Using the Past to Predict the Future
Analytics can be divided into two primary categories: Descriptive Analytics and Predictive Analytics. Descriptive Analytics focuses on historical insights, often handled in Business Intelligence. Predictive Analytics, by contrast, is the domain of Data Scientists, who use historical data to forecast future outcomes.
Data Science involves structuring historical data and feeding it into mathematical models trained to recognize patterns. These models, when applied to new data, can predict outcomes with varying levels of certainty, creating powerful impacts when used effectively.
Terms like Artificial Intelligence (AI) and Machine Learning (ML) have become ubiquitous buzzwords, yet the foundational models remain rooted in decades-old concepts. For example, Deep Neural Networks (DNN) were first conceptualized in the 1960s but only realized their full potential with modern computing. Innovations like Chat GPT and Large Language Models represent the latest iteration of these advancements, demonstrating the ongoing power of deep learning.
The stock market may be inherently unpredictable, but that hasn’t stopped people from trying.
As a personal project, Tony spent months building various Machine Learning models to identify potentially profitable trading strategies. He developed custom software using APIs to automate daily trading and made the tools publicly available for ETrade and TD Ameritrade.
Tony experimented with Linear Regression, Logistic Regression, Decision Trees, and Deep Neural Networks, tracking results daily. While his models didn’t outperform a coin flip (hovering around 51% accuracy), the process provided invaluable lessons. Interestingly, he discovered that his overnight positions, unrelated to the ML models, performed well—a revelation that shifted his focus for future projects.
A major asset manager faced challenges in cross-selling products to customers, struggling to determine which clients would be open to new offerings.
Tony applied K-means clustering to analyze existing customer data and identify patterns in product pairings. This approach, similar to how Netflix recommends movies and shows, provided insights into which products were frequently purchased together. The results empowered sales teams to approach clients with highly relevant and personalized suggestions, improving cross-selling success rates.