The content discusses a real-world problem faced by banks that publish daily lists of available securities/assets (axe lists) to selected clients. This helps clients effectively locate long (buy) or short (sell) trades at reduced financing rates, but also reveals the bank's internal inventory and the trading activity of large ("concentrated") clients.
The key highlights are:
The problem of minimizing the adverse effects of information leakage caused by sharing the axe list with clients, which is important from both a reputational and financial risk management perspective.
The introduction of a new differentially private continual aggregator mechanism that outputs noisy versions of the axe list at every time step, providing strong privacy guarantees while maintaining acceptable P&L costs for the bank.
The definition of obfuscation metrics, including P&L, leakage probability, over-axe frequency, and worst-case cost, to measure the quality of the obfuscation.
The implementation and deployment of the proposed solution, called Atlas-X, in the production environment of a major financial institution (J.P. Morgan) across three major regions, demonstrating its practical utility and success in the real-world.
Benchmarking results using real and synthetic data to showcase the quality of the obfuscation and its effectiveness in production.
To Another Language
from source content
arxiv.org
Djupare frågor