[Paper Review] Predicting bubble bursts in oil prices during the COVID-19 pandemic with mixed causal-noncausal models

Alain Hecq, Elisa Voisin|arXiv (Cornell University)|Jan 1, 2019

Market Dynamics and Volatility4 citations

TL;DR

This paper proposes mixed causal-noncausal autoregressive (MAR) models to detect and predict speculative bubbles in WTI and Brent oil prices during the COVID-19 pandemic. By incorporating both lagged and leaded price dependencies, the model identifies bubble dynamics and estimates crash probabilities, with Monte Carlo simulations showing that appropriate detrending preserves bubble patterns for accurate forecasting.

ABSTRACT

This paper investigates oil price series using mixed causal-noncausal autoregressive (MAR) models, namely dynamic processes that depend not only on their lags but also on their leads. MAR models have been successfully implemented on commodity prices as they allow to generate nonlinear features such as speculative bubbles. We estimate the probabilities of crash in oil prices. To do so we first evaluate how to adequately detrend nonstationary oil price series while preserving the bubble patterns observed in the raw data. The impact of different filters on the identification of MAR models as well as on forecasting bubble events is investigated using Monte Carlo simulations. We illustrate our findings on WTI and Brent monthly series before and during the outbreak of the Covid-19 pandemic.

Motivation & Objective

• To investigate whether mixed causal-noncausal autoregressive (MAR) models can effectively detect speculative bubbles in oil prices.
• To evaluate the impact of different detrending filters on identifying MAR models and forecasting bubble events.
• To assess the robustness of MAR models in capturing nonlinear dynamics such as price bubbles in nonstationary oil price series.
• To compare the performance of MAR models on WTI and Brent oil price data before and during the COVID-19 pandemic.
• To estimate the probability of market crashes by identifying bubble formation and burst patterns in oil prices.

Proposed method

• The study employs mixed causal-noncausal autoregressive (MAR) models that depend on both past (lags) and future (leads) values of oil prices to capture nonlinear dynamics like speculative bubbles.
• Nonstationary oil price series are detrended using various filters, with the goal of preserving bubble patterns while removing deterministic trends.
• Monte Carlo simulations are conducted to evaluate the performance of different detrending methods in identifying MAR models and forecasting bubble events.
• The models are estimated and validated on monthly WTI and Brent oil price series from pre-pandemic to pandemic periods.
• Bubble detection is based on the presence of explosive roots in the MAR model, indicating potential price bubbles.
• Crash probabilities are estimated by analyzing the likelihood of reversal after bubble formation, using statistical properties of the MAR process.

Experimental results

Research questions

• RQ1 How do different detrending filters affect the identification of MAR models and the forecasting of bubble events in oil prices?
• RQ2 To what extent can MAR models detect and predict speculative bubbles in WTI and Brent oil prices during the COVID-19 pandemic?
• RQ3 Which detrending method best preserves bubble patterns while removing trends in nonstationary oil price series?
• RQ4 How do the probabilities of price crashes correlate with the presence of explosive dynamics in MAR models?
• RQ5 Are the bubble patterns in oil prices more pronounced or detectable during the pandemic period compared to pre-pandemic conditions?

Key findings

• The MAR model successfully captures nonlinear features such as speculative bubbles in oil price series, particularly during periods of market stress.
• Proper detrending is critical: certain filters preserve bubble signals better than others, significantly improving model identification and forecasting accuracy.
• Monte Carlo simulations demonstrate that the choice of detrending filter has a substantial impact on the reliability of MAR model estimation and crash probability forecasts.
• Bubble patterns are more evident in both WTI and Brent oil prices during the early stages of the COVID-19 pandemic, indicating heightened speculative behavior.
• The model estimates a measurable increase in crash probability during periods of explosive price behavior, confirming its predictive potential for market reversals.
• The inclusion of leaded (future) price dependencies in the MAR model enhances its ability to detect and forecast bubble bursts compared to purely causal models.