Cormac Kinsella

Cormac Kinsella

Mathematics Master Student and Software Engineer

Blog Posts

2025

Harnessing Deterministic Signals through Stochastic Sampling: A State-Based Approach to Algorithmic Trading

12 minute read

Published:

Abstract: This article details the architecture and methodology of an automated, quantitative trading system designed for cryptocurrency markets. The system’s core is a machine learning strategy that leverages unsupervised clustering to identify distinct market patterns and applies cluster-specific models to generate daily trade signals. Trade execution is managed through a sophisticated two-phase lifecycle, incorporating a momentum-based entry confirmation and dynamic in-trade risk management. The software is designed with a dual-mode operational framework, allowing for both live trading and robust, parallelized backtesting on historical data. This document outlines the system’s modular components, the intricacies of the trading and capital management strategies, and the implementation of its operational modes.

Improving the Odds: A Heuristic Method for Selecting Crypto Breakouts

13 minute read

Published:

Abstract: This document details a proprietary algorithm designed to predict significant upward price movements, or “breakouts,” in crypto-assets. The methodology is built on a multi-stage process that transforms raw market data into a powerful selection heuristic. The algorithm leverages historical price action, market sentiment, and a novel, stability-focused machine learning approach to identify and leverage a more favorable, informed subset of the crypto-asset universe for potential trading opportunities.

Identifying Market Accumulation Patterns with K-Means Clustering

4 minute read

Published:

Abstract: This paper details the “ClusterBuster” algorithm, a methodology for identifying potentially profitable market conditions using unsupervised machine learning. The approach leverages K-Means clustering to group historical financial data, with features primarily derived from rolling modes of the “Fear and Greed Index” over various lookback periods. Each resulting cluster is evaluated using custom performance metrics: the “breakthrough ratio” to quantify upside potential and the “loss ratio” to assess downside risk. The core of the algorithm is an optimization loop that systematically tests different numbers of clusters to identify a model that maximizes the proportion of high-performing, or “breakthrough,” clusters. The final output is an optimized clustering model that effectively isolates historical data patterns correlated with significant positive returns, providing a systematic approach to identifying trading opportunities.

2023

Public Documentation on BCMS Upload Services

5 minute read

Published:

I provide public documentation of my work regarding BCMS Upload Services in this article where you can see how I structured the application and what decisions led me to those structures be implemented in the application. I show my own way of creating an application and apply my principles which are shown in my work in trin-app and trin-react.

2022