← All tasks
T25 · Event-Driven Trading

Short selling efficiency

Yong Chen et al. (3 authors), 2022 · Journal of Financial Economics

0%
pass@1 across 16 systems
0/16
systems with ≥1 successful attempt
0/5
no system produced a successful attempt
80 attempts · 16 systems × 5
Claude Code + Opus 5
0%
Claude Code + Opus 4.8
0%
Claude Code + Opus 4.7
0%
Codex + GPT-5.6-sol
0%
Codex + GPT-5.5
0%
OpenCode + Kimi-K3
0%
OpenCode + Opus 4.7
0%
OpenCode + GPT-5.5
0%
OpenCode + Sonnet 4.6
0%
OpenCode + Qwen3.7-Max
0%
OpenCode + GLM-5.2
0%
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.

Time the equity market using short-selling efficiency, the monthly cross-sectional slope of abnormal short interest on mispricing; low or negative SSE forecasts higher next-month excess market return and turns on a long-market signal, while weak forecasts switch out of the market.

Type
fundamental
Implementation
pandas
Rebalance
monthly
Long-short
no
Declared task setup
Data
One-command fetcher for WRDS-sourced task data
Operational scope
Construct MISP from eight available anomaly signals and use a five-year initial forecasting window.
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

20 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 (10)
  • A1
    no backfill
    32/76
  • A2
    no global normalize
    59/76
  • A3
    no future data access
    75/76
  • A4
    causal consistency
    62/76
  • A5
    no survivorship bias
    60/76
  • A6
    train test temporal order
    74/76
  • A7
    data frequency alignment
    64/76
  • A8
    transaction timing
    36/76
  • A9
    signal delay test
    57/76
  • A10
    e2e metric consistency
    1/76
TYPE B · PAPER-SPECIFIC (10)
  • B1
    asi lookback 12 months
    §§3.1 “we define abnormal short interest for each stock in each month as the value of short interest in the current month minus the average short interest over the past 12 months.”
    provenance · paper-explicit
    58/76
  • B2
    si normalized by shares outstanding
    §§2 “abnormal short interest (i.e., the ratio of shares sold short over shares outstanding)”
    provenance · paper-explicit
    64/76
  • B3
    misp construction compound
    §§3.1 “we adopt the mispricing measure of Stambaugh et al. (2015), constructed from a combination of 11 well-known stock return anomalies. The original measure is a composite rank between 1 and 100 across stocks. To suit our analysis, we rescale and demean the rank measure each month so the most overvalued (undervalued) stock in the cross-section has a score of 0.5 (−0.5).”
    provenance · benchmark-specified
    60/76
  • B4
    sse cross sectional regression
    §§2 “Each month we regress abnormal short interest on the mispricing score of Stambaugh et al. (2015). The regression coefficient of interest is the slope coefficient bt, which captures short selling efficiency in month t.”
    provenance · paper-explicit
    73/76
  • B5
    stochastic detrend expanding
    §§4.2 “Similar to Campbell (1991), we remove the time trend in SSE and SSL stochastically using information up to month t. Specifically, we use the data from January 1974 to December 1975 as the first subsample to remove the time trend. We then standardize the residuals of the time trend regression and retain the last observation to be matched with stock market return over the next month. We extend the subsample by one month at a time.”
    provenance · paper-explicit
    31/76
  • B6
    winsorize insample only
    §§4.2 “we winsorize SSE and SSL at 1% and 99% only using data up to month t”
    provenance · paper-explicit
    21/76
  • B7
    oos expanding window regression
    §§4.2 “We first run the following time-series regression using a subsample with information up to month t: rt = α + βxt−1 + εt, where rt is excess market return for month t, and xt−1 is the one-month lagged value of the predictor. Then, based on the coefficient estimates only using information up to month t, we compute the forecast of the equity premium for month t + 1.”
    provenance · paper-explicit
    26/76
  • B8
    coefficient sign restriction
    §§4.2 “The second case imposes the coefficient sign restriction, which sets an equity premium forecast to the historical average when the coefficient sign is incorrect (e.g., a positive coefficient for SSE).”
    provenance · paper-explicit
    1/76
  • B9
    mid month short interest
    §§3.1 “we use the short interest as of the middle of the month to ensure that it is in investors' information set when forming expectations of next-month market returns”
    provenance · paper-explicit
    39/76
  • B10
    predictor one month lag
    §§4.2 “rt = α + βxt−1 + εt, where rt is excess market return for month t, and xt−1 is the one-month lagged value of the predictor”
    provenance · paper-explicit
    32/76
What this task reveals

The lowest-passing observed requirement is coefficient sign restriction: 1 of 76 observed checker verdicts pass. The full property list above shows whether errors concentrate in domain-level validity or the paper-specific strategy logic.

Open paper → All tasks
← Previous in Event-Driven Trading
T24 · Earnings Extrapolation and Predictable Stock Market Returns