The aim of this training is to introduce the modelling and forecasting with mixed frequency models. The world is indeed mixed-frequent. This not only means that mixed frequency (MF) models constitute a popular and widely studied topic in macroeconomic and financial time series econometrics, it is simply an omnipresent fact that applied and theoretical researchers need to deal with. Indeed, national accounts variables, such as consumption, export or the gross domestic product are available quarterly; inflation, unemployment rate or industrial production indexes are available monthly, whereas most financial time series, such as interest rates or stock prices, are released on a daily basis and are even available at an intraday frequency, minute by minute. When studying the relationship between these series, it has recently become standard to properly account for the mismatch in publication frequencies among variables, instead of aggregating high-frequency observations using predetermined aggregation schemes. Working with genuine variables instead of loosing some important information when aggregating them is very helpful for forecasting or nowcasting macroeconomic indicator as well as for detecting causality. In finance, intraday data are combined to evaluate the uncertainty of daily financial assets (namely their volatility, and hence their risk) and then of portfolios.
The set of mixed frequency models ranges from single-regression models (e.g., the well known MIDAS model) over factor models to vector autoregressive models. While it was relatively tedious to implement those methods with routines written in different programming languages, the new features proposed in EViews11 allow, with some practice, to easily build useful models and to make forecast in a mixed frequency environment.
The course will cover the following:
Introduction to mixed frequency issues, real time data and data availability, nowcasting and forecast evaluation with EViews in single equation and vector autoregressive models, aggregation of stock and flow data. Aggregation and interpolation at different frequencies with EViews.
Comparison between observational and data driven methods and mixed frequency methods. Opening a mixed frequency sheet in EViews. Introduction and estimation with (MI)xed (DA)ta (S)ampling (MIDAS) for stationary time series in single equation. Determination of the choice of the weighting function of the high frequency variables and U-MIDAS. Application to macroeconomics and finance.
In order to look at the impact of low frequency data to the high frequency ones and to forecast at multiple horizons we have to introduce the mixed frequency vector autoregressive models for stationary time series. We focus on testing for Granger causality between real and financial time series.
Additional features: nonstationary mixed frequency model models, and mixed factor models.
|Time||Session / Description|
|08:45-09:15||Arrival & Registration|
Knowledge of time series econometrics, Knowledge of the key features of EViews (e.g. reading data, running regressions).
The number of delegates is restricted. Please register early to guarantee your place.