- The observation_trends block is already specified in terms of linear trends, so it would increase consistency.
- It is unclear how the current system interacts with the loglinear keyword (is it applied automatically when automatic detrending is used?), or with variables specified in logs (and hence having linear trends).
- The natural intuitive approach when faced with a model with exponential growth is to express everything in logs. The cost of doing this in dynare in pretty minimal thanks to model local variables (my model declarations all start with code like #Y = exp( y(0) ); #LAG_Y = exp( y(-1) ); etc.) It is clear that the loglinear keyword is a shorthand for this, and hence that even in this case you're really removing linear, not exponential trends.
Thanks for all your hard work.
Tom