About Me
My name is Marc Wildi. I am a Professor of Econometrics at the Zurich University of Applied Sciences. My primary research interests lie in forecasting and signal extraction. Over the years, I have developed (partly alone, partly in cooperation with fellow researchers) a number of novel forecasting frameworks designed to tackle real-world prediction challenges and to reconcile various — sometimes competing — research objectives within a unified forecast optimization criterion:
a) MDFA (initiated in 2005): available in this repository
b) M-SSA (initiated in 2020): available in a separate GitHub repository
c) Look-Ahead DFP and PCS (initiated in 2024): available in a separate GitHub repository
####
Research:
The theoretical foundations and principles of MDFA are documented in the following references:
I) Books
Wildi, M. (2005). Signal Extraction: Efficient Estimation, ‘Unit Root’-Tests and Early Detection of Turning Points. Lecture Notes in Economics and Mathematical Systems. Springer. DOI: https://doi.org/10.1007/b138291
A new book on MDFA, co-authored with Tucker McElroy, is currently in preparation.
II) Articles
Wildi, M. \& McElroy, T. (2019). The trilemma between accuracy, timeliness and smoothness in real-time signal extraction. International Journal of Forecasting, Volume 35, Issue 3.
Wildi, M. \& McElroy, T. (2016). Optimal real-time filters for linear prediction problems. Journal of Time Series Econometrics, Volume 8, Issue 2.
Wildi, M. \& McElroy, T. (2020). The multivariate linear prediction problem: Model-based and direct filtering solutions. Econometrics and Statistics, Volume 14.
Quast, J., van Norden, S. \& Wildi, M. (2026). Credit cycles and credit crises: Some measurement issues and implications. Working paper submitted to the 2026 SNB Research Conference (Oct. 2–3, 2026).
####
Working paper versions of the above articles can be found in the Literature folder of the MDFA tutorial project. The most recent cited work (2026) analyses DFA in the context of financial crisis prediction, comparing filter characteristics and classification performance against the HP (lambda=400’000), Hamilton (h=20, p=4), and simple difference filters (1-B^20). The DFA design relies on a white noise assumption (flat spectrum), which avoids data snooping, and is applied to differenced Credit-to-GDP data for a selection of approximately 50 countries.