Firm performance¶
This page documents the active firm-MQ analysis that generates Table 7 and Appendix Tables A5–A6, plus detailed coefficient exports used for checking. The older pipelines/firm_performance.R entry point is not the current manuscript pipeline.
Inputs
| Dataset | Location |
|---|---|
| Full-training Compustat–CRSP–MQ panel | /data/ |
| Time-profile Compustat–CRSP–MQ panel | /data/ |
One run attempts three samples: full-training in-sample, time-profile
in-sample, and time-profile out-of-sample. The time-profile panel is split
using sample_status. If an input dataset is unavailable, that sample is
logged and skipped while available samples continue.
Build
| Component | Location |
|---|---|
| Analysis module | analyses/modules/firm_mq_performance.R |
| Runtime | hct-tower-analyses-cpu |
| Active entry point | analyses/pipelines/run_firm_mq_performance.R |
Output
| Table/export | Output file |
|---|---|
| Full-training sample | /artifacts/ |
| Time-profile in-sample | /artifacts/ |
| Time-profile out-of-sample | /artifacts/ |
The time-profile in-sample run also writes two fixed-baseline tables:
firm_mq_performance_2016_baseline_controls_only.tex and
firm_mq_performance_2016_baseline_industry_fe.tex in
/artifacts/<coverage>/analyses/firm_mq_performance/time_in_sample/.
Each folder contains sample-construction, Table 7 specifications, the two horizon tables, and four detailed coefficient exports.
Firm-years require at least 20 scored relationships by default. Firm MQ mean and its within-firm-year standard deviation are included as regressors, along with the outcome baselines and other firm controls. In the paper tables, only the Firm MQ coefficient and standard error are displayed; Firm MQ SD remains in the regression as a control but its coefficient row is suppressed. Table 7 shows the final three next-year specifications: industry and year FE, industry-by-year FE, and firm plus industry-by-year FE. The horizon tables retain their industry-by-year and firm plus industry-by-year FE structures. The four detailed coefficient exports retain all six next-year specifications and their full coefficient rows for checking.
Firm MQ mean, Firm MQ SD, other continuous controls, and outcomes are winsorized at the 1st and 99th percentiles within each regression sample. Winsorization does not change the MQ-score threshold or sample-construction counts.
The analysis includes 2024 baseline years when they have MQ scores and the required outcomes and controls. Each sample’s result files are written separately so the estimates can be reviewed side by side.
The fixed-baseline specification uses only 2016 Firm MQ and control values from the time-profile in-sample panel to explain outcomes beginning in 2017. Its five columns report outcomes through 2017–2021 (t+1 through t+5), with controls only or NAICS 3-digit industry fixed effects. A firm fixed effect is not identified in this single-year cross-section. These two tables are additional outputs; the existing yearly and horizon analyses are unchanged.
Command reference
| Option | Required / default | Effect |
|---|---|---|
| --config PATH | Required | Analyses YAML config, such as config/config_full.yaml. |
| --minimum-mq-scores N | 20 | Minimum scored relationships required for a firm-year. |
The firm-performance task is retained for the superseded
pipelines/firm_performance.R entry point. Use the distinct
firm-mq-performance task for the current pipeline.
The run refits its regressions and rewrites the configured LaTeX output files; this analysis does not currently reuse cached model fits.