Leakage-safe CMBS special-servicing-transfer timeline dataset for the job: inspect, prompt, and model CMBS asset histories that may transfer to special servicing within the next 12 months, using only point-in-time information available from SEC-filed CMBS loan reports.

This release is a timeline-oriented companion to the flat supervised learning table. Each row is one asset-period observation with a compact natural-language timeline_text field, numeric state fields, target label, split, and SEC filing provenance.

import pandas as pd

train = pd.read_parquet("train.parquet")
test = pd.read_parquet("test.parquet")

Listing Terms

Contact: cairn@cmdrvl.com

License: CC-BY-NC-4.0. This dataset is open source for non-commercial use only.

Commercial use: Snowflake Marketplace listing coming soon, contact cairn@cmdrvl.com.

Files

File	Rows	Positives	Positive rate	Notes
`train.parquet`	497,552	10,825	2.18%	`reporting_period_end_date <= 2022-12-31`
`test.parquet`	299,401	6,581	2.20%	`reporting_period_end_date >= 2023-07-01`
`all.parquet`	796,953	17,406	2.18%	Combined file with `split` column

The six-month embargo window from 2023-01-01 through 2023-06-30 is excluded from all published files. Rows whose full 12-month forward label window is not yet observable are also excluded rather than shipped as negatives.

Grain

One row is one CMBS asset observation at:

cik
loannumber
assetnumber
reporting_period_end_date

loannumber is descriptive. assetnumber is part of the row grain.

Label And Split

Target column: transfers_to_special_servicing_within_12m

The target is 1 when the asset's first observed special-servicer transfer date occurs after the observation period end date and within the next 12 months. Rows on or after that asset-level first transfer date are dropped before modeling, so the timeline contains pre-transfer observations only.

Split policy:

Train: reporting periods on or before 2022-12-31.
Embargo: 2023-01-01 through 2023-06-30, excluded.
Test: reporting periods on or after 2023-07-01.

The split is intentionally temporal, not random. It asks models to generalize from the 2020-2022 regime into the 2023+ reporting regime. The embargo is a buffer band discarded between train and test so the two sets do not touch at the boundary.

Timeline Text

timeline_text is a compact per-period rendering suitable for retrieval, review queues, and language-model prompts. Example:

Loan 1 asset 1 in deal CIK 123456 reported 2022-09-30 with Current payment status, Not Reported workout strategy, appointed special servicer Not Reported, and actual balance 850000.

Prompt pattern:

You are reviewing CMBS asset-period histories for special-servicing transfer risk. Given the timeline rows below, identify changes in payment status, workout strategy, balance, delinquency streak, and servicer evidence that could explain near-term transfer risk. Use only the provided rows and cite the source_filing_id values for evidence.

Column Families

Keys and ordering: observation_id, SEC deal identifier, loan number, asset number, reporting period, and loan_period_sequence.
Narrative timeline: timeline_text.
Period state: payment status, workout strategy, loan structure, payment type, balances, interest rate, delinquency streak, seasoning, and maturity distance.
Point-in-time servicer fields: source-visible primary servicer and appointed master, special, and trustee servicer names.
Label and split: target, transfer date, temporal split, and pre-transfer flag.
Provenance: source_filing_id, filing_id, filing_date, source_created_at, source_url, source_table, and source reporting period.

Leakage Verification

This release is gated by a machine-checkable leakage receipt before publication. The receipt checks that no feature row survives on or after its first special-servicer-transfer date, no published row is in the embargo window, point-in-time servicer joins do not use later filings, and the split counts reproduce exactly.

Provenance

The differentiator is provenance to SEC filings. Every published row includes source_filing_id, filing_date, and source_url. Use source_url and source_filing_id to trace an observation back to the SEC archive context for the source filing-derived loan report.

Quickstart

import pandas as pd

train = pd.read_parquet("train.parquet")

timeline_rows = train.sort_values([
    "cik",
    "assetnumber",
    "reporting_period_end_date",
    "loan_period_sequence",
])

grouped_text = (
    timeline_rows
    .groupby(["cik", "assetnumber"])
    .head(12)
    .groupby(["cik", "assetnumber"])["timeline_text"]
    .apply("\\n".join)
)

Downloads last month: 5

Total file size:

123 MB