Update get_predictions_from_sims.jl

[SGHoekstra]

I think I found a bug in the get_predictions_from_sims.jl script. The wrong transformed variable was used in the cumprod function which led to negative numbers and domain errors. Before the variable is used in the cumprod function it should be exponentiated just like other variables in the script. This proposed fix makes sure that the right transformed variable is used.

[AldoGl]

Hi @SGHoekstra, thank you so much for your contribution!

I see you have changed

dict["x"] = exp.(x) .- 1
to
x = exp.(x) .- 1
+
dict["x"] = x

I like it as it makes the code more readable. We will merge this for sure, but is there really a bug? In which line?

In general, this script is still experimental. If you manage to improve it further (e.g., by making it more modular) please do not hesitate to open a PR!

[SGHoekstra]

Hi @AldoGl ,

This was the old situation:

real_government_consumption = sims.real_government_consumption
real_government_consumption_growth_quarterly = diff(log.(real_government_consumption), dims = 1)
model_dict["real_government_consumption_growth_quarterly"] = exp.(real_government_consumption_growth_quarterly) .- 1
real_government_consumption_quarterly =
data["real_government_consumption_quarterly"][data["quarters_num"] .== quarter_num] .*
cumprod(1 .+ real_government_consumption_growth_quarterly, dims = 1)

This is the new situation:

real_government_consumption = sims.real_government_consumption
real_government_consumption_growth_quarterly = diff(log.(real_government_consumption), dims = 1)
real_government_consumption_growth_quarterly = exp.(real_government_consumption_growth_quarterly) .- 1

model_dict["real_government_consumption_growth_quarterly"] = real_government_consumption_growth_quarterly
real_government_consumption_quarterly =
  data["real_government_consumption_quarterly"][data["quarters_num"] .== quarter_num] .*
  cumprod(1 .+ real_government_consumption_growth_quarterly, dims = 1).

In the old situation the variable real_government_consumption_growth_quarterly was used in the cumprod function before the exponent function was applied. So the exponent was applied to model_dict["real_government_consumption_growth_quarterly"] but not to real_government_consumption_growth_quarterly. So If you pay close attention I made sure that the exponent function is applied to the variable before it is being used in the cumprod function. This is not necessarily a programming bug but more a mathematical error. This also happend to some other variables for which I also made the changes. For most other variables like real_gdp this was already implemented correctly. If you do not apply these changes you can get negative growth numbers which do not make sense. Hope that this makes everything more clear!

[AldoGl]

Thanks @SGHoekstra, now it's clear, well spotted!

I am merging this now, welcome to the contributors of the project! For the moment I will add you here here but we might need to add an "Authors" file at some point. Once again, don't hesitate to push further improvements