Next generation stock exchange: Recurrent neural learning model for distributed ledger transactions

Abstract

A distributed stock exchange system encompasses multiple network hosts that participate in the sharing and exchange of resources. In such exchanges, the mediator or stock exchange must manage and delineate all operations in a cohesive manner. Stock exchange (SE) also acts as the transaction manager to provide consistent, isolated, durable, and atomic transactions for participating entities. However, the work for the stock exchange is not so straightforward as it may sound. With multiple transactions happening per second, the global serializability and concurrency control becomes an issue resulting in multiple threats and vulnerabilities. We propose a novel stock exchange that integrates time series prediction to distributed transactions and understanding the rapid global transactions and limitations of resources at the stock exchange. We use distributed acyclic graph (DAG) based distributed ledger technology IOTA to provide security and consensus for independent users. The paper proposes a time-variant model that adjusts its predictions based on transactions, moments of observations, participating entities, and history. We show that our model outcasts other state-of-art schemes in terms of prediction accuracy. Also, the model is fair, fast, and scalable to handle millions of transactions per second.