Time-varying dynamics in panel data sets with stochastic trends

• Applicants: Professor Dr. Matei Demetrescu; Professor Dr. Christoph Hanck
• Subject Area: Statistics and Econometrics
• Term: from 2013 to 2019
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 240888307

Panel data have become pervasive in applied economics - for example, in form of multi-country or of firm-level data - and, compared to time series or cross-sectional data, may provide richer empirical evidence for questions such as the identification of the determinants of economic growth or of stock returns.The empirical analysis needs, as always, to take the stylized facts of the data into account. The initial proposal argued that the interaction of persistence and time-varying volatility in the presence of cross-sectional dependence may take researchers to misleading conclusions when applying standard tools, and set out to provide robust methods.While a large part of the initial project is completed (see Beschreibung_des_Vorhabens" for a detailed list of contributions), we realized during the first funding period that time-varying volatility is often paired with time-varying autocorrelations (dynamics). It is not clear at this time how such time-varying dynamics affect the properties of standard panel inference tools. In view of the experience with time-varying volatility, it is however to be expected that panel methods are particularly badly affected, since small unit-specific distortions typically cumulate over the panel.The extension of the project aims to quantify the distortions induced by time-varying dynamics and to discuss solutions and corrections to take this problem into account. Also, some of the initial research questions still need attention, the initial project having been planned for three years, only two of which were granted.First, we will complete some work already started, but not yet finished, e.g., the paper on panel predictability testing; it will be of particular interest to consider the aspect of time-varying dynamics in the predictor as well. The combination of tests approach shall also receive full attention. This was not possible to the desired extent so far due to time restrictions, not least because we identified and explored some additional fruitful topics within the scope of the project during the past 18 months.Second, we will discuss the monitoring of changes in volatility as well as dynamics. Most current procedures for detecting volatility or autocorrelation changes are applicable in an ex-post fashion only, while practitioners typically require real-time information. Third, we will discuss different approaches for dealing with time-varying dependence. While we strive to provide generic solutions whenever possible, this will not always be the case, and, often, specific solutions to specific problems fare better. Moreover, concrete comparisons are indispensable when both approaches are available.Fourth, we shall provide empirical analyses highlighting the relevance of the theoretical research. We plan to study, among others, the extent to which country-specific indicators of contemporaneous financial instability can be useful as leading indicators of future output and output volatility.

• DFG Programme: Research Grants
• International Connection: United Kingdom
• Cooperation Partner: Professor Dr. Markus Eberhardt