Skip to content

AGENTS.md

Latest commit

 

History

History
68 lines (54 loc) · 3.19 KB

AGENTS.md

File metadata and controls

68 lines (54 loc) · 3.19 KB
Raw

Stock Performance Analysis

Purpose

This project uses monthly CRSP (Center for Research in Security Prices) data to study the long-run return distribution of US-listed stocks. It answers three questions:

  1. 10-year holding period - What percentage of stocks produce a positive return over 10 years? What does the 10-year return distribution look like?
  2. 30-year holding period - What percentage of stocks produce a positive return over 30 years? What does the 30-year return distribution look like?
  3. Full-life holding period - What percentage of stocks produce a positive return over their entire listed life? What does that distribution look like?

Data

  • Source: CRSP Monthly Stock File (data/data.csv, ~2.6 GB, gitignored).
  • Reference: data/Monthly Stock File.pdf documents the schema and field definitions.

Methodology - Survivorship Bias Is The Central Concern

The analysis must include every stock that was ever listed during the period under study, not just stocks still trading at the end. Restricting the sample to survivors would systematically overstate returns.

The universe is US-listed common stocks. The code explicitly filters to common equity and excludes REITs, ETFs, closed-end funds, and other non-common-stock securities.

Delisting rule: if a stock is acquired, merged, liquidated, or delisted for any reason before the holding period ends, its return is computed from the start of the period to the delisting date using the CRSP monthly return stream. The stock is not dropped from the sample.

Holding-period construction:

  • Stock-anchored 10-year and 30-year periods: for each stock, form non-overlapping holding periods starting from its first observed CRSP month. The first observed month return is included in compounding (this convention is reported alongside a skip-first-observed sensitivity). A period that begins but cannot complete because the stock delists early is still included, terminated at the stock's last observed month.
  • Calendar start-cohort windows: fixed windows such as 1925-1935, 1935-1945, and 1985-2015. A stock must exist at the window start to enter that window; stocks listed midway through the window are not added halfway. The window models a buy-at-end-of-start_ymm to sell-at-end-of-target_end_ymm investor, so the start_ymm month return itself is excluded from compounding (it was earned before the assumed entry). Cohort stocks whose only observation is the start_ymm month (delisted during the start month, hence not tradable at end of start_ymm) are flagged via no_post_start_obs and excluded from headline summaries.
  • Active periods that are incomplete only because the sample ends are retained in the detail/audit output but excluded from headline holding-period summaries.
  • Full-life: from first available monthly return to last available monthly return.

Detail outputs include CRSP terminal fields, sparse-history flags, a first-period-per-stock companion summary, and a first-observed-return sensitivity table.

Repository Layout

  • program/ - analysis code
  • data/ - raw CRSP inputs (large CSV gitignored, PDF schema doc tracked)
  • results/ - output tables, charts, summary statistics