDB releases ship an excel extension with an xlsx writer:

INSTALL excel;
LOAD excel;

COPY (SELECT * FROM 'orders.parquet')
TO 'orders.xlsx' WITH (FORMAT xlsx, HEADER true, SHEET 'orders');

From a shell it is one line:

duckdb -c "INSTALL excel; LOAD excel; COPY (SELECT * FROM 'orders.parquet') TO 'orders.xlsx' WITH (FORMAT xlsx, HEADER true);"

Details worth knowing, verified on DuckDB 1.5:

• HEADER defaults to false — forget it and the column names are gone.
• Dates, times and timestamps are written as Excel serial numbers with a date number format, so they behave as dates in Excel. TIMESTAMPTZ is converted to UTC and the zone is dropped.
• Booleans become TRUE/FALSE cells. Lists, structs and other non-numeric types are cast to text.
• SHEET_ROW_LIMIT defaults to 1,048,576, and the writer errors rather than truncating when a query exceeds it.

pandas: to_excel and its two errors

import pandas as pd

df = pd.read_parquet("orders.parquet")
df.to_excel("orders.xlsx", index=False)   # needs openpyxl or xlsxwriter

Error 1: timezones. Spark- and Arrow-written Parquet usually stores timestamps as timezone-aware, and pandas refuses:

ValueError: Excel does not support datetimes with timezones.
Please ensure that datetimes are timezone unaware before writing to Excel.

Convert every such column first:

for col in df.select_dtypes("datetimetz").columns:
    df[col] = df[col].dt.tz_convert(None)   # UTC, zone dropped

Use dt.tz_convert("Asia/Tokyo").dt.tz_localize(None) instead if the spreadsheet should show local time.

Error 2: too many rows. Past 1,048,576 rows pandas stops with ValueError: This sheet is too large!. Split across sheets:

LIMIT = 1_000_000
with pd.ExcelWriter("orders.xlsx", engine="xlsxwriter") as xw:
    for i, start in enumerate(range(0, len(df), LIMIT)):
        df.iloc[start:start + LIMIT].to_excel(xw, sheet_name=f"part{i + 1}", index=False)

xlsxwriter is considerably faster than openpyxl for large writes; either way, expect Excel output to be an order of magnitude slower than CSV, so filtering before exporting pays off.

Two type traps that survive the conversion

Long integer IDs. Excel keeps 15 significant digits in any number cell, xlsx included, so an int64 order ID like 4523891027365412789 gets its trailing digits zeroed. Cast such columns to text before exporting — CAST(order_id AS VARCHAR) in DuckDB or .astype(str) in pandas.

Nested columns. Lists and structs become one text cell per row in both tools. If you need their fields as columns, flatten first (see flattening nested Parquet columns).

Check the file before converting

Drop the Parquet file into the Parquet Viewer: the schema tab shows which columns are timestamps with a zone, which are nested and how many rows there are — all read from the footer in under a second. If the row count is above the limit, do not fight it: run an aggregate or a filter in the SQL Workbench and export the slice a spreadsheet can actually use.