How to backfill FX rates without look-ahead bias

Use GET /v1/{date}/{base} to retrieve the rate as published on a specific past date, going back to 1999. The /v1/latest endpoint always returns today's value, so it has no place in a historical series. Check `is_forward_filled` on every row: when it is true, no rate was published that day and the prior value was carried forward.

What look-ahead bias means for FX data

Look-ahead bias occurs when your historical series contains a value that was not actually published on the date you are simulating. With FX rates this is easy to introduce accidentally: if you query /v1/latest/USD today and store the result against a past date, you are using today's rate in a historical slot. The data will appear self-consistent but it describes a world that did not exist on that date.

The fix is to query each date individually via the point-in-time path. The response reflects what was published that day, not a later revision.

The dated endpoint

Put the date in the path: GET /v1/{date}/{base}. The date must be ISO 8601 (YYYY-MM-DD) and the base is a three-letter currency code.

$ curl -s https://api.exchangerate.dev/v1/2026-03-14/USD \
    -H "Authorization: Bearer exr_live_..."

The response shape is the same as /v1/latest with one addition: is_forward_filled. When that field is true, no rate was published on the requested date and the API returned the most recent prior value.

Reading is_forward_filled

For reference-rate sources (ECB and Federal Reserve), rates are only published on business days. Requesting a Saturday, Sunday, or public holiday returns the last published value with is_forward_filled: true. The flag lets you treat those rows deliberately rather than silently treating a carried value as a freshly published one.

Python: looping dates to build a series

The example below requests USD rates for a week, printing each date's EUR rate and whether it was forward-filled. Throttle to stay within the 12-request-per-minute free-tier limit.

import time
from datetime import date, timedelta
import requests

API = "https://api.exchangerate.dev/v1"
KEY = "exr_live_..."   # free key: https://exchangerate.dev/signup

start = date(2026, 6, 9)
end   = date(2026, 6, 15)

current = start
while current <= end:
    r = requests.get(
        f"{API}/{current}/USD",
        headers={"Authorization": f"Bearer {KEY}"},
        timeout=10,
    )
    r.raise_for_status()
    data = r.json()

    eur  = data["rates"]["EUR"]
    ffwd = data.get("is_forward_filled", False)
    print(f"{current}  EUR={eur}  forward_filled={ffwd}")

    current += timedelta(days=1)
    time.sleep(5)  # stay under 12 req/min (free tier)

On June 14 (Saturday) and June 15 (Sunday) you will see forward_filled=True. Both are non-publication days for reference-rate sources.

Cross-checking with data_updated_at

The response also carries data_updated_at, the timestamp when the underlying rate was last written. On a forward-filled day this will point to the previous business day's publication time, giving you an additional signal about the age of the value you received.

data = r.json()
print(data["data_updated_at"])   # e.g. 2026-06-13T16:15:00Z on a weekend date
print(data["is_forward_filled"]) # True

A note on rate type