FAQ: How do I extract the output gap?
**Fabio Canova – **
Abstract
I investigate the properties of gaps and potentials in a variety of DSGE models and their relationship with estimates obtained with standard approaches. Gaps display low frequency variations, have similar frequency domain representation as potentials, and are correlated with them. Transitory and permanent fluctuations display similar features, but are uncorrelated. I use a number of procedures to estimate the latent components. All approaches generate distortions. Gaps are best estimated with a polynomial filter; transitory fluctuations with a differencing approach. Explanations for the outcomes are given. I design a procedure that reduces the biases of existing methods.
Summary
TK
https://youtu.be/nWeL1lKugik?t=5393
Typical statistical approaches for extracting or removing trends make several assumptions that don’t necessarily hold in the underlying DGP:
- Trend isn’t necessarily uncorrelated with cycle
- Cycles are not necessarily transitory
- True cycles do not necessarily happen at business cycle frequencies
Polynomial detrending / filtering is Lindy: it is the oldest procedure and arguably one of the best.
”Gap” isn’t a meaningful concept without some sort of structural theory for “potential”. Given that I mostly avoid deep theoretical models in my work, I should be careful about any suggestion of “potential” in venture capital data.