- Food Security and COVID-19 Employment Shock in Nigeria: Any ex-ante Mitigating Effects of Past Remittances?, with Al-mouksit Akim and Jeffrey Kouton
[ PDF ] Food Policy, Forthcoming
This paper examines the role of past remittances in mitigating the adverse effects of COVID-19 employment shocks on food security in Nigeria. We formally define the mitigating effects parameter as the difference in the shock impact between households that received remittances and those that did not. Leveraging pre- and post-COVID-19 surveys, we employ a triple-difference strategy to estimate the mitigating effects parameter. Our results suggest that past remittances can alleviate the negative consequences of COVID-19 employment shocks, particularly in the short term. However, the mitigation effect is limited to the early stages of the pandemic, as the negative effects of the shock persist over time. Additionally, we find that the impact of remittances on mitigating the shock varies based on the origin of remittances, recipients’ area of residence, and poverty status. Furthermore, our study highlights the importance of the capital channel in explaining the mitigating role of past remittances. Our findings demonstrate that formal financial inclusion, capital ownership such as livestock, and rental earnings amplify the impact of remittances in mitigating the negative consequences of COVID-19 employment shocks on food security.
• Presentations: SCSE Conference 2023, ICDE 2023: International Conference on Development Economics, CSAE Conference 2022:Economic Development in Africa, 2021 Africa Meeting of the Econometric Society, 2021 International Conference in Development Economics, GLAD 2021, Pan-African Scientific Research Council 2021, Quebec Social Sciences PhD Students Seminar 2021
- High-Frequency Inflation Expectations from Big Data: A Natural Language Approach (Job Market Paper) [ PDF ]
- Can Media Narratives Predict House Price Movements?, with Christopher Rauh
- Identification and Estimation of Common Factors in Group Factor Models
- Regional and Sectoral News-Based Indicators for Macroeconomic Forecasting
In this study, I leverage large language models (LLMs) in natural language processing to scrutinize a comprehensive dataset of more than 2 million newspaper articles and 40 million tweets across Canadian provinces. This method is employed to develop novel high-frequency and real-time indicators of consumer inflation expectations at both national and subnational levels. I first identify news articles and tweets related to inflation or prices. Additionally, I apply deep learning methods, particularly LLMs to extract information specifically related to future price dynamics. Then, I construct daily measures of text-based inflation expectations as the difference between the number of news articles or tweets about inflation and the number of news articles or tweets about deflation. The results indicate a high correlation between the resulting text-based inflation expectations indices with consumers’ survey-based inflation expectations and realized inflation. Subsequently, I use a mixed-frequency machine learning approach to generate nowcasts/forecasts of quarterly inflation expectations and actual inflation based on large sets of text indicators and Google Trends search volume data for inflation-related terms. The analysis demonstrates that news and social media data contain valuable information regarding inflation dynamics and my newly developed indicators effectively anticipate consumer inflation expectations and actual inflation. The paper further explores the application of Shapley additive explanations (SHAP) values to enhance the interpretability of complex, nonlinear models. The findings suggest that newspaper and social media data can serve as a timely source for market participants and policymakers to elicit beliefs on inflation or future price dynamics.
• Presentations: 2023 NBER-NSF Time Series Conference, poster session, 1st CIREQ Interdisciplinary Conference on Big Data and Artificial Intelligence (2023), SCSE Conference (2023), 3rd GREDI/CREATE/CIREQ PhD Student Research Workshop, UdeM Department of Mathematics and Statistics Brownbag Seminar (2023), HEC Montreal*, Concordia University, UdeM Macroeconomic Brown Bag, 18th CIREQ PhD Students Conference (2023), Quebec Social Sciences PhD Students Seminar (2023)
This paper investigates how the housing market, a major asset in household wealth, mirrors broader economic trends and presents a predictive model for housing price movements in Canada at both local and national levels. Our methodology unfolds in two distinct stages: initially, we process over two million newspaper articles through advanced natural language processing techniques to extract media narratives, analyze sentiments, and sort articles according to their focus on past, present, or future events. Subsequently, we implement mixed-frequency machine learning methods to generate a sequence of predictions for quarterly housing prices. The predictions are based on linear models estimated via the LASSO, Ridge, and Elastic net, nonlinear models based on Random Forests, Extreme Gradient Boosting, Artificial Neural Networks, and ensembles of linear and nonlinear models. The results indicate that news data contain valuable information about the housing market's direction. Furthermore, we identify the economic drivers of our machine learning models by applying a novel framework based on SHAP values, uncovering nonlinear relationships between the predictors and house prices.
This paper examines the comovement among factors extracted from two distinct large panels (or groups) of variables. I show that estimating factors introduces a bias in the estimated correlation between factors, which becomes negligible if the factors are estimated from panel data sets containing a large number of cross-sectional series. I show that a modified version of the wild bootstrap algorithm proposed by Gonçalves and Perron (2014) can correct the bias and provide reliable inference on the correlation of interest. Additionally, I apply my modified wild bootstrap method to analyze the influence of institutional factors on economic growth, as examined by Deniz et al. (2018), and the degree of synchronization of business cycles in developed and emerging economies, as explored in Kose et al. (2013) and Aastveit et al. (2015).
• Presentations: 17th International Conference on Computational and Financial Econometrics (Berlin, 2023)*, Annual Meetings of the Canadian Economics Association (CEA, 2022), Conference in Honor of Eric Renault 2022, SCSE Conference 2021, CIREQ Ph.D. Students’Conference 2021, Quebec Social Sciences PhD Students Seminar 2021
This paper evaluates the informational content of sentiment extracted from news articles about the state of the economy. First, I apply deep learning and lexical-based techniques to construct a new high-frequency measure of sentiment indices embodied in a vast news corpus covering economic and financial articles in Canada from January 1977 to October 2022. These sentiment indices are constructed at the sectoral (or 6-digit NAICS), provincial, and national levels. Second, I document that the sentiment indices significantly correlate with contemporaneous key economic and financial variables such as GDP, inflation, housing prices, and unemployment. Third, I use an advanced machine learning method to isolate information about future, current, and past sentiments. Finally, this paper provides novel evidence of how news sentiment tracks current and future economic and financial conditions and significantly enhances predictive power in forecasting models using shrinkage methods and nonlinear machine learning techniques, ensembles of linear and nonlinear models.
• Presentations: Annual Meetings of the Canadian Economics Association (CEA, 2023), Annual Toronto Machine Learning Summit 2022, IVADO Digital October 2022, SCSE Conference 2022, Quebec Social Sciences PhD Students Seminar 2021
Work in Progress
- (Almost) 50 Years of Signals? How Media Deciphers Central Bank Messages
- Monetary Policy Narratives and House Price Expectations, with Juste Djabakou
- Deep Dynamic Factor Models in a Data‐Rich Environment
- Power Blackout ‘Pandemic' and Social Media Voice, with Joseph Agossa
In this paper, I use an advanced computational linguistics approach to examine several ways to extract timely economic signals from a 47-year span of Bank of Canada (BoC) news media coverage. I demonstrate how such data can enhance the assessment of the economy's health and improve macroeconomic forecasts, including CPI inflation and GDP growth. Exploiting BoC newspaper text can improve economic forecasts both unconditionally and when conditioning on other relevant information, but the performance of the latter varies according to the method used. Incorporating text into forecasts by combining forward-looking time dimension with supervised machine learning delivers the highest forecast improvements relative to existing text-based methods. These improvements are most pronounced during periods of economic stress when, arguably, forecasts matter most. My results have two significant implications for monetary policy. First, my text measures can serve as real granular macroeconomic expectations indicators of banks' staff economic forecasts. Second, I shed some light on the links between BoC news coverage and consumer inflation expectations, facilitating the study of the transmission of monetary shocks.
This paper examines the impact of Central Bank narratives on house price expectations using a unique dataset from three different textual sources: direct central bank communication (monetary policy reports and speeches), newspaper articles, and Twitter posts. Leveraging advanced computational linguistics and machine learning techniques, we analyze the narrative tone in monetary policy reports, speeches, news articles, and tweets related to the monetary policy decisions of the Bank of Canada (BoC). Our findings reveal that narrative sentiment expressed in these sources significantly shapes expectations for future house prices. Furthermore, we observe that sentiment related to credit, financial conditions, and housing narratives holds considerable predictive power in shaping house price expectations. Additionally, we employ deep learning methods to extract information specifically related to the forward-looking aspects of sentiment in monetary policy narratives. These results highlight the pronounced impact of forward-looking narrative sentiment on house price expectations. The study suggests that social and news media can serve as valuable tools for central banks in managing economic expectations, with significant implications for the housing market.
- Economic Government Support and Lockdown-Compliance in Africa, with A., Akim [SSRN]
- More Than Words: A Textual Analysis of MEFP, with J., Andritzky, and H., Hesse
- Network Effects and IMF Program Review Teams, with J., Andritzky, and H., Hesse
- Fiscal Vulnerabilities and the Role of Fiscal Policy in Commodity-Exporting Countries, with C., Richaud, S., Essl, A., Mendes, and S., Matta [pdf]
- Regional Debt Market in the Waemu: Curse Or Blessing?, with B., Loko, and C., Richaud
"...When we use very large datasets, it can be dangerous to rely on standard methods for statistical inference. In addition, we need to worry about computational issues. We must be careful in our choice of statistical methods and the algorithms used to implement them.", (James G. MacKinnon, JoE 2023)