Skip to content
Projects / Phase 4.3 · JD to Projects

Intraday Cross-Sectional Statistical Arbitrage: From Signal to Monetised P&L

Do intraday cross-sectional signals survive costs, and how much does a cost-aware optimiser add?

Open the projectSee all projects

Headline result

A 5-minute cross-sectional book on Binance perpetuals and S&P 500 stocks. The crypto signal is real (30-minute rank IC 0.027, positive in all 33 test months) but worth about 1 bp against 5–6 bp per trade. Rank-and-trade was ruined within weeks to months; the cost-aware optimiser traded 93–96% less and still lost 57–58% of capital.

Figures

  • Development net Sharpe of the optimiser book against mean daily turnover, crypto and equities.
    In development, net Sharpe rises as the book trades less and crosses zero only when the book barely trades.
  • Equity at $1m on a log scale for rank and optimiser books, crypto and equities, evaluation period.
    Rank books fell below 10% of capital within three weeks in crypto and ten months in equities; the ridge optimiser books ended near 42–43%.
  • Gross P&L against fees, half-spread and impact for four books at $1m, crypto and equities.
    Costs swamp gross P&L in every rank book: crypto fees alone took 63% of capital against a 17% gross gain.
  • Monthly 30-minute rank IC of ridge and gradient-boosted forecasts, crypto and equities.
    Crypto IC stayed positive every month; in evaluation ridge averaged 0.027 and gradient boosting a lower 0.018.

Data used

  • Binance USD-M perpetuals 1-minute bars and funding from data.binance.vision: 900 symbols, Jan 2020 to Sep 2026
  • S&P 500 members’ 5-minute bars from Alpaca’s SIP feed: 723 tickers, Jan 2016 to Sep 2026

Built from Mid-Frequency Quant Researcher job descriptions

The role asks for

  • Intraday alpha signal research
  • Transaction-cost and execution modelling
  • Portfolio construction and optimisation
  • Machine learning and statistics
  • Python and C++
  • Owning strategies in production

This project uses

  • Ridge and gradient-boosted return forecasts
  • Rank IC and signal-decay analysis
  • Square-root market-impact cost model
  • Gârleanu–Pedersen cost-aware optimiser
  • Capacity analysis from $0.1m to $200m
  • Point-in-time universe construction

Members open the full write-up, notebook, tested code and data in the project reader.