Reproducibility Scripts for "The Naive–Power Law Blend as a Robust Baseline for Bitcoin Price Forecasting"
Reproducibility scripts and data for the paper "The Naive–Power Law Blend as a Robust Baseline for Bitcoin Price Forecasting" by Carlos Baquero and Daniel Tinoco. This package contains all evaluation scripts, Bitcoin price data, and on-chain activity signals needed to reproduce the paper's results, including: the multi-holdout evaluation across five non-overlapping market regimes (2016–2026); the formal forecast-comparison tests (Diebold–Mariano, Clark–West, and the Model Confidence Set) together with the return-based and directional analyses; the per-window grid search showing that zero correction is optimal; a pre-2016 pre-sample calibration sensitivity check; the autoresearch experiment log; the expanded Bayesian Structural Time Series (BSTS) evaluation; and all figure-generation scripts. A single entry point, reproduce.sh, regenerates the main tables and statistical tests in a few minutes; see README.md for step-by-step guidance and software versions.