I was asked yesterday about why I am using HMRC aggregate data to estimate the impact of the tax proposals that I am making in the Taxing Wealth Report 2024.
The person asking suggested that I should, instead, be using micro simulations, i.e. economic models. This, they suggested, is the approach used by the IFS, the Resolution Foundation, Compass, and many other organisations who look at some of the issues I am addressing. I am well aware, for example, that is the approach used by Howard Reed, who comments on this site and whose work I much appreciate.
In that case, let me be clear that I do not have a problem with people using microsimulation data. It is an accepted technique, and the data that is used for this purpose is based upon ONS household surveys, which are considered to be statistically valid, although there are always reasonable questions to ask with regard to selection bias whenever they are used.
However, as an accountant and auditor, rather than as an economist on this occasion, I am always inclined to use available aggregate data rather than produce a simulation based upon a survey. The reasons are obvious to me.
Suppose, as an auditor, I have to choose two approaches to verifying data. Suppose that I was trying to establish the figure for sales for a company. Which technique would I prefer? Would I prefer to go to the company's books and records, from which I can extract a figure for aggregate sales which I can then test for credibility and completeness, or would I rather do a survey, asking people whether they have bought from the company, and then extrapolate total potential sales from that? I think I can tell you which one is the accepted audit technique. I can also confirm that only one of those would pass scrutiny if the quality of an audit was being reviewed.
Saying that, I wholly accept that there are problems with aggregate data. One of those is that HM Revenue & Customs is sometimes inconsistent with regard to the information it publishes, because revisions are not unknown. That is why I use the most recent verified data in some cases rather than more recent estimated data. I am seeking the most reliable data source that I can, even if it is older as a consequence.
Another problem with this information is with regard to stratification. Sometimes the data is simply not stratified in the way that I would wish. I accept that problem because it is real. However, it is usually not very difficult to estimate data points within the available information to approximate the stratification that is required. So long as caution is used, I think that this is acceptable.
Then there is the obvious problem of behavioural responses, but this is absolutely the same with microsimulation techniques. Why that methodology should better estimate a behavioural response then I can, I do not know. In both cases, the person creating the tool is expressing their opinion on what people might most commonly do in response to a tax change. I explicitly mention my own assumptions on this in every one of the notes that I produce so that no one can miss the issue.
Some of the suggestions I make on behavioural responses may be countercultural to many people. For example, I suggest that many people who are asked to pay more tax work more, rather than less, as a result, although microeconomic theory would usually suggest otherwise. However, microeconomics can be wrong. The reality is that many people live in a world where they have very strong fixed financial commitments, so tax increase requires them to work more, whatever microeconomic theory says.
In other cases, I suggest that although a change might be significant, it will produce very little behavioural response because the structure of society does not really permit that. For example, tax relief on pensions might change, but if contractually a person is obliged to commit to pay a fix percentage of their salary in pension contribution then that is what they will still do so, in most cases. as a consequence. As a result, I very often suggest that the behavioural response to what I am suggesting will be small because that is what a reasonable appraisal of the evidence suggests.
Thirdly with regard to aggregate data, by providing links to the data that I use from HMRC and elsewhere, what I hope I provide is an opportunity for someone to have a go at creating their own estimate if they do not like mine. I usually find that very difficult when looking at microsimulation data.
Finally with regard to aggregate data, it is available for all the many taxes I am looking at, whereas data to allow microsimulation on many of them on a reliable is not, in my opinion. As a result, I am using aggregate data to be consistent throughout the work, which I think is very important.
What I stress is that none of this implies that I think that I am necessarily right when making forecasts. I am presenting estimates. I do not claim that they will be anything more than that. But that is the best that anyone can do. By suggesting that there is an opportunity for someone to have another go and come up with another answer, I hope that I am making clear my own fallibility, or at least subjectivity. That is inherent in all work of this sort: when assumptions change with regard to forecasts, so do the resulting answers.
Overall, I have one objective, and that is to provide an indication of the consequences of plausible changes. Absolutely, no one can say such indications are right, and I do not. But, making such forecasts is a normal part of the life of an accountant or an economist, and so it is an entirely reasonable thing to do, so long as it is understood that the suggestion is based upon plausible, verifiable data, to which reasonable assumptions have been applied, and reasonable behavioural responses have been considered.
Over the whole range of work that I am doing in the Taxing Wealth Report 2024 the use of aggregate data best lets me make the forecasts that I want to present. I will be sticking with my methodology.