When Energy Risk Becomes Salient
Description
Replication data for “When Energy Risk Becomes Salient: Attention, Structural Exposure, and International Energy Equity Revaluation” This dataset contains the replication materials and empirical outputs supporting the manuscript When Energy Risk Becomes Salient: Attention, Structural Exposure, and International Energy Equity Revaluation. The study examines whether energy-related investor attention affects the relative valuation of national energy-sector equities, whether these effects vary across countries and information regimes, and whether persistent structural energy-system exposures become financially relevant when energy risk becomes salient. The archive contains the data, code, figures and test-level outputs corresponding to the final manuscript and supplementary appendix. It includes: Google Trends Web Search and News Search data for the 12 analytical markets and the Worldwide series; structural energy-system inputs including energy-import dependence, energy intensity, energy mix and carbon-pricing variables; macroeconomic, geopolitical-risk, oil-price and market-risk controls used in the analysis; processed structural-exposure datasets; complete monthly and daily empirical result tables; country-heterogeneity and slope-equality tests; structural attention-by-exposure interaction results; robustness, nonlinear, regime, falsification and influence tests; publication-timing and pseudo-real-time information-set tests; out-of-sample forecasting and economic-value results; publication-quality figures and the underlying source tables for Figures 1–10; analysis scripts for the monthly, daily and information-set pipelines; editable versions of the principal manuscript tables; the full supplementary-results appendix. The principal outcome is the Energy Revaluation Spread (ERS), defined as the local-currency return on a country's energy-sector equity index minus the return on its corresponding broad domestic equity market. Raw national energy-sector and broad-market index histories are not redistributed because the underlying source files remain subject to source licensing restrictions. To support reproducibility, the archive provides the exact benchmark definitions, raw source filenames, sample windows, file sizes and SHA-256 hashes of the market files used in the analysis. Researchers with legally obtained copies of the same series can therefore verify and reconstruct the financial-data inputs. Google Trends histories were downloaded in 2026 and may reflect Google's historical sampling and normalization procedures. Accordingly, the information-set exercises should be interpreted as publication-lag-adjusted or pseudo-real-time tests rather than true historical-vintage reconstructions. The archive is organized so that every principal result reported in the manuscript can be linked to its underlying test-level output through
Files
Steps to reproduce
1. **Extract the archive** while preserving the folder structure. `00_documentation/MANUSCRIPT_ANALYSIS_MAP.csv` links the main manuscript analyses and figures to their underlying output files. 2. **Restore the licensed equity-index data.** Raw national energy-sector and broad-market histories are not redistributed because of source licensing restrictions. The exact benchmark names, source filenames, file sizes and SHA-256 hashes used in the study are provided in `00_documentation/RESTRICTED_SOURCE_FILE_HASHES.csv`. The country benchmark mapping is also reported in `01_manuscript_tables/Table_1.csv`. 3. **Install the Python environment** using `06_code/requirements.txt`. The original scripts retain `/mnt/data/...` paths from the research environment; replace these with the local extraction path before execution. 4. **Run the monthly pipeline** in `06_code/monthly_pipeline/` in numerical order, beginning with `01_build_long_panel.py` and ending with `06_falsifications_extensions.py`. This reproduces the measurement diagnostics, monthly local projections, country heterogeneity, structural interactions, robustness tests and monthly out-of-sample analysis. 5. **Run the daily pipeline** in `06_code/daily_pipeline/`, beginning with `07a_build_daily_response_panel.py` and continuing through the numbered 07–08 scripts. This reproduces the daily price-discovery paths, country heterogeneity, structural interaction timing, return decomposition and daily forecasting tests. 6. **Run the information-set pipeline** in `06_code/information_set_pipeline/` in this order: `10a_build_information_set_panel.py`, `10b_information_set_tests.py`, `10c_publication_lag_oos.py`, and `10d_info_set_summary_figures.py`. These reproduce the recursive transformations, publication-lag assumptions, stale-exposure tests, delayed-attention tests and strict 2024+ out-of-sample exercises. 7. **Validate results** against the exact test-level CSVs in `04_results/`. Final Figures 1–10 are in `05_figures/`, with the corresponding source table for each figure in `05_figures/source_tables/`. 8. **Check integrity** using `FILE_MANIFEST.csv`. Exact replication requires the same licensed equity-index histories. Google Trends data may be resampled or renormalized in later downloads, so the supplied 2026 Trends files should be treated as the replication vintage. The information-set tests are therefore publication-lag-adjusted/pseudo-real-time rather than true historical-vintage reconstructions.