Tax Gap: Preposterous nonsense or valid estimate?
Wendy Bradly has re-examined the methodology behind HMRC's tax gap figures and concludes that the calculations are based on some very dubious reasoning.
Last month I looked at HMRC’s tax gap documents and argued that they did not justify the administrative burden of Making Tax Digital. It was suggested on Twitter that I may have misunderstood the way the figures were put together in respect of the direct tax gap, so I went back to check.
The direct tax calculation is based on the amount of extra tax found to be due from an extremely small number of random tax enquiries; see table H1 in the Methodology document. Also I know from my experience of working tax enquiries as an HMRC officer that not all cases picked for enquiry are fully worked.
I asked HMRC whether I was correct to say that it:
- Investigates a small number of random cases;
- Assumes all unworked random cases have the same ratio of errors as the worked cases;
- Multiplies step 1 number using figures taken from US Random Enquiry Programs (para H28 of the methodology doc) and then
- extrapolates the result to the full UK taxpayer population.
The suggestion was that step 3 and step 4 were the same process, so that in essence I was accusing HMRC of "multiplying by the number they first thought of" not once, but twice.
Not so. HMRC confirmed my understanding of the third step is correct. The process really does seem to be investigate > extrapolate > multiply > extrapolate again.
Take it apart
Let's look at those steps in reverse.
“Extrapolate the result to the full population" is a reasonably scientific method, except you have to remember that you're starting from a very small number and assuming the results apply to a very large number. It's like battleships –one space to the left in your random selection and you hit a giant battleship, one step to the right and there's nothing at all.
Adjusting the figure by a multiplier derived from the US random enquiry programme troubles me. The US tax system, and particularly the US tax administration, are quite different from the UK system. Just about every taxpayer in the US is required to complete a personal tax return, where as only around one quarter of UK taxpayers are required to complete a self assessment return. There is an obvious "not comparing like with like" issue.
As for using a multiplier derived from a supposed metric of tax officers' success rates, I would need to be persuaded that US tax officials and UK tax professionals were looking for the same things in the same way after the same kind of training and experience. It's like looking at the number of bad eggs you get in your grocery shop and assuming you'll need to discard the same number of oranges.
The second "adjustment", assuming that any unworked random enquiries have the same proportion of errors as the fully worked ones is, as I discussed in my earlier article, at best a dubious proposition. Random cases are selected at random and are are de-selected for a reason. The reasoning for deselection is not in the public domain, but some possible reasons are not hard to fathom: this person is too old, frail, sick… Does that really sound as though the worked and unworked random cases will have the same profile?
Finally, back to the starting point: the small number of random cases. 2763 selections (table H1 in the Methodology document) of how many returns (10 million in HMRC's "record breaking" stats here, though that's all SA returns, not just businesses).
What does HMRC say?
When I asked HMRC to check my workings I also asked it for a quote and this was offered from Jon Thompson, HMRC’s Chief Executive:
“The UK is the only country in the world to regularly publish their tax gap in detail and at 5.7%, it remains at its lowest for five years. I am pleased that the downward trend shows HMRC and HM Treasury’s continued hard work to tackle evasion and avoidance is working.
“HMRC is also working hard to help taxpayers get their tax right by offering support and investing in digital services to improve businesses’ record keeping and reduce errors.”
This doesn't address my point. If the intermediate stages were cut out of the direct tax gaps calculation and the figure was simply the results of the random enquiries extrapolated to the rest of the taxpayer population, would the "downward trend" continue on the same trajectory? If so, why bother fiddling about with the headline figure at the risk of your credibility?