MPC Transcripts Reveal How Inflation and Growth Views Shaped Bank Rate Decisions

Bank Rate decisions reflect more than the economic data available at the time. A new analysis of Bank of England Monetary Policy Committee (MPC) discussions from 2015 to 2017 finds that members’ own assessments of inflation and growth were related to their choices: both measures were linked to a greater likelihood of supporting a rate increase, while growth views alone helped explain support for a cut.

The work by Dragos Gorduza, Livia Silva Paranhos and Ivan Yotzov uses language models to turn transcripts into structured measures of individual views and the arguments behind them. The findings offer a way to examine how policymakers’ reasoning developed through a period that included the Brexit decision, heightened uncertainty and a return to policy normalisation as inflation rose and the economy proved more resilient than anticipated.

Turning discussion into comparable measures

The researchers assembled 258 statements by MPC members from 29 meeting transcripts spanning 2015–17. The records are released with an eight-year lag; 2017 is the latest year available in this data set. The committee has nine seats, and changes in membership during the period mean the analysis covers 14 speakers: seven internal and seven external members. Three meetings in 2017 had eight members.

For each statement, the team measured how the speaker described UK inflation relative to the Bank’s 2% CPI target and output growth relative to potential over the year ahead. A score of -5 represents inflation judged very far below target, 0 means close to target and +5 means very far above. Growth receives an analogous score against potential output. The researchers used a separate OpenAI gpt-5-mini call for each measure, with outputs restricted to a single integer from -5 to +5.

They also extracted the causal reasoning in each statement and assigned it to one or more of 14 subjects. These include inflation, the labour market, output, exchange rates and the policy rate. The rate-setting category is tightly defined: it includes direct discussion of the rate decision, while comments about factors that might affect it are classified under their immediate subject. The article focuses on six of the 14 categories.

The method is designed to capture differences that a tally of committee decisions can miss. Two members can reach the same choice while weighing inflation, growth and uncertainty differently. These text-based measures make it possible to compare those assessments across speakers and over time, though they remain estimates derived from what members chose to say.

Growth concerns fell, then recovered

The growth measure weakened in the months before the June 2016 Brexit decision and dropped sharply afterwards. It improved during 2017 as the economy performed better than initially expected. The authors connect this pattern both to lagged movements in GDP growth and to uncertainty about the near-term outlook. Financial markets moved sharply after the decision, business surveys deteriorated, and forecasts at the time warned of a substantial slowdown and potentially severe recession risks.

Inflation assessments followed a different pattern, tracking annual CPI inflation closely across the period. The reported correlation between perceived inflation relative to target and CPI inflation is 0.75. The authors also find that the relationship between the growth and inflation measures became more positive and statistically stronger after the Brexit decision. They say this is consistent with a greater demand-side reading of economic developments in that part of the sample.

Chart 1: Trends in text-based inflation and growth metrics over 2015–17

What members discussed also shifted

Output was the most frequently discussed subject, appearing in 19% of statements, followed by inflation at 15% and direct rate-setting at 14%. After the Brexit decision, the mix shifted towards rate-setting, the trade-off between inflation and output, and political events. References to inflation and financial markets became less frequent. This change suggests that the discussions’ focus moved as members assessed a more uncertain outlook and its implications for the policy choice.

Figure 1: Extracting causal arguments and topics from MPC transcripts

Chart 2: Distribution of topic shares in MPC transcripts before and after the Brexit decision

Different views mattered for increases and cuts

To test whether the extracted measures added information about Bank Rate choices, the researchers modelled the likelihood of supporting an increase, holding the rate or cutting it. The analysis excludes asset-purchase decisions. It controls for each speaker’s typical pattern, annual CPI inflation, annual GDP growth and the Bank’s year-ahead inflation and growth forecasts. The question is whether members’ expressed assessments help explain their choices beyond those aggregate measures.

Across the full sample, stronger views that inflation was above target and growth above potential were each associated with a higher likelihood of supporting a rate increase. Both relationships were statistically significant at the 5% level. A stronger above-target inflation assessment was also linked to a lower likelihood of supporting a hold; the growth measure had no statistically significant relationship with holding.

For rate cuts, the result was distinct. Members who viewed growth as further above potential were less likely to support a cut, while the inflation measure was small and statistically insignificant. Put another way, weaker growth assessments were associated with a greater likelihood of supporting a reduction, while inflation views did not show the same relationship in the model.

Chart 3: Rate decisions and members’ views on inflation and growth

The pattern also appears in separate models for individual members, after excluding those who took part in fewer than 10 meetings and those who never supported an increase or cut. The estimated relationship between both measures and support for an increase is positive for essentially every member, and statistically strong for several. For cuts, the growth coefficients are negative across the members studied; inflation coefficients are smaller and more dispersed around zero.

That consistency suggests the pooled result is not simply the product of a few members’ positions. But the strength of the relationships differs across individuals, and the analysis captures association rather than establishing that a particular expressed view caused a particular choice. Its scope is also limited to Bank Rate decisions in the available 2015–17 transcripts.

A record of reasoning, with limits

The transcripts provide evidence about how members framed the outlook, alongside the choices recorded at each meeting. The analysis finds that those frames changed over time and carried information about policy preferences beyond observed economic data and the Bank’s forecasts. The next useful comparison would be with members’ public speeches from the same periods, to see how the reasoning in private deliberations compares with explanations offered in public.

Gorduza and Paranhos work in the Bank’s Advanced Analytics Division; Yotzov works in its Structural Economics Division. The analysis is based on transcripts made publicly available after the eight-year release delay.