Paper: arXiv 2507.01979

Authors: Adam Nelson-Archer, Aleia Sen, Meena Al Hasani, Sofia Davila, Jessica Le, Omar Abbouchi

Abstract

We present a deep learning approach for forecasting short-term employment changes and assessing long-term industry health using labor market data from the U.S. Bureau of Labor Statistics. Our system leverages a Long- and Short-Term Time-series Network (LSTNet) to process multivariate time series data, including employment levels, wages, turnover rates, and job openings. The model outputs both 7-day employment forecasts and an interpretable Industry Employment Health Index (IEHI). Our approach outperforms baseline models across most sectors, particularly in stable industries, and demonstrates strong alignment between IEHI rankings and actual employment volatility. We discuss error patterns, sector-specific performance, and future directions for improving interpretability and generalization.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced deep learning architectures (LSTNet with CNNs, GRUs, and skip connections) and includes formal mathematical definitions for evaluation metrics like SMAPE and RMSE. It demonstrates strong empirical rigor with specific data sources (BLS), preprocessing steps (interpolation, sliding windows), sector-level performance analysis, and clear quantitative results.

Research Flowchart

  flowchart TD
  Goal["Research Goal: Forecast employment & assess industry health"] --> Data["Data: BLS multivariate time series<br>(employment, wages, turnover, openings)"]
  Data --> Method["Methodology: Long- and Short-Term<br>Time-series Network LSTNet"]
  Method --> Compute["Computational Process:<br>7-day forecast + Industry<br>Employment Health Index IEHI"]
  Compute --> Outcomes["Outcomes: Outperforms baselines<br>Strong IEHI alignment with volatility<br>Sector-specific insights"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a deep learning approach using a Long- and Short-Term Time-series Network (LSTNet) to forecast short‑term employment changes and assess long‑term industry health.

Source sentence
We present a deep learning approach for forecasting short‑term employment changes and assessing long‑term industry health using labor market data from the U.S. Bureau of Labor Statistics.

Assumes 3

  • Weekly employment data are obtained by interpolating monthly BLS reports. data
    Source sentence
    The BLS reports this data in monthly intervals, which have been interpolated to obtain weekly approximations for a higher level of granularity.
  • A 30‑day sliding window captures both immediate trends and longer‑term patterns. model
    Source sentence
    The model processes this multivariate time series through a 30‑day sliding window, allowing it to capture both immediate trends and longer‑term patterns in the employment data.
  • The multivariate input series are complete after preprocessing for missing values and temporal consistency. data
    Source sentence
    The dataset has been preprocessed to handle missing values and to ensure temporal consistency.

Data

  • BLS weekly employment indicators · U.S. Bureau of Labor Statistics macro public
    March 2006 to December 2024 · ten industries
    Source sentence
    to forecast employment trends. The dataset consists of weekly employment indicators spanning from March 2006 to December 2024, showcased in Table 1.

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • The LSTNet approach outperforms baseline models across most sectors, especially in stable industries. performance
    Source sentence
    Our approach outperforms baseline models across most sectors, particularly in stable industries
  • IEHI rankings align strongly with actual employment volatility. predictive-accuracy
    Source sentence
    demonstrates strong alignment between IEHI rankings and actual employment volatility

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 14 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (33)

  1. DeepAR: Probabilistic forecasting with autoregressive recurrent networks (2020)
  2. Modeling long-and short-term temporal patterns with deep neural networks (2018)
  3. Multivariate time series data prediction based on att-lstm network (2021)
  4. Unemployment rate forecasting: LSTM-GRU hybrid approach (2023)
  5. M5 accuracy competition: Results, findings, and conclusions (2022)
  6. Another look at measures of forecast accuracy (2006)
  7. Root mean square error (RMSE) or mean absolute error (MAE)?—Arguments against avoiding RMSE in the literature (2014)
  8. Current Employment Statistics (2025)
and 25 more
  1. Current Population Survey (2025)
  2. Job Openings and Labor Turnover Survey (2025)
  3. Business Employment Dynamics (2024)
  4. Productivity (2023)
  5. Forecasting output and inflation: The role of asset prices (2003)
  6. Building a composite help-wanted index (2010)
  7. Measuring business cycles (1946)
  8. Alternative Measures of Labor Underutilization (U-6) (2025)
  9. Temporal fusion transformers for interpretable multi-horizon time series forecasting (2021)
  10. Deep vector autoregression for macroeconomic data (2021)
  11. Making a city: Urbanity, vitality and urban design (1998)
  12. Behavioral, psychological, and clinical outcomes of Arabic-speaking people with type 2 diabetes during COVID-19 pandemic (2022)
  13. Comparative study of machine learning methods for COVID-19 transmission forecasting (2021)
  14. Advancing Financial Forecasting: A Comparative Analysis of Neural Forecasting Models N-HiTS and N-BEATS — → 2409.00480
  15. Time-series neural network: a high-accuracy time-series forecasting method based on Kernel filter and time attention (2023)
  16. Wage growth and labor market tightness (2024)
  17. The E-Rule: A Novel Composite Indicator for Predicting Economic Recessions (2025) — link
  18. Development of a CNN+ LSTM hybrid neural network for daily PM2. 5 prediction (2022)
  19. The M3-Competition: results, conclusions and implications (2000)
  20. Forecasting: principles and practice (2018)
  21. Long-range forecasting: From crystal ball to computer (1985)
  22. A pragmatic view of accuracy measurement in forecasting (1986)
  23. Inferring causal impact using Bayesian structural time-series models (2015)
  24. Mismatch unemployment (2014)
  25. Forecasting with judgment (1999)

1 of 33 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.