← All tasks
T15 · Factor Investing

Growing the Efficient Frontier on Panel Trees

Lin William Cong et al. (4 authors), 2025 · 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.

Boosted Panel Trees grow economically guided cross-sectional splits with a Sharpe-maximizing global criterion to span the efficient frontier from 61 firm characteristics.

Type
ml
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

29 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)
  • A1
    no backfill
    61/74
  • A2
    no global normalize
    62/74
  • A3
    no future data access
    68/74
  • A4
    causal consistency
    70/74
  • A5
    no survivorship bias
    68/74
  • A6
    train test temporal order
    67/74
  • A9
    signal delay test
    69/74
  • A10
    e2e metric consistency
    1/72
TYPE B · PAPER-SPECIFIC (21)
  • B1
    characteristic normalization 01
    §§2.2 “Firm characteristics are uniformly normalized cross-sectionally to the range of [0, 1] within each period.”
    provenance · paper-explicit
    64/74
  • B2
    missing value imputation 05
    §§3.1 (footnote 12) “Missing values of characteristics are imputed as 0.5, which implies the firm is neutral in the security sorting.”
    provenance · paper-explicit
    62/74
  • B3
    return winsorization 1 99
    §§3.1 “We apply cross-sectional winsorization at 1% and 99% to mitigate the impact of outliers on individual stock returns.”
    provenance · paper-explicit
    72/74
  • B4
    num characteristics 61
    §§3.1 “We use 61 firm characteristics with monthly observations for each stock, covering six major categories: momentum, value, investment, profitability, frictions (or size), and intangibles.”
    provenance · paper-explicit
    8/74
  • B5
    quintile split points
    §§2.2 “We assess different split threshold options c_m for characteristics z_{·,k}, such as quintile splits c_m ∈ [0.2, 0.4, 0.6, 0.8].”
    provenance · paper-explicit
    70/74
  • B6
    split criterion sharpe maximization
    §§2.2 “Our split criterion aims to maximize the squared Sharpe ratio of the MVE portfolio formed by the leaf basis portfolios.”
    provenance · paper-explicit
    55/74
  • B7
    leaf basis portfolio value weighted
    §§2.1 “The leaf basis portfolios are (value-weighted) portfolios of stocks within each leaf node, and their time series of returns are saved in a vectorized leaf parameter.”
    provenance · paper-explicit
    49/74
  • B8
    num leaves per tree 10
    §§2.2 “We consider P-Trees with J + 1 = 10 leaf nodes in the baseline specification.”
    provenance · paper-explicit
    58/74
  • B9
    minimum leaf size 20
    §§2.2 “Furthermore, we mandate a minimum leaf size of 20 for growing the tree because the leaf needs to serve as a basis portfolio, and any leaves that do not meet this criterion are not subjected to further splitting.”
    provenance · paper-explicit
    71/74
  • B10
    shrinkage parameter 1e4
    §§2.2 (footnote 10) “Following the regularization approaches in Kozak et al. (2020) and Bryzgalova et al. (2023), we utilize small shrinkage parameters, γ_Σ = 10^{−4}, for robustly estimating efficient portfolio weights.”
    provenance · paper-explicit
    62/74
  • B11
    mve weight normalization
    §§2.2 “Also, portfolio weights are normalized such that the sum of absolute weights equals one.”
    provenance · paper-explicit
    69/74
  • B12
    regularized covariance formula
    §§2.2 “w^(j) = [Cov(R_t^(j)) + γ_Σ I_{k+1}]^{-1} E(R_t^(j))”
    provenance · paper-explicit
    24/74
  • B13
    num boosted trees 20
    §§3.2 “ranging from 6.37 for a single P-Tree (with 10 test assets) to 15.63 for 20 boosted P-Trees (with 200 test assets)”
    provenance · paper-explicit
    26/74
  • B14
    boosting split criterion joint
    §§2.3 “The second factor f_{2,t} is generated by the second P-Tree to complement the first factor and maximize the squared Sharpe ratio. The second tree growth follows a similar boosting split criterion: L(c̃_{k,m}) = μ'_F Σ_F^{-1} μ_F, where F = [f_{1,t}, f_{2,t}].”
    provenance · paper-explicit
    41/74
  • B15
    two step mve process
    §§2.3 “The boosted P-Tree utilizes a two-step process to produce either the MVE portfolio or the SDF by incorporating multiple sets of leaf basis portfolios. Initially, the tree-specific MVE portfolio of all leaf basis portfolios within each P-Tree is computed to determine the current P-Tree factor. Subsequently, this factor is merged with all previous P-Tree factors to create an all-tree MVE portfolio.”
    provenance · paper-explicit
    32/74
  • B16
    universe filter compound
    §§3.1 “This universe includes only stocks listed on NYSE, AMEX, or NASDAQ for more than one year and uses those observations for firms with a CRSP share code of 10 or 11. We exclude Stocks with negative book equity or lag market equity.”
    provenance · paper-explicit
    13/74
  • B17
    time period 1981 2020
    §§3.1 “The monthly data ranges from 1981 to 2020.”
    provenance · paper-explicit
    59/74
  • B18
    train test half half split
    §§3.3 “We follow Kozak et al. (2020) and perform a half-half split to construct the training and test samples.”
    provenance · paper-explicit
    64/74
  • B19
    macro variables 10 rolling standardization
    §§3.1 “In addition, we use 10 macro variables to capture potential regime switches... We standardize these macro predictor data by the historical percentile numbers for the past 10 years.”
    provenance · paper-explicit
    62/74
  • B20
    monthly rebalancing
    §§3.1 “We use 61 firm characteristics with monthly observations for each stock.”
    provenance · paper-explicit
    71/74
  • B21
    global split criterion evaluation
    §§2.2 “We assess the split criteria for all candidate splits for both leaf nodes and choose the split with the largest value of the criterion irrespective of which node the split occurs.”
    provenance · paper-explicit
    22/74
What this task reveals

The lowest-passing observed requirement is e2e metric consistency: 1 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.

Open paper → All tasks
← Previous in Factor Investing
T14 · Timing the Factor Zoo