Vera Data Timeseries Generating

Time Series — Full Analysis & Manuscript Generation

Table of Contents

• Workflow
• Additional Inputs
• Output Structure
• Key References (read before generation)
• Reporting Standards
• Cross-Validation for Time Series (Mandatory)
• Cross-Skill Interface

Open-source skill. Read reference/specs/output-variation-protocol.md before every generation — apply all variation layers.

Workflow

Continues from where vera-data-timeseries-reviewing stopped (PART 0-2 done).

Additional Inputs

Collect if not already provided:

• Target discipline (for reporting conventions)
• Target journal or style (APA 7th, STROBE, etc.)
• Research question / hypothesis
• Forecast horizon (if not set in initial testing)
• Whether volatility modeling is relevant
• Whether cross-series relationships exist

Output Structure

output/
├── methods.md
├── results.md
├── tables/             ← Markdown + CSV per table
├── figures/            ← PNGs, 300 DPI
├── references.bib
├── code.R              ← Style-varied
└── code.py             ← Style-varied

Key References (read before generation)

Reporting Standards

Same as vera-data-timeseries-reviewing, plus:

• Model order notation: always ARIMA(p,d,q)(P,D,Q)[s] for seasonal models
• AIC/BIC: report for all fitted models for comparability
• Ljung-Box: report on residuals for every fitted model
• GARCH: report ARCH-LM test before fitting, conditional variance equation
• VAR: report lag selection criteria (AIC, BIC, HQ), Granger causality p-values
• Spectral: report dominant frequency, corresponding period, and power
• Forecast accuracy: RMSE, MAE, MAPE on hold-out set — frame as "which assumptions fit" not "which model wins"
• Tree-based with time series: frame as "exploratory"; never claim superiority over statistical models

Cross-Validation for Time Series (Mandatory)

Standard k-fold CV leaks future into past and MUST NOT be used on temporal data. All validation and hyperparameter tuning for this skill uses one of:

1. Rolling-origin (fixed-size) forecast evaluation: at each split, fit on data[t−w : t] (fixed window w) and forecast data[t+1 : t+h]; slide t forward. Recommended when the underlying process is approximately stationary.
2. Expanding-window forecast evaluation: fit on data[0 : t] (growing) and forecast data[t+1 : t+h]. Recommended when more history helps the model.

Implementations: sklearn.model_selection.TimeSeriesSplit (Python), tscv::rolling_origin or tsibble::stretch_tsibble (R). Report the scheme used, window size w, horizon h, and number of splits. Never shuffle time-indexed data prior to split.

Cross-Skill Interface

Method Unit Contract:
├── code_r           → .R script (style-varied)
├── code_python      → .py script (style-varied)
├── methods_md       → methods.md (varied structure)
├── results_md       → results.md (varied phrasing)
├── tables/          → Markdown + CSV
├── figures/         → PNGs 300 DPI (varied layout)
├── references_bib   → .bib with cited references
└── comparison       → cross-method narrative (in results.md)

Invoked directly after vera-data-timeseries-reviewing or orchestrated by vera-data-application-pipelining.