Take Back Control: Economic Impact Analysis
By Saskia Poulter
Introduction
Announced in the 2024 Autumn Statement, the Growth and Skills Levy (GSL) is a hypothecated tax raised on the payrolls of businesses with annual revenue in excess of £3mn; set at 0.5% of a firm’s pay bill, it is worth approximately £3.9bn annually. Levy-paying firms may then use the money which they contribute to fund skills training programmes; that which they do not spend is redistributed – usually to non-paying SMEs – by the Treasury.
The GSL replaces the Apprenticeship Levy, which was introduced in 2017 and has been widely criticised for its rigidity, requiring firms to spend their training budgets on tightly regulated apprenticeships rather than shorter courses which may be better suited to their needs. Apprenticeship completion rates have been chronically low since the 2017 reforms, averaging 55.4% between 2022 and 2024. Research finds that apprentices often drop out because their employers fail to give them sufficient time to complete the training elements of their programmes, or because they realise that the content of their training is poorly matched to the requirements of their target sector.
Under the 2024 reforms, enterprises can access 50% of the funds which they pay into the GSL to spend on apprenticeships, and can spend the remainder on a broader range of training activities such as short courses or Bootcamps. Given the strong evidence that shorter training courses provide signifi cant boosts to productivity and output, with higher completion rates than apprenticeships and close matching to labour market demands, these changes should be celebrated.
We believe, however, that there are significant gains to be made from further increasing the flexibilisation of GSL spending in sectors which show exceptional promise for economic growth, and/or in occupations in which there are acute skills shortages. Specifically, we argue that firms in the IS8 industries should be able to spend 100% of their GSL contributions flexibly; spending on employees in the top ten ‘occupations in most critical and elevated demand’ (as defined by Skills England) could also be made completely flexible.[1] Targeting reforms at these sectors should deliver particular benefits by allowing the increased matching of skills to needs in firms for which skills access is a major bottleneck, and by further boosting productivity in areas in which the UK is seeking to cultivate a competitive advantage internationally. Importantly, the risks of deadweight loss – whereby government-funded training generates outcomes that are not additional to what would have occurred in absence of such funding (for example, where firms simply relabel training which they would have provided in the absence of GSL ringfencing) – are also likely to be low in these sectors: facing the problem of tight labour markets, it is strongly in firms’ interests to invest productively. This is particularly true given that, with reforms to visa eligibility conditions, it is becoming increasingly difficult to substitute domestic labour with skilled labour from abroad.
[1] Note that there is substantial overlap among these groups: many sectors in the IS8 – such as Clean Energy Industries and Digital and Technologies – employ more manufacturing labourers and scientific/technical professionals than the national average, for example.
For similar reasons, we further suggest that the newly increased Immigration Skills Charge (ISC) should be distributed among IS8 sectors, proportionally to their outcome. ISC revenue raised around £667mn annually prior to the 32% increase. Multiplying the current value of the ISC by this increase, we assume in what follows that this could result in revenue of £880m; we acknowledge that this is likely to be an over-estimate, however, given that the charge increase will likely reduce migration flows. We leave accounting for this nuance to future models.
The following analysis models the potential benefits to output and fiscal headroom resulting from described reforms to the skills spending in the IS8. It begins by setting out the models’ structures, before discussing parameter values.
Model Structure
Skills increases deliver productivity benefits both by increasing the efficiency of workers already in a sector, and by increasing the supply of effectively available skilled labour. These mechanisms are reflected in Model 1, where increases in skills are essentially represented as augmenting Total Factor Productivity:
Model 1a:
Ys, t+1= [1+α*Fs*(1-θs,A)*ρs,A*δs,A +(1-α)*Fs*(1−θs,S)*ρs,S*δs,S ]*Ys,t + E
or equivalently:
Ys, t+1= [α*Fs*(1-θs,A)*ρs,A*δs,A]Ys,t +[(1-α)*Fs*(1−θs,S)*ρs,S*δ s,S ]Ys,t + Ys,t + E
Model 1b (equivalent, with prior output represented as a Cobb-Douglas production function):
Ys, t+1= [1+α*Fs*(1-θs,A)*ρs,A*δs,A +(1-α)*Fs*(1−θs,S)*ρs,S*δs,S ]*(AKβL1-β)s,t + E
or equivalently:
Ys, t+1= [α*Fs*(1-θs,A)*ρs,A*δs,A]*(AKβL1-β)s,t +[(1-α)*Fs*(1−θs,S)*ρs,S*δs,S ]*(AKβL1-β)s,t + (AKL1-)s,t + E
Model Terms
Where the subscript ‘s’ denotes a parameter applying specifically to a sector of the economy (e.g. Defence or Life Sciences), and the subscripts ‘A’ and ‘S’ respectively signify that a parameter applies to Apprenticeships and Short Courses (a stand-in for training on which GSL funds can be spent aside from apprenticeships). To calculate the total effects on output, we sum the output changes calculated for each of the IS8 industries.
Model terms are defined as follows:
● Ys : output for industry s;
● α: the proportion of available GSL funds allocated by firms in a sector to apprenticeship training;
● Fs : annual GSL funds available to firms in sector s for training;
● θs: deadweight loss in sector s (percentage of GSL/ISC funds which are used for training which would have taken place in absence of the GSL/ISC intervention);
● ρs,A , ρs, S : completion rates for each training type (assumed to be homogenous across a sector);
● δs,A , δs,S : average productivity gain in the sector overall per unit spent on completed training in a sector;
● A: total factor productivity before GSL spending;
● K: capital;
● L: labour;
● β: output elasticity to capital — typically assumed to be ⅓; and
● E : error term — assumed to equal zero.
In words, Model 1 represents the effect on output in a sector from the augmentation of productivity, quantified as the share of GSL and/or ISC funds which are allocated to apprenticeships, multiplied by all available GSL and/or ISC funds, multiplied by the completion rate, multiplied by the average productivity boost generated by a one unit (£1) increase in spending apprenticeships which are completed (including spillovers), less the effects of that spending which is deadweight, multiplied by output before GSL investment, plus the effects of ‘short course’ (other skills training) spending, calculated in the same way.
We calculate the total economic effects by summing the equations calculated for each of the IS8 industries.
Assumptions and Limitations
In the models, we assume that all GSL/ISC funds available within the sector are spent: a reasonable simplification given that the Treasury reallocates any available funds which are not used by the firms which pay them through the levy. That is, their opportunity cost for spending their funds is unlikely to be high.
The model’s structure has some limitations, which we intend to account for in future versions. The first is that we do not assume diminishing marginal returns to skills spending; this could be incorporated by working with logarithmic functions. Nevertheless, we believe that given that skills shortages in the IS8 are acute, many firms will not reach the point of diminishing marginal returns to skills spending, which diminishes the effect which this assumption would have on our estimations. Moreover, many of the IS8 heavily involve frontier/advanced technologies, where there are likely to be complementarities with new innovations in capital/TFP, which keep the marginal skill’s/worker’s productivity high. Secondly, we observe that skills improvements could be modelled as augmentations to human capital as well as – or rather than – augmentations to TFP.
We use the latter approach primarily out of pragmatism, having found more existing research which helps to quantify the overall productivity effects of labour than that which helps calculate returns isolated to human capital alone. Thirdly, we note that the model does not explicitly account for the fact that the benefits from training will take time to come to fruition; we model annual returns as if taken at a point where the policy had ‘matured’ and courses were being steadily completed throughout the year.[2] Fourthly, we note that the economy – and especially the IS8 – are expected to grow over time, though we do not yet account for this when summing the accumulation of the policy’s effects over multiple years.
[2] That is, given that the benefits from longer training courses such as apprenticeships will typically take a multi-year period to be observed fully, we measure the annual effects of the training as if from a period after the first round of GSL/ISC spending, when the effects of the training have started to feed completely through to the economy.
We also note that the current model does not explicitly represent the effects of new labourers in industry s reducing labour supply in the sectors in which they previously worked. This is a reasonable simplifying assumption given that the roles in which workers were previously employed are likely to have required less specialised skills, so are likely to be re-filled. Some workers will also have been NEET prior to their training. The additive model also assumes that both types of training (A and S) are equally likely to bring a person into the labour force per unit of efficiency increase they deliver per currency unit spent – an assumption which can be justified by noting that both types can be tailored to bring in new workers in different ways. For example, apprenticeships might have a strong effect on new labour supply because they are often explicitly aimed at (close to-) fully training people with little experience in a sector, whilst short courses might also be effective at bringing people in given that they require less commitment from employers and trainees alike.
Estimating Parameters
a:
We assume that when firms are able to allocate their GSL/ISC spending flexibly between A and S, they do so in a way which maximises their expected productivity gain per unit spending, subject to constraints (e.g. quality preferences, hiring needs, capacity and switching costs). This assumption is particularly plausible for the IS8 given the potential benefits to them of higher productivity, and that they often face bottlenecks relating to a lack of skilled labour.
This is recognised in the Industrial Strategy itself, as well as other research. The former notes that ‘there are significant skills shortages across our frontier and foundational industries, and only 9% of Secondary vocational learners are studying in the in-demand sectors of engineering, manufacturing, and construction, compared to the OECD average of 32%’.[3] Additionally, over a quarter (27%) of all vacancies in the UK were associated with skills shortages according to the Employer Skills Survey 2024.
[3] The UK’s Modern Industrial Strategy, Pg. 64-5.
We assume that under the status quo, firms spend 50% of their GSL/ISC funding on apprenticeships and 50% on short courses ( α= 0.5), and that this is homogenous across IS8 industries. Under the proposed counterfactual, we assume they spend between 20% and 40% of their funding on apprenticeships; in calculations, we use a midpoint as a first approximation ( α= 0.3). The assumption that αstatus quo > αcounterfactual is partly justified by the fact that the government has anticipated a need to place a floor on apprenticeship spending: this suggests that the government expects that αwithout this requirement would be lower than 0.5. The fact that a significant proportion (£1.1bn in 2022) of Apprenticeship Levy funds have not historically been spent at all further suggests a strong revealed preference for lower apprenticeship spending, since this implies that some firms value apprenticeship delivery negatively (there would have been no direct opportunity cost to them to that spending given that it was ringfenced for apprenticeships).
In future versions of the model, we hope to provide more a precise and empirically grounded estimate for a plausible range of α, perhaps drawing on comparisons with other high income economies which have skills funding which can be spent flexibly.
Fs:
We assume that the value of GSL spending available to a sector is its share of the economy multiplied by the £3.9bn total GSL funding. This will be true if the proportion of the sector’s value which is delivered by firms earning above £3mn is similar in the IS8 as it is across the economy as a whole. We add to this a share of the £880mn raised from the ISC proportional to its output as a fraction of the total IS8’s, to give Fs.
Currently the IS8 comprises approximately 32% of GDP, meaning that collectively these industries would receive around £1.248bn from GSL; the sum of the eight Fs factors, with the additional ISC funding also incorporated, is therefore £2.128bn.
θs:
Existing work shows that the anticipated deadweight from skills funding interventions – here, the percentage of GSL/ISC funds which are used for training which would have taken place in absence of the funding interventions – differ dramatically depending on the design of the interventions and the market conditions in which they are implemented. Importantly, for example, the exact magnitude and balance of deadweight between apprenticeships and other skills spending will depend heavily on how tightly the government chooses to define ‘flexible spending’ (e.g. whether and how ‘flexible’ courses will have to be accredited – the government is yet to publish plans on this), since this affects both the perceived value of training to firms and the relative ease of paying for deadweight training through GSL/ISC funding pots.
In 2024, the national What Works Centre for youth employment found that 54% of Apprenticeship Levy payers reported having converted some of their existing training programmes into apprenticeships in order to use their allowance – though the research did not record what proportion of those firms’ spending was deadweight. We assume that deadweight loss under our policy counterfactual is likely to be lower than the level implied by this, given that our policy applies to the IS8. In large part this is because, as described above, we expect IS8 firms to buy in heavily to the importance of training, given the acute skills shortages which they face, the fact that they often require highly specialised workers, and that they can rationally expect high growth in the short- and mid-terms.
Our analysis draws on estimates of the deadweight associated with skills spending, as set out in a Civil Service report, but we have adapted these assumptions to fit two distinct models:
Scenario 1: Apprenticeship deadweight is assumed to be 0.05, and short course deadweight is 0.24.
Scenario 2: Apprenticeship deadweight is assumed to be 0.2, and short course deadweight is 0.24.
The short course deadweight figure of 0.24 is based on the Train to Gain programme, a large-scale skills funding initiative from the mid 2000s, which assisted firms in procuring a diverse spread of training – at 0.24 for firms judged to have ‘strong buy-in’ to the funding.
We assume that this figure is comparable to the deadweight associated with short course spending under our policy, given that that spending can similarly be allocated flexibly, and that firms are likely to see the benefits of training spending, for the reasons outlined above. . The 0.05 figure for Apprenticeships is also sourced from this Train to Gain programme Civil Service report, which found a full deadweight range of 0.05 to 0.5. We have taken the lowest end to offer a conservative estimate, assuming that the strict framework of apprenticeships adopted in recent years resulted in a lower deadweight. The 0.2 figure used in Scenario 2 for Apprenticeships is taken from the same report and is closer to the 0.24 deadweight associated with the Train to Gain programme for firms with ‘strong buy-in’. This figure represents a more optimistic scenario, where a moderate amount of deadweight is still present - this seems reasonable given the What Works Centre report (referenced above) which found 54% of Levy Payers reported rebadging at least some of their existing training as apprenticeships.
For simplicity, we assume that θs is the same across IS8 sectors. We also note that it is theoretically possible that, contrary to the empirically-derived figures given above, θs, S > θs, A , since it could require less effort to pass offexisting training as ‘new’ GSL/ISC training. Future versions of the model should seek to refine or provide a wider estimate range for the deadweight assumptions given here; in absence of existing empirical work to support this, however, we use figures directly from the literature.
There is a reasonable assumption to be made that this deadweight may indeed be lower than we have modelled. This is both because of the tight nature of labour markets in IS-8 sectors and importantly because current private sector spending on skills has been falling significantly and consistently, both in aggregate (19%) and per employee (31%) according to the IFS. We nonetheless have chosen a conservative assumption.
ρs:
Under the status quo, completion rates for apprenticeships ρs, A are notoriously low, at 0.554 between 2022 and 2024. Under flexible spending this is likely to improve, since more firms will offer apprenticeships only when they believe that doing so is a good use of their resources, and so will be more inclined to ensure that they are delivered well. Training delivered in formats other than apprenticeships have historically had higher completion rates than this: in 2022-2023, Skills Bootcamps had a completion rate of 0.65. Elsewhere, rates among more technical courses such as Green and Engineering – which are likely to be particularly useful in the IS8 – were higher, at 76% and 67% respectively. We therefore assume a range of 0.55 for ρs,A and 0.65-0.76 for ρs,S. As illustrative point estimates, we take the midpoint of the latter range: ρs,A = 0.55 and ρs,S = 0.705. It is noted this is the Wave 3 (most recent Bootcamps) rate, while Bootcamps in earlier waves have had significantly higher completion rates of up to 0.90, but our assumptions have utilised conservative estimates based on the most recent data.
We again assume that these values will be the same across IS8 sectors – a reasonable simplification, given that sectors will be able to tailor the kinds of flexible training in which they are investing in order to meet their needs and incentivise apprenticeships to follow through with their courses.
δs:
There is general consensus that training delivers substantial returns to productivity, allowing already-employed workers to produce greater output through the more efficient combination of factor resources, and by allowing the unplugging of bottlenecks. Stronger employee retention – resulting from higher job satisfaction as a result of having completed training – is a further factor which drives productivity elevation, reducing the need to retrain new staffand stemming the loss of institutional knowledge through staffturnover.
Nevertheless, estimating programmes’ average effect is difficult given selection effects into training and the diversity of the skills-related courses available. Leading research by Dearden, Reed, van Reenen (2006) estimates that a 1% increase in the proportion of workers who are trained in an industry is associated with a 0.6% increase in value added per hour; while returning efficiency effects in a similar region to the Dearden et al. (ibid.) result, Cavaglia et al. (2020) find that the wage premium from completing an apprenticeship in England varies widely depending on the trainee’s industry.[4]
[4] Nevertheless, in the analysis of Fersterer et al. (2008), for Level 2 Apprenticeships, those authors never put the wage premium below 3%, and find that it is never below 12% for Level 3 Apprenticeships. They find that premia are particularly high for many courses which closely map to the IS8: Level 3 apprenticeships in energy and water delivered a 23.4% wage return, whilst manufacturing apprenticeships delivered premia of 14.8%.
To arrive at a figure for δs, we use the following formula:
(labour productivity increase from training completion * effect of labour productivity on overall productivity * spillovers) / (cost * number of workers in the sector)
The first term is evaluated as proportional to the wage increases associated with training; this approximation will be most reasonable where, as in the IS8, labour markets are tight, so employees have greater bargaining power to demand that a greater proportion of their productivity increases are reflected in their wages. Cavaglia et al. (2020) provide estimated wage premia across a range of apprenticeship levels and subjects which are likely to be of use in the IS8. Among Level 3 apprenticeships, they find Engineering to deliver 40% returns, ICT 32% returns, Construction 26% returns, and Performing Arts 30% returns. For Level 2 courses, Manufacturing Technologies delivers a premium of 13%, Engineering of 18%, and Creative Arts and Design 23%.
As an approximation, we assume a point estimate of δs,A = 0.2; constructing a more precise estimate, tailored across industries, would be a useful extension in future versions of the model. We adjust this to the cost of obtaining the listed productivity improvements. In this case, the IFS prices an apprenticeship at approximately £9000 per year; if the average duration of the apprenticeships we use to derive estimations of efficiency improvements is 18 months, then these premia cost £13,500. We further assume, in line with NAO analysis, that training produces productivity spillover boosts to other workers in a firm, equivalent to an augmentation of 25%. We adjust the resultant value by the effect of labour productivity on overall productivity, according to a published estimation of 6%. Finally, we adjust this by the estimated total size of the labour force in the IS8:892000. The calculation (0.2*0.581[5]*1.25)/(13500*892000) returns the estimate that δ s,A = 1.2041189e-11.
[5] Based on 2025 Review of European Economic Policy analysis of impact of skills on TFP.
We assume that the effects on productivity per pound spent through flexible (e.g. short course) spending is 30% higher than for apprenticeship spending: δs,S = 1.5653546e-11. This is grounded in the observation that various short courses such as Skills Bootcamps have delivered very high productivity returns for in a matter of weeks or months rather than years; in 2022, for example, the government allocated £34mn to Heavy Goods Vehicles (HGC) ‘Skills Bootcamps’, which trained around 11,000 HGV drivers. The Government has not released data for other sectors as of 2022-2023.
In future versions of the model, we hope to provide an estimation for δs,S which is more strongly grounded in existing empirical estimates of returns to comparable skills policies, rather than theoretical intuition. Obtaining this kind of estimate would involve developing a theory of the kinds of skills and training demanded across IS8 sectors. Moreover, the model could be further improved by making explicit the role of growth in the labour supply over time. To sense-check the results delivered by our model, we could compare our final figures with estimates produced using the benchmark from Dearden et al. (2006), cited above.
Aggregating Effects across IS-8 Sectors
Using the above assumptions, the total change in output from the proposed policy in one representative year is therefore:
Yt+1l= 𝑠=18Σ[1+α*Fs*(1-θs,A)*ρs,A*δs,A +(1-α)*Fs*(1−θs,S)*ρs,S*δs,S ]*Ys,t + E
We use this for the following scenarios.
Status Quo Alternative: Apprenticeship Levy only
As a first approximation, we can also model a situation akin to the Apprenticeship Levy (we take the average of the three most recent course completion figures: ρs = 0.554):
Scenario 1 (Apprenticeship deadweight of 0.05):
Δ𝑌IS8, AL = £9,155,848,673.64
Removing the injection gives a return of £7,907,848,673.64
Scenario 2 (Apprenticeship deadweight of 0.2)
Δ𝑌IS8, AL = £7,710,188,356.75
Removing the injection gives a return of £6,462,188,356.75
Status quo for current GSL
Under the currently announced GSL policy (‘the status quo’, where 50% of the GSL is ringfenced for apprenticeships) we obtain:
Scenario 1 (Apprenticeship deadweight of 0.05 and short courses 0.24):
Δ𝑌IS8, SQ = £10,680,713,652.74
Removing the injection gives a return of £9,432,713,652.74
Scenario 2 (Apprenticeship deadweight of 0.2 and short courses 0.24)
Δ𝑌IS8, SQ = £9,957,883,494.30
Removing the injection gives a return of £8,709,883,494.30
Flexible GSL
Under the counterfactual policy (where GSL is distributed freely), we obtain: Scenario 1 (Apprenticeship deadweight of 0.05 and short courses 0.24):
Δ𝑌IS8, CF = £11,290,659,644.38
Removing the injection gives a return of £10,042,659,644.38
Scenario 2 (Apprenticeship deadweight of 0.2 and short courses 0.24)
Δ𝑌IS8, CF = £10,856,961,549.32
Removing the injection gives a return of £9,608,961,549.32
Flexible GSL PLUS ISC
Under the counterfactual policy (where the GSL and ISC is distributed freely), we obtain:
As a point estimate,
Scenario 1 (Apprenticeship deadweight of 0.05 and short courses 0.24):
Δ𝑌IS8, CF+ISC = £19,252,022,214.14
Removing the injection gives a return of £17,124,022,214.14
Scenario 2 (Apprenticeship deadweight of 0.2 and short courses 0.24)
Δ𝑌IS8, CF+ISC = £18,512,511,359.73
Removing the injection gives a return of £16,384,511,359.73
Mapping to Fiscal Headroom
These increases are expected to feed through to fiscal headroom forecasts through two channels in the short term. Firstly, the increase in GDP should increase tax receipts approximately in line with the tax-to-GDP ratio specified by the OBR of 37.7%. Secondly, assuming that a significant portion of workers in the IS8 move into that industry from economic inactivity prior to undertaking training, the policy will reduce state expenditure by reducing demand for Universal Credit and other welfare support.
Since we have not found any existing research which estimates the magnitude of workers who are likely to be newly taken into work through skills support in the IS8, we leave modelling this channel of feedthrough to fiscal headroom to future work.
Assuming that the 37.7% tax ratio applies directly to the estimated output increase, then a movement from Apprenticeship Levy to the proposed policy in the first year of maturation (i.e. when all of the productivity increases from the investment are observed) should equal between £805m and £3.74bn.
Flexible GSL compared to Apprenticeship Levy only
Scenario 1 (Apprenticeship deadweight of 0.05 and short courses 0.24):
Fiscal headroom = £2,134,810,970.74 * 0.377 = £804,823,735.97
Scenario 2 (Apprenticeship deadweight of 0.2 and short courses 0.24):
Fiscal headroom = £3,146,773,192.57 * 0.377 = £1,186,333,493.60
Flexible GSL PLUS ISC compared to Apprenticeship Levy only
Scenario 1 (Apprenticeship deadweight of 0.05 and short courses 0.24):
Fiscal headroom = £9,216,173,540.50 * 0.377 - £3,474,497,424.77
Scenario 2 (Apprenticeship deadweight of 0.2 and short courses 0.24):
Fiscal headroom = £9,922,323,002.98 * 0.377 = £3,740,715,772.12
This estimate does not account for the potential opportunity costs of the GSL/ISC being spent on skills training, nor does it account for annual GDP growth caused by factors other than skills increases.
