By Josh Martin
The coronavirus pandemic caused major disruption to the UK economy, leading to a rapid increase in homeworking and online retail, amongst other things. It also led to faster deployment of digital-enabled business practices, such as digital healthcare, online ordering in restaurants (remember the QR-code menus?), and remote teaching. One thing all of these activities surely have in common is the need for specialist software.
It seems strange then, that investment in software in the official business investment statistics from the ONS appears so low during and after the coronavirus pandemic.
In these figures, there are two main types of software investment: “purchased” and “own-account”. Purchased software relates to purchases from another business, either as a license to use an off-the-shelf software (such as Microsoft office), or a bespoke software product developed by a specialist company. Either way, a transaction occurs, and this makes it possible for businesses to report this investment on business surveys run by ONS.
The other type of software investment, so-called “own-account” investment, is software created in-house by employees of the business. Rather than buy software on the market, a business might opt to create it in-house, giving it more flexibility and agility, and perhaps also making it a better fit for its needs. This also carries a cost but doesn’t come with a sticker price – instead, the costs are the wages and salaries of the software developers and IT professionals in the business, as well as a share of the overheads and use of computers.
Following international guidance, investment in own-account software is estimated using a “costs of production” or “sum of costs” method – that is, estimating all the costs involved in producing the software in-house, and adding them up. The largest cost, and the one most easily identified, is the labour costs – this is some fraction of the wages of the relevant workers in the business. To do this, the ONS identifies workers with relevant occupations, assumes a fraction of their time is spent producing in-house software, and capturing that fraction of their wages as investment costs. On top of this, the ONS adds non-wage labour costs (pensions etc.), costs of goods and services used, a share of overheads, costs of using capital assets, and an assumed profit margin (to mirror the market price). This is the standard approach in National Accounts, and the ONS is amongst the best in the world in this area.
The data source for the labour costs used by ONS is the Annual Survey of Hours and Earnings (ASHE) – a business survey, collecting all the necessary information, and with a large sample. Since the survey is completed by employers, the pay data and industry identifier are thought to be of a high quality.
However, ASHE is an annual survey, collecting a snapshot of information in early April of each year. Most of the time, early April is probably a fair reflection of the rest of the year when it comes to software professionals.
But during the pandemic things changed quickly, and April 2020 was anything but normal. Data on employment and pay of software professionals in April 2020, when the country was in lockdown and millions of workers were on furlough, might not be representative of the rest of 2020, let alone 2021.
Given lags with the survey collection and National Accounts process, ONS is currently using April 2020 ASHE data to estimate own-account software investment all the way to the end of 2021. Most of the time this wouldn’t be a problem, but the rapid changes during and after the pandemic made this normal process problematic.
First, rather than treating data from April 2020 as representative of the rest of 2020 and 2021, the quarterly LFS allows up to date estimates of investment every quarter. So, when there is a big burst in investment (and it seems there was), we know about it straight away.
Second, using a quarterly datasource allows a truly quarterly estimate. Using an annual source like the ASHE might be good as an annual benchmark, but cannot tell us about within-year changes. And given the big quarterly swings in economic activity during 2020, having a genuinely quarterly pattern is important.
Third, the LFS may even be a better data source than the ASHE in normal times. A crucial part of the own-account software method is the identification of relevant workers by their occupation. The coding of occupations on ASHE is partly automated and partly manual, based on job titles written by survey respondents on paper forms. Job titles can be tricky to interpret out of context, especially for fast-changing fields like IT. Despite the best efforts of the ONS, the allocated occupation codes on ASHE might misclassify IT workers. Meanwhile the occupation coding on the LFS is done by interviewers who have the relevant context, which might make them more accurate. Changes to methods implemented by ONS in future may change this judgement.
The chart below shows the ONS estimate of own-account software investment (based on ASHE), and the estimates from my new paper (based on LFS). They track each other closely between 2002 and 2019, in both level and trend. However, in 2020 they diverge – with the ONS estimates roughly flat, and the LFS-based estimates rising quickly. In 2021 that divergence continues, with the ONS estimates falling about 15%, and the LFS-based estimates continuing to increase before levelling off.
Rapid software investment at a time when the economy was rapidly adopting digital business models seems appropriate. The paper also demonstrates that this is more in line with estimates in other countries, and that the increase in cost is mirrored by increases in hours worked.
Because own-account software investment reflects non-market transactions, it captures not only business investment but also additional output. In fact, all three measures of GDP are affected simultaneously, so revising own-account software investment up would also revise GDP by the same.
As well as these quarterly estimates, the paper also explores other methodological improvements for estimates of own-account software investment, including the treatment of the self-employed, and the calculation of sales adjustment factors. I also proposed some further applications of the approach introduced in the paper, such as the development of quarterly estimates of R&D, and other own-account intangible investments. These provide a roadmap for future research.
 The comparison is complicated by an earlier processing error in the official estimates, which has been corrected in Blue Book 2022, although the necessary detailed breakdowns have not yet been published.
Read the full ESCoE Discussion Paper here.
ESCoE blogs are published to further debate. Any views expressed are solely those of the author(s) and so cannot be taken to represent those of the ESCoE, its partner institutions or the Office for National Statistics.
The views expressed here are those of the author and should not be taken as the views of the Bank of England or any of its committees, or the ONS.