Goals#
Replicate the three core exhibits of Stambaugh (1999, Journal of Financial Economics 54):
Table 1 — the finite-sample distribution of the OLS predictive slope (bias, skewness, kurtosis, and the “true” p-value), simulated at parameters estimated from CRSP data for the paper’s four subsamples.
Table 2 — Bayesian posterior moments for the slope under four prior/likelihood specifications, including the exact-likelihood cases that require Metropolis-Hastings sampling.
Figure 1 — every estimate plotted in (persistence, slope) space, showing that all corrections move the slope down and the persistence up together.
Then update all of it through 2024, and ship two educational products: a guided walkthrough notebook and an interactive playground.
Success criterion: match the paper’s published values within a documented
tolerance (exact digits are unattainable — CRSP has revised 27 years of
history), with every number generated by code and the whole pipeline rebuilt by
one doit command.