Energy Sector Stock Price Forecasting Around COVID-19
A time-series forecasting analysis of energy sector stock prices using three complementary modeling approaches to capture market dynamics before and after the COVID-19 shock.
Overview
This project forecasts weekly stock prices for energy sector companies, examining how different time-series modeling approaches perform in capturing market behavior before and after the COVID-19 pandemic. The analysis uses three complementary forecasting frameworks — Univariate models, Vector Autoregression (VAR), and Vector Error Correction Model (VECM) — implemented in Gretl with supporting visualizations generated in Excel.
Motivation
The COVID-19 pandemic created an unprecedented shock to global energy markets, disrupting supply chains, collapsing demand, and creating extreme price volatility. Understanding how different forecasting models handle such structural breaks is valuable both for academic research and practical investment decision-making.
Methodology
- Univariate Model: A baseline autoregressive approach using each stock's own historical price series. Captures simple persistence and trend patterns.
- Vector Autoregression (VAR): Models the joint dynamics of multiple energy stocks simultaneously, allowing each stock to respond to its own history as well as the histories of other stocks in the system.
- Vector Error Correction Model (VECM): An extension of VAR for non-stationary series that are cointegrated. Decomposes movements into short-term adjustments and long-term equilibrium corrections, making it particularly suited for modeling how energy stocks revert to fundamental relationships after shocks.
Data & Preprocessing
Weekly stock price data for multiple energy companies was collected and preprocessed. Key steps included computing log differences to achieve stationarity, generating correlograms (ACF/PACF) to identify appropriate lag structures, and testing for cointegration relationships among the series before fitting VECM models.
Key Findings
Multivariate models (VAR and VECM) generally provide more accurate forecasts than univariate baselines, particularly during periods of high volatility when cross-stock relationships become more pronounced. The VECM model proved especially useful for capturing the long-run equilibrium dynamics among energy stocks as markets adjusted to post-pandemic conditions.
Visualizations & Results
