Could knowledge about Central banks impact households’ expectations? Technical Annex

The Bank Underground blog Could knowledge about Central banks impact households’ expectations? used an institutional knowledge score and an economic knowledge score constructed from the Bank/TNs Inflation Attitudes Survey to estimate the impact of institutional knowledge on households’ expectations. This technical annex describes how each knowledge score was constructed, and presents the full estimation results of the regressions described in the main blog post.

Institutional Knowledge Score

As in Haldane and McMahon, the institutional knowledge score is constructed from three questions about the institutional structure of the Bank. The below table summarises how scores are awarded for each of the answers to these three questions. The scores for each question are then summed to produce an aggregate score (with possible values 0-6). Scores are only available for February quarters of the survey, as these are the only surveys in which Q11-Q13 are asked.

Table 1: Questions in the institutional knowledge score

In constructing the knowledge score, those who answered ‘Don’t know’ scored higher than those who answered incorrectly, on the grounds that these respondents were at least aware of their lack of knowledge. I also created a variant of the knowledge score where ‘don’t know’ instead scored less than a wrong answer. This produced lower knowledge scores on average, but did not materially change the central findings.

Economic Knowledge Score

The economic knowledge score is constructed from two related questions on the operation of monetary policy and the transmission mechanism. Given the lag in the transmission of monetary policy (for example because it takes time for changes in Bank Rate to feed through into commercial interest rates, and for shops to change their prices), a rise in interest rates would not be expected to have much of an impact on high street prices within a month. Batini and Nelson estimate that for the UK, it takes over a year for monetary policy to have its peak impact on inflation. So disagree and strongly disagree were the correct answers for Q9a. However, higher interest rates would be expected to weigh on price rises in the medium term, so agree and strongly agree were the correct answers for Q9b.

Table 2: Questions in the economic knowledge score

Regression Results

The main blog presents simplified results from the regressions of households’ expectations on their knowledge scores at the 1 year and 5 year horizon. The tables below present the full outputs for these regressions, including demographic controls. They also show the results at the 2 year horizon, which tell a very similar story to those at a 5 year horizon.

Table 3 shows the results of regressing households’ expectations on their institutional knowledge scores, without controlling for their economic knowledge. As one  can see from the full results, not only are the coefficients on institutional knowledge statistically significant, they also appear to be economically significant in the context of the coefficients on other variables. For example, age is often thought to be an important determinant of households’ inflation expectations (as older households remember the periods of much higher inflation in the 1970s). And indeed, there is a statistically significant positive coefficient on age – for example, for 5 year expectations, all else equal you would expect someone aged 45-54 (the age group with the largest deviations in inflation expectations, all else equal) to have inflation expectations 0.19pp further from the 2% target. However, the gap in inflation deviations between those with the highest and lowest knowledge scores is far larger. All else equal, these results suggest that someone with a knowledge score of 6 will have inflation expectations 0.56pp closer to target than someone with a knowledge score of 0.

Table 4 shows the results of the same regressions with households’ economic knowledge scores included as a control. As shown in the blog, all coefficients on institutional knowledge remain statistically significant – this is also true for expectations at the 2 year horizon.

Table 3: Regression results excluding economic knowledge

Table 4: Regression results controlling for economic knowledge