Skip to content

Firm-outcome sample description

This analysis produces full-training descriptive statistics and a separate sample-construction table for the firm-performance, innovation, and M&A analyses. The Compustat-based sections use the current full-training panel, including every MQ year available in that file. Time-profile in-sample and out-of-sample construction counts are produced by the corresponding regression scripts and remain in their profile-specific output folders.

flowchart LR
    firm[(Compustat--CRSP--MQ<br/>firm-year panel)]
    deals[(XBRL deal regression data)]
    counterfactual[(XBRL target<br/>counterfactual MQ)]
    summarize[[Apply analysis screens,<br/>construct outcomes, and summarize]]
    csv[(firm_outcome_sample_description.csv)]
    latex[/firm_outcome_sample_description.tex/]
    construction[/firm_outcome_sample_construction.tex/]
    metadata[(metadata.yaml)]

    firm --> summarize
    deals --> summarize
    counterfactual --> summarize
    summarize --> csv
    summarize --> latex
    summarize --> construction
    summarize --> metadata

    classDef input fill:#e8eef7,stroke:#52739e,stroke-width:2px
    classDef code fill:#f3f3f3,stroke:#666,stroke-width:2px
    classDef support fill:#fff0cc,stroke:#a66b00,stroke-width:2px
    classDef tableOutput fill:#e2f2e7,stroke:#3b7a4b,stroke-width:2px
    class firm,deals,counterfactual input
    class summarize code
    class csv,metadata support
    class latex,construction tableOutput
Inputs

The script reads these configured full-training datasets. The paths are taken from data.full_training in the analyses config. It does not read the time-profile panel because the time-profile regression runners write their own in-sample and out-of-sample construction tables.

Required dataset Config key Location Used for
Compustat--CRSP--MQ firm-year panel data.full_training.compustat_crsp_mq_path /data/<coverage>/analyses/full_training/compustat_crsp_mq.parquet Firm MQ, performance outcomes and controls, patent outcomes, and innovation controls
XBRL deals regression data data.full_training.xbrl_deals_regression_data_path /data/<coverage>/analyses/full_training/xbrl_deals_regression_data.parquet Goodwill deals, acquisition price, mappings, and M&A controls
XBRL target counterfactual MQ data.full_training.xbrl_target_counterfactual_mq_path /data/<coverage>/analyses/full_training/xbrl_target_counterfactual_mq.parquet Counterfactual target MQ and Delta MQ

The script validates the required columns before constructing any table. Firm-performance and innovation samples use all observed full-training MQ years. The M&A panel retains deals whose observed_mq_year is no later than 2024, matching the active XBRL analysis.

Build
Component Location
Self-contained builder analyses/modules/firm_outcome_sample_description.R
Direct pipeline entry point analyses/pipelines/firm_outcome_sample_description.R
Slurm entry point slurm/analyses/firm_outcome_sample_description.slurm
Slurm task firm-outcome-sample-description
Runtime hct-tower-analyses-cpu
Output
Dataset File / location Contents
Descriptive-statistics table firm_outcome_sample_description.tex in /artifacts/<coverage>/analyses/firm_outcome_sample_description/ Full-training firm characteristics and the current performance, innovation, and M&A outcomes
Sample-construction table firm_outcome_sample_construction.tex in the same directory Common screen and final-sample counts for each analysis
Machine-readable summaries firm_outcome_sample_description.csv in the same directory Construction rows and variable-level N, mean, SD, and median
Run metadata firm_outcome_sample_description_metadata.yaml in the same directory Config path, MQ threshold, observed firm-MQ year range, M&A MQ-year ceiling, winsorization, and output paths

Sample and statistics

The table applies the current analysis restrictions rather than using the broader Revelio sample. Firm-year regressions require at least 20 observed MQ scores by default. The firm-performance panel uses next-year log revenue, employment, and revenue-per-employee growth, plus the change from current to next-year ROA. The innovation panel uses cumulative issued-patent outcomes through t+5; citation quality requires positive cumulative patents, and value quality also requires positive cumulative real value. Its regressions require observed prior three-year issued-patent stock. The M&A section applies the observed-MQ-year, goodwill, Revelio mapping, control, acquisition-price, and minimum target-MQ-score restrictions.

All continuous descriptive variables are winsorized at the 1st and 99th percentiles within the relevant analysis sample, matching the regression conventions. Innovation quality statistics use the logged outcomes from the regressions: log(1 + citations per issued patent) and log real value per issued patent. Binary indicators are reported as shares.

Command reference
Option Required / default Effect
--config PATH analyses/config/config_full.yaml Reads the analyses data paths and coverage configuration.
--output-dir PATH Derived from the configured sample-description artifact directory Overrides the output directory.
--min-mq-scores N 20 Changes the minimum observed MQ-score threshold for firm-year and target-deal screens.
Option Required / default Effect
--task firm-outcome-sample-description Required Selects the CPU analyses task.
--config CONFIG.yaml Required Resolves the analyses config passed to the pipeline.
--partition, --time, --mem, --cpus, --job-name, --dependency, --dry-run Wrapper defaults Standard Slurm overrides; default is 8 CPUs, 64 GB, and 4 hours.
cd /hct-tower/analyses
Rscript --vanilla pipelines/firm_outcome_sample_description.R \
  --config config/config_full.yaml
cd /hct-tower
slurm/submit.sh --task firm-outcome-sample-description \
  --config config_full.yaml

The builder does not fit regressions or alter existing analysis artifacts; it only reads the configured inputs and writes the descriptive-statistics, sample-construction, CSV, and metadata outputs.

The construction table reports firm-years and firms for the firm-performance and innovation panels. The M&A panel uses the same table but redefines the numeric columns as deals and acquirers.