SecretMatch

Abstract

Inventory matching is a process by which a broker or bank pairs buyers and sellers, without revealing their respective orders in a public exchange. Banks often undertake to match their clients, so that these clients can trade securities without incurring adverse price movements. If a bank finds matches between clients, it may execute them at reduced rates; if no matches are found, the clients must trade in a public market, which introduces costs for both parties. This problem is distinct from that solved by dark pools or public exchanges, which implement Continuous Double Auctions (CDAs). CDAs incorporate both price and volume. Inventory matching incorporates volume alone, and extracts price from an external source (such as a public market). As it is currently conducted, inventory matching requires that clients share their intentions to buy or sell certain securities---along with the sizes of their positions---with the bank. Clients worry that if this information were to "leak" in some way, other market participants could become aware of their intentions, and cause the price to move adversely against them before they trade. A solution to this problem promises to enable more clients to match their orders more efficiently---with reduced market impact---while also eliminating the risk of information leakage. We present a cryptographic approach to multi-client inventory matching, which preserves the privacy of clients. Our central tool is threshold fully homomorphic encryption; in particular, we introduce an efficient, fully-homomorphic integer library which combines GPU-level parallelism with insights from digital circuit design. Our solution is also post-quantum secure. We report on an implementation of our protocol, and describe its performance.