To-Dos
January 1, 2022
It’s like painting the Forth Bridge, this. I’ve got to get back to the day-job stuff that pays the bills. Here’s a rough list of what I think currently needs to be done. It’s a lot. There are lots of TODOS and FIXMES dotted through the code - I count 256 of them. Often these are notes about duplicate code, bad naming practices, or things that I’ve hard-coded which should be parameterised. Often things that need checked again.
Here are the main things, I think:
Verification
The tests all pass, and we’ve got plenty of them.But we’re at least £1bn over in income tax revenues and inequality is too low. And I haven’t checked benefits against XPlore yet.
Takeup
Well, I wrote the book. With the wonderful Vaness Fry. Needed for inequality especially, I think. I wonder if this stuff has moved on? Or Backwards?
Matching
The matching in of Scottish Household Survey is very ad-hoc, and I’ll likely need to go again with LCF data. So this all needs sorting out. There’s an R implementation of the EuroStat approach that I need to either get working or (better) port to Julia.
Pensions Data
FRS has almost zero employer’s pension contributions. That alone is likely enough to account for the extra billion in tax revenues. There are a couple of sources that I could use to impute something here. probably need a literature search as well - someone must have noticed.
Uprating, weighting and timing
- All the model data is uprated and weighted to a point in time. Which means that effectively we just skip over the pandemic (or would be if there wasn’t Omicron). I’m thinking of uprating and weighting from the raw data to the corresponding month of the current year.
- The SFC budget 2022 forecasts need added.
Coding
Lots needs done especially to get this into a state where other people could work on the code:
- naming function and structure names are inconsistent, the names of enumerations are often way too long since they were autogenerated from the UKDS raw documentation;
- structures - in particular the results and household structs are constructed and named quite differently;
- code style: I claim to follow blue style but clearly don’t;
- tests there are plenty of them, but they don’t follow the code structure anything like closely enough.
Data
- I need to add LCF data, through matching I think - Engel curves/a complete demand system is not a goer I think - I should write about why.
- document the many problems I’ve found in the FRS and submit to someone;
- There’s a year of FRS to add - there’ll always be an extra year.
But I suspect the TODOs will increase ..