2. Survey data

We use survey data from the Michigan Survey. As of 1978, the Survey Research Center at the University of Michigan conducts approximately 500 telephone interviews with randomly selected persons at a monthly frequency. While the majority of respondents is only interviewed once, approximately 40% of the respondents are interviewed a second time 6 months after the initial interview.Footnote ⁵ The random selection of households is designed so that the sample should be representative for the US population.

The Michigan Survey contains quantitative as well as qualitative questions. For expected inflation, we use answers to the following two questions:

(A12) “During the next 12 months, do you think that prices in general will go up, or go down, or stay where they are now? (A12b) By about what percent do you expect prices to go (up/down) on the average, during the next 12 months?”

And to infer respondents’ readiness to spend concerning durable consumption goods, we use the following question:

(A18) “About the big things people buy for their homes – such as furniture, a refrigerator, stove, television, and things like that. Generally speaking, do you think now is a good or bad time for people to buy major household items?”

To include the survey data in the VAR analysis, we aggregate the data in a way that allows us to characterize the joint behavior of inflation expectations and consumers’ readiness to spend. If higher expected inflation has a stimulating effect in line with the expected inflation channel, then for those survey respondents who expect higher inflation, the incidence of a higher readiness to spend should increase and the incidence of a lower readiness to spend should decline. In other words, the share of respondents who indicate that they expect inflation to rise and at the same time indicate a higher readiness to spend, which we denote by $\text{Share(INFEX+,RTS+)}$ , should increase relative to the share of respondents who expect higher inflation, but have a lower readiness to spend, $\text{Share(INFEX}+,\text{RTS}{-})$ . Using these shares, we construct the following measure of conditional readiness to spend (CRTS) conditional on higher expected inflation:Footnote ⁶

(1) \begin{equation} \text{CRTS}({+})=\text{Share(INFEX+,RTS+)} - \text{Share(INFEX}+,\text{RTS}{-}). \end{equation}

Analogously to equation (1), we calculate a similar measure for respondents’ readiness to spend conditional on lower expected inflation:

(2) \begin{equation} \text{CRTS}({-})=\text{Share(INFEX}-,\text{RTS}{-}) - \text{Share(INFEX}-,\text{RTS+)}. \end{equation}

Increases in these measures in response to a policy shock, regardless of the sign of the shock, suggest that inflation expectations and readiness to spend are updated in a way that is consistent with standard theoretical predictions. In other words, higher (lower) expected inflation coincides with higher (lower) readiness to spend in the aftermath of the shock. In addition, the expected inflation channel holds that expansionary monetary policy stimulates consumption through an increase in expected inflation and contractionary shocks result in a downward revision of expected inflation. Thus, the expected inflation channel predicts that $\text{CRTS}({+})$ increases in response to an expansionary policy shock and that $\text{CRTS}({-})$ declines.

Since survey respondents provide quantitative inflation forecasts, we transform the quantitative inflation expectation into an ordinal measure, which indicates whether the expected inflation rate goes down, remains constant, or goes up, before we calculate the shares and the CRTS measures. To do so, we subtract the past realized inflation rate of the consumer price index (CPI) from the inflation rate forecast, which is provided in integers. The past realized inflation rate is calculated as the 12-month backward-looking moving average up to period $t-1$ .Footnote ⁷ We round the resulting difference to the nearest integer, and the sign of the difference indicates the expected direction of change (see Carvalho and Nechio (Reference Carvalho and Nechio2014); Dräger et al. (Reference Dräger, Lamla and Pfajfar2016); Geiger and Scharler (Reference Geiger and Scharler2021)). If the rounded difference is equal to zero, the respondent is classified as expecting constant inflation.

Figure 1. Time series plots of survey and conditional readiness to spend measures (1991M2-2019M6).

Notes: Shaded areas indicate economic downturns as classified by the NBER Business Cycle Dating Committee.

In Figure 1, we plot the shares of respondents grouped according to expected inflation and their readiness to spend in Panels (A) to (D), as well as the CRTS measures in Panel (E). Shaded areas indicate NBER recessions.

We see from Panel (C) that conditional on higher expected inflation, relatively more respondents indicate a higher readiness to spend, on average. While $\text{Share(INFEX+,RTS+)}$ is approximately 33% on average, $\text{Share(INFEX}+,\text{RTS}{-})$ is on average only about 11%.Footnote ⁸ Consequently, $\text{CRTS(+)}$ in Panel (D) is positive for most of the sample. $\text{Share(INFEX}-,\text{RTS+)}$ typically exceeds $\text{Share(INFEX}-,\text{RTS}{-})$ , 30% versus 9% on average, and $\text{CRTS}({-})$ is negative during most of the sample. Thus, lower inflation expectations are associated with a higher readiness to spend in general. We also see that $\text{Share(INFEX+,RTS+)}$ tends to decline and $\text{Share(INFEX}-,\text{RTS}{-})$ tends to increase during the onset of recessions. Although the patterns are less clear for $\text{Share(INFEX+,RTS}{-})$ and $\text{Share(INFEX}-,\text{RTS+)}$ , $\text{CRTS(+)}$ tends to drop during recessions while $\text{CRTS}({-})$ rises.

While these observations suggest that inflation expectations may be related to how consumers update their readiness to spend in general, we now turn to the structural analysis to study how inflation expectations and readiness to spend respond jointly to monetary policy announcements.

3. Estimation

We estimate reduced-form VAR models of the type:

(3) \begin{equation} Y_{t}=c+\sum _{p=1}^{P}A_{p}Y_{t-p} + e_{t}, \end{equation}

where $Y_t$ is the vector of endogenous variables, $A_p$ is the coefficient matrix at lag $p$ , $c$ is a vector of constants, and $e_t\sim N(0,\Sigma )$ is a vector of white noise residuals. As it is standard when using monthly data, we set $P=12$ (see e.g. Jarocinski and Karadi (Reference Jarocinski and Karadi2020)).

The vector of endogenous variables includes industrial production (IP), the CPI, the 1-year treasury bill rate (1Y TB Rate), which we obtain from the FRED database, the excess bond premium (EBP) from Gilchrist and Zakrajsek (Reference Gilchrist and Zakrajsek2012) as a proxy for financial market tensions (Caldara and Herbst (Reference Caldara and Herbst2019)), the high-frequency change in the 3-month ahead futures rate that occurs between 10 min before and 20 min after the FOMC announcement, as in Gertler and Karadi (Reference Gertler and Karadi2015), and the CRTS measures discussed in Section 2. IP and the CPI enter the VAR in log levels times 100 (see e.g. Ramey (Reference Ramey2016)). In our baseline, we use monthly data running from February 1990 to June 2019, where the sample period is determined by the availability of the interest rate surprises. Table A2 in the Appendix provides a detailed description of the data and lists the data sources.

Since the policy announcement should dominate financial markets within the tight window around the announcement, it is plausible that the change in the 3-month ahead futures rate is correlated with a monetary policy shock but uncorrelated with other macroeconomic shocks (see e.g. Gertler and Karadi (Reference Gertler and Karadi2015); Caldara and Herbst (Reference Caldara and Herbst2019)). Thus, we interpret this high-frequency change in the futures rate as a monetary policy-induced surprise that is exogenous and, following Plagborg-Møller and Wolf (Reference Plagborg-Møller and Wolf2021), we impose a recursive identification scheme where we order the surprise variable first in the vector of endogenous variables.Footnote ⁹ Thus, rather than imposing restrictions motivated by theoretical models, which is necessary when, for example, sign restrictions are used for identification, our identification approach relies on the assumption that the high-frequency change in the futures rate is exogenous.

FOMC announcements do not only reveal information about the implementation of monetary policy measures, but also information about the Fed’s economic outlook. To ensure that our results are not driven by these information effects, we replicate our estimations in the robustness analysis with policy surprise measures that are orthogonal to information effects (Jarocinski and Karadi (Reference Jarocinski and Karadi2020); Miranda-Agrippino and Ricco (Reference Miranda-Agrippino and Ricco2021)). Nevertheless, since orthogonalizing the surprise measures requires additional assumptions, we use surprises that incorporate information effects in our baseline specification.

We estimate the VAR with Bayesian methods to increase efficiency and to take into account that IP and consumer prices are nonstationary. Specifically, we use a standard Minnesota prior (see e.g. Litterman (Reference Litterman1986)), with the tightness and decay hyperparameters set to 0.05 and 0.5, respectively. The posterior distribution is simulated with 4500 Gibbs sample draws, where the first 500 draws are dropped as burn in and every fourth draw is saved.

5. Robustness analysis

We present a series of robustness checks to evaluate potential sensitivities with respect to the construction of CRTS measures and the estimation approach. Figures 9 and 10 show the responses of CRTS measures for the various robustness checks for the two subsamples.Footnote ¹⁵

Figure 9. Robustness checks of the response of the conditional readiness to spend measures to an expansionary monetary policy shock estimated with the pre-ZLB sample (1990M2-2008M10).

Notes: The figure shows point-wise median responses (blue lines) as well as 68% and 90% credible sets (dark and light blue shaded areas) to an expansionary monetary policy shock. The conditional readiness to spend measures are defined as CRTS $({+})=\text{Share}(\text{INFEX}+,\text{RTS}{+}) - \text{Share}(\text{INFEX}+,\text{RTS}{-})$ in the first and second line, and as CRTS $({-})=\text{Share}(\text{INFEX}-,\text{RTS}{-}) - \text{Share}(\text{INFEX}-,\text{RTS}{+})$ in the third and fourth line, times 100 plus 100. The various robustness checks are described in Section 5.

Figure 10. Robustness checks of the response of the conditional readiness to spend measures to an expansionary monetary policy shock estimated with the ZLB sample (2008M11-2015M11).

Notes: The figure shows point-wise median responses (blue lines) as well as 68% and 90% credible sets (dark and light blue shaded areas) to an expansionary monetary policy shock. The conditional readiness to spend measures are defined as CRTS $({+})=\text{Share}(\text{INFEX}+,\text{RTS}{+}) - \text{Share}(\text{INFEX}+,\text{RTS}{-})$ in the first and second line, and as CRTS $({-})=\text{Share}(\text{INFEX}-,\text{RTS}{-}) - \text{Share}(\text{INFEX}-,\text{RTS}{+})$ in the third and fourth line, times 100 plus 100. The various robustness checks are described in Section 5.

We first consider alternative CRTS measures that exploit the rotating panel dimension in the Michigan Survey. For the respondents who are interviewed a second time after 6 months, we use changes in inflation expectations and readiness to spend between the two interviews to construct the measures of CRTS. The responses are shown in the columns labeled “Change” in Figures 9 and 10. We see from Figure 9 that the responses of these alternative conditional measures in the pre-ZLB period are in line with the results reported in our baseline analysis. At the ZLB, $\text{CRTS}+$ declines, although less markedly than in the baseline, and $\text{CRTS}-$ increases according to Figure 10, which confirms our baseline results.

Next, we take into account that information conveyed about the state of the economy, which is disclosed when policy actions are announced, may confound the effects of policy shocks. We use the approaches suggested by Jarocinski and Karadi (Reference Jarocinski and Karadi2020) and Miranda-Agrippino and Ricco (Reference Miranda-Agrippino and Ricco2021) to obtain monetary policy surprises that are orthogonal to information effects, and, in addition, consider only a subset of FOMC meetings for which information effects are less likely to play a role (Nakamura and Steinsson, Reference Nakamura and Steinsson2018). The “No Info (JK)” columns in Figures 9 and 10 show responses of the conditional measures (as constructed in the baseline) obtained with Jarocinski and Karadi’s (Reference Jarocinski and Karadi2020) “poor man’s” approach, which uses stock market surprises in addition to interest rate surprises to disentangle policy shocks and information effects. Responses to policy surprises purged from information effects as suggested in Miranda-Agrippino and Ricco (Reference Miranda-Agrippino and Ricco2021) are shown in columns “No Info (MRA)” in Figures 9 and 10. Specifically, we purge the surprises in the 3-month ahead futures rate (i.e. surprises in the fourth federal funds futures; FF4 $_m$ ) from the Federal Reserve’s internal Greenbook forecasts:

(4) \begin{align} FF4_m & = \alpha + \sum _{i=-1}^3 \gamma _i \widetilde{\Delta y}_{m,i} + \sum _{i=-1}^2 \lambda _i\! \left(\widetilde{\Delta y}_{m,i}-\widetilde{\Delta y}_{m-1,i}\right) + \sum _{i=-1}^3 \phi _i \tilde{\pi }_{m,i} \nonumber \\ & \quad + \sum _{i=-1}^2 \theta _i \!\left(\tilde{\pi }_{m,i}-\tilde{\pi }_{m-1,i}\right) + \rho \tilde{u}_{m0} + u_m, \end{align}

where the index $m$ corresponds to the FOMC meetings, and $\widetilde{\Delta y_{m,i}}$ , $\tilde{\pi }_{m,i}$ , and $\tilde{u}_{m,i}$ are Greenbook forecasts of real output growth, inflation, and the unemployment rate for the previous ( $i=-1$ ), current ( $i=0$ ), or subsequent ( $i=1,2,3$ ) quarters at each FOMC meeting $m$ . The purged policy surprises $u_m$ are aggregated to a monthly frequency as suggested in Gertler and Karadi (Reference Gertler and Karadi2015). In columns “Scheduled” in Figures 9 and 10, we only consider scheduled FOMC meetings in the construction of the monetary policy surprise measure as in Nakamura and Steinsson (Reference Nakamura and Steinsson2018).Footnote ¹⁶ The intuition is that unscheduled meetings are usually due to abrupt macroeconomic developments. In these meetings, the monetary authority reveals information to the markets that pertain to signaling effects. Our main conclusions remain unchanged when we take information effects explicitly into account.

Next, we evaluate the robustness of our results with respect to the specification of prior parameters. In the baseline, we set the tightness and decay hyperparameters to 0.05 and 0.5. We now consider alternative values of 0.2 and 2 (as in e.g. Jarocinski and Karadi, Reference Jarocinski and Karadi2020). The responses of conditional measures are shown in columns labeled “Alternative Prior” in Figures 9 and 10 and are largely unaffected by the choice of prior parameters.

To take unconventional policy measures that directly target the longer end of the yield curve more explicitly into account, we consider surprises in either the 2-year government bond yield (“2Y Surprise”) or the 10-year government bond yield (“10Y Surprise”).Footnote ¹⁷

We use long-term inflation expectations from the Michigan Survey to construct our measure of inflation expectations (“Long-Term Inf. Ex.”), and we calculate the ordinal measure of expected inflation using the average expected inflation from the previous month instead of the realized inflation rate (“Past Av. Inf. Ex.”).

Finally, we have re-estimated the model with CRTS measures constructed based on a spending question that considers a longer horizon (“Alternative CRTS”). Specifically, we use Question A19: “Speaking now of the automobile market—do you think the next 12 months or so will be a good time or a bad time to buy a new vehicle, such as a car, pickup, van or sport utility vehicle?.” Although this question is more specific about the automobile market, rather than durable consumption goods in general, it has a longer horizon than the question that we use in our baseline, which asks about current readiness to spend.

Overall, we obtain similar results as in our baseline analysis.