Point-in-Time Forecasting from SEC Disclosures
Open Access DepositedIntegrating 10-Q/10-K Fundamentals, MD&A Narrative, 8-K Events, and Lagged Macroeconomics for Multi-Horizon Revenue Forecasting and Next-Quarter Layoff Early Warning
models which primarily depend on existing historical numeric data may be insensitive to managerial, event-driven and narrative cues that are announced in SEC filings far earlier than they are reflected in quarterly outcomes. To deal with that issue, the study constructs a point-in-time forecasting model that combines 10-Q and 10-K systematic fundamentals, Management Discussion and Analysis (MD&A) narrative aspects, 8-K event signals and macroeconomic indicators lagged by one quarter. The praxis tests four hypotheses based on blocked forward validation with time delay to avoid leakage. Hypothesis 1 aims at testing whether the LightGBM model constructed with structured SEC and macroeconomic inputs significantly decreases the mean absolute error (MAE) as compared to the Recursive Rolling-4 benchmark model over all 4 forecast horizons (h2 h3 h4). Hypothesis 2 will evaluate whether or not residual learning, development-set calibration and safeguard logic can make benchmark-anchored revenue forecasts even better. Hypothesis 3, tests the hypothesis that MD&A-derived narrative features provide more incremental predictive power relative to numeric-only models, and whether the latter has more predictive power than transformer-based sequence comparators. Hypothesis 4 examines how Form 8-K Item 2.05 signals, together with macroeconomic and structured context, can provide next-quarter early warnings of layoff-related events by use of significance indicators such as Average Precision, ROC-AUC, and calibration diagnostics. The findings indicate that structured LightGBM reduces the average h2-h4 revenue MAE by 5. 11 percent relative to Recursive Rolling-4, while the calibrated residual learning is statistically supported in multiple horizon enhancement under cluster-conscious inference. The strongest matched-sample revenue performance is achieved by LightGBM with narrative features, outperforming both Recursive Rolling-4 and numeric-only LightGBM. In the case of the layoff task, the most effective classifiers not only meet the minimum average-precision threshold, but also do so at reasonable calibration, albeit sparse positive events are to be handled with care. In general, the praxis has shown that systematic, narrative and occurrence-based SEC disclosures improve the multi-horizon revenue predictions and enhance the early warning of layoff-related events.
Integrating 10-Q/10-K Fundamentals, MD&A Narrative, 8-K Events, and Lagged Macroeconomics for Multi-Horizon Revenue Forecasting and Next-Quarter Layoff Early Warning This praxis explores the question of whether point-in-time disclosures by the Securities and Exchange Commission (SEC) can enhance the forecasting of firms in the presence of the structured filing information and lagged macroeconomic conditions. The study overcomes one of the core limitations of traditional firm-level revenue forecasts namely
Point-in-Time Forecasting from SEC Disclosures
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Pangaribuan_gwu_0075A_17909.pdf | 2026-06-24 | Open Access |
|