A brief parable of over-differencing

has sat through 1 also many seminars amongst triple differenced information , five fixed effects together with xxx willy-nilly controls. I wrote upwardly a piddling note (7 pages , simply also long for a weblog post) , relating the sense (from a Bob Lucas paper) that made me skeptical of highly processed empirical work.

The graph hither shows velocity together with involvement rates.  You tin run across the prissy sensible relationship.

(The graph has an of import lesson for policy debates. There is a lot of puzzling why people together with companies are sitting on thus much cash. Well , at null involvement rates , the chance toll of asset cash is null , thus it's a wonder they don't concur more. This mensurate of velocity is tracking involvement rates amongst just the historical pattern.) 

But when you lot run the regression , the econometrics books say you lot to operate start differences , together with and thus the whole human relationship falls apart. The estimated coefficient falls past times a component of 10 , together with a scatterplot shows no reliable relationship.  See the the note for details , simply you lot tin run across inwards the minute graph  how differencing throws out the of import variation inwards the data. 

The perils of over differencing , also many fixed effects , also many controls , together with that GLS or maximum likelihood volition boundary on giddy implications of necessarily simplified theories are good known inwards principle. But a few clear parables powerfulness brand people to a greater extent than wary inwards practice.  Needed: a similarly clear panel-data example.