Skip to content

Reprocessed XBRL deal panel

This builder assembles a clean issuer–target panel from the current reprocessed SEC 10-K and 10-Q XBRL fact files. It is the current XBRL input to the M&A analysis; the older curated CSV is used only as an additional legacy source when constructing the reviewable target and acquirer mappings.

flowchart LR
    k[(10-K numeric + text facts)]
    q[(10-Q numeric + text facts)]
    build[Identify BusinessAcquisition contexts<br/>select deal facts]
    panel[(Clean XBRL deal panel)]

    k --> build
    q --> build
    build --> panel

    classDef input fill:#e8eef7,stroke:#52739e,stroke-width:2px
    classDef process fill:#fff0d8,stroke:#b7791f,stroke-width:2px
    classDef output fill:#e2f2e7,stroke:#3b7a4b,stroke-width:2px
    class k,q input
    class build process
    class panel output
Inputs
Input Location Purpose
Reprocessed 10-K numeric facts /data/xbrl/form_10K_notes_numeric_long.parquet Acquisition contexts and accounting facts
Reprocessed 10-Q numeric facts /data/xbrl/form_10Q_notes_numeric_long.parquet Quarterly fallback when an annual fact is unavailable
Reprocessed 10-K text facts /data/xbrl/form_10K_notes_text_long.parquet Resolve the registrant name from the first acquisition filing
Reprocessed 10-Q text facts /data/xbrl/form_10Q_notes_text_long.parquet Same name lookup for 10-Q first filings
Build
Component Location Purpose
Builder shared/data/xbrl_deals/build_xbrl_deals.py Creates the deal spine and selects goodwill, consideration, and related XBRL facts
Output config shared/config/config_full.yaml, data.xbrl_deals_path /data/full/processed/raw/xbrl_deals.parquet
Slurm wrapper None Run the Python module in an environment with DuckDB and pandas
Output
Dataset Location Grain
Clean XBRL deal panel /data/full/processed/raw/xbrl_deals.parquet One issuer CIK × reported target label

Construction and timing

The builder unions the 10-K and 10-Q numeric fact files and retains contexts whose segment identifies a BusinessAcquisition target. The source files are not restricted to a hard-coded year range; the panel's date coverage depends on the current contents of the staged Parquets.

The deal key is (cik, target_firm). The builder identifies the first filing for that issuer–target pair and stores its first_filing_date and first_fiscal_year. It then selects point-in-time facts (qtrs = 0), preferring 10-K/10-K/A over quarterly filings when an annual fact is available. Selected facts must be filed within 365 days after the first acquisition filing. Among the preferred facts, it chooses the earliest reference date, with deterministic rules for dimensions and matching amendments. This 365-day rule is a fact-selection window per deal, not a global sample cutoff.

The output retains acquisition candidates and records whether a goodwill fact was found. The downstream M&A regression-data builder filters to has_goodwill = true, attaches selected Revelio mappings, observed pre-deal match quality, and as-of Compustat data, and keeps the deal rows for analysis. The later regression sample applies positive-goodwill, both-firms-mapped, available-controls, and positive-consideration restrictions.

The reported xbrl_year used for pre-deal MQ is first_fiscal_year; the year fixed effect in the regression is derived from the selected goodwill fact's reference date (goodwill_ddate). Those year fields need not be identical.

python -m shared.data.xbrl_deals.build_xbrl_deals --overwrite

The command reads the four default inputs above and writes to the path in shared/config/config_full.yaml. --input, --text-input, --config, --output, and --overwrite can be used to change the inputs, config, or destination. The builder refuses to replace an existing output unless --overwrite is supplied.

Next steps are the target mapping, acquirer mapping, and the analysis-layer XBRL regression panel.