Handling of UKMod Outputs as Policy Schedule Inputs in SimPaths

[AndreasxHoehn]

I am currently stitching together a pipeline to allow for a smooth run from scenario definition to UKMod (Training Data & FRS-based) to SimPaths (Training Data & UKHLS/WAS) to post simulation analyses.

I have now noticed that SimPaths seems to require UKMod outputs for all years (e.g. 2011-2026) - even though these years are not considered in the simulated period. Running SimPaths from command line without all (currently required) UKMod output files makes it hang.

I wonder if it needs changes to the model code as this behaviour is not properly controllable with the file "EUROMODpolicySchedule.xlsx" in the input folder.

[pbronka]

The simulation requires a UKMOD output file corresponding to the base price year used in the simulation, which is 2015. If it is not provided, it will fail because of the following runtime exception:

https://github.com/centreformicrosimulation/SimPaths/blob/a4f7864caab3e5787acfbccda17ea6494adc3df1/src/main/java/simpaths/data/Parameters.java#L1954-L1955

Edit: It might be helpful to look at the messages displayed in the console, for example in this case:
Exception in thread "main" java.lang.RuntimeException: Must include a tax database generated for the base price year assumed for analysis (2015) at simpaths.data.Parameters.calculateEUROMODpolicySchedule(Parameters.java:1953) at simpaths.model.taxes.database.TaxDonorDataParser.constructAggregateTaxDonorPopulationCSVfile(TaxDonorDataParser.java:446) at simpaths.experiment.SimPathsStart.runGUIdialog(SimPathsStart.java:408) at simpaths.experiment.SimPathsStart.main(SimPathsStart.java:70)

[AndreasxHoehn]

Many thanks @pbronka. I got things all to run just now thanks to @andrewbaxter439 !

Some background;
The idea was to use FRS-based UKMod Output for 2019-2026 only (based on FRS UK_2019_b1.txt)
With UKHLS/WAS based initial population for 2019. (population_initial_UK_2019.csv)

The problem was that, when creating scenarios, changes are also required in;

1. “parameters.java” (MIN_START_YEAR, MAX_START_YEAR, BASE_PRICE_YEAR)
2. recompile SimPaths
3. In the shell command add “--rewrite-policy-schedule” to the single run.

It's not super obvious from the documentation and there might be a better way going forward by feeding in .yaml files that will do this (means no re-compiling required and better for pipeline mode operation).

[pbronka]

Important to note, however, that all monetary values used to estimate the transition probabilities are in 2015 prices. Changing the base price year most likely requires all processes to be re-estimated. I think it would be better to keep the base price year set to 2015 and provide that extra policy file.

[andrewbaxter439]

Ah that's good to know, thanks Patryk!

[AndreasxHoehn]

Thanks for clarifying @pbronka !

Is there a case to provide MIN_START_YEAR, MAX_START_YEAR through a yaml file rather than changing it in a java file + recompiling? The alternative is to make fully clear in the documentation that UKMod outputs need to be provided for period X .... Y

[justin-ven]

I think the MIN_START_YEAR and MAX_START_YEAR parameters are mostly to ensure the existence of input data for the starting population. I'd be inclined to keep the YAML files to parameters that can be varied by users between simulations. Alternatively, we could adjust the code to suppress the need for the two parameters listed here - the model could be directed to use whatever cross-sectional data are closest to the requested start year and work from there...

[AndreasxHoehn]

I have edited the Wiki to reflect the need for a UKMod output file for the base price year (2015). As the transition probabilities are based on this, there is likely no other way really that is pragmatic.

It might make sense to keep at least the MAX_START_YEAR parameter flexible as there are more recent initial population files available already.

[andrewbaxter439]

Skimming through issues to check if anything I can do to help tidy up. This one could be closed but giving a suggested, more focussed issue:

• Suppress need for MIN_START_YEAR and MAX_START_YEAR parameters to match supplied initial populations

The reasons for this would be:

• Not all starting populations from 2011 to 2021 are needed for every run
• Changing this currently needs either adjust parameters/recompile, or setting in yaml (neither of which are inherently very tidy)
• Detecting files and dynamically setting min/max would suit all use cases?

[andrewbaxter439]

Closing as potentially resolvable through #229 and linked PR #248