Cartea, Cucuringu & Jin (2023) — SPONGE / SPONGEsym
Authors: Álvaro Cartea (University of Oxford), Mihai Cucuringu (University of Oxford / The Alan Turing Institute), Qian Jin.
Venue: SSRN Electronic Journal, 2023; doi:10.2139/ssrn.4560455. Working paper, "Correlation matrix clustering for statistical arbitrage portfolios".
Contribution
Introduced the SPONGE (Signed Positive Over Negative Generalized Eigenproblem) family of signed graph clustering algorithms, with the symmetric variant SPONGEsym as the recommended default. See SPONGEsym for the algorithm.
Reported performance: long/short baskets rebalanced every 3 days on S&P 500 produced 12.2% ARC with Sharpe 1.1, before transaction costs. Profitability collapses under modest transaction costs — the gap that motivates Korniejczuk Slepaczuk 2024 Graph Clustering Stat Arb and Fanshawe Masih Cameron 2025 THGNN Correlation Forecasting.
Importance to this wiki
This is the canonical citation for graph-based clustering of stocks into long/short baskets — every newer StatArb paper builds on SPONGEsym's decomposition or compares against it.