Review and re-ingest IMF WEO data

[angelicalmcgowan]

1. Reviewing the IMF WEO data to figure out what to use for our models

In the querysets for fatalities002, there are four IMF WEO predictors related to annual growth of GDP included:

• ngdp_rpch_tcurrent
• ngdp_rpch_tplus1
• ngdp_rpch_tplus2
• ngdp_rpch_tminus1

Remco provided the following description of what these are/how the IMF WEO data is ingested to Maxine :

“The fetcher goes over every biannual WEO report since October 2007, and per report gets:

• All values of the included variables for that year (“_tcurrent”)
• The adjusted estimate for the prior year (“_tmin1", so 2006 for October 2007)
• The forecast for next year (“_tplus1”, 2008 for October 2007)
• And the forecast for the year after (“_tplus2").
  This is all put together and then staged to country-month. After that, extrapolated like:
  October 2007 --> April 2008 --> October 2008 --> etc.
  The “tcurrent” should be that of the April 2020 report, from month 484 (April 2020) forwards. “tplus” actually refers to the whole of 2021 in this case, which makes things a bit confusing. So “tplus” between April 2020 and October 2020 will be the forecast of 2021 as of April 2020"

From Sofia's research on this:

"The data has been moved from Views2 so there is no ingester what I can find. The latest data is from a report in April 2020. In Views we have monthly data. However, the data is actually on country-year level, but updated twice a year. We therefore have values for April and October each year. When I query the different tcurrent, tplus1 etc for 2018 and then look at the actual report from October 2018 the numbers for October in the different variables correspond to to the original data. So, in Views tcurrent for October 2018 is the value for 2018 in the October 2018 report, tplus1 is the forecasted value for 2019 and tmin1 in October is the the value for for 2017 according to the October 2018 report."

This seems like a rather confusing set-up, given that the IMF WEO now offers a single dataset with all historic GDP data + growth forecasts for the next five years (using the variable ngdp_rpch).

1. Can we simply this set-up by discarding the old data and instead using the dataset that IMF now makes available, then also gaining access to their forecasts for the next 5 years? See: https://www.imf.org/external/datamapper/NGDP_RPCH@WEO/OEMDC/ADVEC/WEOWORLD
2. Since they update their dataset with new actuals + new forecasts once or twice a year, do we then replace the old data with the new and retrain, or is there another solution to ensure that we use the latest data?

2. Reviewing the IMF WEO data to figure out what to share as raw predictor data in the API

In addition to the four variables above, there are also two more IMF variables in our database, both of which we have yet to figure out the definition of:

• f_ngdp_rpch
• s_ngdp_rpch

f_ngdp_rpch is what we used to share with ESCWA via the API, then calling it "annual GDP growth", presented as original data from IMF. I am not sure what this has been done correctly; I can't figure out how this variable is created and the data does not match what the IMF makes available. It's probably due to extrapolation or some other transform we've applied to the original data, which is not ideal for this purpose.

1. What are the two variables above?
2. We need to provide raw data from IMF on historic GDP growth per year + their most recent growth forecasts. Which database variable do we use for this? (If it even exists, given the complex set-up of the IMF data we currently have ingested).

3. Reviewing the IMF WEO data to figure out what to use as a surrogate model

What variable would we use to create a surrogate model based on GDP growth/growth forecasts? (Assuming ESCWA will want one)