Transparency data

Insolvency Statistics Seasonal Adjustment Review April 2025

Updated 25 April 2025

1. Background

Seasonal adjustment is the estimation and removal of effects on a time series that are a result of the time of year, such as the calendar month or Easter. Accounting for these effects makes it possible to analyse the underlying trend in the data.

A common example is retail sales, which peak as a result of the Christmas shopping period. However, this rapid increase in sales does not indicate a sustained upwards trend. Seasonal adjustment removes the effect estimated to be as a result only of Christmas, leaving the actual story in the data to be assessed.

If a data series is seasonal but is not adjusted to account for this, it is not valid to compare the current time period to the previous one.

The purpose of this annual review was to determine which of the data series included in the Insolvency Service’s Accredited Official Statistics publications, the monthly Company and Individual Insolvency Statistics, should be adjusted to account for seasonality.

2. Scope of the review

Prior to this review, three company insolvency time series were seasonally adjusted:

  • Compulsory liquidations
  • Creditors’ voluntary liquidations (CVLs)
  • Administrations

Eight individual insolvency time series were seasonally adjusted:

  • Total bankruptcy orders
  • Bankruptcy orders – Creditor petitions
  • Bankruptcy orders – Debtor petitions
  • Consumer bankruptcy
  • Trader bankruptcy
  • Individual voluntary arrangements (IVAs) by registration date
  • Individual voluntary arrangements (IVAs) by approval date
  • Debt relief orders (DROs)

These series were reviewed to determine whether it was still appropriate to apply seasonal adjustment and, if so, whether the model being used was still applicable.

Data series for Scotland and Northern Ireland were outside the scope of this review.

The data used in this review can be found in the long-run CSV files for the Company and Individual Insolvency Statistics.

The revisions policy for seasonally adjusted series is set out in a separate document on our website. Changes to the revisions policy were not considered as part of this review, but the Statistics Team would welcome comments and feedback on this approach.

3. Analysis

The data series published by the Insolvency Service contain monthly and quarterly data. They are series of differing lengths. Numbers for compulsory and creditors’ voluntary liquidations exist as far back as Q1 1960, compared to IVAs from 1987 and administrations from 1993. For the purposes of seasonal adjustment, monthly data since January 2000 has been used where possible, that is 302 full months of data at the time of this review. A few time series have a different start date. Accurate information relating to employment status in bankruptcy is only available starting from 2007, so the Consumer bankruptcy and Trader bankruptcy series start from January 2007. DROs were introduced in April 2009, so the series for seasonal adjustment starts at this time.

The Insolvency Service constrains the seasonally adjusted series to match the annual calendar year totals in the original data. While this may distort the seasonally adjusted totals, it is a common approach, and doing so aids interpretation of the data. The seasonally adjusted numbers for bankruptcies were derived by adding the adjusted creditor and debtor series. Total individual insolvency numbers were derived by adding the seasonally adjusted IVA, DRO and bankruptcy series. Total company insolvency numbers were derived by adding the seasonally adjusted compulsory liquidation, creditors’ voluntary liquidation (CVL) and administration numbers, and adding on the unadjusted company voluntary arrangement (CVA) and receivership numbers. Company voluntary arrangements and receiverships were not seasonally adjusted due to very small volumes.

For each series, the following have been reviewed:

  • Tests for the presence of seasonality
  • regARIMA model, which considers the ARIMA model, as well as features (prior adjustments) such as outliers, level shifts, temporary changes, seasonal breaks as well as Easter and trading day effects
  • The optimal choice of filters for the seasonal moving average and trend moving averages

The analysis was carried out using X13-ARIMA-SEATS, as implemented in the ‘seasonal’ package in R. This is the recommended programme for seasonal adjustment for Official Statistics, and the pick model function was used to avoid over fitting.

Detailed comments on each series are given in the following sections.

4. Summary of outcomes

For all of the series where significant seasonality was detected, the seasonality was multiplicative and therefore a log transformation was applied to the data.

For most personal insolvency series, the “pick model” function selected the ARIMA model (0,1,1)(0,1,1), a 3x5 seasonal moving average and a 13-term Henderson moving average to account for the trend. The exceptions were the debtor petition bankruptcy, consumer bankruptcy and individual voluntary arrangement by approval date series, where the ARIMA model (2,1,2)(0,1,1) was chosen, indicating a slightly more complex model for the trend. After comparing the seasonally adjusted results from the ARIMA(2,1,2)(0,1,1) and the previously used model for each series, it was found that the differences between them were minimal and did not significantly impact the reported figures. In line with guidance from the ONS, and to maintain consistency with previous years, the decision was made not to change the model specifications for these series. For the debtor petition bankruptcy and consumer bankruptcy series the ARIMA(0,1,1)(0,1,1) is used. For the individual voluntary arrangement by approval date series, the ARIMA (0,1,2)(0,1,1) is used. This decision will be reviewed again in April 2026.

For all company insolvency series, the “pick model” function selected the ARIMA model (0,1,1)(0,1,1), a 3x5 seasonal moving average and a 23-term Henderson moving average to account for the trend. The 23-term Henderson moving average suggests that the non-seasonally adjusted data contains more short-term fluctuations or noise. After comparing the results from the 23-term and 13-term moving averages, the differences were minimal and did not significantly impact the reported figures. The 13-term filter is the standard choice for monthly economic data, providing a balance between smoothing and retaining key trends. In line with ONS guidance and to maintain consistency with previous years, the 13-term Henderson moving average filter was retained for all seasonally adjusted company insolvency series. This will be reviewed again as part of the regular assessment in April 2026.

Series Decomposition Model Priors Seasonal moving average Trend moving average
Compulsory liquidations Multiplicative Log(0,1,1)(0,1,1) Weekday, Easter[1] 3x5 13-term
Creditors’ voluntary liquidations Multiplicative Log(0,1,1)(0,1,1) Weekday, Easter[8] 3x5 13-term
Administrations Multiplicative Log(0,1,1)(0,1,1) Weekday 3x5 13-term
Receiverships Seasonality not tested due to very small number of cases        
Company voluntary arrangements Seasonality not tested due to very small number of cases        
Bankruptcy - Creditor Petitions Multiplicative Log(0,1,1)(0,1,1) Trading day, Easter[8] 3x5 13-term
Bankruptcy - Debtor Petitions Multiplicative Log(0,1,1)(0,1,1) Trading day, Easter[1] 3x5 13-term
Consumer bankruptcies Multiplicative Log(0,1,1)(0,1,1) Trading day, Easter[1] 3x5 13-term
Trader bankruptcies Multiplicative Log(0,1,1)(0,1,1) Trading day, Easter[1] 3x5 13-term
Individual voluntary arrangements by registration date Multiplicative Log(0,1,1)(0,1,1) Weekday 3x5 13-term
Individual voluntary arrangements by approval date Multiplicative Log(0,1,2)(0,1,1) Trading day, Easter[15] 3x5 13-term
Debt relief orders Multiplicative Log(0,1,1)(0,1,1) Trading day, Easter[8] 3x5 13-term

‘Easter[n]’ means an Easter effect lasting n days.

‘Weekday effect’ refers to an adjustment based on the number of weekdays in each month, assuming all weekdays have a similar impact.

‘Trading day effect’ allows each day of the week to have a different impact on the data.

Compulsory Liquidations

Similarly to last year’s review, this review found that the monthly series demonstrated significant seasonality.

Prior adjustments that were identified for the series included weekday trading effects and an Easter effect of one day. Additive outliers were detected for May and June 2020 and a level shift in September 2020, which corresponded to lower numbers during the coronavirus pandemic. A level shift was also detected for January 2022, when compulsory liquidations started to increase following the dip during the pandemic.

Creditors’ Voluntary Liquidations

Similar to the previous reviews, the series demonstrated significant seasonality.

Prior adjustments that were identified for the series included weekday trading effects and an Easter effect of eight days. Temporary changes were identified for June 2020 and January 2021, during the coronavirus pandemic.

The adjustment review for this series uses numbers after bulk insolvencies are removed. Bulk insolvencies were a large number of connected personal services companies entering liquidation, primarily between 2016 and 2019, following changes to claimable expense rules. See the Company Statistics Methodology and Quality Document for more information.

Administrations

In previous reviews, with the exception of 2019, the administrations time series demonstrated significant seasonality. This review found that the series again demonstrated significant seasonality.

Prior adjustments that were identified for the series included weekday trading effects and multiple additive outliers for August 2000, October 2001, November 2006, October 2008 (when 728 managed service companies entered administration on the same day in September 2008) and December 2020.

Receiverships

Due to the very small numbers of receiverships, we did not test for seasonality in the data.

Company Voluntary Arrangements

Due to the very small numbers of receiverships, we did not test for seasonality in the data.

Bankruptcy Orders - Creditor Petitions

Similar to the previous reviews, the series demonstrated significant seasonality.

Prior adjustments that were identified for the series included trading day effects and an Easter effect of eight days. A temporary change was identified for April 2020, which corresponded to lower numbers at start of the coronavirus pandemic.

Bankruptcy Orders - Debtor Petitions

Similar to the previous reviews, the series demonstrated significant seasonality. The algorithm selected an ARIMA(2,1,2)(0,1,1) model; however, the decision was made to retain the ARIMA(0,1,1)(0,1,1) specification, as previously used, since the new model did not result in a meaningful improvement to the seasonal adjustment and maintaining consistency with previous publications was considered more appropriate.

Prior adjustments that were identified for the series included trading day effects and an Easter effect of one day. An additive outlier was identified for May 2011. As with creditor petitions, a level shift was detected for April 2020, which corresponds to lower numbers at the start of the coronavirus pandemic.

Consumer Bankruptcy Orders

Similar to the previous reviews, the series demonstrated significant seasonality. As with debtor bankruptcies, the algorithm selected an ARIMA(2,1,2)(0,1,1) model; however, the decision was made to retain the ARIMA(0,1,1)(0,1,1) specification for consistency.

Prior adjustments that were identified for the series included trading day effects and an Easter effect of one day. An additive outlier was detected for May 2011. A level shift and an additive outlier were detected for April and June 2020 respectively, which corresponded with lower numbers at the start of the coronavirus pandemic.

Sole Trader Bankruptcy Orders

Similar to the previous reviews, the series demonstrated significant seasonality.

Prior adjustments that were identified for the series included trading day effects and an Easter effect of one day. A level shift and additive outlier were detected for April and May 2020 respectively, which corresponded with lower numbers during the start of the pandemic. A temporary change was detected for September 2020, which saw higher numbers than earlier in the pandemic. An additive outlier for December 2020 corresponded to a decline in numbers back to levels seen during the early part of the pandemic.

Individual Voluntary Arrangements (IVAs) by registration date

Similar to the previous reviews, the series demonstrated significant seasonality.

Prior adjustments that were identified for the series included weekday trading effects and an Easter effect of one day. Additive outliers were detected for March and May 2020, corresponding with a large number of delayed registrations of IVAs agreed before the start of the coronavirus pandemic. A temporary change was detected for October 2020, when a similar registration spike occurred.

Individual Voluntary Arrangements (IVAs) by approval date

Similar to the previous reviews, the series demonstrated significant seasonality. The algorithm selected an ARIMA(2,1,2)(0,1,1) model; however, the decision was made to retain the ARIMA(0,1,2)(0,1,1) specification for consistency.

Prior adjustments that were identified for the series included trading day effects and an Easter effect of 15 days. A temporary change was detected for June 2002. A level shift and a temporary change were detected for April and May 2020, respectively, corresponding with lower volumes at the start of the coronavirus pandemic.

Debt relief orders (DROs)

Similar to the previous reviews, the series demonstrated significant seasonality.

Prior adjustments that were identified for the series included trading effects, an Easter effect of eight days and multiple outliers. An additive outlier and temporary change were detected for April and May 2009, corresponding with the introduction of DROs. An additive outlier for September 2015 and and a level shift for July 2021 both corresponded with increases of eligibility limits of debt, assets and income for DROs. A level shift in April 2020 corresponded with lower numbers at the start of the coronavirus pandemic. Another level shift in March 2023 corresponded with the introduction of DRO hubs. A level shift in April 2024 corresponded with the abolition of the £90 administration fee to obtain a DRO.