A note on the impact of news on US household inflation expectations
Ben Zhe Wang, Jeffrey Sheen, Stefan Trück, Shih-Kang Chao, Wolfgang Karl Härdle
AA note on the impact of news on US householdinflation expectations (cid:73)
Ben Zhe Wang a, ∗ , Jeffrey Sheen a , Stefan Tr¨uck a , Shih-Kang Chao c , WolfgangKarl H¨ardle b a Macquarie University b Humboldt-Universit¨at zu Berlin and Sim Kee Boon Institute for Financial Economics,Singapore Management University c University of Missouri (cid:73)
We thank the editor and the two anonymous referees for their constructive comments.This is a post-peer-review, pre-copyedit version of an article published in Macroeconomic Dynamics. Thefinal authenticated version is available online at: http://dx.doi.org/10.1017/S1365100518000482 ∗ Corresponding author, Email: [email protected]
September 25, 2020 a r X i v : . [ q -f i n . GN ] S e p unning title: News Impact on US inflation expectations
Corresponding author:
Name: Ben Zhe WangEmail: [email protected]: +61 2 98508500 Address: 4ER 432, Department of Economics, Macquarie University,North Ryde, 2109, NSW, Australia 2 bstract
Monthly disaggregated US data from 1978 to 2016 reveals that exposure to news on inflationand monetary policy helps to explain inflation expectations. This remains true when control-ling for household personal characteristics, perceptions of government policy effectiveness,future interest rates and unemployment expectations, and sentiment. We find an asymmetricimpact of news on inflation and monetary policy after 1983, with news on rising inflation andeasier monetary policy having a stronger effect in comparison to news on lowering inflationand tightening monetary policy. Our results indicate the impact on inflation expectationsof monetary policy news manifested through consumer sentiment during the lower boundperiod.
Keywords
Inflation expectations; news impact; monetary policy signalling; unconventional monetarypolicy
JEL Classification:
C81, D83, D84, E31 3 . Introduction
Inflation expectations play a major role in modern macroeconomics, with rational expecta-tions ubiquitous as the modelling device for a representative agent. However, the literatureprovides both theoretical models and empirical observations that can explain how differenteconomic agents form inflation expectations and why they might disagree on their forecasts.For example, Mankiw et al. (2004) document a considerable degree of disagreement in sur-veys of US inflation expectations. This disagreement is time-varying and exhibits covariationwith macroeconomic variables. Mankiw and Reis (2002) construct a formal model and at-tribute disagreements to information rigidity. The idea is that the dissemination of newinformation occurs gradually between people.One way households acquire information is through media reports, which we refer to as‘news’ in this paper. News can directly impact on household inflation expectations by di-rectly informing the consumer about the possible future path of inflation (e.g. throughexpert forecasts), or indirectly through impacting on household perceptions of current infla-tion. Lamla and Maag (2012) find that the disagreement in household inflation expectationsin Europe depends on the reporting intensity and the ‘tone’ of the news about inflation,while Dr¨ager (2015) finds that the media has a small but significant impact on inflation ex-pectations in Sweden. Carroll (2003) uses an epidemiology model and finds that professionalforecasts as a proxy for news have predictive power for household forecasts in the US.All the aforementioned studies use aggregated news measures obtained from a separate sourcethan that for the measure of inflation expectations. One drawback with this approach is thatthe news measures do not necessary reflect the news heard by the individual household, andthus may not necessarily be attributable to household inflation expectation formation. Inthis paper, we use the Michigan Survey of Consumers from 1978 to 2016, which allows us to4xamine the direct impact of news on individual households.There is an emerging literature on investigating the effect of perceived news using the Michi-gan Survey of Consumers data. For example, utilising the panel structure of the MichiganSurvey of Consumer data, , Pfajfar and Santoro (2013) test the epidemiology model of Car-roll (2003) using an aggregate measure of news and household perceived news, and find atbest weak support for the epidemiology model. Although hearing inflation news increases theprobability of updating inflation expectations, it enlarges the forecast gap between house-holds’ inflation expectation and those of professional forecasts, as well as the gap betweenhouseholds’ inflation expectation and actual realized inflation. Similarly, Dr¨ager and Lamla(2017) find the hearing of news on inflation increases the chance of households updating theirinflation expectations, irrespective of whether it is favourable or unfavourable news. Pfajfarand Santoro (2009) find households with different socioeconomic background form inflationexpectation differently in response to inflation news, and they exhibit different degrees ofinformation stickiness when updating their inflation expectations. In addition, Ehrmannet al. (2017) find households tend to forecast inflation higher if they have financial difficul-ties or are pessimistic about major purchases, income developments or the unemploymentrate—however, their bias shrinks by more than the average household in response to inflationnews. Lahiri and Zhao (2016) also find consumer sentiment responds to perceived news andZhang et al. (2016) find stock markets react to news through its impact on sentiment.In this paper, we contribute to the literature by considering monetary policy news alongwith inflation news, and evaluate whether favourable or unfavourable news have asymmetric Each month, about 40 per cent of the households are randomly chosen to be re-interviewed six monthsafter their initial interview. This rotating panel feature is useful for analysing how consumers update theirinflation expectations. Our results are robust after controllingfor household demographics, their perception of the effectiveness of government policies,their expectations of future interest and unemployment rates, and their sentiment. We alsofind an asymmetric impact of news on rising inflation (contractionary monetary policy) com-pared to news on falling inflation (expansionary monetary policy). Our results indicate thatthis asymmetric impact started to become significantly stronger in the early 1990s. Wefind that the absolute impact of news on higher inflation became statistically greater thannews on lower inflation after 1991, while after 1999 news on easing monetary policy had asignificantly greater impact on inflation expectations than contractionary monetary policy.Finally, during the zero lower bound period after 2008, news about monetary policy becomesan imperfect signal for inflation expectations formation. This signal manifested through con-sumer sentiment, which implies central banks should pay attention to consumer sentimentwhen communicating monetary policies.The subsequent paper is organized as follows. Section 2 describes the applied model and Our paper is also related to a growing theoretical literature that shows monetary policy could have realeffects even in the absence of nominal rigidities, if we are willing to not assume rational expectations. Thetransmission channels may arise from information rigidities (Woodford, 2001), rational inattention (Adam,2007) and potential signalling effects (Melosi, 2017).
2. The model and data
Since 1978, around 500 adults in households have been surveyed each month on their one-year-ahead inflation expectations by the University of Michigan (Survey of Consumers). Thesurvey asks respondents to provide a numerical answer to the following question:By about what percent do you expect prices to go (up/down) on the average,during the next 12 months?The data exhibits a considerable degree of disagreement among these US households inany month. In addition to inflation expectations, the survey also asks respondents whetherthey have heard news about current economic conditions, and also for their evaluations ofcurrent and expected future paths of the economy as well as their personal financial situation.We test if news plays a role in explaining household inflation expectations by estimatingequation (1) using pooled ordinary least squares (OLS): π eit = α + T D t θ (cid:48) + φ π N πit + φ r N rit + C it γ (cid:48) + (cid:15) it (1)where π eit is the one-year ahead inflation expectation of household i at time t , α is a con-stant, and T D t collects monthly time dummies that are invariant among households at agiven month. Since our focus is on investigating the impact of news on individual inflationexpectations, we include these time dummies to account for aggregate developments of theeconomy in each month that might have an impact on household inflation expectations.7 πit and N rit indicate whether household i has been exposed to any news of inflation andmonetary policy, respectively. The survey asks respondents to indicate whether they haveheard news of changes in business conditions:During the last few months, have you heard of any favorable or unfavorablechanges in business conditions? What did you hear?The respondents may indicate they have heard news on rising or falling prices, which we useto approximate inflation news, and lower or higher interest rates or easier or tighter creditconditions, which we use to approximate monetary policy news.If no particular news has been heard, the respective variable has a value of 0; N πit is set to avalue of 1 if household i has been exposed to news about higher inflation, and − N rit takes on a value of 1 if household i has heard news about higher interest rates or tighter credit conditions, mostly associatedwith tighter monetary policy, and − φ π and φ r measure the impact ofinflation news and monetary policy news on household inflation expectations, which is a keyfocus of this paper. C it ∈ [ D it , P it , E it , CS it ] represents control variables for the characteristics of household i .Hereby, D it denotes economic and demographic variables for respondent i , including log in-come, age, gender (1 for a female), and level of education (measured on a scale between 1to 6, with 6 indicating the highest level of education). P it denotes household perceptionson the effectiveness of government policies in managing inflation or unemployment, taking A value of 1 indicates Grade 0-8 without high school diploma; 2 indicates grade 9-12 without high schooldiploma; 3 indicates grade 0-12 with high school diploma; 4 indicates grades 13-17 without a college degree;5 indicates grade 13-16 with a degree and 6 indicates grade 17 with a college degree. − E it collects household expectations on the future course of interest ratesand unemployment over the next year, with 1 indicating that household i expects the futurerespective rate will increase, 0 that it stays the same, and -1 indicates that the householdexpects the rate to be reduced. CS it is the measure from the Michigan Survey for consumersentiment. It is constructed from five qualitative questions about the household’s currentand future expected personal financial situation, its current buying attitude regarding largeticket household items, and its expectation of short and medium term business conditions. Appendix A shows the pairwise correlations among the explanatory variables. Apart fromlog income and education, and the perception of government policies and consumer senti-ment, all explanatory variables are not highly correlated.Our monthly sample starts from January 1978 and ends in February 2016, containing 208,777individual records in total. The cross-sectional inflation expectations data ranges from 0 to(a cap at) 50 percent. We follow the literature, see e.g. Curtin (1996), and restrict our sampleto those respondents who gave inflation expectations below 30 percent, on the grounds that The survey asks respondents to provide their opinion on the following question:As to the economic policy of the government – I mean steps taken to fight inflation orunemployment – would you say the government is doing a good job, only fair, or a poor job? The survey asks respondents to provide their forecast of interest rate and unemployment:No one can say for sure, but what do you think will happen to interest rates for borrowingmonetary during the next 12 months–will they go up, stay the same, or go down?How about people out of work during the coming 12 months–do you think that these willbe more unemployment than now, about the same or less? More details about the calculation can be found at: https://data.sca.isr.umich.edu/fetchdoc.php?docid=24770
Median inflation expectation and actual realised inflation
78 80 82 84 86 88 90 92 94 96 98 00 02 04 06 08 10 12 14 16024681012
Inflation expectation(t|t+12)Realised inflation(t+12)
Fraction of households that has heard a particular news item
78 80 82 84 86 88 90 92 94 96 98 00 02 04 06 08 10 12 14 160102030
Inflation newsMonetary policy news
Figure 1: The top panel shows the median 12-month ahead household inflation expectation and the realized12-month ahead inflation. The bottom panel shows the fraction of households that have heard inflation newsor monetary policy news.
The top panel of Figure 1 shows the 12-month ahead median household inflation expecta-tions (blue solid line) and the actual realized 12-month ahead inflation (red dashed line), withshaded areas indicating NBER-dated recessions and the vertical line showing our structural10reak date. Both actual and expected inflation were high in the late 1970s but graduallydecreased during the two recessions in the early 1980s. Both remained relatively low through-out the 1990s and early 2000s. It is interesting that households on average also expectedhigher inflation during and after the financial crisis of 2008. These expectations remainedmuch greater than realized inflation, while deflation was likely more of a concern to policymakers.The bottom panel shows the fraction of households that had heard inflation news (blue solidline) or monetary policy news (red dashed line). As illustrated by the figure, the fraction ofhouseholds who had heard inflation news is quite volatile. As proposed by Ehrmann et al.(2017), this fraction is often driven by people who have heard that prices are higher. Theauthors also find the fraction to be highly correlated with retail gasoline price inflation ingeneral, suggesting that frequently purchased items shape households’ inflation (news) per-ceptions. The spikes in the series could be related to economic recessions or actual low orhigh inflation rates. For example, the high percentage of households who had heard inflationnews in the period March-May 1986 can be explained by the fact that in 1986 inflation rateshad reached levels below 2% for the first time in 20 years, a situation that was frequentlydiscussed in the media. The spike in the series in September-November 1990 is most likelya result of the US entering into recession in July 1990, lasting until March 1991. The reces-sion was at least partially related to the restrictive monetary policy enacted by the FederalReserve throughout 1989 and 1990, when the stated policy was to reduce inflation. The The median inflation expectation in the first sub-sample (1978:01-1983:09) was 6 per cent, comparedwith 3 per cent in the second sub-sample (1983:10-2016:02). This reduction in the median expectation wasaccompanied by a reduction in the heterogeneity of inflation expectations, with the variance of the cross-sectional distribution decreasing from 34.3 in the first sub-sample to 15.2 in the second sub-sample. Thisreduction in the heterogeneity of inflation expectations was likely due to the low and stable inflation rate inthe second sub-sample, and a stronger emphasis placed by the FED on maintaining low and stable inflation. The fraction of household that had heard inflation (monetary policy) news was 11.33(12.47) percent onaverage for the first sub-sample, decreasing to 5.61(5.96) per cent in the second sub-sample. . News impact on household inflation expectations We test if exposure to direct news of inflation and monetary policy affects household inflationexpectations. Model 1 considers only the impact of news on inflation and Model 2 considersboth news on inflation and monetary policy without controlling for characteristics of house-hold i , thus serving as a benchmark. The subsequent five models extend the benchmarkspecification: Model 3 controls for the additional impact of demographic characteristics D it ;Model 4 controls for the additional impact of perceptions of the effectiveness of governmentpolicies P it ; Model 5 controls for the additional impacts of expectations about interest ratesand unemployment; Model 6 controls for the additional impacts of consumer sentiment oninflation expectations ICS it ; and Model 7 considers all explanatory variables jointly. Wereport results in Table 1, with the top panel of the table showing the results for January1978 to September 1983, the middle panel focuses on October 1983 to February 2016. Sincewe have 457 monthly time dummies and their interpretations do not necessarily relate tonews effects, we omit results for the time dummies in the table. Results for the first sub-sample show news of inflation and monetary policy having a strongimpact on household expectations. In this relatively high inflation period, Model 1 showsthat hearing news of higher inflation led an average household to increase their inflationexpectations by 1.03 per cent. When jointly considered with news of monetary policy, theimpact of news of inflation reduces to 0.98.News of monetary policy changes can affect inflation expectations in two opposite ways. Oneis if households understand the transmission mechanism of monetary policy to future infla-tion, in which case news of tighter monetary policy implies lower expected future inflation. The monthly dummies capture the effect of common factors on household inflation expectation. One ofthese factors may be the objective intensity of news reporting on inflation and monetary policy. ub-sample 1: 1978:01 – 1983:09Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7Constant 4.45*** 4.48*** 4.43*** 4.56*** 4.47*** 6.08*** 4.04***News: inflation( φ π ) 1.03*** 0.98*** 0.94*** 0.83*** 0.83*** 0.87*** 0.67***News: monetary policy( φ r ) – 0.49*** 0.47*** 0.37*** 0.35*** 0.40*** 0.22***Log income – – 0.07 – – – 0.21***Age – – -0.03*** – – – -0.03***Female – – 0.26*** – – – 0.06Education – – 0.15*** – – – 0.19***Perception: government policy – – – -1.08*** – – -0.73***Expectation: interest rate – – – – 0.82*** – 0.72***Expectation: unemployment rate – – – – 0.82*** – 0.48***Consumer sentiment – – – – – -0.02*** -0.01***Adjusted- R φ π ) 0.64*** 0.62*** 0.61*** 0.52*** 0.50*** 0.43*** 0.39***News: monetary policy( φ r ) – 0.35*** 0.32*** 0.26*** 0.18*** 0.17*** 0.08**Log income – – -0.45*** – – – -0.35***Age – – -0.01*** – – – -0.01***Female – – 0.66*** – – – 0.54***Education – – -0.14*** – – – -0.11***Perception: government policy – – – -0.75*** – – -0.33***Expectation: interest rate – – – – 0.47*** – 0.40***Expectation: unemployment rate – – – – 0.73*** – 0.35***Consumer sentiment – – – – – -0.02*** -0.01***Adjusted- R The other is if households do not understand the transmission mechanism but understandthat the central bank targets inflation, in which case news of contractionary monetary policyis a signal that inflation is higher than previously expected. Model 2 shows that hearing newsof contractionary monetary policy induced households to expect 0.49 per cent higher infla-tion, indicating households understood that higher interest rates are a result of the centralbank’s concern about higher future inflation . Therefore the average household expecta- A potential endogeneity problem arises for the two types of news used in this paper. For example, in appear to be informed as a signal by the central bank’s response function, for examplea Taylor rule (as suggested by Carvalho and Nechio (2014)), rather than households concern-ing themselves with the expected future contractionary effect of the interest rate change oninflation. The results from Model 1 and 2 are consistent with earlier findings using aggre-gate data from the media having a role in driving household inflation expectations (Lamlaand Maag, 2012; Dr¨ager, 2015). Our results are also consistent with the information rigidityhypothesis, which suggests that households’ private information sets (through news) play arole in explaining disagreements in inflation expectations (Mankiw and Reis, 2002; Madeiraand Zafar, 2015).Model 3 confirms results in the earlier literature that households with different demo-graphic backgrounds disagree on inflation expectations, see, for example, Bryan and Venkatu(2001a,b) for the US, Blanchflower and MacCoille (2009) for the UK, Easaw et al. (2013) forItaly, and Jonung (1981) for Sweden. Households with different demographics may purchasedifferent consumption bundles. In particular, we find that those who are younger, female,and better educated tended to forecast higher inflation levels. Households may form infla- response to tighter monetary policy there may be a perception in this news that inflation prospects will bemitigated. To address this issue, we re-ran the estimation excluding those households that indicated theyhave heard news on lower inflation and tightening monetary policy and those households that indicatedthey have heard news on higher inflation and easing monetary policy. There are only marginal changes tothe estimated coefficients and all results remain qualitatively the same. The results are available from theauthors upon request. We cannot rule out the possibility that households form higher inflation expectations when hearing ofcontractionary monetary policy because they may think the monetary policy is too accommodative, see, forexample, Clarida et al. (2000) and Gertler et al. (1999). A consensus about accommodative monetary policycontributing to high inflation was achieved much later (in the 1980s) and in the first sample period it wassurely not well understood when households formed their inflation expectation. Therefore we do not expectthe average household would form inflation expectation in this sophisticated way.
Many of the aforementioned results remain true in the second sub-sample. But there aresome notable differences. First, the impacts on inflation expectations of news about bothinflation and monetary policy were smaller, although they remain significant. This lowerimpact of news in the second sub-sample reflects the fact that inflation had fallen andstabilised during this period, making news of inflation and monetary policy less salient forhouseholds, and thus reducing their impact on inflation expectations. Second, the signs ofthe impact of household income changed to be negative, and was much larger in absolutesize. Third, the sign on education also reversed so that now better than average educatedhouseholds expected lower inflation. These two sign reversals mean that households withhigher income and better education forecasted lower inflation than the average household inthis sub-period. Finally, gender played a much larger role, with the difference between maleand female expectations becoming significantly larger.In summary over the whole sample, our results indicate that exposure to news of higherinflation and contractionary monetary policy significantly increased household inflation ex-pectations. This result is robust across sample periods, and holds even after controlling forhousehold demographic characteristics, their perceptions on the effectiveness of governmentpolicies, their expectations about interest rates and unemployment, and their sentiment.Among macroeconomic theories, information rigidity models have been widely used to ex-plain cross-sectional disagreements of inflation expectations, see, for example Mankiw andReis (2002) and Mankiw et al. (2004). These models typically assume information is costlyto acquire, so people have a different information set when forecasting future paths of the18conomy. Our results indicate that households who had a larger news exposure expecteddifferent inflation rates ( ceteris paribus ), thus supporting the information rigidity theory.The fact that the estimated news effect ( φ π and φ r ) between Model 2 and 3 are very simi-lar suggests that household demographics and news almost independently explain inflationexpectations. This means that the demographic impacts on inflation expectations were notdue to the different demographic groups’ exposure to news.Controlling for the perception of the effectiveness of government policies, for expectationson future interest and unemployment rates and for consumer sentiment reduces the impactof news on household inflation expectations. These findings suggest that news of inflationand monetary policy impacted on inflation expectations partially through these householdperceptions about policy effectiveness, their expectations of future interest rates and unem-ployment, and their sentiments about current economic conditions.
4. The asymmetric impact of news
News on the movements of underlying economic variables may have an asymmetric impacton inflation expectations. This may arise if one particular direction of movement of thevariable has a more salient effect on expectations than the other at the time of making theexpectation decision. For example, due to diminishing marginal utility, higher inflation canerode household wealth and reduce utility more than it would increase it, if inflation fell bythe same amount–households may thus pay more attention to news of higher inflation thanof lower inflation. Households may also have experienced the high inflation episodes in the1970s and understand high inflation may indicate unsuccessful policies and have long lastingeffects on future paths of inflation (Madeira and Zafar (2015)) compared to lower inflation.Thus, it may be reasonable to assume a bigger impact of high inflation on future inflationexpectations. Using aggregated data, Lamla and Maag (2012) and Dr¨ager (2015) indeed findthe content of media reports have an asymmetric impact on inflation expectations.19tilising the cross-sectional nature of the MSC inflation expectations and news data, weinvestigate whether news content has an asymmetric impact on household inflation expec-tations at the disaggregated level. For each of the news variables N πit and N rit considered,we construct two dummy variables according to the content of the news. An upward arrow ↑ denotes news that corresponds to an increasing value of the underlying variable, whilea downward arrow ↓ relates to news decreasing the value of the underlying variable. Forexample, news of rising inflation would result in a value of 1 for N πit ↑ and a value 0 for N πit ↓ ; N rit ↓ = 1 indicates news about easing monetary policy and N rit ↑ = 1 indicates newsabout contractionary monetary policy. We thus replace N πit and N rit in Equation (1) with N πit ↓ , N rit ↓ and N πit ↑ , N rit ↑ : π eit = α + φ π ↓ N πit ↓ + φ π ↑ N πit ↑ + φ r ↓ N rit ↓ + φ r ↑ N rit ↑ + C it γ (cid:48) + T D t θ (cid:48) + (cid:15) it (2)For j ∈ { π, r } , we expect φ j ↓ to have the opposite impact on inflation expectations to φ j ↑ .We are interested in testing whether or not increases and decreases have the same absoluteimpact on inflation expectations. To test this, we calculate the z -score between the twoestimated parameters and test whether the null hypothesis φ j ↓ = − φ j ↑ is rejected: z j = φ j ↓ − ( − φ j ↑ ) (cid:113) Var( φ j ↓ − ( − φ j ↑ )) (3)Table 2 shows the estimation results of equation (2) with the columns giving the estimationresults for the two sub-samples. The results are broadly consistent with those presented inTable 1. Consistent with our expectations, the two directions of news content had oppositeeffects on household inflation expectations. This result is robust across both inflation andmonetary policy news and across sample periods.Both news of rising and declining inflation had significant impacts on household inflation ex-pectations across both sub-samples. Hearing news of higher inflation in the first sub-sample20 ub-sample 1 Sub-sample 21978:01 – 1983:09 1983:10 – 2016:02Constant 4.07*** 8.84***News: lower inflation( φ π ↓ ) -0.49** -0.22***News: higher inflation( φ π ↑ ) 0.70*** 0.46***News: easing monetary policy( φ r ↓ ) -0.08 -0.15***News: tightening monetary policy( φ r ↑ ) 0.31*** -0.00Perception: government policy -0.73*** -0.33***Expectation: interest rate 0.72*** 0.40***Expectation: unemployment rate 0.48*** 0.35***Consumer sentiment -0.01*** -0.01***Log income 0.20*** -0.35***Age -0.03*** -0.01***Female 0.07 0.54***Education 0.19*** -0.11***Adjusted- R φ π ↓ = − φ π ↑ φ r ↓ = − φ r ↑ increased inflation expectations by 0.70 per cent on average, and hearing news of lower infla-tion reduced inflation expectations by 0.49 per cent on average in this high inflation period.Hearing news on higher inflation in the second sub-sample increased household inflation ex-pectations by 0.46 per cent on average, but being exposed to news on lower inflation onlyreduced inflation expectations by 0.22 per cent. This result indicates that households re-spond to news on higher inflation more than to news on lower inflation in general, though theeffect of inflation news was much weaker in the second sub-sample. This is especially truefor news on lower inflation, where the impact is more than halved in the second sub-sample.The second row of the lower panel of Table 2 shows the the significance of the z-score test ofequation (3) in terms of inflation news ( φ π ↓ = − φ π ↑ ). Consistent with previous results, the testindicates that the symmetric effect of news on inflation cannot be rejected for the first sub-sample period, while the effect became significantly asymmetric in the second sub-sample,21here news of higher inflation had a bigger absolute impact than news of lower inflation.News on easing monetary policy did not significantly alter household inflation expectationsin the high inflation period (sub-sample 1), but significantly reduced inflation expectationsin the second sub-sample. On the other hand, news of tightening monetary policy signif-icantly increased household inflation forecasts in the first sub-sample but was irrelevant inthe second sub-sample. The third row of the lower panel of Table 2 shows the significanceof the z-score test of φ r ↓ = − φ r ↑ : we find that the symmetric effect of news on monetarypolicy could not be rejected for the first sub-sample, even though only tightening monetarypolicy was significant. However news on monetary policy became asymmetric in the secondsub-sample, when easing monetary policy had a much bigger absolute impact on inflationexpectations than tightening monetary policy.Since the asymmetries became significant in the second sub-sample, we are interested toknow how they evolved over time. We do this by conducting an expanding window estima-tion of z j (equation (3)) starting in October 1983. Figure 2 shows the evolution of z j forboth inflation and monetary policy news, with the horizontal black line indicating signifi-cance at the 10 per cent level. It is interesting that both news of increase and decrease oninflation (monetary policy) had similar absolute impacts on household inflation expectationsfor most of the 1980s. However, both news on inflation and monetary policy started to be-come increasingly asymmetric in the early 1990s, with the absolute impact of news on higherinflation becoming statistically greater than news on lower inflation after 1991 (top panel ofFigure 2), and news on easing monetary policy having a greater impact than contractionary Our credibility interpretation remains valid after distinguishing easing and tightening of monetary policynews in the second sub-period. Though the average household reduces inflation expectations when hearingnews on easing monetary policy, those who perceive effective government policies understand the implicationof monetary policy and forecast higher inflation. The detailed results on this are available on request. Figure 2: Time-varying asymmetries—z-score tests. monetary policy after 1999 (bottom panel).One explanation for this interesting evolution of asymmetric news may be rational inatten-tion to information (Sims, 2003). Since information is costly to process, households mayonly pay attention to news information that they regarded as relatively important. A gen-eral consensus developed in the 1980s and 1990s was that high inflation was bad and neededto be avoided. Presumably then, high inflation news came to represent unfavourable infor-mation for households. As a consequence, low and stable inflation became a norm in thelate 1980s and households inflation expectations became firmly anchored around 3 per cent.Even though inflation became a lesser concern, household paid disproportionate attentionto news on higher inflation that was regarded as unfavourable. Households may also con-23ider that higher inflation (above the norm) tends to be more persistent compared to lowerinflation, thus regarding higher inflation as unfavourable. After 2008, however, there maywell have been a growing relative unease about the risks of deflation, but we see no evidenceof that in Figure 2, since the z-score tests remained flat. With regards to monetary policy,we find that there is consistently significant evidence since 2007 of a greater impact of newson easing monetary policy in comparison to news on contractionary monetary policy. Oneinterpretation of this result could be that news on upcoming cuts in the federal funds rate(that started to occur in late 2007) as well as on quantitative easing by the Federal Reserve(that started in November 2008) had a noticeably strong impact on inflation expectations.In summary, we find evidence that rising inflation news and easing monetary policy impactson household inflation expectations significantly more than does lower inflation and tight-ening monetary policy. This is true in particular for the relatively lower inflation period(sub-sample 2: 1983:10-2016:02). Extending window estimation shows that the impact ofnews on higher inflation (easing monetary policy) increasingly became bigger compared tolower inflation (contractionary monetary policy) during the 1990s. These asymmetries onboth news persisted through the remaining sample.
5. The impact of news under the zero lower bound
Does the impact of inflation and monetary policy news change under the zero lower boundfrom 2008? Table 3 shows the estimates of Model 1-7 for the period between June 2008 andFebruary 2016. Compared with the second sub-sample in Table 1, inflation news has a muchbigger impact on household inflation expectations for all models. This may reflect the factthat households in this period realize that the FED had lost the effectiveness of its conven-tional instrument (the federal funds rate) in managing inflation (deflation). Therefore, inthis period households may have reacted more sensitively to any news on inflation. Lookingagain at Figure 1, note that median inflation expectations were almost always greater than24 ample period: 2008:06 – 2016:02Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7Constant 4.04*** 4.04*** 10.30*** 3.79*** 3.80*** 5.84*** 10.53***News: inflation( φ π ) 1.22*** 1.21*** 1.08*** 1.01*** 0.95*** 0.80*** 0.67***News: monetary policy( φ r ) – 0.52*** 0.52*** 0.35*** 0.20* 0.14 0.07Log income – – -0.49*** – – – -0.44***Age – – -0.00 – – – -0.01***Female – – 0.58*** – – – 0.48***Education – – -0.26*** – – – -0.17***Perception: government policy – – – -0.90*** – – -0.37***Expectation: interest rate – – – – 0.35*** – 0.35***Expectation: unemployment rate – – – – 1.14*** – 0.58***Consumer sentiment – – – – – -0.02*** -0.01***Adjusted- R realized inflation from 2008.During this period, the FED could not cut the current federal funds rate any further, thoughit was able to and did use extensively forward guidance, aiming to influence expectations of future interest rates and inflation to try to stimulate the economy. Forward guidance canbe either Odyssean—when the Feb publicly commits monetary policy to a future action, orDelphic—when the policy states the likely future policy actions based on the policymaker’spotential private information about macroeconomic fundamentals, see, for example, Camp-bell et al. (2012). In addition, the FED undertook 3 rounds of large scale asset purchasesfrom 2008 to 2014, otherwise known as “quantitative easing”, leading to a significant expan-sion of its balance sheet with bank debt, Treasury securities and mortgage-backed securities.Similar to the earlier results, the effectiveness of such unconventional policies relies cruciallyon how those economic agents respond on hearing the news of these policies. However, com-pared with the second sub-sample of Table 1, there are two noticeable differences. First,the25mpact of monetary policy news on expected inflation appears strengthened (Model 2-3),even when the households’ demographic backgrounds are jointly considered, thus seeming tostrengthen the signalling role of the FED’s policy. Second, jointly considering consumer sen-timent (Model 6) makes the impact of monetary policy small and statistically insignificant.The comparison of this to the results in Table 1 is indicative of the different implicationsof unconventional and conventional monetary policy news. The (unconventional) monetarypolicy news estimate of Model 2 in Table 3 may be seen as a proxy for consumer sentimentin relation to inflation expectations formation.Regressing consumer sentiment on monetarypolicy news yields a significant coefficient (at the 1% level) of -7.56. Therefore, hearingnews on monetary policy contraction would be associated with a 7.56 reduction in consumersentiment during the zero lower bound period. Consumer sentiment fell significantly in 2008to 2009, but improved consistently thereafter. The significant negative estimate of consumersentiment in Model 6 suggests that those households hearing news on monetary policy eas-ing and thus credit easing, recognised this as a signal to lower their inflation expectationsand to expect easier conditions that improved consumer sentiment. Households not hearingthis news had no signal, and were responsible for maintaining inflation expectations aboverealized inflation.In summary, these results suggest in a consistent way that monetary policy news provided asignal about future inflation. This signalling effect manifests through consumer confidenceduring the zero lower bound period. This implies that central banks should pay particularattention to the impact of their policy communications on consumer sentiment to maximisethe impact of asset purchases and forward guidance on inflation expectations. We thank the referee for suggesting this. . Conclusions We have examined the impact of news on household inflation expectations. Using monthlyUS consumer inflation expectations data between January 1978 and February 2016, we findthat, in general, exposure to news on inflation and monetary policy significantly helps toexplain household inflation expectations. This remains true even after controlling for house-holds demographic characteristics, their perception of the effectiveness of government policiesin managing business cycles, their expectations of future interest and unemployment rates,and their sentiment. This result tells us that the average effect of news is unaffected by thecontrols. To understand better other distributional aspects of the response, we would needto consider empirical non-linearities, which we leave for future research.We find evidence that news on inflation and monetary policy had an asymmetric impacton household inflation expectations. In particular, households responded to news of higherinflation and easing monetary policy significantly more than news of lower inflation and tight-ening monetary policy. This was especially true in the relatively low inflation period after1983, and probably was a result of the broad persuasion by public figures about the dangersof high inflation. The more unfavourable perception of risks of higher inflation remainedvalid also after 2008, even though, also, the impact of a deflation threat has likely increasedsince then. If this deflation threat were to become more relevant and households did not yetrealize its significance, policy-makers, political leaders and opinion-makers to become moreactive to warn about its consequences. The significant asymmetric impact favouring easieras opposed to tighter monetary policy suggests that the aggressive unconventional monetarypolicy pursued by the FED from 2008 worked in the desired direction.From 2008, expected inflation became persistently higher than realized inflation. We findnews of unconventional monetary policy acts as an imperfect signalling device for household27nflation expectations, which may be seen as a proxy for consumer sentiment in relation toinflation expectation formation. Weak consumer sentiment through perceived credit marketconditions may have played an important role in understanding the relatively high inflationexpectations in this period. 28 . Reference
Adam, Klaus (2007) Optimal monetary policy with imperfect common knowledge. Journalof monetary Economics 54 (2), 267–301.Bai, Jushan. and Pierre Perron (2003) Computation and analysis of multiple structuralchange models. Journal of Applied Econometrics 18 (1), 1–22.Blanchflower, David. and MacCoille, Conall (2009) The formation of inflation expectations:an empirical analysis for the UK. Tech. rep., National Bureau of Economic Research.Bryan, Michael. and Guhan Venkatu (2001a) The curiously different inflation perspectivesof men and women. Economic Commentary.Bryan, Michael. and Guhan Venkatu (2001b) The demographics of inflation opinion surveys.Economic Commentary.Campbell, Jeffrey, Charles Evans, Jonas Fisher, and Alejandro Justiniano (2012) Macroeco-nomic effects of federal reserve forward guidance. Brookings Papers on Economic Activity2012 (1), 1–80.Carroll, Christopher D (2003) Macroeconomic expectations of households and professionalforecasters. The Quarterly Journal of Economics 118 (1), 269–298.Carvalho, Carlos and Fernanda Nechio (2014) Do people understand monetary policy? Jour-nal of Monetary Economics 66, 108 – 123.Clarida, Richard, Jordi Gali and Mark Gertler (2000) Monetary Policy Rules and Macroeco-nomic Stability: Evidence and Some Theory. The Quarterly Journal of Economics 115 (1),147–180.Curtin, Richard (1996) Procedure to estimate price expectations. University of MichiganSurvey Research Center. 29r¨ager, Lena (2015) Inflation perceptions and expectations in Sweden–Are media reportsthe missing link? Oxford Bulletin of Economics and Statistics 77(5), 681–700.Dr¨ager, Lena and Michael J Lamla (2017) Imperfect information and consumer inflationexpectations: Evidence from microdata. Oxford Bulletin of Economics and Statistics.Easaw, Joshy, Roberto Golinelli and Marco Malgarini (2013) What determines householdsinflation expectations? theory and evidence from a household survey. European EconomicReview 61, 1–13.Ehrmann, Michael, Damjan Pfajfar, and Emiliano Santoro (2017) Consumer attitudes andtheir inflation expectations. International Journal of Central Banking, 225–259.Gertler, Mark, Jordi Gali and Richard Clarida (1999) The Science of Monetary Policy: ANew Keynesian Perspective. Journal of Economic Literature 37 (4), 1661–1707.Jonung, Lars (1981) Perceived and expected rates of inflation in Sweden. The AmericanEconomic Review, 961–968.Lahiri, Kajal and Yongchen Zhao (2016) Determinants of consumer sentiment over businesscycles: Evidence from the us surveys of consumers. Journal of Business Cycle Research,1–29.Lamla, Michael and Thomas Maag (2012) The Role of Media for Inflation Forecast Disagree-ment of Households and Professional Forecasters. Journal of Money, Credit and Banking44 (7), 1325–1350.Madeira, Carlos and Basit Zafar (2015) Heterogeneous inflation expectations and learning.Journal of Money, Credit and Banking 47(5), 867–896.Malmendier, Ulrike and Stefan Nagel (2016) Learning from inflation experiences. The Quar-terly Journal of Economics 131 (1), 53–87.30ankiw, N. Gregory and Ricardo Reis (2002) Sticky Information Versus Sticky Prices: AProposal To Replace The New Keynesian Phillips Curve. The Quarterly Journal of Eco-nomics 117 (4), 1295–1328.Mankiw, N. Gregory, Ricardo Reis and Justin Wolfers (2004) Disagreement about InflationExpectations. In: NBER Macroeconomics Annual 2003, Volume 18. NBER Chapters.National Bureau of Economic Research, Inc, pp. 209–270.Melosi, Leonardo (2017) Signalling effects of monetary policy. The Review of EconomicStudies 84 (2), 853–884.Pfajfar, Damjan and Emiliano Santoro (2009) Asymmetries in inflation expectations acrosssociodemographic groups. Tech. rep., mimeo.Pfajfar, Damjan and Emiliano Santoro (2013) News on inflation and the epidemiology ofinflation expectations. Journal of Money, Credit and Banking 45 (6), 1045–1067.Sims, Christopher (2003) Implications of rational inattention. Journal of Monetary Eco-nomics 50 (3), 665–690.Woodford, Michael (2001) Imperfect common knowledge and the effects of monetary policy.Tech. rep., National Bureau of Economic Research.Zhang, Junni, Wolfgang K H¨ardle, Cathy Y. Chen and Elisabeth Bommes, (2016) Distillationof news flow into analysis of stock reactions. Journal of Business & Economic Statistics34 (4), 547–563. 31 ppendix A. Pairwise correlation
Pairwise Correlation
V1 V2 V3 V4 V5 V6 V7 V8 V9 V10V1V2V3V4V5V6V7V8V9V10 -1-0.8-0.6-0.4-0.200.20.40.60.81