← All tasks
T08 · Mean Reversion

Empirical Investigation of an Equity Pairs Trading Strategy

Huafeng (Jason) Chen et al. (4 authors), 2019 · Management Science

29%
pass@1 across 16 systems
11/16
systems with ≥1 successful attempt
4/5
4/5 attempts passed for the best system
80 attempts · 16 systems × 5
Claude Code + Opus 5
60%
Claude Code + Opus 4.8
60%
Claude Code + Opus 4.7
80%
Codex + GPT-5.6-sol
20%
Codex + GPT-5.5
20%
OpenCode + Kimi-K3
60%
OpenCode + Opus 4.7
40%
OpenCode + GPT-5.5
20%
OpenCode + Sonnet 4.6
40%
OpenCode + Qwen3.7-Max
40%
OpenCode + GLM-5.2
20%
OpenCode + DeepSeek V4-Pro
0%
OpenCode + GLM-5.1
0%
OpenCode + Qwen 3.6 Plus
0%
OpenCode + MiniMax-M3
0%
OpenCode + MiniMax M2.7
0%
all checkers pass Type-A clean, Type-B fail fail checking timeout no result / incomplete
Strategy

What the paper asks for.

Each trade year y, for every CRSP stock i, take the 50 most-correlated stocks (Pearson correlation of monthly returns over [Jan(y-5), Dec(y-1)], 60 months) as i's pair set P_i; fit OLS slope beta_C of R_i on Cret_i over the same window. Each month m of year y, form the signal RetDiff = beta_C * (Cret - Rf) - (Lret - Rf), sort the cross section into deciles, value-weight LONG D10 / SHORT D1 (zero cost), one-month holding, monthly rebalance.

Type
technical
Implementation
pandas
Rebalance
monthly
Long-short
yes
Declared task setup
Data
One-command fetcher for WRDS-sourced task data
Operational scope
The paper-defined methodological core is retained; the fixed instruction instantiates only the data mappings, interfaces, and evaluation conventions needed for execution.
Six-stage quantitative strategy pipeline
  1. 01 Data loading
  2. 02 Feature engineering
  3. 03 Model fitting
  4. 04 Signal generation
  5. 05 Portfolio construction
  6. 06 Execution
Checker suite

18 properties · 80 attempts.

Each row is one requirement the strategy must satisfy. The bar reports the share of observed checker verdicts that pass; missing checker output is not converted into a property verdict.

TYPE A · DOMAIN-LEVEL (8)
  • A2
    no global normalize
    stage · feature eng
    71/72
  • A3
    no future data access
    stage · feature eng
    64/72
  • A4
    causal consistency
    stage · signal gen
    69/72
  • A5
    no survivorship bias
    stage · portfolio
    65/72
  • A7
    data frequency alignment
    stage · data loading
    67/72
  • A8
    transaction timing
    stage · execution
    67/72
  • A9
    signal delay test
    stage · signal gen
    60/72
  • A10
    e2e metric consistency
    26/72
TYPE B · PAPER-SPECIFIC (10)
  • B1
    pairs count top 50
    §Section 2.1 “We then find 50 stocks with the highest correlations to stock i as its pairs”
    provenance · paper-explicit
    stage · signal gen
    68/72
  • B2
    pairs set excludes self
    §Section 2.1 “We then find 50 stocks with the highest correlations to stock i as its pairs”
    provenance · paper-explicit
    stage · signal gen
    69/72
  • B3
    beta c ols slope 60 months
    §Section 2.1 “beta^C is the regression coefficient of firm i's monthly return on its pairs-portfolio return using monthly data between year t-4 and t”
    provenance · paper-explicit
    stage · feature eng
    59/72
  • B4
    retdiff formula exact
    §Section 2.1 “RetDiff = beta^C * (Cret - Rf) - (Lret - Rf)”
    provenance · paper-explicit
    stage · feature eng
    67/72
  • B5
    long d10 short d1
    §Section 2.1 “the zero-cost portfolio Decile 10 - Decile 1 (i.e., longing Decile 10 and shorting Decile 1) generates a return of 1.40% per month”
    provenance · paper-explicit
    stage · signal gen
    72/72
  • B6
    value weighted within leg primary
    §Section 2.1, Table 1 Panel A “Panel A reports value-weighted portfolios formed using all stocks with 60 monthly returns in the last five years.”
    provenance · paper-explicit
    stage · portfolio
    66/72
  • B7
    annual pair set reformation
    §Section 2.1 “For each stock i in year t+1, we compute the Pearson correlation coefficients between the returns of stock i and all other stocks in the CRSP using monthly data from January of year t-4 to December of year t”
    provenance · paper-explicit
    stage · feature eng
    68/72
  • B8
    lret is previous month stock return
    §Section 2.1 “Lret is the previous month's stock return”
    provenance · paper-explicit
    stage · feature eng
    68/72
  • B9
    zero cost long short dollar neutral
    §Section 2.1 “the zero-cost portfolio Decile 10 - Decile 1”
    provenance · paper-explicit
    stage · portfolio
    70/72
  • B10
    ret diff sort uses previous month
    §Section 2.1 “we form 10 portfolios, Decile 1 through Decile 10, based on the previous month's RetDiff; the holding period is one month”
    provenance · paper-explicit
    stage · signal gen
    71/72
What this task reveals

The lowest-passing observed requirement is e2e metric consistency: 26 of 72 observed checker verdicts pass. The full property list above shows whether errors concentrate in domain-level validity or the paper-specific strategy logic.