08 Implementation

재현 가능성 (reproducibility)

1 Implementation Notes

This section summarizes the technical implementation details of the TBTF strategy, including programming environment, file structure, and reproducibility instructions. The entire empirical workflow is based on Quarto and organized into modular .qmd documents with embedded Python code blocks. This ensures seamless rendering of figures and tables alongside text, and supports reproducible research.

2 Programming Environment

All analysis was conducted using Python 3.x. Core packages used include:

• Data Processing: pandas, numpy
• Statistical Modeling: statsmodels, scipy, sklearn
• Visualization: matplotlib, seaborn, plotly
• Rendering: quarto, jupyter, ipykernel

Additional utilities like joblib and multiprocessing were used for robustness tests and performance acceleration.

3 Document Structure

The analysis is broken into modular .qmd files, each corresponding to a specific stage of the empirical framework. These files are rendered either as part of the full manuscript or as standalone reports.

QMD File	Description	Related Section
02_data.qmd	Load and clean CRSP & FF data, construct derived variables	Section 2 – Data
03_ff_compare.qmd	FF size decile analysis, Sharpe dynamics comparison	Section 3 – FF Benchmark
04_structure.qmd	Capital convexity, mixture modeling, Markov transition structure	Section 4 – Structure
05_strategy.qmd	TBTF logic: What–How–When framework, weight estimation	Section 5 – Strategy
06_performance.qmd	Performance comparison: distribution, price level, metrics	Section 6 – Performance
07_robustness.qmd	Robustness checks: size, frequency, weights, sample splits	Section 7 – Robustness

All figures and tables are embedded directly within these .qmd documents using Quarto-native fig-cap and tbl-cap options.

4 Output Artifacts

The results are organized in structured folders and linked automatically to each section:

• figs/: Figures (e.g., PNG/HTML via matplotlib or plotly)
• tables/: Summary tables and evaluation metrics (CSV or HTML)

Each artifact is generated at render time and versioned with appropriate suffixes. Example:

• sharpe_curve_s10_b10.png, return_dist_tbtf_post2010.png
• performance_summary_pre2010.csv, weight_table_exp_fit.html

5 Reproducibility

To reproduce the full pipeline:

1. Clone the project repository and install dependencies:

pip install -r requirements.txt

2. Download CRSP/Compustat data via WRDS (institutional access required)
3. Edit configuration or .qmd arguments if needed
4. Render the full manuscript or individual parts:

quarto render

Optional automation scripts using make, snakemake, or papermill are available for batch processing.

6 Appendix

6.1 requirements.txt

pandas>=1.5
numpy>=1.22
scikit-learn>=1.3
matplotlib>=3.6
seaborn>=0.12
statsmodels>=0.13
scipy>=1.10
plotly>=5.10
quarto>=1.3