STRAPSim: A Portfolio Similarity Metric for ETF Alignment and Portfolio Trades

Battle Count: 77

Highly relevant to institutional quantitative trading, particularly in fixed-income and ETF markets. STRAPSim directly addresses portfolio trade execution (matching custom baskets to benchmark ETFs), pricing efficiency, hedging, and systematic execution. The paper discusses portfolio trading constituting ~20% of US IG credit markets and ~11% of European markets. The metric enables automated basket matching, reducing reliance on manual heuristic methods. However, it is more of an infrastructure/tooling contribution than a direct alpha-generating strategy.

The core STRAPSim algorithm is relatively straightforward: compute pairwise similarity matrix, sort by similarity, iteratively match and update weights. The main complexity lies in (1) selecting and computing an appropriate constituent-level similarity function (e.g., training a Random Forest for bond proximity), (2) handling the data pipeline for portfolio holdings and weights, and (3) integrating with existing trading/ETF infrastructure. The algorithm itself is O(n*m*log(n*m)) for sorting plus O(n*m) for matching, which is computationally efficient compared to solving OT linear programs.

The paper provides detailed algorithmic steps, mathematical formulations (Eq. 1a-1b), and experimental setup descriptions. Public datasets (Iris, Breast Cancer, Big Mac, Movie Rating) are available via UCI and Kaggle. However, the corporate bond ETF dataset (20 ETFs, 6870 bonds) is proprietary BlackRock data. No code repository is mentioned. The random forest hyperparameters and exact feature engineering details are partially described but not fully reproducible without the proprietary data.