Annual analysis of revisions to money, credit and effective interest rates data (2013-15)

Each year the Bank publishes an analysis of revisions to its monthly data on money, credit, and effective interest rates produced by the Data and Statistics Division (DSD).
Published on 29 June 2018

By Paige Benattar, Hannah Phaup and Alister Ratcliffe


Each year the Bank publishes an analysis of revisions to its monthly data on money, credit, and effective interest rates produced by the Data and Statistics Division (DSD). This year’s analysis shows that the revisions have been immaterial for most series tested. This is the same broad conclusion as the 2017 analysis of 2012-14 data and the 2016 analysis of 2011-13 data.

Revisions are a normal part of the data production process. There are several reasons why data might be revised after initial publication. Reporters of data to the Bank may submit corrections to earlier data if they discover errors or make improvements to their data systems. In addition, the Bank might change the methodology it uses to produce the data. Also, the seasonal adjustment process can lead to revisions to an entire series, as each new data point provides new information about the seasonal pattern of the data.

Revisions analysis gives users an indication of how much weight to place on data when it is first released. Data that is not usually revised much can be regarded as less noisy and more reliable.

Revisions are measured in different ways: the average size of the revisions, bias, and their variability. We consider each of these in turn below.

Average revision — Tables 1 and 2

The mean revision is the arithmetic average of the revisions to a data series over the time period of interest. It includes the effects of both positive and negative values (upward and downwards revisions). Average upward or downward revisions will be due to changes to submitted data or methodology; revisions due to seasonal adjustment will average zero over a year. The mean absolute revision disregards the sign of revision in this calculation – all revisions are counted as positive. So this will be larger than the mean revision.

The tables include the average published outturn for each series, so the size of revision can be compared to the data. If the average revision is low compared to the average outturn, the initial data can be considered as more reliable.

For 2013-15 money and credit data, the largest revisions were to NIOFC (non-intermediate other financial corporations) deposits (Chart 1) and lending, lending to businesses (Chart 2), and consumer credit (Charts 3 and 4). Deposits from (Chart 5) and lending to households were revised the least. For data on effective interest rates, the largest average revisions were to credit cards (Chart 6) and new unsecured (other) loans to households. While revisions to credit card rates are infrequent, they can be fairly large, as shown in chart 6. Revisions to seasonally adjusted (SA) data tend to be slightly larger than non-seasonally adjusted (NSA).

Around three quarters of the series in Tables 1 and 2 were revised up on average, a similar proportion to previous years. This suggests that revisions tend to be upwards, but our more detailed tests (in tables 3 and 4) show that this bias is statistically significant for very few series.


Checks for bias — Tables 3 and 4

A check for bias shows whether series are mostly revised up, or mostly revised down.

Two tests of bias are used – a t-test, and a Newey-West test. If both tests find bias, the conclusion is that there is bias in the revisions for that series. If only one test finds bias, the result is inconclusive. And if neither test finds bias, the revisions are considered unbiased.

For most of the money and credit data, the tests reach the same conclusion and the series are considered unbiased. Exceptions include both series on lending to individuals (for mortgage and consumer credit), where revisions are more likely to be upwards. M4 lending to households, (a wider measure of lending to households) is also biased in seasonally adjusted space.

In the effective rates data, most series were found to have have no bias, though the results were inconclusive for 4 of the 18 series considered.

Checks for variability — Tables 5 and 6

These measure how volatile revisions are. Tables 5 and 6 use the root mean square revision (RMSR), which is an alternative measure of the average size of revisions, and the mean square revision (MSR), the square of this quantity. A high RMSR means that revisions are quite volatile, ranging from large increases to large decreases.

This measure of the volatility of revisions is put into context by comparing the RMSR with the average data outturn (‘b’ columns in Tables 5 and 6), and the MSR with the variance of the data (‘c’ columns in Tables 5 and 6). For both measures smaller ratios mean that revisions tend to be of similar size and more predictable. How much weight users can put on early estimates depends on how the data is being used. This measure should be considered alongside the average revision and bias metrics.

Some series have larger ratios – M4, NIOFC M4, lending to PNFCs and NIOFCs, consumer credit to individuals, credit card rates and new other household loan rates. In general this reflects the fact that the average or variance of the revised data is quite low, and the revisions are relatively large in comparison.

For example, the NSA M4 mean/RMSR ratio is -4.71. This reflects an RMSR of 0.07, but a mean outturn of -0.01.

To view all related tables, please download the full article:

PDF Annual analysis of revisions to money, credit and effective interest rates data (2013-15) 

For questions relating to this article please contact  or call +44 (0) 20 3461 5361.


Give your feedback

Was this page useful?
Add your details...