The ZEW Financial Market Survey Panel

2.1 Historical Response Behavior

Figure 1 provides details about the size of the ZEW FMS panel and the historical response behavior over the lifetime of the ZEW FMS. Whether panel members are active or have left the panel is not documented for survey waves before 2011. For these survey waves, we thus estimate the status of panel members with the help of response data. We assume that these panel members were active in the periods between their first responses and their last responses. Note that the resulting values are only rough approximations of the actual number of active panel members. For example, if panel members first participated after the month they entered the panel, the estimated number of active panel members is understated for all periods between the month they entered and the month of their first participation.

Figure 1:

Panel size and response behavior. These figures show the size of the panel and historical response behavior over the lifetime of the ZEW FMS. The vertical dashed lines in Figures 1a and 1b separate the periods for which the number of active panel members is documented and for which it is not. The values for number of active panel members and the response rate before 2011 are approximated from response data.

Figure 1a shows the number of active members of the ZEW FMS panel since December 1991. The time series reveals two phases of rapid panel expansion. The first phase took place in the first year of the survey’s existence, in which the number of active participants increased from 141 in December 1991 to around 350. The second phase took place in August 1998, when the number of active panel members increased to 420 from 335 in September 1998. With an average of 380, the size of the ZEW FMS panel was relatively stable in the 2000s. The local maximum of 407 panel members in March 2011 marks the start of a downward trend. In September 2021, the ZEW FMS panel had 271 active panel members and the total number of panel members since the start of the survey was 2002. Figure 1b reveals that the response rate of the ZEW FMS has been trending down since the start of the ZEW FMS and is highly volatile. In the survey’s first year, the estimated response rate fluctuated around about 90 percent. Between September 2020 and September 2021, the response rate was on average 65.7 percent and fluctuated between 60.5 and 72.3 percent.

The cumulative distribution of the number of participations by panel member depicted in Figure 2 reveals that the ZEW FMS dataset features long individual time series. The number of responses by ZEW FMS panel members ranges from 1 to 350 with an average of 49.45. While about 20 percent answered only 1 or 2 times, about 30 percent of the respondents participated more than 50 times, and about 20 percent more than a 100 times. Half of the panel members responded more than 18 times. More detailed information on e.g. the composition of the panel, the development of the cross-section over time, information on the professional experience of the participants etc. can be found in Section 2.2 of Brückbauer and Schröder (2021).

Figure 2:

The distribution of the number of participations by panel member.

4.1 Using ZEW Variables in Forecasting Exercises

Are the expectations collected by the ZEW Financial Market Survey useful to forecast the future value of selected target variables? The papers outlined in this section examine the predictive performance of different expectation series of the ZEW FMS.

Five papers are primarily interested in business cycle forecasting. Benner and Meier (2004), Breitung and Jagodzinski (2001), Hüfner and Schröder (2002a, 2002b) investigate the forecasting performance of the ZEW Economic Expectations and other leading indicators for Germany, whereas Carstensen et al. (2011) focus on the euro area.

Hüfner and Schröder (2002a, 2002b) concentrate on the forecast performance evaluation of the IFO business cycle expectations and the ZEW Economic Expectations index for predicting the future change in the German industrial production index. Benner and Meier (2004) also take the IFO and the ZEW expectations indices and additionally analyse the “Earlybird” indicator of the journal Wirtschaftswoche. Going beyond Huefner and Schröder, Benner and Meier also determine the model recursively, and not only the forecasts. Carstensen et al. (2011) compare the forecast performance of seven indicators for the industrial production in the euro area. The indicators are, for example, the OECD composite leading indicator, the European Sentiment Indicator (ESI), the FAZ Euro Indicator, and the ZEW Economic Expectations index.

Four publications examine the forecast performance of exchange rate expectations. These are Bofinger and Schmidt (2003), Dick et al. (2015), Leitner and Schmidt (2006), and MacDonald et al. (2009). Bofinger and Schmidt (2003) and Leitner and Schmidt (2006) take the aggregated USD/EUR expectations of the ZEW FMS and assess the forecast performance for the future USD/EUR exchange rate. Dick et al. (2015) and MacDonald et al. (2009) use the microdata, i.e. the expectations on the individual level and try to find out, why some forecasters are better than others. They conclude, inter alia, that persons with a superior forecast performance for exchange rate fundamentals are also better in forecasting exchange rates.

Another four studies concentrate on stock and bond markets. Entorf and Steiner (2007), Entorf et al. (2012), and Hess and Niessen (2010) examine whether a new release of the ZEW Economic Expectations index has impact on capital markets in Germany. Entorf and Steiner (2007) use high-frequency data for the Xetra DAX and estimate the market impact of the release of the ZEW index. This analysis is extended to the impact of the IFO Business Climate index in Entorf et al. (2012). Hess and Niessen (2010) investigate the impact of ZEW and IFO index on the German bund futures market in the context of a Bayesian learning model using high-frequency data. Schmeling and Schrimpf (2011) examine the predictive power of inflation expectations on future stock returns for France, Germany, Italy, Japan, the US, and the UK. For all six countries, expected inflation is measured by the inflation expectations from the ZEW FMS.

One paper (Krüger et al. 2011) assesses the forecast performance of the ZEW expectations for the short-term interest rates in the euro area for the future EURIBOR. This paper uses the Carlson and Parkin (1975) method to estimate quantitative short-term interest forecasts using the qualitative expectations of the ZEW FMS.^[4] In Krüger et al. (2011), the quantified EURIBOR survey forecasts are combined with forecasts from a time series model applying a time-varying weighting scheme. Scheufele (2011) uses the quantified ZEW inflation expectations to forecast consumer price inflation. He finds, inter alia, that the inflation expectations, when added to forecasting models, improves forecast accuracy.

There are three studies which are primarily methodologically oriented. Nolte and Pohlmeier (2007) test the forecasting performance of traditional time series methods and compare them with the results using different quantification methods applied to the qualitative expectations from the ZEW FMS. They make use of the following data from the survey (for the period December 1991 until April 2004): German, US and Japanese inflation rates; German, US, and Japanese short-term interest rates; DAX index, Dow Jones Industrial Index, Nikkei 225 Index, FTSE 100 Index, CAC 40 Index; the USD/EUR and the GBP/EUR exchange rate. Mokinski (2016) develops a nowcasting model with the aim to estimate the (latent) daily responses using the irregularly distributed responses within the monthly survey interval. This is possible because the time stamps (i.e. the time of arrival) of the individual responses are available. The results are interesting for improving event studies based on the individual responses of the ZEW FMS. Fastrich and Winker (2014) use a portfolio-based approach to improve the point forecasts of the ZEW FMS participants regarding the DAX30 index. The authors construct forecast-portfolios in which the analysts are like “assets” with specific characteristics that are used to form the portfolios. The results show, for example, that the forecast-portfolios exhibit smaller forecast errors compared to the most successful single forecaster on horizons smaller than six months.

4.2 Expectation Formation

Twenty papers use data from the ZEW FMS to analyze expectation formation. Many of these studies examine specific behavioral finance topics like overconfidence, the effect of framing on expectations, herding behavior or differences between “chartists” and “fundamentalists”. Three papers are concerned with the effect of monetary policy on inflation expectations.

The earlier studies like Marnet (1996), König et al. (1998, 1999), and Szczesny et al. (1997), are using the the expectations only in form of aggregates. Almost all of the other, more recent, papers use the expectations data of the ZEW FMS on the individual level, i.e. the individual responses to the questions of the survey.

Seven of the papers focus on stock expectations and in particular the expectations on the German DAX. Deaves et al. (2010) is the first paper to use the DAX point and interval forecasts. Since mid-2003, the regular questionnaire of the ZEW FMS has asked for the six-month-ahead forecast of the DAX index and a 90 per cent confidence interval for the point forecast. Deaves et al. (2010) and Deaves et al. (2019) use these data for analyzing the statics and dynamics of overconfidence and for the effect of overconfidence on the ability to forecast future stock returns. In Deaves et al. (2021), the DAX point forecasts are combined with the responses on the question regarding the fundamental value of the DAX (”Is the DAX currently overpriced, fairly priced, or underpriced?”). Using these data the study investigates whether the survey respondents believe in stock market efficiency. Brückbauer (2020) examines how the DAX expectations, in particular the DAX point forecasts, are formed. He uses not only DAX point forecasts, but also the qualitative expectations on the DAX, the future economic situation in Germany, inflation, and short- and long-term interest rates. He investigates which factors drive the variation in expected DAX returns and he also evaluates the precision of DAX return forecasts. Hoffmann et al. (2017) take the DAX point forecasts and investigate if the DAX returns the respondents witnessed in the past influence how they form expectations on the future DAX. As control variables they use (qualitative) expectations from the ZEW FMS on the economic situation. They make also use of personal characteristics of the respondents like gender, age, and year of career start.

Breitung and Schmeling (2013) focus on methods which are used to quantify qualitative survey data, for example the Carlson and Parkin (1975) method and more flexible methods with time-varying parameters. The authors use DAX point forecasts and qualitative DAX expectations to evaluate different econometric procedures to quantify the qualitative DAX expectations. The comparison of the quantified qualitative expectations with the quantitative DAX forecasts (from the same respondents and for the same time horizon) makes it possible to analyze the characteristics of the quantification procedures in greater detail.

The paper of Rangvid et al. (2009) focuses on so called “higher-order expectations”. If higher-order expectations persist in a market, an investor does not only invest in those assets they prefer as best performing assets, but also takes into account what other investors will choose to invest in. They will also take into account that other investors will act in the same way (see Rangvid et al. (2009: 2)). In their study, the authors try to differentiate the hypothesis of higher-order expectations from the hypothesis of herding behavior. Their study is based on the qualitative expectations from the ZEW FMS on the stock markets of France, Germany, Italy, Japan, the US, and the UK.

A paper that also deals with herding is Lux (2009). The author focuses on the response behavior for the question on the future economic situation in Germany. He uses neither the individual data nor the aggregate indices but the ratio of replies for the three categories “increase”, “no change”, and “decrease”. The author uses a stochastic dynamic model on social interaction and finds that the respondents of the ZEW FMS show a strong tendency to follow the opinion of their peers.

Another seven papers deal with the expectations on exchange rates.^[5] The majority of these papers investigate the expectations on the USD/EUR exchange rate, for earlier periods also on the USD/DM exchange rate. One of the older publications, Schröder and Dornau (2002), investigates whether the respondents have structural economic exchange rate models (flexible- or sticky-price model, Mundell Fleming model) in mind when forming expectations on exchange rates, future economic conditions, inflation, and interest rates.

Menkhoff and Rebitzky (2008) estimate the long-term relationships between USD/EUR expectations and (relative) prices and investigate how deviations from long-run purchasing power parity (PPP) are reflected in the exchange rate expectations. Whereas this paper uses the expectations index, the individual responses are the main data source for Menkhoff et al. (2008). In this second study, the authors investigate if and how strongly the two groups of “chartists” and “fundamentalists” are driven by the PPP model in contrast to so called technical forecasting models. They find, inter alia, that fundamentalists believe too much in mean reversion, which leads to a poor forecasting performance, whereas “chartists” rely too much on extrapolating short-term trends.

Menkhoff et al. (2009) examine the sources of heterogeneity in exchange rate expectations for USD/EUR GBP/EUR, and JPY/EUR applying the chartist-fundamentalist model. Some of the findings are that heterogeneity decreases when exchange rates deviate relatively strongly from their fundamentals, the group of fundamentalists expects a mean-reverting exchange rate behavior. Strong movements in the exchange rates lead fundamentalists to shift temporarily into the group of chartists.

Dick and Menkhoff (2013) investigate the different behaviors of chartists and fundamentalist for the USD/EUR in more detail. They make use of the individual USD/EUR expectations from the ZEW FMS and combine this dataset with the answers to a special question in which the participants had been asked for a self-assessment regarding the use of charts and fundamentals. They find, inter alia, that the choice of forecasting tools is influenced by recent experience: when exchange rates exhibit trends, the respondents tend to switch toward Chartism; in contrast, they move away from Chartism when the exchange rate deviates substantially from its longer-term average.

Gloede and Menkhoff (2014) deal with the USD/DM and USD/EUR expectations and investigate the overconfidence of the respondents. In addition to the regular ZEW FMS data, they make use of responses to special question regarding a self-assessment on how strong they see their own forecasting precision compared to the average of all survey participants. To account for person-specific characteristic Gloede and Menkhoff (2014) include data on gender, age, and job function into their analysis. They find that high overconfidence goes along with a high self-rating and a low forecasting performance.

Glaser et al. (2019) study the so called framing effect. In particular, the authors investigate in three sub-studies if professional forecasters think of asset prices and asset returns in the same way. In the third sub-study data from an experiment, conducted within the ZEW FMS, are used. This survey experiment consisted of 12 waves between September 2012 and June 2015. The respondents received an additional questionnaire to collect the data needed for the experiment. One main result of the paper is that it actually significantly matters if forecasters are asked to form expectations on prices or on returns.