Subjective Models of the Macroeconomy: Evidence From Experts and Representative Samples

Associations in the federal funds rate vignette: examples of households’ open-text responses

Thoughts about a cost channel	Thoughts about demand-side channels
“If the cost to borrow funds goes up, then a business will have to pay more to pay back a loan. Thus, businesses will have to raise prices. This will result in inflation. A business may not be able to pay employees and have to let them go or a business will not be able to pay back the load and the business will fail. The employees will lose their jobs and raises unemployment” “I believe if the fed rate increases, the inflation rate will as well because companies will be paying more on their credit and they will pass that on to consumers. Do not think it will affect unemployment.” “If the Fed rate is increased, the following usually happens—the cost of borrowing money for businesses increases—the business has to raise prices—there is usually a corresponding effect on the unemployment rate as employers find they have to cut staff to remain competitive ” “The higher federal funds rate causes the cost of borrowing to rise. As a result, prices are raised. And employment is lowered to cover cost of borrowing.” “When the interest rate rises that would mean that it would cost more for companies to borrow money and so they would charge more for their products (inflation would go up) and they would not have money to expand and hire more people (unemployment would go up). I really don’t know if the exact amounts of the inflation and unemployment rises would be the same as the \|$\%$\| that the inflation rate rose but I thought maybe it would.” “The cost of business goes up so business will try to raise prices to make a profit. Business will try to cut costs by employing fewer workers.”	“with change in fed funds rate upward, unemployment is likely to rise (as cost to business to borrow increases and invest less in expansion) and inflation should in theory be kept in check and even fall.” “Interest rates rising will increase the cost of investment. This will make companies lay people off. However, with higher interest rates, less money will be invested and it will cause inflation to fall.” “when the interest rate goes up I believe the unemployment rate goes up as well. Inflation will also hurt the job market. If people are not buying the jobs decrease.” “the demand will decrease and the investment will be less than usual also saving will be increased” “With the target rate going up, money will become more expensive to borrow, consumer credit rates will rise. This will cause consumer demand to drop and possibly put people out of work” “when interest rates increase there is less spending no new jobs” “Interest rate hike will cause less overall spending slightly more unemployment and greater inflation as prices adjust to this rate hike.”

Thoughts about a cost channel

Thoughts about demand-side channels

“If the cost to borrow funds goes up, then a business will have to pay more to pay back a loan. Thus, businesses will have to raise prices. This will result in inflation. A business may not be able to pay employees and have to let them go or a business will not be able to pay back the load and the business will fail. The employees will lose their jobs and raises unemployment”

“I believe if the fed rate increases, the inflation rate will as well because companies will be paying more on their credit and they will pass that on to consumers. Do not think it will affect unemployment.”

“If the Fed rate is increased, the following usually happens—the cost of borrowing money for businesses increases—the business has to raise prices—there is usually a corresponding effect on the unemployment rate as employers find they have to cut staff to remain competitive ”

“The higher federal funds rate causes the cost of borrowing to rise. As a result, prices are raised. And employment is lowered to cover cost of borrowing.”

“When the interest rate rises that would mean that it would cost more for companies to borrow money and so they would charge more for their products (inflation would go up) and they would not have money to expand and hire more people (unemployment would go up). I really don’t know if the exact amounts of the inflation and unemployment rises would be the same as the |$\%$| that the inflation rate rose but I thought maybe it would.”

“The cost of business goes up so business will try to raise prices to make a profit. Business will try to cut costs by employing fewer workers.”

“with change in fed funds rate upward, unemployment is likely to rise (as cost to business to borrow increases and invest less in expansion) and inflation should in theory be kept in check and even fall.”

“Interest rates rising will increase the cost of investment. This will make companies lay people off. However, with higher interest rates, less money will be invested and it will cause inflation to fall.”

“when the interest rate goes up I believe the unemployment rate goes up as well. Inflation will also hurt the job market. If people are not buying the jobs decrease.”

“the demand will decrease and the investment will be less than usual also saving will be increased”

“With the target rate going up, money will become more expensive to borrow, consumer credit rates will rise. This will cause consumer demand to drop and possibly put people out of work”

“when interest rates increase there is less spending no new jobs”

“Interest rate hike will cause less overall spending slightly more unemployment and greater inflation as prices adjust to this rate hike.”

Notes: This table displays examples for households’ responses to the open-text question, focusing on the monetary policy vignette. The left-hand side focuses on responses explicitly referring to a cost channel and neglecting demand-side mechanisms. The right-hand side focuses on responses pointing to demand-side channels.

Table 1.

Associations in the federal funds rate vignette: examples of households’ open-text responses

Thoughts about a cost channel	Thoughts about demand-side channels
“If the cost to borrow funds goes up, then a business will have to pay more to pay back a loan. Thus, businesses will have to raise prices. This will result in inflation. A business may not be able to pay employees and have to let them go or a business will not be able to pay back the load and the business will fail. The employees will lose their jobs and raises unemployment” “I believe if the fed rate increases, the inflation rate will as well because companies will be paying more on their credit and they will pass that on to consumers. Do not think it will affect unemployment.” “If the Fed rate is increased, the following usually happens—the cost of borrowing money for businesses increases—the business has to raise prices—there is usually a corresponding effect on the unemployment rate as employers find they have to cut staff to remain competitive ” “The higher federal funds rate causes the cost of borrowing to rise. As a result, prices are raised. And employment is lowered to cover cost of borrowing.” “When the interest rate rises that would mean that it would cost more for companies to borrow money and so they would charge more for their products (inflation would go up) and they would not have money to expand and hire more people (unemployment would go up). I really don’t know if the exact amounts of the inflation and unemployment rises would be the same as the \|$\%$\| that the inflation rate rose but I thought maybe it would.” “The cost of business goes up so business will try to raise prices to make a profit. Business will try to cut costs by employing fewer workers.”	“with change in fed funds rate upward, unemployment is likely to rise (as cost to business to borrow increases and invest less in expansion) and inflation should in theory be kept in check and even fall.” “Interest rates rising will increase the cost of investment. This will make companies lay people off. However, with higher interest rates, less money will be invested and it will cause inflation to fall.” “when the interest rate goes up I believe the unemployment rate goes up as well. Inflation will also hurt the job market. If people are not buying the jobs decrease.” “the demand will decrease and the investment will be less than usual also saving will be increased” “With the target rate going up, money will become more expensive to borrow, consumer credit rates will rise. This will cause consumer demand to drop and possibly put people out of work” “when interest rates increase there is less spending no new jobs” “Interest rate hike will cause less overall spending slightly more unemployment and greater inflation as prices adjust to this rate hike.”

Thoughts about a cost channel

Thoughts about demand-side channels

“The higher federal funds rate causes the cost of borrowing to rise. As a result, prices are raised. And employment is lowered to cover cost of borrowing.”

“The cost of business goes up so business will try to raise prices to make a profit. Business will try to cut costs by employing fewer workers.”

“when the interest rate goes up I believe the unemployment rate goes up as well. Inflation will also hurt the job market. If people are not buying the jobs decrease.”

“the demand will decrease and the investment will be less than usual also saving will be increased”

“With the target rate going up, money will become more expensive to borrow, consumer credit rates will rise. This will cause consumer demand to drop and possibly put people out of work”

“when interest rates increase there is less spending no new jobs”

“Interest rate hike will cause less overall spending slightly more unemployment and greater inflation as prices adjust to this rate hike.”

Moreover, across all vignettes, sizable fractions of households (about 20|$\%$| to 30|$\%$|⁠) indicate thoughts that firms react to reductions in demand by increasing prices to maintain profit levels—a channel that has no role in standard models, and which is selected by almost none of the experts. Households’ positive predicted inflation response to interest rate or income tax hikes—the most striking deviation from experts’ forecasts—could thus be partially driven by (1) relatively higher attention to supply-side factors and (2) a different view on how firms adjust their prices in response to changes in product demand.

In the government spending shock, households select channels working through increases in product demand much less frequently than experts (between 25|$\%$| and 33|$\%$| among households compared to between 53|$\%$| and 63|$\%$| among experts). By contrast, households are almost twice as likely as experts to indicate reductions in household spending due to an increase in expected future taxes (29|$\%$| vs. 14|$\%$|⁠). Together, these patterns could explain households’ more muted average prediction about the unemployment response to higher government spending.

Finally, we use the open-ended data to document that experts are more likely to account for general equilibrium effects in their forecasts than households based on two facts. First, 10|$\%$| and 6|$\%$| of experts refer to endogenous reactions of the central bank to the oil shock and to the government spending shock, respectively, in the open-text question (see Figure A.7). Virtually none of the households refer to reactions by the Fed to these shocks. Second, 22|$\%$| of the experts explicitly refer to an economic model (such as the New-Keynesian model), compared to none of the households, suggesting that experts are more likely to think about the shocks through the lens of economic theories. These theories in turn account for general equilibrium effects of the shocks.¹⁹

Discussion. Taken together, we find strong heterogeneity in the propagation mechanisms respondents think about, both within and between our samples of households and experts. The responses by experts suggest that many experts retrieve textbook models when making their forecasts. These models in turn account for general equilibrium effects of the shocks. Heterogeneity within the expert sample could, for instance, be driven by differences in academic backgrounds or fields of expertise.²⁰ Households frequently choose channels that are less important in textbook models and often neglect mechanisms that are commonly considered to be central. Their forecasting seems to be based on a patchwork of partial equilibrium responses that strongly differs across contexts and individuals. Households often do not account for second-round effects, such as policy responses, or disagree on their direction, such as for the pricing response of firms to changes in product demand. We explore the role of heterogeneous personal experiences as one driver of differences in associations within the household sample in Section 4.5.2 below.

Taking together the evidence presented above, our second main result is the following:

Result 2

The propagation channels that are on top of respondents’ minds vary systematically within and across our samples of households and experts. Experts tend to recall channels that are central in textbook models, while households in many cases neglect these channels and think of channels that are conventionally seen as less important.

Robustness: open-ended responses. We also leverage responses to the open-text question eliciting participants’ thoughts on the prediction screen to demonstrate the robustness of our findings to a different measurement technique. First, Supplementary Appendix Figure A.7 highlights how frequently different word groups are mentioned in the open-ended question across vignettes and samples. While naturally the levels are not comparable between structured and unstructured data of thoughts, we replicate differences between households and experts in terms of the relative importance of different mechanisms. Second, in Supplementary Appendix F, we develop a coding scheme to manually categorize open-ended responses into thoughts of different mechanisms. Each response is independently coded by two coders, with high inter-rater reliability. The hand-coded measures of thoughts are strongly correlated with our main measures based on the structured question (see Supplementary Appendix Tables A.19 and A.20), and are similarly distributed across vignettes (see Supplementary Appendix Figure A.11). These findings validate our measures based on the structured questions and mitigate concerns related to ex post rationalization of forecasts in the structured questions.

4.4. Correlations between associations and predictions

Is heterogeneity in thoughts about propagation channels driving heterogeneity in inflation and unemployment forecasts? Table 2 shows that the propagation mechanisms selected in the structured question are strongly associated with inflation and unemployment forecasts in both the expert and the household sample across all four vignettes. For presentational convenience, we use dummies indicating whether a respondent selects at least one (positive/negative) demand-side or supply-side channel, respectively.²¹

Table 2.

Thoughts of propagation channels correlate with predictions

Oil price
	Households		Experts
	\|$\Delta\pi$\|	\|$\Delta u$\|	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)
Supply (–)	0.343^***	0.145^***	–0.010	0.329^***
	(0.047)	(0.046)	(0.081)	(0.063)
Demand (–)	0.069^*	0.230^***	0.174^**	0.165^**
	(0.040)	(0.041)	(0.077)	(0.076)
Constant	0.173^***	0.090^**	0.296^***	–0.076^*
	(0.044)	(0.042)	(0.065)	(0.046)
Observations	557	557	91	91
R²	0.113	0.078	0.058	0.150
R² (all 7 channel indicators)	0.168	0.214	0.095	0.440
Government spending
Crowding-out	0.140^***	0.236^***	–0.036	0.057
	(0.050)	(0.055)	(0.071)	(0.046)
Demand (+)	–0.067	–0.249^***	0.076	–0.299^***
	(0.045)	(0.047)	(0.076)	(0.057)
Constant	0.329^***	0.080^**	0.195^***	0.009
	(0.038)	(0.037)	(0.067)	(0.051)
Observations	519	519	88	88
R²	0.023	0.102	0.014	0.266
R² (all 7 channel indicators)	0.062	0.180	0.178	0.438
Federal funds target rate
Supply (–)	0.188^***	0.142^***	0.090	–0.094
	(0.041)	(0.044)	(0.075)	(0.063)
Demand (–)	–0.053	0.088^**	–0.324^***	0.340^***
	(0.040)	(0.044)	(0.096)	(0.068)
Constant	0.229^***	0.068^*	0.068	–0.012
	(0.034)	(0.039)	(0.084)	(0.063)
Observations	520	520	92	92
R²	0.041	0.032	0.175	0.199
R² (all 8 channel indicators)	0.088	0.068	0.167	0.206
Income taxes
Supply (–)	0.217^***	0.188^***	0.018	0.004
	(0.041)	(0.044)	(0.074)	(0.074)
Demand (–)	0.024	0.054	–0.150^***	0.212^***
	(0.041)	(0.043)	(0.046)	(0.038)
Constant	0.254^***	0.130^***	–0.035	0.041
	(0.032)	(0.034)	(0.041)	(0.030)
Observations	530	530	100	100
R²	0.053	0.039	0.095	0.169
R² (all 8 channel indicators)	0.128	0.129	0.375	0.277

Oil price
	Households		Experts
	\|$\Delta\pi$\|	\|$\Delta u$\|	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)
Supply (–)	0.343^***	0.145^***	–0.010	0.329^***
	(0.047)	(0.046)	(0.081)	(0.063)
Demand (–)	0.069^*	0.230^***	0.174^**	0.165^**
	(0.040)	(0.041)	(0.077)	(0.076)
Constant	0.173^***	0.090^**	0.296^***	–0.076^*
	(0.044)	(0.042)	(0.065)	(0.046)
Observations	557	557	91	91
R²	0.113	0.078	0.058	0.150
R² (all 7 channel indicators)	0.168	0.214	0.095	0.440
Government spending
Crowding-out	0.140^***	0.236^***	–0.036	0.057
	(0.050)	(0.055)	(0.071)	(0.046)
Demand (+)	–0.067	–0.249^***	0.076	–0.299^***
	(0.045)	(0.047)	(0.076)	(0.057)
Constant	0.329^***	0.080^**	0.195^***	0.009
	(0.038)	(0.037)	(0.067)	(0.051)
Observations	519	519	88	88
R²	0.023	0.102	0.014	0.266
R² (all 7 channel indicators)	0.062	0.180	0.178	0.438
Federal funds target rate
Supply (–)	0.188^***	0.142^***	0.090	–0.094
	(0.041)	(0.044)	(0.075)	(0.063)
Demand (–)	–0.053	0.088^**	–0.324^***	0.340^***
	(0.040)	(0.044)	(0.096)	(0.068)
Constant	0.229^***	0.068^*	0.068	–0.012
	(0.034)	(0.039)	(0.084)	(0.063)
Observations	520	520	92	92
R²	0.041	0.032	0.175	0.199
R² (all 8 channel indicators)	0.088	0.068	0.167	0.206
Income taxes
Supply (–)	0.217^***	0.188^***	0.018	0.004
	(0.041)	(0.044)	(0.074)	(0.074)
Demand (–)	0.024	0.054	–0.150^***	0.212^***
	(0.041)	(0.043)	(0.046)	(0.038)
Constant	0.254^***	0.130^***	–0.035	0.041
	(0.032)	(0.034)	(0.041)	(0.030)
Observations	530	530	100	100
R²	0.053	0.039	0.095	0.169
R² (all 8 channel indicators)	0.128	0.129	0.375	0.277

Notes: This table shows data from Wave 3. It regresses the predicted inflation (⁠|$\Delta \pi$|⁠) and unemployment (⁠|$\Delta u$|⁠) changes on the propagation channels that were on respondents’ minds while they made their predictions (see Figure 3). Each panel presents results for a different vignette. In each panel, Columns (1) and (2) present results for households, Columns (3) and (4) present results for experts. “Supply (–)” takes value 1 for respondents who choose a negative supply-side propagation channel. “Demand (–)” and “Demand (+)” take value 1 for respondents choosing a negative or positive demand-side propagation channel, respectively. In the government spending vignette, “Crowding-out” takes value 1 for respondents who select the channel that demand falls due to higher expected future taxes (see Figure 3 for more details). Robust standard errors are in parentheses. * denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. level.

Table 2.

Thoughts of propagation channels correlate with predictions

Oil price
	Households		Experts
	\|$\Delta\pi$\|	\|$\Delta u$\|	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)
Supply (–)	0.343^***	0.145^***	–0.010	0.329^***
	(0.047)	(0.046)	(0.081)	(0.063)
Demand (–)	0.069^*	0.230^***	0.174^**	0.165^**
	(0.040)	(0.041)	(0.077)	(0.076)
Constant	0.173^***	0.090^**	0.296^***	–0.076^*
	(0.044)	(0.042)	(0.065)	(0.046)
Observations	557	557	91	91
R²	0.113	0.078	0.058	0.150
R² (all 7 channel indicators)	0.168	0.214	0.095	0.440
Government spending
Crowding-out	0.140^***	0.236^***	–0.036	0.057
	(0.050)	(0.055)	(0.071)	(0.046)
Demand (+)	–0.067	–0.249^***	0.076	–0.299^***
	(0.045)	(0.047)	(0.076)	(0.057)
Constant	0.329^***	0.080^**	0.195^***	0.009
	(0.038)	(0.037)	(0.067)	(0.051)
Observations	519	519	88	88
R²	0.023	0.102	0.014	0.266
R² (all 7 channel indicators)	0.062	0.180	0.178	0.438
Federal funds target rate
Supply (–)	0.188^***	0.142^***	0.090	–0.094
	(0.041)	(0.044)	(0.075)	(0.063)
Demand (–)	–0.053	0.088^**	–0.324^***	0.340^***
	(0.040)	(0.044)	(0.096)	(0.068)
Constant	0.229^***	0.068^*	0.068	–0.012
	(0.034)	(0.039)	(0.084)	(0.063)
Observations	520	520	92	92
R²	0.041	0.032	0.175	0.199
R² (all 8 channel indicators)	0.088	0.068	0.167	0.206
Income taxes
Supply (–)	0.217^***	0.188^***	0.018	0.004
	(0.041)	(0.044)	(0.074)	(0.074)
Demand (–)	0.024	0.054	–0.150^***	0.212^***
	(0.041)	(0.043)	(0.046)	(0.038)
Constant	0.254^***	0.130^***	–0.035	0.041
	(0.032)	(0.034)	(0.041)	(0.030)
Observations	530	530	100	100
R²	0.053	0.039	0.095	0.169
R² (all 8 channel indicators)	0.128	0.129	0.375	0.277

Oil price
	Households		Experts
	\|$\Delta\pi$\|	\|$\Delta u$\|	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)
Supply (–)	0.343^***	0.145^***	–0.010	0.329^***
	(0.047)	(0.046)	(0.081)	(0.063)
Demand (–)	0.069^*	0.230^***	0.174^**	0.165^**
	(0.040)	(0.041)	(0.077)	(0.076)
Constant	0.173^***	0.090^**	0.296^***	–0.076^*
	(0.044)	(0.042)	(0.065)	(0.046)
Observations	557	557	91	91
R²	0.113	0.078	0.058	0.150
R² (all 7 channel indicators)	0.168	0.214	0.095	0.440
Government spending
Crowding-out	0.140^***	0.236^***	–0.036	0.057
	(0.050)	(0.055)	(0.071)	(0.046)
Demand (+)	–0.067	–0.249^***	0.076	–0.299^***
	(0.045)	(0.047)	(0.076)	(0.057)
Constant	0.329^***	0.080^**	0.195^***	0.009
	(0.038)	(0.037)	(0.067)	(0.051)
Observations	519	519	88	88
R²	0.023	0.102	0.014	0.266
R² (all 7 channel indicators)	0.062	0.180	0.178	0.438
Federal funds target rate
Supply (–)	0.188^***	0.142^***	0.090	–0.094
	(0.041)	(0.044)	(0.075)	(0.063)
Demand (–)	–0.053	0.088^**	–0.324^***	0.340^***
	(0.040)	(0.044)	(0.096)	(0.068)
Constant	0.229^***	0.068^*	0.068	–0.012
	(0.034)	(0.039)	(0.084)	(0.063)
Observations	520	520	92	92
R²	0.041	0.032	0.175	0.199
R² (all 8 channel indicators)	0.088	0.068	0.167	0.206
Income taxes
Supply (–)	0.217^***	0.188^***	0.018	0.004
	(0.041)	(0.044)	(0.074)	(0.074)
Demand (–)	0.024	0.054	–0.150^***	0.212^***
	(0.041)	(0.043)	(0.046)	(0.038)
Constant	0.254^***	0.130^***	–0.035	0.041
	(0.032)	(0.034)	(0.041)	(0.030)
Observations	530	530	100	100
R²	0.053	0.039	0.095	0.169
R² (all 8 channel indicators)	0.128	0.129	0.375	0.277

Most of the correlational patterns uncovered in Table 2 go into the expected direction. For example, households thinking of negative supply-side propagation channels expect higher increases of inflation (⁠|$p<0.01$|⁠) and unemployment (⁠|$p<0.01$|⁠) in response to an oil price shock. Experts choosing supply-side propagation channels also expect higher increases in unemployment (⁠|$p<0.01$|⁠) in response to oil price hikes, but do not expect higher levels of inflation. In the context of the government spending shock, we uncover robust negative correlations between choosing propagation channels related to positive demand-side shocks and expected changes in unemployment rates (⁠|$p<0.01$|⁠). Among households, we also find a strong positive association between choosing channels related to crowding-out and predicted increases in inflation (⁠|$p<0.01$|⁠) in response to a government spending increase, while for experts this association is more muted. For households, we document strong positive associations between choosing supply-related propagation mechanisms and predicted increases in inflation (⁠|$p<0.01$|⁠) and unemployment (⁠|$p<0.01$|⁠) in response to both an interest rate hike and an increase in income taxes, while for experts these patterns are less pronounced. For experts, on the other hand, we find that choosing demand-related mechanisms is associated with lower inflation (⁠|$p<0.01$|⁠) and higher unemployment (⁠|$p<0.01$|⁠) predictions in response to both an interest rate and an income tax hike.

Table 2 illustrates that, across shocks, dummies for thoughts about different propagation channels have significant explanatory power for forecasts. Regressing forecasts on dummies for all vignette-specific channels gives an R-squared between 6|$\%$| and 21|$\%$| for households, and between 10|$\%$| and 44|$\%$| for experts. These values are sizeable given the low R-squared often documented in studies of the determinants of survey expectations, such as individual characteristics or experiences (Malmendier and Nagel, 2011; Kuchler and Zafar, 2019; Das, Kuhnen and Nagel, 2020; Giglio et al., 2021). The actual explanatory power of associations is likely even larger than measured in our survey given (1) the potential measurement error in associations, (2) the fact that we do not measure the perceived strength of the different channels, and (3) the possibility that we do not capture all relevant channels that respondents have on their minds.

Can differences in associations account for differences in average predictions between households and experts? Table 3 examines the extent to which the gap in predictions between experts and households can be explained by differences in responses to the structured question on propagation mechanisms. Our analysis zooms in on the three predictions for which the average gap between households and experts is most pronounced. Columns 1 and 2 show that the average differences in unemployment predictions in the government spending vignette are fully explained by differences in the selected propagation mechanisms. Columns 3 and 4 show that the propagation channels explain approximately one third of the gap in inflation predictions in the interest rate vignette. Finally, they explain about one third of the prediction gap in the tax vignette (Columns 5 and 6). Taking together the evidence presented above, our third main result is the following:

Table 3.

Thoughts of propagation channels account for differences between experts’ and households’ predictions

	Government spending		Federal funds rate		Income taxes
	Unemployment \|$\boldsymbol{\Delta u}$\|		Inflation \|$\boldsymbol{\Delta \pi}$\|		Inflation \|$\boldsymbol{\Delta \pi}$\|
	(1)	(2)	(3)	(4)	(5)	(6)
Expert	–0.215^***	–0.003	–0.462^***	–0.323^***	–0.517^***	–0.347^***
	(0.036)	(0.035)	(0.037)	(0.048)	(0.030)	(0.041)
Constant	0.013	0.040	0.297^***	0.207^***	0.368^***	0.248^***
	(0.025)	(0.035)	(0.020)	(0.030)	(0.021)	(0.030)
\|$\boldsymbol{p_F}$\|⁠: Expert coeff. equal		\|$\boldsymbol{<}$\|0.001		\|$\boldsymbol{<}$\|0.001		\|$\boldsymbol{<}$\|0.001
Channels	–	\|$\checkmark$\|	–	\|$\checkmark$\|	–	\|$\checkmark$\|
Observations	608	607	614	612	631	630
R\|$^{2}$\|	0.020	0.203	0.127	0.199	0.152	0.258

	Government spending		Federal funds rate		Income taxes
	Unemployment \|$\boldsymbol{\Delta u}$\|		Inflation \|$\boldsymbol{\Delta \pi}$\|		Inflation \|$\boldsymbol{\Delta \pi}$\|
	(1)	(2)	(3)	(4)	(5)	(6)
Expert	–0.215^***	–0.003	–0.462^***	–0.323^***	–0.517^***	–0.347^***
	(0.036)	(0.035)	(0.037)	(0.048)	(0.030)	(0.041)
Constant	0.013	0.040	0.297^***	0.207^***	0.368^***	0.248^***
	(0.025)	(0.035)	(0.020)	(0.030)	(0.021)	(0.030)
\|$\boldsymbol{p_F}$\|⁠: Expert coeff. equal		\|$\boldsymbol{<}$\|0.001		\|$\boldsymbol{<}$\|0.001		\|$\boldsymbol{<}$\|0.001
Channels	–	\|$\checkmark$\|	–	\|$\checkmark$\|	–	\|$\checkmark$\|
Observations	608	607	614	612	631	630
R\|$^{2}$\|	0.020	0.203	0.127	0.199	0.152	0.258

Notes: This table uses data from Wave 3 of the expert and household surveys. It tests whether thoughts of different propagation channels (see Figure 3) can account for the differences in experts’ and households’ predicted inflation (⁠|$\Delta \pi$|⁠) and unemployment (⁠|$\Delta u$|⁠) changes. We consider the three cases for which large differences in experts’ and households’ predictions can be found: Unemployment in the government spending vignette (columns 1–2), inflation in the federal funds rate vignette (columns 3–4), and inflation in the income tax rate vignette (columns 5–6). “Expert” takes value 1 for respondents from the expert sample. Results in columns (2), (4), and (6) control for the selected propagation channels (7–8 indicators, depending on the vignette, see Figure 3 for all propagation channels). p-values result from an F-test of equality of the “Expert” coefficient with and without channel controls (estimated using seemingly unrelated regressions). * denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. level.

Table 3.

Thoughts of propagation channels account for differences between experts’ and households’ predictions

	Government spending		Federal funds rate		Income taxes
	Unemployment \|$\boldsymbol{\Delta u}$\|		Inflation \|$\boldsymbol{\Delta \pi}$\|		Inflation \|$\boldsymbol{\Delta \pi}$\|
	(1)	(2)	(3)	(4)	(5)	(6)
Expert	–0.215^***	–0.003	–0.462^***	–0.323^***	–0.517^***	–0.347^***
	(0.036)	(0.035)	(0.037)	(0.048)	(0.030)	(0.041)
Constant	0.013	0.040	0.297^***	0.207^***	0.368^***	0.248^***
	(0.025)	(0.035)	(0.020)	(0.030)	(0.021)	(0.030)
\|$\boldsymbol{p_F}$\|⁠: Expert coeff. equal		\|$\boldsymbol{<}$\|0.001		\|$\boldsymbol{<}$\|0.001		\|$\boldsymbol{<}$\|0.001
Channels	–	\|$\checkmark$\|	–	\|$\checkmark$\|	–	\|$\checkmark$\|
Observations	608	607	614	612	631	630
R\|$^{2}$\|	0.020	0.203	0.127	0.199	0.152	0.258

	Government spending		Federal funds rate		Income taxes
	Unemployment \|$\boldsymbol{\Delta u}$\|		Inflation \|$\boldsymbol{\Delta \pi}$\|		Inflation \|$\boldsymbol{\Delta \pi}$\|
	(1)	(2)	(3)	(4)	(5)	(6)
Expert	–0.215^***	–0.003	–0.462^***	–0.323^***	–0.517^***	–0.347^***
	(0.036)	(0.035)	(0.037)	(0.048)	(0.030)	(0.041)
Constant	0.013	0.040	0.297^***	0.207^***	0.368^***	0.248^***
	(0.025)	(0.035)	(0.020)	(0.030)	(0.021)	(0.030)
\|$\boldsymbol{p_F}$\|⁠: Expert coeff. equal		\|$\boldsymbol{<}$\|0.001		\|$\boldsymbol{<}$\|0.001		\|$\boldsymbol{<}$\|0.001
Channels	–	\|$\checkmark$\|	–	\|$\checkmark$\|	–	\|$\checkmark$\|
Observations	608	607	614	612	631	630
R\|$^{2}$\|	0.020	0.203	0.127	0.199	0.152	0.258

Result 3

Thoughts of specific propagation channels are correlated with forecasts of the effects of macroeconomic shocks on inflation and unemployment in the expected directions, and account for part of the differences in forecasts between households and experts.

Other drivers of forecasts. In Supplementary Appendix Section H.1, we provide evidence on other potential drivers of respondents’ forecasts. That section illustrates a quantitatively important role of associations about propagation channels relative to other potential drivers of forecasts, including respondents’ (1) beliefs about historical correlations of different macroeconomic variables, (2) perceived importance of macroeconomic knowledge for making good economic decisions, (3) actual knowledge of different aspects of the economy, (4) numeracy, and (5) background characteristics.

4.5. Memory and heterogeneity in associations

Human memory is known to be associative, selective, and to draw on personal experiences (Kahana, 2012; Bordalo, Gennaioli and Shleifer, 2020b; Enke et al., 2020). Is associative memory driving heterogeneity in thoughts of propagation channels and forecasts? In this section, we test two key predictions of theories of associative memory. First, theories of associative memory predict that contextual cues affect the forecasts people make by changing the associations that come to their minds. Second, different personal experiences in the memory database should be reflected in differences in associations and forecasts. We test these conjectures based on a priming intervention and an additional correlational exercise exploiting rich data on people’s personal experiences.

4.5.1. The effect of contextual cues

To shed light on the role of contextual cues, we conduct a simple experiment. The experiment aims to provide a proof of concept that an exogenous change in contextual cues can change people’s selective retrieval of propagation mechanisms and, thereby, causally affect their forecasts of the effects of macroeconomic shocks. If an exogenous change in attention to specific aspects of the prediction problem changes respondents’ forecasts, this would suggest that individuals do not hold a “fixed” subjective model, but instead form their models “on the fly,” depending on the associations triggered by the context.

Sample. We conduct this experiment with a sample of 1,521 respondents provided by Lucid in February 2021 (Wave 4 of the household survey). Our sample is again broadly representative of the US population in terms of a set of basic demographic variables (see Supplementary Appendix Table A.1).

Design. Our design closely follows the descriptive survey on associations, except that it only focuses on inflation expectations and the interest rate vignette (see Supplementary Appendix Figure A.1 for a visual summary). We focus on beliefs about the effect of a federal funds rate hike on the inflation rate as this is one of the cases where predictions differ the most between households and experts.²² Moreover, monetary policy innovations are the most studied type of shock in the theoretical and empirical literature.

In the experiment, we randomize respondents into one of three arms: Respondents in the “cost treatment” are asked two additional questions on firms’ costs of doing business before making their inflation prediction. First, they are asked whether US firms face higher or lower costs of doing business when the federal funds rate rises. Second, they are asked to describe their main considerations in making their prediction about costs in an open-text box. In the “demand treatment,” respondents are asked about the demand for firms’ products before they forecast effects on inflation. First, they are asked whether firms face higher or lower demand for their goods and services when the federal funds rate rises. Second, as in the cost treatment, they describe their main considerations in making the prediction about demand in an open-ended question. Respondents in the “control treatment” do not receive any additional prompt before they make their inflation prediction. Respondents in all three groups report in an open-text box what considerations are on their mind while they make their inflation prediction.²³

At the end of the survey, respondents in the control treatment are asked either the same two additional questions on costs (“cost control group”) or the same two additional questions on demand (“demand control group”). This allows us to characterize heterogeneity in beliefs and to study whether the effects of our priming treatments depend on participants’ beliefs about the direction of the effect of the federal funds rate hike on costs or demand.²⁴

The purpose of asking respondents to forecast the response of costs or demand to the shock before they make their inflation forecast is to exogenously expose respondents to different contextual cues, which in turn point to different aspects of the shock. For instance, if cues drive associations and forecasts, then our demand treatment should reduce respondents’ inflation forecasts by increasing their retrieval of demand-side mechanisms. We believe that asking an additional question on the decision screen is a relatively subtle way of changing the contextual cues our respondents are exposed to, which mitigates concerns about experimenter demand effects (de Quidt, Haushofer and Roth, 2018). In addition to providing evidence on the role of contextual cues, our experiment allows us to test for a causal relationship between associations and forecasts, thereby validating our evidence presented in Section 4.4.

Results. We leverage the text data in which respondents describe what is on their mind while making the inflation prediction to shed light on the effects of our treatments on selective retrieval of propagation mechanisms. Columns 1 and 2 of Table 4 present the effects of the treatments on the words that respondents use to describe their thoughts.²⁵ Respondents in the “cost treatment” arm are 8.6 p.p. (⁠|$p<0.01$|⁠) more likely to use words related to firms’ costs (control mean: 9.3|$\%$|⁠). The demand treatment increases the use of words related to demand by 7.7 p.p. (⁠|$p<0.01$|⁠)—a 75|$\%$| increase compared to the control group mean of 10.6|$\%$|⁠. There are no spillovers of the cost treatment on the use of demand-related words, or vice versa. The overall small fractions mentioning such words should be viewed in light of the unstructured nature of the open-text data. Taken together, the contextual cues in our treatments draw respondents’ attention to specific propagation mechanisms of the shocks.

Table 4.

Results of the priming study (households only)

	Word usage (open-text data)		Inflation prediction
	Cost-related words	Demand-related words	\|$\Delta \pi$\|
	(1)	(2)	(3)
Costs prime	0.086^***	0.007	0.021
	(0.023)	(0.020)	(0.031)
Demand prime	–0.021	0.077^***	–0.057^**
	(0.017)	(0.023)	(0.029)
Constant	0.093^***	0.106^***	0.366^***
	(0.010)	(0.011)	(0.017)
\|$p$\|⁠: Costs = Demand	<0.001	0.007	0.028
Observations	1,521	1,521	1,521
R\|$^{2}$\|	0.017	0.010	0.004

	Word usage (open-text data)		Inflation prediction
	Cost-related words	Demand-related words	\|$\Delta \pi$\|
	(1)	(2)	(3)
Costs prime	0.086^***	0.007	0.021
	(0.023)	(0.020)	(0.031)
Demand prime	–0.021	0.077^***	–0.057^**
	(0.017)	(0.023)	(0.029)
Constant	0.093^***	0.106^***	0.366^***
	(0.010)	(0.011)	(0.017)
\|$p$\|⁠: Costs = Demand	<0.001	0.007	0.028
Observations	1,521	1,521	1,521
R\|$^{2}$\|	0.017	0.010	0.004

Notes: This table presents results from the priming study which focuses on the interest rate vignette (Wave 4 of the household survey). “Costs prime” takes value 1 for respondents randomly assigned to be primed on the costs of production. “Demand prime” takes value 1 for respondents randomly assigned to be primed on product demand. Columns (1) and (2) show effects on word usage in the open-text responses, and Column (3) presents the effects on the inflation forecast. The variable “Cost-related words” takes value 1 for responses which include the word (stem) “cost.” “Demand-related words” takes value 1 for responses which use the words or word stems “demand,” “buy,” “purchas,” “invest,” “spend,” and “consum.” |$\Delta \pi$| denotes the perceived reaction of the inflation rate. Robust standard errors are in parentheses. * denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. level.

Table 4.

Results of the priming study (households only)

	Word usage (open-text data)		Inflation prediction
	Cost-related words	Demand-related words	\|$\Delta \pi$\|
	(1)	(2)	(3)
Costs prime	0.086^***	0.007	0.021
	(0.023)	(0.020)	(0.031)
Demand prime	–0.021	0.077^***	–0.057^**
	(0.017)	(0.023)	(0.029)
Constant	0.093^***	0.106^***	0.366^***
	(0.010)	(0.011)	(0.017)
\|$p$\|⁠: Costs = Demand	<0.001	0.007	0.028
Observations	1,521	1,521	1,521
R\|$^{2}$\|	0.017	0.010	0.004

	Word usage (open-text data)		Inflation prediction
	Cost-related words	Demand-related words	\|$\Delta \pi$\|
	(1)	(2)	(3)
Costs prime	0.086^***	0.007	0.021
	(0.023)	(0.020)	(0.031)
Demand prime	–0.021	0.077^***	–0.057^**
	(0.017)	(0.023)	(0.029)
Constant	0.093^***	0.106^***	0.366^***
	(0.010)	(0.011)	(0.017)
\|$p$\|⁠: Costs = Demand	<0.001	0.007	0.028
Observations	1,521	1,521	1,521
R\|$^{2}$\|	0.017	0.010	0.004

We next turn to the effects on respondents’ inflation forecasts. Column 3 of Table 4 shows that while the cost prime increases inflation predictions insignificantly by 0.021 p.p. (⁠|$p=0.50$|⁠), the demand prime significantly decreases inflation predictions by 0.057 p.p. (⁠|$p<0.05$|⁠). The weaker response of inflation forecasts to the cost treatment could be due to the fact that many households already predict a positive inflation response by default, potentially due to higher attention to supply-side mechanisms. This could limit the scope for further increases in inflation forecasts.

Despite the relatively large effects on associations, the effects on inflation forecasts we uncover are relatively small in magnitude. There are at least three potential explanations. First, the effect of attention to changes in costs or product demand on inflation forecasts should depend on respondents’ beliefs about the direction of changes in costs or product demand in response to the rate hike. If there is disagreement on the directions of these changes, this will attenuate the average effects of attention to costs or demand on inflation forecasts. Consistent with this conjecture, Supplementary Appendix Table A.6 shows that the demand treatment decreases inflation forecasts by 0.10 p.p. (⁠|$p<0.05$|⁠) among respondents expecting a decrease in demand, while it has no significant effect among those expecting an increase. Similarly, the cost treatment increases inflation predictions by 0.05 p.p. among respondents who expect an increase in costs (⁠|$p=0.20$|⁠), while it decreases predictions by 0.15 p.p. among respondents who expect a decrease in costs (⁠|$p=0.14$|⁠). Second, even among respondents with beliefs about changes in costs or changes in demand in the same direction, there could be disagreement about the direction of firms’ pricing response to a given change in costs or demand. Indeed, as documented in Section 4.3, households seem to disagree about the direction in which firms adjust their prices in response to decreases in demand. Such disagreement implies that higher attention to demand or costs shifts different households’ inflation forecasts in different directions, which may further attenuate the average effects on inflation forecasts. Third, inattention to the demand- or supply-side may only be part of the story, i.e., people could hold differential beliefs about the importance of demand- and supply-side channels in the transmission. Hence, even if respondents are made attentive to these channels, only part of them might think this is important for inflation.

4.5.2. The role of experiences

Associative memory also predicts that different personal experiences in the memory database should be reflected in differences in associations and forecasts. We use an additional data collection on the government spending vignette among households (Wave 5) and data on the oil price vignette (from Wave 3 of the household survey) to shed light on this conjecture.

Experiences with the propagation channels of military spending. In an additional data collection (Wave 5 of the household survey, |$n=486$|⁠), we collect data on the government spending vignette using identical baseline instructions as in Wave 3.²⁶ In addition, we include two main sets of variables to gauge the role of personal experiences.

First, we ask respondents to assess their overall experience with the mechanisms that we listed in our structured question on propagation channels, such as an increase in household spending due to higher incomes (see Figure 3). Respondents rate the extent to which they themselves or their family and friends have been part of each mechanism on a five-point scale ranging from “no experiences” to “a lot of experiences.” For the analysis, we compute two summary indices, namely the standardized sum of experiences with positive demand-side channels and a standardized version of experience with “crowding-out” channels. The two indices provide measures of respondents’ cumulative first-hand and second-hand experiences with propagation channels.

Second, we also zoom in on a more specific experience by eliciting whether the respondent or anyone among their friends and family members has ever been employed by a company receiving contracts from the US military. This, in turn, allows us to capture one specific way in which a respondent could have direct personal experience with the demand-side mechanisms and, in particular, the potential labour market effects of military spending increases.

Panel A of Table 5 shows that respondents who indicate to have more experiences with positive demand-side mechanisms are more likely to choose positive demand channels (⁠|$p<0.01$|⁠) and somewhat less likely to choose channels related to crowd-out (⁠|$p<0.10$|⁠) in the structured question, and are more likely to mention words related to product demand (⁠|$p<0.10$|⁠) and labour demand (⁠|$p<0.10$|⁠) in the open-text question. Conversely, respondents who have more experiences with crowd-out channels are more likely to choose propagation channels related to crowd-out (⁠|$p<0.01$|⁠) and less likely to choose channels related to increases in demand (⁠|$p<0.01$|⁠) in the structured question, and somewhat more likely to mention words related to costs (⁠|$p<0.10$|⁠) and less likely to mention words related to labour (⁠|$p<0.05$|⁠) in the open-text question. These differences in the propagation channels respondents think of are reflected in a more negative predicted unemployment response to the spending program among those with positive demand-side experiences (⁠|$p<0.01$|⁠) and a more positive predicted unemployment response among those with crowd-out experiences (⁠|$p<0.01$|⁠).

Table 5.

Households’ experiences correlate with mechanism associations and forecasts

(A) Government spending: experience with propagation channels (std. indices)
	Propagation channels		Word usage (open-text data)			Predictions
	Crowding-out	Demand (+)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Experienced crowd.-out	0.093^***	–0.081^***	0.023^*	–0.034	–0.050^**	0.004	0.106^***
	(0.026)	(0.026)	(0.012)	(0.025)	(0.025)	(0.026)	(0.029)
Experienced demand +	–0.046^*	0.107^***	0.003	0.044^*	0.048^*	0.038	–0.109^***
	(0.026)	(0.026)	(0.014)	(0.025)	(0.026)	(0.025)	(0.030)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	483	483	483	483	483	483	483
R\|$^{2}$\|	0.113	0.076	0.038	0.191	0.092	0.142	0.180
(B) Government spending: ever worked for military supplier (self/friend, binary indicator)
	Propagation channels		Word usage (open-text data)			Predictions
	Crowding-out	Demand (+)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Yes	–0.010	0.081^*	–0.005	0.036	0.121^***	–0.024	–0.101^**
	(0.043)	(0.046)	(0.020)	(0.043)	(0.042)	(0.045)	(0.049)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	483	483	483	483	483	483	483
R\|$^{2}$\|	0.088	0.050	0.025	0.187	0.098	0.137	0.155
(C) Oil price: experienced OPEC crisis (born before 1962, binary indicator)
	Propagation channels		Word usage (open-text data)			Predictions
	Supply (–)	Demand (–)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Yes	0.114^***	0.036	0.100^**	0.039	0.011	0.208^***	0.202^***
	(0.040)	(0.045)	(0.040)	(0.031)	(0.041)	(0.044)	(0.043)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	521	521	521	521	521	521	521
R\|$^{2}$\|	0.064	0.040	0.053	0.020	0.026	0.080	0.074

(A) Government spending: experience with propagation channels (std. indices)
	Propagation channels		Word usage (open-text data)			Predictions
	Crowding-out	Demand (+)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Experienced crowd.-out	0.093^***	–0.081^***	0.023^*	–0.034	–0.050^**	0.004	0.106^***
	(0.026)	(0.026)	(0.012)	(0.025)	(0.025)	(0.026)	(0.029)
Experienced demand +	–0.046^*	0.107^***	0.003	0.044^*	0.048^*	0.038	–0.109^***
	(0.026)	(0.026)	(0.014)	(0.025)	(0.026)	(0.025)	(0.030)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	483	483	483	483	483	483	483
R\|$^{2}$\|	0.113	0.076	0.038	0.191	0.092	0.142	0.180
(B) Government spending: ever worked for military supplier (self/friend, binary indicator)
	Propagation channels		Word usage (open-text data)			Predictions
	Crowding-out	Demand (+)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Yes	–0.010	0.081^*	–0.005	0.036	0.121^***	–0.024	–0.101^**
	(0.043)	(0.046)	(0.020)	(0.043)	(0.042)	(0.045)	(0.049)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	483	483	483	483	483	483	483
R\|$^{2}$\|	0.088	0.050	0.025	0.187	0.098	0.137	0.155
(C) Oil price: experienced OPEC crisis (born before 1962, binary indicator)
	Propagation channels		Word usage (open-text data)			Predictions
	Supply (–)	Demand (–)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Yes	0.114^***	0.036	0.100^**	0.039	0.011	0.208^***	0.202^***
	(0.040)	(0.045)	(0.040)	(0.031)	(0.041)	(0.044)	(0.043)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	521	521	521	521	521	521	521
R\|$^{2}$\|	0.064	0.040	0.053	0.020	0.026	0.080	0.074

Notes: This table presents results from Wave 3 (Panel C) and Wave 5 (Panels A and B) of the household survey. In Columns (1) and (2), it asks whether respondents who made experiences related to the vignettes think about different propagation mechanisms (binary indicators; see Figure 3). In Columns (3) to (5), it tests whether respondents with vignette-related experiences use different word (stems) in their open-text responses (binary indicators; “Costs”: cost; “Demand”: demand, buy, purchas, invest, spend, consum; “Labour”: layoff, fire, hire, labour, work, job). In Columns (6) and (7), it tests whether they make different forecasts (inflation: |$\Delta \pi$|⁠, unemployment: |$\Delta u$|⁠). The right-hand-side experience variable varies across panels. In Panel A, “Experienced crowding-out” and “Experienced demand (+)” are standardized indices of self-rated experiences (familiarity) with crowding-out and positive demand-side channels, respectively. In Panel B, “Yes” is a binary dummy taking value 1 if respondents themselves or friends/family of them ever worked for a company that sells to the US military. In Panel C, “Yes” is a binary dummy taking value 1 if respondents were born before 1962, a proxy that they experienced the OPEC crisis. Control variables comprise age (except for Panel C), log income, inflation, and unemployment forecasts in the baseline scenario, as well as binary indicators for gender, college education, being a Republican, having taken an economics course at the college level, and census regions. Robust standard errors are in parentheses. * denotes significance at 10 pct., ** at 5 pct., and *** at 1 pct. level.

Table 5.

Households’ experiences correlate with mechanism associations and forecasts

(A) Government spending: experience with propagation channels (std. indices)
	Propagation channels		Word usage (open-text data)			Predictions
	Crowding-out	Demand (+)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Experienced crowd.-out	0.093^***	–0.081^***	0.023^*	–0.034	–0.050^**	0.004	0.106^***
	(0.026)	(0.026)	(0.012)	(0.025)	(0.025)	(0.026)	(0.029)
Experienced demand +	–0.046^*	0.107^***	0.003	0.044^*	0.048^*	0.038	–0.109^***
	(0.026)	(0.026)	(0.014)	(0.025)	(0.026)	(0.025)	(0.030)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	483	483	483	483	483	483	483
R\|$^{2}$\|	0.113	0.076	0.038	0.191	0.092	0.142	0.180
(B) Government spending: ever worked for military supplier (self/friend, binary indicator)
	Propagation channels		Word usage (open-text data)			Predictions
	Crowding-out	Demand (+)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Yes	–0.010	0.081^*	–0.005	0.036	0.121^***	–0.024	–0.101^**
	(0.043)	(0.046)	(0.020)	(0.043)	(0.042)	(0.045)	(0.049)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	483	483	483	483	483	483	483
R\|$^{2}$\|	0.088	0.050	0.025	0.187	0.098	0.137	0.155
(C) Oil price: experienced OPEC crisis (born before 1962, binary indicator)
	Propagation channels		Word usage (open-text data)			Predictions
	Supply (–)	Demand (–)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Yes	0.114^***	0.036	0.100^**	0.039	0.011	0.208^***	0.202^***
	(0.040)	(0.045)	(0.040)	(0.031)	(0.041)	(0.044)	(0.043)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	521	521	521	521	521	521	521
R\|$^{2}$\|	0.064	0.040	0.053	0.020	0.026	0.080	0.074

(A) Government spending: experience with propagation channels (std. indices)
	Propagation channels		Word usage (open-text data)			Predictions
	Crowding-out	Demand (+)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Experienced crowd.-out	0.093^***	–0.081^***	0.023^*	–0.034	–0.050^**	0.004	0.106^***
	(0.026)	(0.026)	(0.012)	(0.025)	(0.025)	(0.026)	(0.029)
Experienced demand +	–0.046^*	0.107^***	0.003	0.044^*	0.048^*	0.038	–0.109^***
	(0.026)	(0.026)	(0.014)	(0.025)	(0.026)	(0.025)	(0.030)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	483	483	483	483	483	483	483
R\|$^{2}$\|	0.113	0.076	0.038	0.191	0.092	0.142	0.180
(B) Government spending: ever worked for military supplier (self/friend, binary indicator)
	Propagation channels		Word usage (open-text data)			Predictions
	Crowding-out	Demand (+)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Yes	–0.010	0.081^*	–0.005	0.036	0.121^***	–0.024	–0.101^**
	(0.043)	(0.046)	(0.020)	(0.043)	(0.042)	(0.045)	(0.049)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	483	483	483	483	483	483	483
R\|$^{2}$\|	0.088	0.050	0.025	0.187	0.098	0.137	0.155
(C) Oil price: experienced OPEC crisis (born before 1962, binary indicator)
	Propagation channels		Word usage (open-text data)			Predictions
	Supply (–)	Demand (–)	Costs	Demand	Labour	\|$\Delta\pi$\|	\|$\Delta u$\|
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Yes	0.114^***	0.036	0.100^**	0.039	0.011	0.208^***	0.202^***
	(0.040)	(0.045)	(0.040)	(0.031)	(0.041)	(0.044)	(0.043)
Controls	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|	\|$\checkmark$\|
Observations	521	521	521	521	521	521	521
R\|$^{2}$\|	0.064	0.040	0.053	0.020	0.026	0.080	0.074

Panel B of Table 5 shows that respondents who were either personally employed by a company receiving contracts from the US military or have someone among their friends and family members who was employed by such a company are somewhat more likely to choose propagation channels related to demand in the structured question (⁠|$p<0.10$|⁠), and are more likely to use words related to labour in the open-ended question (⁠|$p<0.01$|⁠) when they make their forecasts. They also predict a stronger decrease in the unemployment rate in response to the increase in government spending (⁠|$p<0.01$|⁠).²⁷

Experiences with oil supply shocks. To provide further evidence on the role of personal experiences, we leverage variation in whether respondents lived through the OPEC crisis in the 1970s—a singular and particularly memorable event. Building on prior work by Binder and Makridis (2022), we proxy personal experiences of the 1970s oil crisis with an indicator for whether the respondent was born before 1962 (teenagers by the late 1970s). Given that the oil price shocks of the 1970s are conventionally seen as supply-side shocks, we would expect respondents with personal experiences of the OPEC crisis to be more likely to recall channels related to production cost increases.

Panel C of Table 5 shows that individuals born before 1962 are indeed more likely to choose propagation channels related to the supply-side (⁠|$p<0.01$|⁠) and more likely to use words related to costs (⁠|$p<0.05$|⁠) when making predictions in the oil vignette. Consistent with the associations on top of their mind, respondents who experienced the OPEC crisis predict stronger increases in unemployment and inflation (⁠|$p<0.01$|⁠) (Panel C of Table 5).²⁸

Taking together the evidence presented in Section 4.5, our fourth and final result can be summarized as follows:

Result 4

Contextual cues causally affect households’ predictions by changing the associations that come to their minds. Moreover, personal experiences are correlated with selective recall of specific propagation mechanisms, which is reflected in individuals’ beliefs about the effects of macroeconomic shocks. These two findings are in line with an important role of associative memory in shaping individuals’ forecasts about the effects of macroeconomic shocks.

In this view, households do not form their expectations based on a fixed subjective model. Instead, individuals form their subjective models “on the fly,” in line with the associations that come to their mind, which depend on both the context and prior experiences. This suggests that news or actual events in the economy systematically affect which models people entertain. Rather than sticking to one particular model, economic agents retrieve specific experiences when cued by events, which in turn shape the economic mechanisms they think of.

Other sources of associations. Personal experiences typically vary widely across individuals and are hence likely to be a key driver of heterogeneity in associations regarding macroeconomic shocks. At the same time, personal experiences are likely not the only source of households’ associations. For instance, individuals could retrieve things they have recently heard in the news, recall things about economics they learned in college or school, or think of the immediate consequences of a shock for themselves.

Supplementary Appendix Table A.8 uses responses to a question on which approaches households followed in making their forecasts (see Supplementary Appendix Figure A.5) to examine how thoughts of different channels vary across different sources of associations. Households that use knowledge of economics in their predictions are more likely to have associations of channels that are important in textbook models, moving their thoughts closer to those of experts. Respondents whose predictions are shaped by their personal situations are more likely to think of demand-side channels, such as changes in household spending, across the different vignettes. Finally, retrieving macroeconomic experiences or things heard in the news is significantly associated with having more thoughts of both supply-side and demand-side channels in the different vignettes.

Future research could provide more systematic evidence on how personal experiences or media exposure trigger different associations across contexts. Such an exercise could be guided by a model of memory that makes predictions on how experiences or things heard in the news affect associations across contexts.

5. Implications

In this section, we discuss the broader implications of our findings for understanding macroeconomic expectation formation and for modelling choices.

5.1. Understanding disagreement in expectations

One of the most well-documented empirical facts on the macroeconomic expectations of households, firms, and experts is that there is a substantial amount of disagreement about the future development of the economy (Mankiw et al., 2003; Coibion and Gorodnichenko, 2012; Dovern et al., 2012; Link et al., 2021; Giglio et al., 2021). This evidence is at odds with traditional models of full information and rational expectations. There are two broad views on the origins of disagreement in macroeconomic expectations. The most prominent explanation for belief disagreement brought forward by the theoretical literature is that agents have different information on the current state of the economy, which may be driven by infrequent updating of information sets (Mankiw and Reis, 2002; Reis, 2006) or by noise in private signals about the economy (Sims, 2003; Woodford, 2003). According to such explanations, if agents have the same information sets, they fully agree on how the economy responds to shocks. In contrast to this view, we document strong heterogeneity in unemployment and inflation forecasts even in a setting where all individuals observe the same shock and hold similar information about current realizations of macroeconomic variables. This finding is more in line with the alternative view that dispersion in expectations is (partially) due to individuals relying on different subjective models of the economy (Bray and Savin, 1986; Marcet and Sargent, 1989; Andrade et al., 2016; Molavi, 2019; Angeletos et al., 2020). Accordingly, economic agents evaluate the same news about the economy through the lens of their own model. Since there is strong heterogeneity in these models, disagreement about the future arises even when agents have comparable information about current realizations of macroeconomic variables and shocks.

5.2. Relation to existing theories featuring disagreement about the model

Can existing theories featuring disagreement about the model of the economy explain our findings? For instance, in theories of learning and model misspecification, agents may disagree about structural parameters of the economy, such as the persistence of inflation (Bray and Savin, 1986; Marcet and Sargent, 1989; Orphanides and Williams, 2005; Milani, 2007; Evans and Honkapohja, 2012; Bhandari, Borovicka and Ho, 2019; Molavi, 2019; Angeletos et al., 2020). In models of learning from experience (Malmendier and Nagel, 2016), individuals only use realizations of macroeconomic variables observed during their lifetimes to estimate the data-generating process, leading to disagreement in inflation expectations across cohorts even if everyone observes the same current realization. While heterogeneous beliefs about structural parameters from this literature find support in our results, these models cannot quantitatively account for the large heterogeneity in beliefs about the impact of the shocks we document, including disagreement even about the directional responses to shocks. More importantly, our priming evidence that changes in attention to different aspects of the problem affect forecasts is at odds with these models.²⁹

5.3. Associative recall and subjective models

Instead, our evidence is consistent with the idea that heterogeneity in macroeconomic expectations is partially due to associative recall of different propagation mechanisms of shocks (Gennaioli and Shleifer, 2010; Bordalo et al., 2020a). In this view, heterogeneity in the models individuals rely on is not fully stable, but depends on what is cued by the context and on individuals’ past experiences. Given our evidence, we believe that incorporating associative recall could be a fruitful avenue for macroeconomic modelling.

While formulating a model of associative recall as a driver of heterogeneity in subjective models is beyond the scope of our article, in Supplementary Appendix I, we compare the predictions of a canonical sticky-information model (Mankiw and Reis, 2002) with those of a basic framework that features heterogeneity in beliefs about the effects of macroeconomic shocks, being agnostic about the sources of heterogeneity in these beliefs. We calibrate both models to the empirical results from our vignettes and show that subjective models can produce either an under- or over-reaction of expectations relative to the true inflation response to shocks, and a rise in disagreement of comparable magnitude to that of the sticky-information model. However, unless some frictions in observation of shocks (i.e. sticky information) are assumed, the subjective models framework cannot explain empirical evidence on the persistence in forecast errors by Coibion and Gorodnichenko (2012). Hence, a subjective models approach does not fully substitute but rather complements information frictions.

5.4. Disagreement in more natural settings

How do our findings speak to disagreement in more natural settings? Our vignettes describe different hypothetical shocks with a small number of parameters, including previous realizations of the shock variable, unemployment and inflation, as well as the duration of the shock and the information structure. Real-world macroeconomic shocks likely feature a higher number of relevant parameters. Moreover, the simplifying common knowledge assumption about the duration of changes in government spending or taxes in our vignettes will rarely be fulfilled in the real world. These points suggest that disagreement about the effects of shocks in more natural settings, both among households and among experts, may be even larger than measured in our surveys.

6. Conclusion

Using samples of about 6,500 households representative of the US population and samples of about 1,500 experts, we use a new vignette-based approach to measure individuals’ subjective models of the economy and investigate their attentional foundations. We document substantial disagreement, even about the directional effects of macroeconomic shocks, both within and between samples of households and experts, in a setting where individuals have similar information about previous realizations of macroeconomic variables. Part of this disagreement seems to be due to selective recall of different propagation mechanisms of the shocks. While experts tend to retrieve textbook models, households often neglect channels that are commonly viewed as central to the transmission of a shock. Finally, we provide evidence on the role of associative memory in shaping heterogeneity in thoughts of propagation channels in two ways. First, contextual cues causally affect forecasts by changing the propagation channels individuals retrieve. Second, prior experiences are reflected in the retrieved propagation channels and forecasts. Our findings highlight associative memory as a source of disagreement in macroeconomic expectations.

We believe that our approach of measuring beliefs about the effects of shocks can be applied to many other questions in macroeconomics. For example, it could be fruitful to apply our approach to other structural shocks that are commonly found to be quantitatively important, such as total factor productivity or sentiment shocks. In addition, we believe that our approach of measuring what is on top of people’s mind while they make their predictions is a widely applicable tool that could help to better understand how associations drive belief formation.

Our findings also have several implications for policymakers.³⁰ First, in recent years policy institutions have made efforts to reach broader groups with their communication to increase the effectiveness of fiscal and monetary policy (Haldane and McMahon, 2018). Such efforts could be less fruitful if households disagree about the direction in which policy shocks affect macroeconomic outcomes. Second, our evidence suggests that the way a policy is communicated—for example, whether demand-side implications rather than supply-side implications are emphasized—could substantially alter its effect on individuals’ expectations. Finally, our finding of substantial heterogeneity in households’ beliefs about macroeconomic relationships implies a large degree of variation in the effectiveness of monetary policy and fiscal policy in shifting expectations and behaviour for different sub-populations of interest.

The editor in charge of this paper was Nicola Gennaioli.

Supplementary Data

Supplementary data are available at Review of Economic Studies online. And the replication packages are available at https://dx.doi.org/10.5281/zenodo.5744016.

Acknowledgements

The views expressed in this article are those of the authors and do not necessarily represent those of the IMF or its board of directors. We thank the editor (Nicola Gennaioli) and three anonymous referees for extremely useful suggestions. We also thank Rudi Bachmann, Carola Binder, Pedro Bordalo, Benjamin Born, Felix Chopra, Olivier Coibion, Francesco D’Acunto, Stefano DellaVigna, Thomas Dohmen, Armin Falk, Andreas Fuster, Xavier Gabaix, Dimitris Georgarakos, Yuriy Gorodnichenko, Thomas Graeber, Michael Haliassos, Lukas Hensel, Chi Hyun Kim, Gizem Kosar, Michael Kosfeld, Fabian Krüger, Matt Lowe, Michael McMahon, Valerie Ramey, Aakaash Rao, Sonja Settele, Andrei Shleifer, Uwe Sunde, Johannes Stroebel, Giorgio Topa, Egon Tripodi, Michael Weber, Mirko Wiederholt, Basit Zafar, Florian Zimmermann, as well as participants at various conferences and seminars. We thank Dorine Boumans, Johanna Garnitz, Andreas Peichl, and the ifo Institute for including our module in the World Economic Survey and Valentin Reich for help with the analysis. We are grateful to the data services of the IDSC of IZA. We thank the Joachim Herz Foundation as well as the Fritz-Thyssen Foundation for financial support. Funding by the Deutsche Forschungsgemeinschaft (DFG) through CRC TR 224 (Project A01) is gratefully acknowledged. The activities of the Center for Economic Behavior and Inequality (CEBI) are financed by the Danish National Research Foundation, Grant DNRF134. Support from the Danish Finance Institute (DFI) is gratefully acknowledged. Roth: Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy – EXC 2126/1-390838866. Hrishikesh Iyengar, Apoorv Kanoongo, Melisa Kurtis, Anna Lane, and David Zeimentz provided excellent research assistance. We received ethics approval from Goethe University Frankfurt and the University of Warwick. The experimental instructions can be found at the following link: https://osf.io/6mxaz/.

Data Availability Statement

The data underlying this article are available in Zenodo at https://doi.org/10.5281/zenodo.5744016 (Andre, Pizzinelli, Roth and Wohlfart, 2022), with the exception of Wave 2 of the experts survey, which is stored at and owned by the ifo Institute (ifo Institute, 2019). Wave 2 can be accessed remotely upon request to the ifo Institute.

Footnotes

More generally, we contribute to a growing literature studying the formation of macroeconomic expectations of experts, households and firms, and the role of these expectations in economic and financial decisions (Fuster, Laibson and Mendel, 2010; Malmendier and Nagel, 2011; Coibion and Gorodnichenko, 2012; Fuster, Hebert and Laibson, 2012; Armantier, Bruine de Bruin, Topa, Klaauw and Zafar, 2015; Bachmann, Berg and Sims 2015; Coibion and Gorodnichenko, 2015a,b; Malmendier and Nagel, 2016; Binder and Rodrigue, 2018; Bordalo, Gennaioli and Shleifer, 2018; Acosta and Afrouzi, 2019; Afrouzi, 2019; Kamdar, 2019; Vellekoop and Wiederholt 2019; Binder and Makridis, 2022; Bordalo, Gennaioli, Ma and Shleifer, 2020c; Goldfayn-Frank and Wohlfart 2020; Bachmann, Carstensen, Lautenbacher and Schneider, 2021; Roth, Settele and Wohlfart, 2022).

For details on the conferences considered, see Supplementary Appendix J. We also invited a few Ph.D. students, experts from several policy institutions, as well as several experts working in the broader areas of expectation formation and macroeconomic forecasting.

A series of papers uses hypothetical vignettes to study belief formation in contexts such as human capital (Wiswall and Zafar, 2017; Delavande and Zafar, 2019) or consumption behaviour (Christelis, Georgarakos, Jappelli, Pistaferri and Van Rooij, 2019; Fuster, Kaplan and Zafar, 2020a).

The median household respondent spends about 14 min to complete the survey (10th percentile: 7–8 min, 90th percentile: 27–33 min, depending on the wave). The median expert in wave 1 needs 5 min to complete the shorter expert survey (10th percentile: 3 min, 90th percentile: 14 min). The survey completion rates are close to 80|$\%$|⁠. See Supplementary Appendix Table A.4 for further details. Supplementary Appendix Figure A.1 summarizes the structure of both surveys. The full set of experimental instructions for Wave 1 and Wave 2 of the surveys can be found under the following link: https://osf.io/6mxaz/.

In Wave 2 of the expert survey, it was not feasible to randomize the order of vignettes. Instead, the vignettes were ordered as follows: (1) income tax shock, (2) federal funds rate, (3) government spending shock, and (4) oil supply shock. Respondents received two randomly selected vignettes.

In Supplementary Appendix Section D.1, we compare predictions across the rise and fall scenarios. Asymmetries occur more often for households than for experts but are mostly minor.

Finally, to account for potential order effects, we cross-randomize whether respondents first receive the question on the inflation rate or the question on the unemployment rate. For each participant, the order of the inflation and unemployment questions is identical across all scenarios.

The last sentence of the vignette was not included in Wave 2.

Part of the heterogeneity in forecasts in our vignettes could reflect measurement error. However, much of our descriptive analysis in Section 3 focuses on directional predictions, for which measurement error should be strongly mitigated. In addition, in our analysis of the role of thoughts of different propagation channels in Section 4, forecasts are used as dependent variables, so (classical) measurement error should not bias coefficient estimates.

10.

We do not fix beliefs about the duration of the change in interest rates under the monetary policy shock, since the interest rate should react endogenously to changes in inflation and unemployment in response to the shock through the Taylor rule.

11.

We found no established benchmark estimate for the inflation response to the income tax shock.

12.

Given the large sample size, even minor differences in households’ and experts’ directional predictions are statistically different (⁠|$p<0.01$|⁠, |$\chi^2$|-tests). Moreover, disagreement is always significantly larger among households than among experts (see Supplementary Appendix Table A.5). We also confirm the robustness of our results in several checks. Supplementary Appendix D.3 discusses order effects and the effect of incentives on predictions of households. Supplementary Appendix Figure A.2 showcases the stability of the expert results in different subsamples of experts.

13.

Bordalo et al. (2020c) propose a framework to study over- and under-reaction of individual and consensus forecasts to news.

14.

In Supplementary Appendix Section H.3, we show that only a very small fraction of households seem to misperceive the interest rate hike as the Fed’s endogenous reaction to a higher inflation outlook.

15.

These patterns also become apparent if we study the predictions of the joint response of inflation and unemployment (see Supplementary Appendix D.2.1). For instance, 55|$\%$| of experts express the conventional view that the interest rate shock increases unemployment and decreases inflation, compared to 11|$\%$| of households.

16.

We replicate our main results for both the directional and the quantitative predictions (see Supplementary Appendix Figures A.3 and A.4). This highlights the robustness of our findings across time and to changes in the design, such as the prediction scales or the simultaneous measurement of thoughts.

17.

The order of response options is randomized across individuals to address potential order effects.

18.

Supplementary Appendix Figure A.6 shows that thoughts of the different propagation channels are very similar across different subgroups in the expert sample. For instance, experts that are PhD students think of very similar channels as non-PhD student experts.

19.

These findings are in line with participants’ responses to a question about the approach they pursued in their forecasts. Supplementary Appendix Figure A.5 shows that 88|$\%$| of experts report that they drew on their knowledge of economics compared to only 29|$\%$| of households. This is consistent with the notion that experts are more likely to think about the shocks through the lens of textbook models. In contrast to experts, households are relatively more likely to rely on their memories of past economic events and their gut feeling when making their predictions.

20.

Our surveys are not tailored to study the drivers of heterogeneity in associations within the expert sample due to space constraints.

21.

In Supplementary Appendix F, we demonstrate robustness of these correlations to using the hand-coded measures of thoughts based on the open-text data.

22.

Given the nature of attention, focusing on one macroeconomic variable (inflation) gives us more control over the respondents’ thoughts while they make their predictions.

23.

Supplementary Appendix Section G provides an overview of the prediction screens across all three treatment arms.

24.

For this analysis to be valid, beliefs about the directions in which costs and demand change need to be balanced between the treatment and control groups, which is the case in our data.

25.

In Supplementary Appendix F, we show similar patterns using measures of thoughts based on hand-coding of the open-text data. We do not use structured measures as those were not included in this data collection.

26.

Our respondents in this sample are on average somewhat older and more educated compared to our other data collections (see Supplementary Appendix Table A.1).

27.

Supplementary Appendix Table A.7 shows that we obtain similar results using alternative measures of personal employment experience with government suppliers.

28.

In Supplementary Appendix F, we show similar patterns for the effect of experiences on associations using measures of thoughts based on hand-coding of the open-text data.

29.

A literature in behavioural macroeconomics has proposed |$k$|-level thinking (Farhi and Werning, 2019), a lack of common knowledge (Angeletos and Lian, 2017), or myopia (Gabaix, 2020) in macroeconomic expectation formation to explain muted responses of output and consumption to shocks. These models mostly do not directly speak to disagreement in expectations. Moreover, in models of diagnostic expectations, disagreement arises from economic agents’ use of the representativeness heuristic to learn from noisy private signals about the economy (Bordalo et al., 2018, 2020c). In our survey, we provide agents with identical information.

30.

The role of macroeconomic expectations in households’ spending decisions is still being debated in the literature. Some studies find a positive association of inflation expectations with consumption (D’Acunto, Hoang and Weber, 2021a), while others document a muted (Bachmann et al., 2015; Galashin, Kanz and Perez-Truglia, 2021) or negative relationship (Coibion et al., 2019). The evidence on the role of expectations about aggregate unemployment and growth is more limited, but there is some evidence suggesting a role in households’ spending decisions (Roth and Wohlfart, 2020; Coibion, Georgarakos, Gorodnichenko, Kenny and Weber, 2021).

REFERENCES

ACOSTA,

and

AFROUZI,

(

2019

), “

Rationally Confused: On the Aggregate Implications of Information Provision Policies

” (

Working Paper

AFROUZI,

(

2019

), “

Strategic Inattention, Inflation Dynamics and the Non-Neutrality of Money

” (

Working Paper

ANDRADE,

CRUMP,

R. K.

EUSEPI,

and

MOENCH,

(

2016

), “

Fundamental Disagreement

”,

Journal of Monetary Economics

106

–

128

ANDRE,

and

FALK,

(

2021

), “

What’s Worth Knowing? Economists’ Opinions about Economics

” (

Working Paper

ANDRE,

PIZZINELLI,

ROTH,

and

WOHLFART,

(

2022

), “

Replication Data for: Subjective Models of the Macroeconomy: Evidence From Experts and Representative Samples

”,

Review of Economic Studies

doi:

10.5281/zenodo.5744016

ANGELETOS,

G.-M.

and

LIAN,

(

2017

), “

Dampening General Equilibrium: From Micro to Macro

” (

Working Paper

ANGELETOS,

G.-M.

and

LIAN,

(

2018

), “

Forward Guidance without Common Knowledge

”,

American Economic Review

108

2477

–

2512

ANGELETOS,

G.-M.

HUO,

and

SASTRY,

K. A.

(

2020

), “

Imperfect Macroeconomic Expectations: Evidence and Theory

”, in

Eichenbaum,

and

Hurst,

(eds)

NBER Macroeconomics Annual 2020

vol. 35

NBER Chapters

(

National Bureau of Economic Research, Inc

ARMANTIER,

NELSON,

TOPA,

VAN DER KLAAUW,

and

ZAFAR,

(

2016

), “

The Price Is Right: Updating Inflation Expectations in a Randomized Price Information Experiment

”,

Review of Economics and Statistics

503

–

523

ARMANTIER,

BRUINE DE BRUIN,

TOPA,

KLAAUW,

and

ZAFAR,

(

2015

), “

Inflation Expectations and Behavior: Do Survey Respondents Act on their Beliefs?

”,

International Economic Review

505

–

536

ARMONA,

FUSTER,

and

ZAFAR,

(

2018

), “

Home Price Expectations and Behaviour: Evidence from a Randomized Information Experiment

”,

The Review of Economic Studies

1371

–

1410

AUERBACH,

A. J.

GORODNICHENKO,

and

MURPHY,

(

2020

), “

Local Fiscal Multipliers and Fiscal Spillovers in the United States

”,

IMF Economic Review

195

–

229

BACHMANN,

CARSTENSEN,

LAUTENBACHER,

and

SCHNEIDER,

(

2021

), “

Uncertainty and Change: Survey Evidence of Firms’ Subjective Beliefs

” (

Working Paper

BACHMANN,

BERG,

T. O.

and

SIMS,

E. R.

(

2015

), “

Inflation Expectations and Readiness to Spend: Cross-Sectional Evidence

”,

American Economic Journal: Economic Policy

–

BALL,

GREGORY MANKIW,

and

REIS,

(

2005

), “

Monetary Policy for Inattentive Economies

”,

Journal of Monetary Economics

703

–

725

BARTH,

M. J. III

and

RAMEY,

(

2002

), “

The Cost Channel of Monetary Transmission

”, in

Bernanke,

Rogoff,

(eds),

NBER Macroeconomics Annual 2001

(

Cambridge, MA

MIT Press

)

199

–

240

BASSO,

H. S.

and

RACHEDI,

(

2019

), “

The Young, the Old, and the Government: Demographics and Fiscal Multipliers

” (

Working Paper

BHANDARI,

BOROVICKA,

and

HO,

(

2019

), “

Survey Data and Subjective Beliefs in Business Cycle Models

” (

Federal Reserve Bank of Richmon Working Paper Series, (19-14)

BINDER,

and

RODRIGUE,

(

2018

), “

Household Informedness and Long-Run Inflation Expectations: Experimental Evidence

”,

Southern Economic Journal

580

–

598

BINDER,

and

MAKRIDIS,

(

2022

), “

Stuck in the Seventies: Gas Prices and Macroeconomic Expectations

”,

Review of Economics and Statistics

104

293

–

305

BLANCHARD,

and

GALÍ,

(

2010

), “

The Macroeconomic Effects of Oil Price Shocks: Why Are the 2000s so Different from the 1970s?

”, in

Galí,

and

Gertler,

M. J.

(eds)

International Dimensions of Monetary Policy

(

Chicago

University of Chicago Press

BORDALO,

COFFMAN,

GENNAIOLI,

and

SHLEIFER,

(

2016

), “

Stereotypes

”,

The Quarterly Journal Economics

131

1753

–

1794

BORDALO,

COFFMAN,

GENNAIOLI,

SCHWERTER,

and

SHLEIFER,

(

2020a

), “

Memory and Representativeness

”,

Psychological Review

128

BORDALO,

GENNAIOLI,

and

SHLEIFER,

(

2020b

), “

Memory, Attention, and Choice

”,

The Quarterly Journal of Economics

135

1399

–

1442

BORDALO,

GENNAIOLI,

and

SHLEIFER,

(

2018

), “

Diagnostic Expectations and Credit Cycles

”,

The Journal of Finance

199

–

227

BORDALO,

GENNAIOLI,

MA,

and

SHLEIFER,

(

2020c

), “

Overreaction in Macroeconomic Expectations

”,

American Economic Review

110

2748

–

BOUMANS,

and

GARNITZ,

(

2017

), “

ifo World Economic Survey Database—An International Economic Expert Survey

”,

Jahrbücher für Nationalökonomie und Statistik

237

–

BRAY,

M. M.

and

SAVIN,

N. E.

(

1986

), “

Rational Expectations Equilibria, Learning, and Model Specification

”,

Econometrica

1129

–

1160

CARROLL,

C. D.

(

2003

), “

Macroeconomic Expectations of Households and Professional Forecasters

”,

The Quarterly Journal of Economics

118

269

–

298

CARVALHO,

and

NECHIO,

(

2014

), “

Do People Understand Monetary Policy?

”,

Journal of Monetary Economics

108

–

123

CAVALLO,

CRUCES,

and

PEREZ-TRUGLIA,

(

2017

), “

Inflation Expectations, Learning and Supermarket Prices: Evidence from Field Experiments

”,

American Economic Journal: Macroeconomics

–

CHRISTELIS,

GEORGARAKOS,

JAPPELLI,

PISTAFERRI,

and

VAN ROOIJ,

(

2019

), “

Asymmetric Consumption Effects of Transitory Income Shocks

”,

The Economic Journal

129

2322

–

2341

COIBION,

and

GORODNICHENKO,

(

2012

), “

What Can Survey Forecasts Tell us about Information Rigidities?

”,

Journal of Political Economy

120

116

–

159

COIBION,

and

GORODNICHENKO,

(

2015

), “

Information Rigidity and the Expectations Formation Process: A Simple Framework and New Facts

”,

The American Economic Review

105

2644

–

2678

COIBION,

and

GORODNICHENKO,

(

2015

), “

Is the Phillips Curve Alive and Well after All? Inflation Expectations and the Missing Disinflation

”,

American Economic Journal: Macroeconomics

197

–

232

COIBION,

GEORGARAKOS,

GORODNICHENKO,

and

VAN ROOIJ,

(

2019

), “

How Does Consumption Respond to News about Inflation? Field Evidence from a Randomized Control Trial

” (

National Bureau of Economic Research Working Paper

COIBION,

GEORGARAKOS,

GORODNICHENKO,

KENNY,

and

WEBER,

(

2021

), “

The Effect of Macroeconomic Uncertainty on Household Spending

” (

Working Paper

COIBION,

GORODNICHENKO,

and

KUMAR,

(

2018

), “

How Do Firms Form Their Expectations? New Survey Evidence

”,

American Economic Review

108

2671

–

2713

COIBION,

GORODNICHENKO,

, and

ROPELE,

(

2020

), “

Inflation Expectations and Firm Decisions: New Causal Evidence

”,

The Quarterly Journal of Economics

135

165

–

219

D’ACUNTO,

HOANG,

and

WEBER,

(

2021a

), “

Managing Households’ Expectations with Simple Economic Policies

”,

Review of Financial Studies

D’ACUNTO,

HOANG,

PALOVIITA,

and

WEBER,

(

2021b

), “

IQ, Expectations, and Choice

”,

Review of Economic Studies

DAS,

KUHNEN,

C. M.

and

NAGEL,

(

2020

), “

Socioeconomic Status and Macroeconomic Expectations

”,

The Review of Financial Studies

395

–

432

DE QUIDT,

HAUSHOFER,

and

ROTH,

(

2018

), “

Measuring and Bounding Experimenter Demand

”,

American Economic Review

108

3266

–

3302

DELAVANDE,

and

ZAFAR,

(

2019

), “

University Choice: The Role of Expected Earnings, Nonpecuniary Outcomes, and Financial Constraints

”,

Journal of Political Economy

127

2343

–

2393

DELLAVIGNA,

and

POPE,

(

2018

), “

What Motivates Effort? Evidence and Expert Forecasts

”,

Review of Economic Studies

1029

–

1069

DOVERN,

FRITSCHE,

and

SLACALEK,

(

2012

), “

Disagreement Among Forecasters in G7 Countries

”,

The Review of Economics and Statistics

1081

–

1096

DRÄGER,

LAMLA,

M. J.

and

PFAJFAR,

(

2016

), “

Are Survey Expectations Theory-consistent? The Role of Central Bank Communication and News

”,

European Economic Review

–

111

ENKE,

SCHWERTER,

and

ZIMMERMANN,

(

2020

), “

Associative Memory and Belief Formation

” (

Working Paper

EVANS,

G. W.

and

HONKAPOHJA,

(

2012

)

Learning and Expectations in Macroeconomics

(

Princeton and Oxford

Princeton University Press

FARHI,

and

WERNING,

(

2019

), “

Monetary Policy, Bounded Rationality, and Incomplete Markets

”,

American Economic Review

109

3887

–

3928

FUSTER,

HEBERT,

and

LAIBSON,

(

2012

), “

Natural Expectations, Macroeconomic Dynamics, and Asset Pricing

”,

NBER Macroeconomics Annual

–

FUSTER,

LAIBSON,

and

MENDEL,

(

2010

), “

Natural Expectations and Macroeconomic Fluctuations

”,

Journal of Economic Perspectives

–

FUSTER,

KAPLAN,

and

ZAFAR,

(

2020a

), “

What Would You Do With $500? Spending Responses to Gains, Losses, News and Loans

”,

Review of Economic Studies

doi:

10.1093/restud/rdaa076

FUSTER,

PEREZ-TRUGLIA,

WIEDERHOLT,

and

ZAFAR,

(

2020b

), “

Expectations with Endogenous Information Acquisition: An Experimental Investigation

”,

The Review of Economics and Statistics,

–

GABAIX,

(

2019

), “

Behavioral Inattention

”, in

Bernheim,

B. D.

DellaVigna,

and

Laibson,

(eds)

Handbook of Behavioral Economics: Applications and Foundations 1

, Vol. 2 (Elsevier)

261

–

343

GABAIX,

, (

2020

), “

A Behavioral New Keynesian Model

”,

American Economic Review

110

2271

–

2327

GALASHIN,

KANZ,

and

PEREZ-TRUGLIA,

(

2021

), “

Macroeconomic Expectations and Credit Card Spending

” (

Working Paper

GENNAIOLI,

and

SHLEIFER,

(

2010

), “

What Comes to Mind

”,

The Quarterly Journal of Economics

125

1399

–

1433

GIGERENZER,

and

TODD,

P. M.

(

1999

Simple Heuristics that Make Us Smart

(

New York

Oxford University Press

GIGLIO,

MAGGIORI,

STROEBEL,

and

UTKUS,

(

2021

), “

Five Facts about Beliefs and Portfolios

”,

American Economic Review

111

1481

–

1522

GOLDFAYN-FRANK,

and

WOHLFART,

(

2020

), “

Expectation Formation in a New Environment: Evidence from the German Reunification

”,

Journal of Monetary Economics

115

301

–

320

GORDON,

and

DAHL,

G. B.

(

2013

), “

Views among Economists: Professional Consensus or Point-Counterpoint?

”,

American Economic Review

103

629

–

635

GRAEBER,

(

2021

), “

Inattentive Inference

” (

Working Paper

. https://www.aeaweb.org/articles?id=10.1257/jel.20211658&from=f.

HAALAND,

ROTH,

and

WOHLFART,

(

2021

), “

Designing Information Provision Experiments

”,

Journal of Economic Literature

HALDANE,

and

MCMAHON,

(

2018

), “

Central Bank Communications and the General Public

”, in

Johnson,

W. R.

and

Markel,

(eds)

AEA Papers and Proceedings

, Vol. 108.

578

–

583

IFO

INSTITUTE

(

2019

), “

ifo World Economic Survey (July 2019 Wave)

”.

KAHANA,

M. J.

(

2012

Foundations of Human Memory

(

Oxford

Oxford University Press

KAMDAR,

(

2019

), “

The Inattentive Consumer: Sentiment and Expectations

” (Meeting Paper No. 647, SED).

KUCHLER,

and

ZAFAR,

(

2019

), “

Personal Experiences and Expectations about Aggregate Outcomes

”,

The Journal of Finance

2491

–

2542

LACETERA,

POPE,

D. G.

and

SYDNOR,

J. R.

(

2012

), “

Heuristic Thinking and Limited Attention in the Car Market

”,

American Economic Review

102

2206

–

2236

LEISER,

and

AROCH,

(

2009

), “

Lay Understanding of Macroeconomic Causation: The Good-begets-good Heuristic

”,

Applied Psychology

370

–

384

LEISER,

and

KRILL,

(

2017

), “

How Laypeople Understand the Economy

”, in

Ranyard,

(ed)

Economic Psychology

(

Hoboken, NJ

Wiley-Blackwell

)

139

–

154

LINK,

PEICHL,

ROTH,

and

WOHLFART,

(

2021

), “

Information Frictions Among Firms and Households

” (

CESifo Working Paper Series 8969, CESifo

MAĆKOWIAK,

and

WIEDERHOLT,

(

2015

), “

Business Cycle Dynamics under Rational Inattention

”,

The Review of Economic Studies

1502

–

1532

MALMENDIER,

and

NAGEL,

(

2011

), “

Depression Babies: Do Macroeconomic Experiences Affect Risk-taking?

”,

The Quarterly Journal of Economics

126

373

–

416

MALMENDIER,

and

NAGEL,

(

2016

), “

Learning from Inflation Experiences

”,

The Quarterly Journal of Economics

131

–

MALMENDIER,

NAGEL,

and

YAN,

(

2021

), “

The Making of Hawks and Doves

”,

Journal of Monetary Economics

117

–

MANKIW,

N. G.

and

REIS,

(

2002

), “

Sticky Information Versus Sticky Prices: A Proposal to Replace the New Keynesian Phillips Curve

”,

The Quarterly Journal of Economics

117

1295

–

1328

MANKIW,

N. G.

REIS,

and

WOLFERS,

(

2003

), “

Disagreement About Inflation Expectations

”,

NBER Macroeconomics Annual

209

–

248

MARCET,

and

SARGENT,

T. J.

(

1989

), “

Convergence of Least Squares Learning Mechanisms in Self-referential Linear Stochastic Models

”,

Journal of Economic Theory

337

–

368

MILANI,

(

2007

), “

Expectations, Learning and Macroeconomic Persistence

”,

Journal of Monetary Economics

2065

–

2082

MOLAVI,

(

2019

), “

Macroeconomics with Learning and Misspecification: A General Theory and Applications

” (

Working Paper

NAKAMURA,

and

STEINSSON,

(

2014

), “

Fiscal Stimulus in a Monetary Union: Evidence from US Regions

”,

American Economic Review

104

753

–

ORPHANIDES,

and

WILLIAMS,

J. C.,

(

2005

), “

The Decline of Activist Stabilization Policy: Natural Rate Misperceptions, Learning, and Expectations

”,

Journal of Economic Dynamics and Control

1927

–

1950

PACIELLO,

and

WIEDERHOLT,

(

2014

), “

Exogenous Information, Endogenous Information, and Optimal Monetary Policy

”,

Review of Economic Studies

356

–

388

REIS,

(

2006

), “

Inattentive Consumers

”,

Journal of Monetary Economics

1761

–

1800

ROTH,

and

WOHLFART,

(

2020

), “

How Do Expectations About the Macroeconomy Affect Personal Expectations and Behavior?

”,

Review of Economics and Statistics

102

731

–

748

ROTH,

SETTELE,

and

WOHLFART,

(

2022

), “

Risk Exposure and Acquisition of Macroeconomic Information

”,

American Economic Review: Insights

–

SAPIENZA,

and

ZINGALES,

(

2013

), “

Economic Experts versus Average Americans

”,

American Economic Review

103

636

–

642

SIMS,

C. A.

(

2003

), “

Implications of Rational Inattention

”,

Journal of Monetary Economics

665

–

690

STANTCHEVA,

(

2020

), “

Understanding Tax Policy: How Do People Reason?

” (

Technical Report, National Bureau of Economic Research

TVERSKY,

and

KAHNEMAN,

(

1973

), “

Availability: A Heuristic for Judging Frequency and Probability

”,

Cognitive Psychology

207

–

232

VELLEKOOP,

and

WIEDERHOLT,

(

2019

), “

Inflation Expectations and Choices of Households

” (

Working Paper

WISWALL,

and

ZAFAR,

(

2017

), “

Preference for the Workplace, Investment in Human Capital, and Gender

”,

The Quarterly Journal of Economics

133

457

–

507

WOODFORD,

(

2003

), “

Imperfect Common Knowledge and The Effects of Monetary Policy

”, in

Aghion,

Frydman,

Stiglitz,

J. E.

, and

Woodford,

(eds)

Knowledge, Information, and Expectations in Modern Macroeconomics: In Honor of Edmund S. Phelps

(

Princeton, NJ

Princeton University Press