JOB MARKET PAPER

• Identifying OPEC News Shocks: The Impact of OPEC Announcements Using Textual Data

Working Paper

This paper presents a novel method for analyzing the macroeconomic effects of oil price shocks by integrating textual analysis of news articles related to OPEC announcements. Using text data from major newspapers, I extract features that capture shifts in oil price expectations driven by OPEC news. These features function as external instruments in a proxy SVAR model. The analysis reveals that 91.3% of oil price surprises are attributable to OPEC-related factors, with 85.6% stemming from actual supply decisions and 5.7% from market sentiment and demand expectations. Historical decomposition indicates that supply news dominates during geopolitical crises, such as the Iranian Revolution and the Gulf War, while demand-driven sentiments play a larger role during economic downturns like the Global Financial Crisis and the COVID-19 pandemic. Additionally, experiments with synthetic news data confirm that real news narratives have a significant impact on market behavior, unlike artificially generated content. These findings suggest that OPEC news shocks primarily reflect both supply-side decisions and demand-side expectations.

WORKING PAPER

• Beyond News Headlines and TF-IDF: Enhancing Text-Based Forecasting Models with Validated Collocations and Improved Attention. (Revise and Resubmit at International Journal of Forecasting)

Working Paper  Replication codes 

This paper proposes a method for improving text-based forecasting models, specifically focusing on forecasting crude oil prices. Utilizing advanced techniques, including pattern validation and attention mechanisms, the study demonstrates notable improvements in predictive power over traditional approaches. One key finding is that considering the full text of news articles, rather than limiting the analysis to news headlines, leads to significant gains in forecasting accuracy. Furthermore, the model featuring verb-noun and noun-verb collocation pattern validation consistently outperforms benchmarks and models based solely on news headlines across various forecasting horizons. The results suggest that the presence of such collocations as 'price fell,' 'prices tumbled,' and 'price dropped' in crude oil-related news articles is associated with a decrease in oil price returns. Additionally, a novel experiment was conducted to test the uniqueness of the text data, operating under the hypothesis that if the text data is uninformative and not unique to crude oil prices, it should not perform differently from any synthetically generated text. Using language model-generated synthetic news data on crude oil, a similar forecasting exercise was conducted. Despite the synthetic data undergoing the same forecasting procedure, it yielded poor results, indicating the authentic text's ability to capture market dynamics.

WORKS IN PROGRESS

• Forecasting Crude Oil Price Returns: A Principal Weighted-Elastic Net (PW-EN) Approach (with Andrew Hanson)

Slides

In this paper, we propose an approach for handling large, complex datasets when forecasting oil price returns. Traditional methods like Principal Component Analysis (PCA) and LASSO regression often face limitations. PCA struggles with interpretation, while LASSO has trouble with highly correlated variables. To overcome these issues, we introduce a new model, the Principal Weighted-Elastic Net (PW-EN). By combining PCA's ability to reduce dimensions with the feature selection power of Elastic Net, PW-EN creates a more accurate and efficient forecasting model. It selectively weighs predictors based on their contribution to total variance enhancing both model performance and interpretability.

• Welfare Implications of Bubble Dynamics (with Andrew Hanson)

• A Regularized Scaled-Sliced Inverse Regression (REGs-SIR) for High Dimensional Data

• The Role of Expert Opinion and Analysis in the Crude Oil Market