Home / Pillar 3 Data Hub

Getting started with the p3dh-data endpoint

The p3dh endpoint gives you access to the Pillar 3 disclosures of hundreds of European banks, as published on the EBA's Pillar 3 Data Hub. These are the raw figures that the /v1/metrics/ endpoint computes its ratios from. This guide shows you how to access the underlying data.

Reach for /v1/p3dh-data/ instead of /v1/metrics/ when you're rolling your own ratio, running analytics that need a nominal amount rather than a computed metric, or loading the full dataset into a data warehouse for analysts across your organization to build on. If you're looking for a specific ratio that relies on a specific definition, please reach out to have it added to the /v1/metrics/ endpoint.

Dimensions

Every data point sits at the intersection of five things: a bank, a reference period, a table, a row, and a column.

Example

Let's say you're looking for Deutsche Bank's key metrics template:

python
url = API_URL + "v1/p3dh-data/"

data = requests.get(
    url=url,
    headers={"X-API-Key": API_KEY},
    params={"entity_name": "DEUTSCHE BANK AKTIENGESELLSCHAFT", "table_code": "K61.00"},
)

data.json()

Each record in the response is one cell:

python
{
  "dp_id": "dp143076",
  "entity_lei": "7LTWFZYICNSX8D621K86.CON",
  "entity_name": "DEUTSCHE BANK AKTIENGESELLSCHAFT",
  "country": "Germany",
  "ref_period": "2025-12-31",
  "base_currency": "EUR",
  "table_code": "K 61.00",
  "table_name": "EU KM1 - Key metrics template",
  "row_code": "0320",
  "row_label": "17. Liquidity coverage ratio (%)",
  "col_code": "0010",
  "col_label": "a. T",
  "value": 1.3722,
  "string_value": None,
  "value_type": "numeric",
  "unit": None
}

value holds numeric cells; string_value holds text cells (narrative disclosures, mostly). Check value_type before reading either. Note that exactly one of the two is populated per row, never both.

Other dimensions

You can also filter on entity_type (Banking Group, CRD Credit Institution, PILLAR3 Sub Group), institution_type (Large highest EEA, Other highest EEA, Large subsidiaries), and submission_date, useful if a bank has resubmitted a period and you want a specific vintage rather than the latest one. entity_lei, country, table_code, and ref_period all accept repeated query parameters, so you can compare several banks or periods in one call:

A bank can resubmit a disclosure module, and each resubmission gets its own row rather than overwriting the prior one. only_latest defaults to true: for each individual cell, only the row from its most-recently-submitted module is returned. Pass only_latest=false to get every submission's rows back instead, and only_latest is ignored when submission_date is set, since pinning an exact date is already more specific than "the latest one."

python
url = API_URL + "v1/p3dh-data/"

data = requests.get(
    url=url,
    headers={"X-API-Key": API_KEY},
    params={
        "entity_lei": ["7LTWFZYICNSX8D621K86.CON", "549300NYKK9MWM7GGW15.CON"],
        "table_code": "K71.00",
        "ref_period": ["2025-09-30", "2025-12-31"],
    },
)

Filtering by axis

A few templates can't be a plain row-by-column grid because the same cell repeats once per country, sector or exposure class (their row_code is * and row_label is null). The only thing that tells those rows apart is axis_value, for example qOOR=eba_GA:FR (France) in the geographical breakdown K 83.01.c/.d, or GPR=eba_GA:DE|NAC=eba_NC:I (Germany, NACE sector I) in the climate templates K 45.00.a/.b. It's a |-joined list of DIM=member parts, and it is returned as reported, not resolved to a label.

Use axis_dim to name the dimension and axis_member to pick one or more members (repeat the parameter to combine them):

python
url = API_URL + "v1/p3dh-data/"

data = requests.get(
    url=url,
    headers={"X-API-Key": API_KEY},
    params={
        "table_code": "K 83.01.c",
        "col_code": "0010",  # gross carrying amount
        "ref_period": "2025-12-31",
        "axis_dim": "qOOR",
        "axis_member": ["FR", "DE"],
    },
)

A few things to know:

Finding the table, row, and column you need

If you don't already know the table_code/row_code/col_code you're after, don't guess. Search for it. /v1/tables/ returns the full schema map (one record per unique table/row/column combination) and supports partial matching on labels:

python
url = API_URL + "v1/tables/"

data = requests.get(
    url=url,
    headers={"X-API-Key": API_KEY},
    params={"row_label": "non-performing"},
)

Or, if you don't know the bank or the template by name, /v1/search-p3dh/ searches both at once with a single keyword:

python
url = API_URL + "v1/search-p3dh/"

data = requests.get(
    url=url,
    headers={"X-API-Key": API_KEY},
    params={"keyword": "leverage"},
)

Authentication

All endpoints require authentication through an API key. You need to include it in the X-API-Key header, like this:

python
data = requests.get(
    url=url,
    headers={"X-API-Key": API_KEY},
    params={"metric_name": "cet1_ratio"},
)

Caps, pagination and offsets

Results default to limit=1000, capped at your key's per-request maximum number of rows. Page through a larger result set with offset.

limit and offset are query parameters, so they go in params, next to your filters. Only the API key goes in headers. The API doesn't read limit or offset from a header. If you put them there, you won't get an error, just the default first page (limit=1000, offset=0) on every call. offset is the number of rows to skip, not a page number: the second page of 1,000 rows starts at offset=1000, not offset=1.

python
url = API_URL + "v1/p3dh-data/"

data = requests.get(
    url=url,
    headers={"X-API-Key": API_KEY},  # the API key only
    params={
        "table_code": "K61.00",
        "limit": 1000,   # rows per page
        "offset": 1000,  # rows to skip: this is the second page
    },
)

Every response carries pagination state in its headers, so you don't have to guess whether there's more:

To fetch the whole result, loop until X-Has-More is false. Advance offset by X-Limit, not by the limit you asked for. If your key clamps a limit=5000 down to 1,000 and you advance by 5,000, you skip 4,000 rows each time and nothing tells you:

python
url = API_URL + "v1/p3dh-data/"
headers = {"X-API-Key": API_KEY}
params = {"table_code": "K61.00", "limit": 1000, "offset": 0}

rows = []
while True:
    response = requests.get(url=url, headers=headers, params=params)
    response.raise_for_status()
    rows.extend(response.json())
    if response.headers["X-Has-More"] != "true":
        break
    params["offset"] += int(response.headers["X-Limit"])

From raw cells to your own metric

Data points come back in long format: one row per cell. If you want to compute your own ratio, say a simple CET1 ratio from CET1 capital and total risk exposure, pull both rows and pivot to wide format in pandas:

python
import pandas as pd

url = API_URL + "v1/p3dh-data/"

data = requests.get(
    url=url,
    headers={"X-API-Key": API_KEY},
    params={
        "entity_lei": "7LTWFZYICNSX8D621K86.CON",  # Deutsche Bank AG
        "table_code": "K61.00",
        "ref_period": "2025-06-30",
        "col_code": "0010",  # "a. T": current-period column only, specific to LCR disclosures
    },
)

df = pd.DataFrame(data.json())

wide = df.pivot_table(
    index=["entity_name", "ref_period"],
    columns="row_label",
    values="value",
)

wide now has one row per bank/period and one column per line item, ready to combine into your own formula, or to compare against the equivalent pre-computed metric from /v1/metrics/ as a sanity check.

What's next