Xinyuan Li

Research Interest

econometrics: forecasting, Big Data, shrinkage methods

monetary policy

international finance

Job Market Paper

Nowcasting with Big Data: is Google useful in the presence of other information?

This paper studies the usefulness of one novel type of big data, Google search data, in nowcasting US jobless initial claims and unemployment rate in a factor model. If the model has already included macroeconomic information, which is standard in the nowcasting literature, Google Trends data, as in Choi and Varian (2012) does not improve the out-of-sample performance signifficantly, while Google Correlate data does. This challenges the usual view in the literature that Google search data help predicting macroeconomic variables. The paper makes contributions to the literature with a detailed review of Google search data.

Presentations: DIW Macroeconometrics Workshop 2015, LBS Brownbag 2015, RES Symposium for Junior Researchers 2016, IAAE 2016 Annual Conference Milan, 2016 Trans-Atlantic Doctoral Conference, 2016 Annual Conference on Real-Time Data Analysis at the Federal Reserve Bank of Philadelphia, BOE/FRB/KCL conference Modelling Big Data and Machine Learning 2018

mentioned in Econbrowser

Working Papers

Dilemma or Trilemma: what do forty years of data on cross-border capital flows say?

I study whether an identified US monetary policy shock affects the monetary stance of countries with a floating exchange rate regime and an open capital account, as well as the cross-border capital flows between these countries and the US. I find that an unexpected tightening of US monetary policy causes most currencies to depreciate against the dollar on impact. Short-term interest rates react heterogeneously: short term rates are not affected for some floaters but rise on impact for others. After an unexpected US tightening, gross flows contract while cross-border net flows behave heterogeneously. These findings advance our understanding of monetary transmission between countries with floating exchange arrangements.

Determine the number of factors: an MCMC approach

I propose a new Bayesian perspective to determine the number of factors in a static factor model, which can be used in the macroeconomic or financial data analysis. The determination of the common factors and the factor loadings is framed as a model selection problem, and Bayesian inference is used. This method does not require strong assumptions as needed in the existing literature and endows flexibility when choosing the hyperparameters. The algorithm performs well in the simulation exercise. Implementing this algorithm to the dataset in Stock and Watson (2005), I identify the number of factor that drives the macroeconomic data is one.

Unemployment in real time (with Luca Onorante)

Presentations: BOE/FRB/KCL conference Modelling Big Data and Machine Learning 2018