Integrating emission indicators in investment decisions

: an evaluation of OLS Regression, kNN and Gradient Boosting Classification approaches

Abstract

This dissertation studies the application of ordinary least squares regressions and supervised machine learning classification models on emission indicator integration on listed share investments. A large set of emission and financial variables are gathered from STOXX600 constituents stretching 2011 - 2020. Implementing a backward elimination feature selection narrow down 60 emission indicators to Internal Carbon Pricing and NOx and SOx Emissions Reduction Initiatives showing statistically significant relations with next quarter returns. The selected emission indicators are complemented by a set of control variables and used in three approaches to forming investment portfolios. A comparative analysis of the approaches through - a rolling window OLS regression, kNN classification and Gradient Boosting classification - show that a kNN approach to forming percentile portfolios outperform both the regression and Gradient Boosting approach. Both the kNN and Gradient Boosting approaches provide next quarter Up/Down return signal prediction higher than 50%. No approach outperforms a 1/N strategy composed of the source index constituents and only the best ranked percentile portfolio shows statistically significant 3 and 5 factor model alphas in all portfolio creation approaches.

Date of Award	24 Jan 2022
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Dan Tran (Supervisor)