Research and analysis

Evaluating additional tax revenue from Making Tax Digital for VAT

Published 10 March 2022

1. Glossary of analytical terms

Self-selection bias: This can arise if businesses that voluntarily sign-up to Making Tax Digital have different characteristics from the ones that do not. For example, businesses with more digital capability may not only be more likely to join Making Tax Digital early, but also be more likely to grow faster and pay more tax over time. To produce best estimates, it is essential to reduce self-selection bias.

Randomised control trial: An experimental design that uses randomisation to prevent selection bias. Businesses would be randomly allocated to join Making Tax Digital and are otherwise prevented from joining. In practice, all similar businesses were able to join Making Tax Digital around the same time, so this ideal way of evaluating the revenue impact of Making Tax Digital was not available.

Propensity score matching: An econometric technique to reduce bias from voluntary take-up. It uses a model to create 2 similar groups for comparison with the remaining key difference being whether they are filing through Making Tax Digital.

Logistic regression model: A statistical model that is used to estimate the probability of a business being in Making Tax Digital, given its particular business characteristics and behaviours.

Difference-in-differences: An econometric technique to mitigate the effects of self-selection bias. It is used to compare the difference in the average declared tax liabilities between the sample groups – one in Making Tax digital and one not. The method calculates the average change in declared tax liabilities as businesses sign-up to Making Tax Digital and produces an estimate of additional tax revenue.

P-value: The probability of achieving results at least as extreme as the average additional tax revenue estimate under the assumption that there is zero additional tax revenue. Lower p-values indicate greater evidence for the presence of additional tax revenue from Making Tax Digital.

95% confidence interval: A range for which we have reasonable certainty will contain the true average additional tax revenue generated due to Making Tax Digital. A 95% confidence interval means that if the analysis were to be repeated a reasonably large number of times, then the proportion of calculated confidence intervals containing the true population mean additional tax revenue would tend towards 95%.

2. Executive summary

Making Tax Digital is a key part of HMRC’s transformation of the tax system to make it easier for taxpayers to get their tax right first time; help reduce scope for error and enhance customer experience.

Since 2019, Making Tax Digital has required VAT-registered businesses with taxable turnover above the VAT £85,000 threshold to keep digital records and use compatible software to submit their VAT tax returns directly to HMRC. Making Tax Digital will apply to all VAT-registered businesses, including those with taxable turnover below the VAT threshold, from April 2022. Since late 2018, VAT businesses have been able to sign-up to Making Tax Digital on a voluntary basis.

Qualitative^{[footnote 1]} and quantitative research demonstrates that Making Tax Digital compatible software can improve the accuracy of returns by removing opportunities to make certain types of mistakes when preparing and submitting tax returns. A reduction in errors is expected to result in a net increase in tax revenue known as additional tax revenue. This is one of the Government’s key outcomes for implementing digital record-keeping and digital filing requirements as part of the Making Tax Digital programme.

This paper outlines the methodology and results of the statistical modelling that was used to evaluate the additional tax revenue from Making Tax Digital for VAT.

We cannot directly observe additional tax revenue in VAT returns as it is difficult to isolate from other factors affecting VAT revenue over time. This modelling used the pilot phases of Making Tax Digital to create 2 similar groups for comparison, one in Making Tax Digital and one not. Additional tax revenue is then estimated by comparing the difference in the average tax liabilities between the 2 sample groups

The analysis was conducted on VAT registered businesses that file their returns quarterly. Two sub-populations were modelled separately: businesses above the £85,000 VAT turnover threshold and businesses below this threshold.

The results indicate that there is likely to be a positive additional tax revenue for both populations. For the population below the turnover threshold, the estimated additional tax revenue due to Making Tax Digital is an average of £19 per business for the quarter. This additional tax revenue represents a 2.2% increase from the average liability estimated had the businesses not signed-up to Making Tax Digital.

For the population above the threshold, the estimated additional tax revenue due to Making Tax Digital for the quarter is an average of £57 per business. This represents a 0.9% increase from the average amount estimated had the businesses not signed-up to Making Tax Digital.

We can extrapolate these results to all businesses that had joined Making Tax Digital in the 2019 to 2020 tax year to compare to previously published forecasts. This results in estimated total additional tax revenue from Making Tax Digital in 2019 to 2020 of around £185 million or £195 million depending on the method chosen. This compares with a previously published estimate of £115 million.

The high variation in the value of the tax returns introduces uncertainty in the modelling estimates. This is outlined in detail in the paper.

This research provides strong evidence that Making Tax Digital is achieving its objective of reducing the tax gap by reducing the amount of errors made when filing tax returns.

The additional tax revenue is likely to be positive and of a similar magnitude to previous forecasts included in the public finances for the 2019 to 2020 tax year.

3. Methodology

Making Tax Digital for VAT aims to reduce the tax gap by helping businesses pay the right amount of tax. The tax gap is the difference between the theoretical amount of tax that should be paid and the actual tax receipts^{[footnote 2]}. The difference is caused by several reasons including avoidance, evasion, and calculation errors or failure to take reasonable care when filing returns.

Making Tax Digital tackles the part of the tax gap which is caused by error and failure to take reasonable care. In Making Tax Digital, businesses are required to keep records in digital form and file their VAT returns using software that directly extracts information from these digital records.

This improves accuracy and removes opportunities to make certain types of mistakes in preparing and submitting tax returns, particularly arithmetical and transposition errors. Without Making Tax Digital these errors would either be missed, or they could lead to compliance intervention at an additional cost to HMRC.

Errors when filing tax returns could be in the favour of the business (underpaying tax) or in the favour of HMRC (overpaying tax). Of the 2, the total value of errors in the taxpayer’s favour is the greater. Therefore, by reducing errors Making Tax Digital is expected to result in a net increase in tax revenue known as additional tax revenue.

Since late 2018 VAT businesses have been able to sign-up to Making Tax Digital on a voluntary basis. Making Tax Digital became mandatory for VAT businesses with a taxable turnover of £85,000 or more from April 2019. Making Tax Digital becomes mandatory for VAT businesses below this threshold from April 2022.

Propensity score matching in conjunction with a difference-in-differences model were used to estimate the additional tax revenue from the introduction of Making Tax Digital for VAT taxpayers who file their returns quarterly. The analysis was performed separately for small businesses^{[footnote 3]} above and businesses below the £85,000 VAT threshold, using data from the respective voluntary sign-up periods.

Ideally to evaluate the impact of Making Tax Digital on tax returns we would use a randomised control trial. This method would create a treatment group mandated to join Making Tax Digital, and a control group prevented from joining. However, for clarity and fairness we cannot prevent anyone from joining Making Tax Digital and the mandation dates must be the same for all similar businesses. Therefore, a randomised control trial was not possible.

We can, however, observe businesses that chose to voluntarily sign-up to Making Tax Digital in the pilot phase and compare them with similar businesses that chose not to sign-up.

By observing businesses in the voluntary sign-up period there is the possibility of self-selection bias. For instance, businesses that voluntarily use Making Tax Digital may be more digitally capable on average and therefore benefit from an early move to digital record keeping.

Consequently, there is likely to be pre-existing differences between volunteers and non-volunteers which are indistinguishable from the impact of Making Tax Digital.

The propensity score matching attempts to reduce the impact of this self-selection bias. Businesses that voluntarily chose to sign-up to Making Tax Digital are matched to similar businesses that chose not to sign-up.

This method involves building a logistic regression model to estimate the probability of a business being both signed-up and having submitted a return through Making Tax Digital.

Various business characteristics and behaviours, such as whether the business already uses software and which industry they trade in, were used as independent variables in this model. The logistic regression equations are shown in the Appendix, Equation 4 and Equation 5. The definitions of the independent variables are also given in Table 9 in the Appendix.

From the logistic regression model, the output probability of being in Making Tax Digital is known as the propensity score. A nearest neighbour algorithm was then used to match businesses on their propensity scores.

The algorithm attempted to match each business which voluntarily signed-up to Making Tax Digital to the closest business, in terms of propensity score, that did not sign-up. If a business in Making Tax Digital does not have a close match, then it remains unmatched. This forms the treatment and control groups.

All unmatched businesses are removed from further stages of the analysis. By definition, the unmatched businesses have different characteristics to the matched businesses. Therefore, the final estimate of the Making Tax Digital impact may not reflect the unmatched population.

After matching the businesses, a difference-in-differences model was applied to evaluate the additional tax revenue due to Making Tax Digital. The difference-in-differences model compares the difference in the average tax liability declared between the matched treatment and control groups between 2 time points. The time points were chosen one year apart to account for seasonality.

At the first time point, none of the businesses in the treatment or control groups had signed-up to Making Tax Digital. By the second time point a year later, all the businesses in the treatment group were using Making Tax Digital. We then calculated the difference between the 2 groups and the difference between the 2 time points – hence the difference-in-differences.

Equation 1 defines the difference-in-differences methodology. Here, DiD is the difference-in-differences estimator. T and C refer to the treatment and control groups respectively, and the labels 1 and 0 refer to the most recent return (time = 1) and the return from one year earlier (time = 0).

The difference-in-differences estimator is the isolated average change in tax declared due to Making Tax Digital, controlling for inherent biases between groups as much as we can.

Equation 1 – Difference-in-differences equation

DiD = (T₁ - T₀) - (C₁ - C₀)

Before conducting the matching and difference-in-differences models, both populations were capped at varying degrees to remove outliers and reduce the variance in the tax return data. The capping was conducted on the net tax returns at both time points in the difference-in-differences model. All businesses outside these capping limits were removed from the analysis, thus removing those with more extreme net tax returns that can skew the results.

Capping is suited to this sort of evaluation as the distribution of tax liabilities from VAT businesses is stretched to very large values declared by only a small number of businesses. This means that capping at both ends of the distribution can reduce the variance.

A range of sensitivity tests were performed on the capping of both populations. The caps were chosen such that they balanced reducing the variance whilst still retaining large sample sizes.

4. Results

4.1 Evaluation below the turnover threshold

The population below the £85,000 turnover VAT threshold are still in the voluntary sign-up period of Making Tax Digital, with mandation due from April 2022. Table 1 shows the accounting periods and return due dates at which data were extracted for the propensity score matching. For the difference-in-differences model, data were collected on the net tax for the return due dates in Table 1 and exactly one year before this^{[footnote 4]}.

Table 1 – Observation period - below the threshold

Return stagger	Accounting period	Return due date
1	October 2019 – December 2019	7 February 2020
2	November 2019 – January 2020	7 March 2020
3	December 2019 – February 2020	7 April 2020

Table 2 shows the results for this population and some statistics which are explained here.

Over the 3 accounting periods in Table 1, approximately 177,000 businesses were matched - half in the treatment group and half in the control group.

The 177,000 matched businesses in the treatment and control groups combined represent approximately 18% of the total VAT registered businesses below the threshold (including non-quarterly businesses and other businesses excluded from the model).

The estimate of additional tax revenue due to Making Tax Digital for the matched population is an average of £19 per business for the quarter.

The average declared tax liability for the signed-up population for the quarter is predicted to have been £880 if they had chosen not to sign-up to Making Tax Digital. The additional tax revenue of £19 represents a 2.2% increase in liability.

The p-value calculated for the result of £19 additional tax revenue is 0.09. This means that there would only be a 9% chance of detecting additional tax revenue at least as extreme as £19 (≥ £19 or ≤ -£19), under the assumption that there is no true additional tax revenue from Making Tax Digital.

Table 2 – Results – below the threshold

Population	Sample size	Percent of total population	Average estimate of additional tax revenue	Additional tax revenue as percentage increase in tax liability	p-value
Below the Threshold	177,000	18%	£19	2.2%	0.09

The full difference-in-differences model used to produce these results is shown in the Appendix in Equation 6. Also, in Table 10 is the standard error and 95% confidence interval for the additional tax revenue estimate.

4.2 Evaluation above the turnover threshold

Businesses above the £85,000 threshold were mandated to join Making Tax Digital from the first VAT period starting on or after 1 April 2019. The population was modelled in the quarter before mandation when sign-up was still voluntary. The return due dates at which data were extracted for the propensity score matching are given in Table 3.

Table 3 - Observation period – above the threshold

Return stagger	Accounting period	Return due date
1	January 2019 - March 2019	7 May 2019
2	February 2019 - April 2019	7 June 2019
3	March 2019 - May 2019	7 July 2019

Table 4 shows a summary of the results.

A total of 223,000 businesses were matched over the 3 accounting periods in Table 3. This represents around 16% of the total VAT businesses above the threshold.

The estimate for the quarter shows an average of £57 additional tax revenue due to Making Tax Digital per business.

The average tax liability for the signed-up population for the quarter is predicted to have been £6,300 if they had not signed-up to Making Tax Digital. The additional tax revenue of £57 represents a 0.9% increase in liability for this population.

The p-value calculated for this result is 0.20. This means that under the assumption that there is no true additional tax revenue from Making Tax Digital, then there would be a one in 5 (20%) chance of observing results at least as extreme as £57 (≥ £57 or ≤ -£57).

Table 4 - Results - above the threshold

Population	Sample size	Percent of total population	Average estimate of additional tax revenue	Additional tax revenue as percentage increase in tax liability	p-value
Above the Threshold	223,000	16%	£57	0.9%	0.20

The full difference-in-differences model used here is given in Equation 7, in the Appendix. Table 10 shows the standard error for the additional tax revenue estimate along with the 95% confidence interval.

Several tests were conducted to assess both models’ estimates and our confidence in their validity. These included:

• testing the modelling assumptions
• checking the quality of the matching and testing an alternative matching method
• testing the fit of the difference-in-differences model
• sensitivity tests on the observation period and the population capping

The tests confirm that for both populations the results are valid, and the order of magnitude and direction of impact is correct. However, the sensitivity tests suggest the results are sensitive to changes in the setup of the experiment, such as the period observed.

5. Comparison with forecast additional tax revenue

From the evaluation estimates above, we can conclude it is likely that there is positive additional tax revenue from Making Tax Digital. In this section of the report, we assess whether these estimates are broadly comparable to the additional tax revenue that Making Tax Digital was predicted to generate.

The public finances include the forecast additional tax revenue that Making Tax Digital is expected to raise for the Exchequer^{[footnote 5]}. By converting this yield into an average quarterly amount per business per year, we can approximately compare our evaluation results to the previously forecast additional tax revenue.

For the below the threshold matched population, the evaluation results estimate an average additional tax revenue of £19 per business per quarter towards the end of the 2019 to 2020 tax year. With a 95% confidence interval ranging from -£3 to £42 (see Appendix, Table 10). The 95% confidence interval is a range for which we have reasonable certainty will contain the true average additional tax revenue generated due to Making Tax Digital.

Table 5 below shows a breakdown of the current forecast additional tax revenue per business per quarter for the below the threshold population from the public finances. The forecast additional tax revenue for the 2019 to 2020 year is £5, which lies within the 95% confidence interval of the evaluation estimate.

Table 5 – Forecast additional tax revenue from the public finances - below the threshold

Year	2019-20	2020-21	2021-22	2022-23	2023-24	2024-25	2025-26
Forecast additional tax revenue per business per quarter	£5	£5	£8	£11	£13	£14	£14
Evaluation estimate	£19	-	-	-	-	-	-

For the above the threshold matched population, the evaluation estimate of the average quarterly additional tax revenue observed at the end of the 2018 to 2019 year is £57 per business. The 95% confidence interval for this estimate ranges from -£30 to £144 (see Appendix, Table 10).

Table 6 below shows the forecast additional tax revenue for the above the threshold population. We can see that in the 2018 to 2019 year the forecast quarterly additional tax revenue per business is £6. Although the forecast amount is much smaller than the evaluation estimate, it falls within the 95% confidence interval of this estimate.

Table 6 – Forecast additional tax revenue from the public finances - above the threshold

Year	2018-19	2019-20	2020-21	2021-22	2022-23	2023-24	2024-25
Forecast additional tax revenue per business per quarter	£6	£31	£34	£50	£52	£54	£56
Evaluation estimate	£57	-	-	-	-	-	-

From this we conclude, our evaluation results and the forecast are broadly comparable. In both analyses, forecast amounts of additional tax revenue per quarter lie within the range of the confidence interval of our evaluation results.

Since the additional tax revenue from Making Tax Digital is most frequently reported on a per tax year basis, we can also compare our evaluation estimates in terms of total annual additional tax revenue. HMRC has published a modelled estimate of the forecast revenues from Making Tax Digital for VAT of £115 million^{[footnote 6]} for the 2019 to 2020 tax year. This is the closest year to our below and above threshold estimates.

To extrapolate we assume our model results are representative of all businesses in Making Tax Digital in 2019 to 2020. The estimated additional tax revenue from Making Tax Digital can then be simply extrapolated by grossing up to an annual basis and for the whole population of VAT businesses in Making Tax Digital for this tax year. Below, we have used 2 methods to calculate the approximate additional tax revenue in the 2019 to 2020 tax year on this basis.

The first method is to extrapolate the average additional tax revenue from the evaluation estimate, to the average population in Making Tax Digital over the year^{[footnote 7]}. The following formula is used to calculate the total additional tax revenue (ATR) from Making Tax Digital in the 2019 to 2020 year.

Equation 2 – Method 1: extrapolate average additional tax revenue to 2019 to 2020 Making Tax Digital population

Total ATR in year = ATR per business per quarter × 4 quarters × average MTD population in year

Table 7 shows the components of this formula and the estimated total additional tax revenue, which sums to £185 million.

Table 7 – Estimated additional tax revenue for 2019 to 2020 Making Tax Digital population using average additional tax revenue

Population	Quarterly additional tax revenue per business	Annual additional tax revenue per business	Average number of businesses in Making Tax Digital in 2019-20	Estimated additional tax revenue from Making Tax Digital (Millions)*
Below Threshold	£19	£78	170,000	£15
Above Threshold	£57	£228	761,000	£175
Total	-	-	-	£185

*Final figures rounded to nearest 5 million.

This method assumes that on average the population that were unmatched or excluded from the modelling will bring in the same amount of additional tax revenue as the matched population. This assumption may not hold true as we know that the Making Tax Digital 2019 to 2020 population have lower average tax liabilities than the matched population.

The second method is to extrapolate the relative additional tax revenue as percentage (%) increases in tax liabilities due to Making Tax Digital, to the total VAT yield from the 2019 to 2020 tax year.

For the evaluation we estimated the average tax liability of the treatment group if they had not signed-up to Making Tax Digital. We then calculated the average additional tax revenue per business as a percentage increase in this tax liability.

For this extrapolation method we apply the percentage increase additional tax revenue to all tax returns submitted though Making Tax Digital in the 2019 to 2020 tax year, to get an estimate for the total population. This is done with the following formula, where ATR is the additional tax revenue and MTD is Making Tax Digital.

Equation 3 – Method 2: extrapolate additional tax revenue as a percentage increase in tax liability to 2019 to 2020 total Making Tax Digital VAT liability

Total ATR in year = total MTD tax liability - predicted non MTD tax liability

Where:

predicted non MTD tax liability = (total tax liability)/(1 + percentage increase due to MTD)

Table 8 shows the estimated additional tax revenue from this approach, which sums to £195 million.

Table 8 – Estimated additional tax revenue for the 2019 to 2020 Making Tax Digital population using average additional tax revenue as a percentage increase

Population	Percentage increase in tax liability due to Making Tax Digital	Total tax liability submitted through Making Tax Digital in 2019-20	Estimated additional tax revenue from Making Tax Digital (Millions)*
Below Threshold	2.20%	£0.2 billion	£5
Above Threshold	0.90%	£21.2 billion	£190
Total	-	-	£195

*Final figures rounded to nearest 5 million.

This method assumes that additional tax revenue per business is proportional to the tax liability per business. This has the advantage of factoring in the variation in tax liabilities over the population.

The returns collected for the 2019 to 2020 year were capped at the top and bottom 2% to remove outliers. Any returns submitted outside of the capping limits were set to the value minimum or maximum value at the 2^nd or 98^th percentile.

Both extrapolations to the 2019 to 2020 Making Tax Digital population use the central estimates to calculate the total additional tax revenue. For the first method we can calculate a 95% confidence interval for the estimate of £185 million additional tax revenue, which ranges from -£80 million to £455 million.

From these comparisons, we again conclude that our evaluation results and the expected additional tax revenue are broadly comparable. The extrapolated point estimates of £185 million and £195 million are comparable to the published additional tax revenue modelled estimate of £115 million. Supporting evidence for this is that the 95% confidence interval for the estimate of £185 million encompasses the £115 million forecast.

We are therefore able to conclude that we are likely to receive positive additional tax revenue of a similar magnitude to what has been forecast and included in the public finances.

A range of assumptions have been made to implement the match design methodology and assess additional tax revenue. The next section explains the limitations of the analysis due to these.

6. Limitations of the analysis

Our evaluation results are estimates for the matched populations only. They do not cover the whole population as they do not necessarily reflect:

• businesses that do not file their VAT returns quarterly
• businesses that were not matched in the propensity score matching
• businesses excluded due to the capping
• another time period
• any other tax head of duty

This research examines businesses within their first year of signing-up to Making Tax Digital. If observed for a longer period, the effects of signing-up may change due to other external factors.

There are limitations to the methodology. The propensity score matching attempts to remove the self-selection bias; however, the bias cannot be completely removed, and some is likely to remain.

The difference-in-differences model further removes the static self-selection bias and the shared time effect for the combined treatment and control groups. However, it does not take account of the differing time effects between the treatment and control groups. For example, growing businesses may be more prone to voluntarily signing up to Making Tax Digital.

To assess the additional tax revenue for the 2019 to 2020 Making Tax Digital population we must extrapolate outside of the matched population which adds uncertainty to the estimates. We must assume that the evaluation results do reflect all VAT businesses in Making Tax Digital in the 2019 to 2020 tax year and that the quarterly results hold for the year as a whole.

As previously discussed, these assumptions made when extrapolating may not hold if we were able to verify them. The extrapolations are based on the maximum likelihood estimates (average) only.

7. Conclusions

Both sets of evaluation results show a positive Making Tax Digital impact. For the population below the threshold, this is £19 in additional tax revenue per business for the quarter. There would be only a 9% chance of observing results at least this extreme (≥ £19 or ≤ -£19) if there were no additional tax revenue due to Making Tax Digital.

For the population above the threshold, the model estimates £57 in additional tax revenue per business for the quarter. There would be a one in 5 (20%) chance of observing results at least this extreme (≥ £57 or ≤ -£57) if there were no additional tax revenue due to Making Tax Digital.

When we compare the 2019 to 2020 additional tax revenue from our model to the forecast revenue benefits included in the public finances for Making Tax Digital, we find that the magnitudes are broadly consistent.

Overall, we conclude that there is a high likelihood Making Tax Digital is generating additional tax revenue by reducing errors and making it easier for businesses to get their tax right.

We also conclude that the additional tax revenue from Making Tax Digital is most likely to be of a similar magnitude to what has been forecast and included in the public finances for the 2019 to 2020 tax year.

8. Appendix

8.1 Logistic regression models for matching

Equation 4 and Equation 5 shows the logistic regression equations used in the propensity score matching for the below the threshold and above the threshold populations respectively. The definitions of the independent variables used in the models are given are in Table 9.

Equation 4 - Logistic regression equation to produce propensity scores – below the threshold

Signed up = BTA + Historic Software + Stagger + Industry Classification + Business Type + Encouragement Letter Flag + Direct Debit + Overseas Trader + EU Trade + Repayment Trader + MOSS Trader + FRS Trader

Equation 5 - Logistic regression equation to produce propensity scores – above the threshold

Signed up = BTA + Historic Software + Historic Agent + Stagger + Industry Classification + Business Type + Encouragement Letter Flag + Direct Debit + EU Trade + MOSS Trader + FRS Trader

Table 9 - Independent variables in the logistic regression model – below the threshold

Variables	Interpretation
BTA	Business has used the Business Tax Account in the past
Historic Software	Business has submitted a return using software prior to Making Tax Digital
Historic Agent	Business has used an agent for their accounts in the past
Stagger	Business filing period – 3 options
Industry Classification	Industry classification section – 21 options[footnote 8]
Business Type	Sole proprietor, incorporated, partnership, local authority, non-profit or public corporation
Encouragement Letter Flag	Business was sent extra communication encouraging them to sign-up to Making Tax Digital
Direct Debit	Latest return was paid by direct debit
Overseas Trader	Registered with a non-UK address
EU Trade	Trader has declared EU sales/purchases
Repayment Trader	Traders are classified as repayment traders if their deductible input tax regularly exceeds their output tax
MOSS Trader	Trader is or has been registered for MOSS – Mini One-Stop Shop
FRS Trader	Traders that are signed-up to the VAT Flat Rate Scheme pay a fixed rate of VAT depending on their business type

8.2 Difference-in-differences regression model to calculate additional tax revenue

Equation 6 below shows the regression model used to calculate the additional tax revenue for the below the threshold population.

Equation 6 - Difference-in-differences regression model - below the threshold

Net tax = £653 + £212 × signed up + £16 × time + £19* × time × signed up

*p-value 0.09

We have 2 time points from the difference-in-differences model so time = 0 at the first time point and time = 1 at the second time point. The variable signed up is 0 for the control group and 1 for the treatment group.

At time = 0:

• the control group declared on average £653 in tax liability
• the treatment group declared an additional £212 on average, therefore declaring a total average of £865. This shows that businesses in the treatment group declared a higher average tax liability before the effects of Making Tax Digital

At time = 1:

• businesses declared on average £16 more VAT compared to the same period in the previous year. This could be due to business growth or inflation
• the treatment group at the second time have both time = 1 and signed up = 1. This activates the final term in the difference-in-differences regression model, the term signifying the interaction between time and the business joining Making Tax Digital. This interaction term indicates that the additional tax revenue due to signing up to Making Tax Digital is on average £19 per business for the quarter
• had businesses in the treatment group not signed up to Making Tax Digital they would have been due to pay on average £881 in tax. The additional tax revenue of £19 represent an increase of 2.2% in tax revenue

Equation 7 shows the difference-in-differences regression model for the above the threshold population.

Equation 7 - Difference-in-differences regression model - above the threshold

Net tax = £5847 + £179 × signed up + £278 × time + £57* × time × signed up

*p-value .20

At time = 0:

• the control group declared on average £5,847 in tax liability
• the treatment group declared an additional £179 on average, therefore paying a total average of £6,026

At time = 1:

• businesses declared on average £278 more VAT at compared to the same period in the previous year
• the treatment group at the second time have both time = 1 and signed up = 1. This activates the final term in the difference-in-differences regression model, the term signifying the interaction between time and the business joining Making Tax Digital. This interaction term indicates that the additional tax revenue due to signing up to Making Tax Digital is on average £57 per business for the quarter
• had businesses in the treatment group not signed up to Making Tax Digital they would have been due to pay on average £6,361 in tax. The additional tax revenue of £57 represent an increase of 0.9% in tax revenue

8.3 Additional tax revenue estimates

Table 10 below shows for both populations the estimate of the average additional tax revenue per business per quarter with statistics on the precision of these estimates.

Table 10 - Additional tax revenue estimates

Population	Average additional tax revenue estimate per business per quarter	Standard error	p-value	95% confidence interval
Below the threshold	£19	11.5	0.09	(-£3, £42)
Above the threshold	£57	44.6	0.20	(-£30, £144)

1. Evaluating Making Tax Digital’s impact on record keeping behaviour and scope for error among small businesses; Exploring the costs and benefits of Making Tax Digital for VAT experienced by smaller businesses; Qualitatively assessing the impact of Making Tax Digital ↩

2. The VAT tax gap is calculated using a ‘top-down’ approach by comparing the net VAT total theoretical liability (VTTL) with actual VAT receipts. The latest Tax Gaps publication ↩

3. Businesses with less than 50 employees. ↩

4. These periods all precede the economic shock from COVID-19. Our analysis is therefore not affected by the impact of the pandemic on tax payments. ↩

5. Additional tax revenue forecast from the public finances on the OBR website ↩

6. HMRC spending review with 2019-20 outturn Making Tax Digital estimate of £115 million ↩

7. The average Making Tax Digital population in the 2019 to 2020 year as a percentage of the total VAT population is 15% for the below the threshold and 57% for the above. ↩

8. Companies house: Standard Industrial Classification (SIC) codes ↩