Research interests
- Forecasting with mixed frequency datasets using Mixed Data Sampling (MIDAS) models.
- Forecasting with economic data associated with being of a high dimension.
Current research
My research interests include:
- Forecasting with mixed frequency datasets using Mixed Data Sampling (MIDAS) models. Specifically I propose a modification of the original MIDAS proposed by Ghysels et al. (2002) to improve forecasting accuracy with financial market data.
- Forecasting with economic data associated with being of a high dimension. Modern day econometricians are increasingly facing the challenge of requiring reliable forecasts to be computed from longitudinal datasets where there are more candidate predictors available than there are time point observations. This has serious negative implications for the standard models used in the field creating a need for new and upcoming methods to carry out such prediction whilst maintaining accuracy and reliability.
My current research focuses more specifically on a class of models that have ideal attributes when analysing datasets typically encountered by economists. The Ridge Regression of Hoerl and Kennard (1970) povides a computationally simple alternative to Ordinary-Least-Squares (OLS) when the number of predictors relative to observations is high or the correlation amongst covariates increases OLS coefficient variances to undersirably high levels. Ridge does this by inducing estimation bias whilst reducing the variance, however, there are certain scenarios where this tradeoff is undersirable for the purposes of prediction accuracy.
My newly proposed 'Partial Ridge' seeks to manage this tradeoff more effectivley by making the penalization of different coefficients more flexible to limit bias while still reaping the benefits of significantly reduced variance.
