Title: Part 1: Training Sets & ASG Transforms

Abstract: In this paper, I discuss a method to tackle the issues arising from the small data-sets available to data-scientists when building price predictive algorithms that use monthly/quarterly macro-financial indicators. I approach this by training separate classifiers on the equivalent dataset from a range of countries. Using these classifiers, a three level meta learning algorithm (MLA) is developed. I develop a transform, ASG, to create a country agnostic proxy for the macro-financial indicators. Using these proposed methods, I investigate the degree to which a predictive algorithm for the US 5Y bond price, predominantly using macro-financial indicators, can outperform an identical algorithm which only uses statistics deriving from previous price.
This was an undergraduate project, subsequently the research was not exhaustive.