Assessing Causality and Delay within a Frequency Band (with Jörg Breitung)

In the "Econometrics and Statistics" journal, 2018, vol. 6, pp. 57-73. DOI information: 10.1016/j.ecosta.2017.04.005

Keywords: Granger causality, frequency domain, filter gain

Download: manuscript version April 2017 (PDF). The methods proposed in this paper can be applied with my following contributed function packages for gretl: "BreitungCandelonTest" and "delayspectral". These are available (from within the gretl program) on gretl's function package server, or see this listing.

Abstract: The frequency-specific Granger causality test is extended to a more general null hypothesis that allows causality testing at unknown frequencies within a pre-specified range of frequencies. This setup corresponds better to empirical situations encountered in applied research and it is easily implemented in vector autoregressive models. Furthermore tools are provided to estimate and determine the sampling uncertainty of the phase shift/delay at some pre-specified frequency or frequency band. In an empirical application dealing with the dynamics of CO2 emissions and US temperatures it is found that emissions cause temperature changes only at very low frequencies with more than 30 years of oscillation. In a business cycle application the causality and leading properties of new orders for German industrial production are analyzed at the interesting frequencies.

And here's a quotation from the conclusions: "we already mentioned the case of the term structure of interest rates where such varying connections are also expected. In addition, according to the economic hypothesis of consumption smoothing a similar result about differing impacts of short- versus long- term fluctuations might hold between income and consumption. We believe that many more potential applications in economics and perhaps other disciplines are likely to exist."

(Latest update: September 2019)