Macroeconomic expectations and time varying heterogeneity:evidence from individual survey data

ABSTRACT

The goal of this paper is to investigate forecast heterogeneity and time variability in the formation of expectations using disaggregated monthly survey data on macroeconomic indicators provided by Bloomberg from June 1998 to August 2017. We show that our panel of forecasters are not rational and are moderately heterogeneous and thus confirm that previously well-established results on asset prices hold for macroeconomic indicators. We propose a flexible hybrid forecast model defined at any time as a combination of the extrapolative, regressive, adaptive and interactive heuristics. Controlling for endogenous structural breaks, we find that experts adjust their forecast behaviour at any time with some inertia in extrapolative and adaptive profiles. Changes in the formation of expectations are triggered mostly by financial shocks, and uncertainty is dealt with by using complex processes in which the fundamentalist component overweighs chartist activity. Forecasters whose models combine different relevant rules and display high temporal flexibility provide the most accurate forecasts. Authorities can then stabilize the domestic markets by encouraging fundamentalists’ forecasts through increased transparency policy.

KEYWORDS:

• Dynamic heterogeneity
• flexible forecast model
• formation of expectations
• individual survey data
• macroeconomic indicators

JEL CLASSIFICATION:

• C32
• C52
• D84
• E71

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ These are the extrapolative (bandwagon), adaptive (error-correcting) and regressive (mean-reverting) rules which assume that forecast assessment is based upon past trends in the variable of interest, past forecast errors and the spread between the actual and target values of the variable, respectively.

² These are realistic assumptions since the soar of information systems reduce costs and aversion to making forecast errors is sensitive to the state of the economy.

³ Our emphasis on modelling changes in expectations instead of levels is in line with Pesaran et al. (1985), who show that subjective sociological factors durably affect expectations in levels but this bias is considerably reduced when expected changes are considered.

⁴ Note that each agent $i$ has in mind their own perception of the target, say ${\bar{a}}_t^i$, which can be different from the true fundamentals-based target ${\bar{a}}_t$. Let ${\bar{a}}_t^i={\bar{a}}_t+{\omega }_0^i+{\eta }_t^i$ where ${\omega }_0^i$ is a systematic bias and ${\eta }_t^i$ the stochastic deviation of ${\bar{a}}_t^i$ from ${\bar{a}}_t$ (${\eta }_t^i$ may contain fundamentals that are omitted in ${\bar{a}}_t^i$ or absorb any mismeasurement of their impact). Using the subjective target ${\bar{a}}_t^i$ into Equation (2) in place of the true target still amounts to specifying the regressive component with the true target ${\bar{a}}_t$ as in Equation (2), ${\gamma }_{i1}{\omega }_0^i$ and ${\gamma }_{i1}{\eta }_t^i$ then being captured by the intercept and the error term of the econometric expectation model. Misspecification in the subjective target would lead to autocorrelated residuals in the model, whereas good fits from this model would imply a correct assessment of the true target by the agent.

⁵ Flieth and Foster (2002) use the term ‘interactive process’ in the sense that decision-makers communicate with each other to exchange their opinions so that their expectations are directly influenced by the amount of information shared. Instead, we believe that market participants do not find the opportunity of communicating directly but endeavour to guess others’ average opinion through the past average forecast released by the survey. Acting this way, each agent indirectly interacts with the market because all others’ contributions to the past average forecast influence the agent’s current forecast.

⁶ Note that for many macroeconomic indicators, the announcement release is scheduled at a date after the end of the reference period. New home sales and unemployment rate are typically released in the second half of the month following the reference month and on the first Friday after the reference month, respectively. Even though Bloomberg allows forecasters to update their forecasts up to the week preceding the announcement release, our choice of focusing on one-month ahead forecasts (with respect to the scheduled announcement dates) ensures that these are made within the reference month.

⁷ The very detailed information associated with the disaggregated surveys provided by Bloomberg makes it possible to follow the career of forecasters through several companies and select those who were not involved in institutional mobility for long sample periods.

⁸ However, in order to keep the sample homogeneous, an exception was made for Nomura Securities (individual # 14) who provided a smaller series of answers in predicting the consumer confidence index. Test results are thus not provided for the case of this respondent to this variable (see Tables 1 to 4).

⁹ These are Briefing.com, Deutsche Bank Securities, Credit Suisse, Morgan Stanley, Citi, CIBC World Markets, PNC Bank, Daiwa Securities America, Wrightson ICAP, IDEAglobal, BMO Capital Markets, BofA Merrill Lynch Research, High Frequency Economics, Nomura Securities Intl., First Trust Advisors, Goldman Sachs, Maria Fiorini Ramirez, and IHS Global Insight. These firms will be numbered in this order from 1 to 18 where appropriate.

¹⁰ We choose this specification instead of the reverse equation linking the ex-post change in the observed value to the expected change because such an equation could present an endogenous regressor problem leading to inconsistent estimates.

¹¹ In its simplest version, Ito’s (1990) test compares an agent $i$‘s expectation based on a public information component $f(I_t)$ and a private information component $g_i$, such that $E_t^i{a}_{t+1}=f(I_t)+g_i+{\epsilon }_{it}$, to the market average expectation defined as $E_t^M{a}_{t+1}=f(I_t)+g_M+{\epsilon }_{Mt}$, where $g_M$ is the average of the $g_i$‘s. Subtracting the latter from the former, we obtain the described test equation. If an agent’s private information is equal to the average private information (the intercept equals zero), nothing makes the agent’s expectation different from the market expectation and the null of homogeneity fails to be rejected.

¹² It is worth emphasizing that estimating each agent’s model with endogenous structural changes in parameters makes more sense than imposing exogenously break dates corresponding to major crises in the form of a Chow instability test. This is because all agents do not necessarily respond to a given economic shock at the same time, individual reaction times may differ between the ones who anticipate the event and those who adjust with delay. In this case, setting a break date arbitrarily can lead to inconsistent estimates.

¹³ Note that autocorrelation and heteroskedasticity in residuals may also arise if a structural change is omitted when in fact it has occurred. Because the number of breaks for each forecaster was determined endogenously and without restriction, this risk was excluded.

¹⁴ This is in line with Capistran and Timmermann (2009) who, using survey data on inflation expectations, show that heterogeneity of professional forecasters varies systemically over time.

¹⁵ We are grateful to Nicolas Sopel, expert global macro forecaster at the RHB Research Institute, for providing useful information from their own experience and for confirming the practical relevance of the four forecast methods described in our analysis.