USES OF BUSINESS AND CONSUMER OPINION SURVEY DATA, IMPLICATIONS FOR DATA PRODUCERS Giuseppe Parigi Bank of Italy, Economic Research Department The art and science of short-term analysis of high frequency economic data is extremely important to economic policy decision makers, such as central bankers. “Good diagnosis helps in making predictions” Katona (1957) Among short-term indicators, survey data play a prominent role Nowadays there is an increasing demand of high quality survey data SURVEY DATA might be represented as containing three types of information (see Fuhrer, 1988): Information on current developments Forward-looking information “Animal spirits” information Information on current developments Katona (1957): “Expectations – intentions as well as other notions about the future – are current data which help to understand what is going on at the time when expectations are held.” TIMELINESS Survey data are available soon after the end of the reference period (generally, the month) and are not revised Bridge models Coincident indicators Early estimate of data released with delay NBER and Factor models Nowcasting Help establish initial conditions Bridge Models Publication of quarterly national accounts within 70 days after the end of reference period Flash estimates in 45 days Survey data and other short term (composite) indicators Need timelier information about National accounts BRIDGE MODELS High frequency data National account data Bridge Models: matching variables and indicators Univariate model Retail sales, CSI, UR Business surveys (expected demand), construction comp. Trade variables, real exch. rates, IP, Surveys data Collective consumption Private consumption Gross fixed capital formation Exports of goods and services Forecasts Imports of goods and services IP, Business surveys GDP SUPPLY SIDE DEMAND SIDE GDP, Surveys data Changes in stocks ________________________________________________________________ (GDP+Imports) GDP= CON + COC + INV + EXP - IMP + VSP Coincident indicator – Eurocoin Industrial Production Trade variables 150 series 40 series Total Money Prices 800 variables 160 series 130 series Miscellanea 80 series Labour market 40 series Survey data 200 series 25% Forward-looking and “animal spirit” information Events which are difficult to quantify (tax changes) Expectations with self-fulfilling properties Survey data Forecasting power Leading Indicators Turning points detection Anticipate the evolution of the cycle Estimates of the probability of being in a recession/expansion Theoretical and Empirical Models Interpretations of survey data: what is this thing called confidence? The problem of sometime too vague verbal questions Survey data and Economic analysis Although some consensus emerged in the literature that SD could play a role, this appears to be ad hoc. A convincing representation of SD is needed… SD as an alias of macroeconomic variables? SD as a proxy of non-linearities (shocks)? SD as a proxy of unobserved variables? … but their informative content is still a mystery Survey data and Economic analysis EXPECTATIONS Scepticism of economists to the use of survey data: one should believe only what people do and not what people say. Revealed preference analysis Economists attempt to infer expectations by combining data on realized experience (choice data) with assumptions about the process of expectation formation. Survey data and Economic analysis The prevailing practice has been to assume that agents have expectations that are objectively correct (i.e. rational). But lack of empirical evidence on the validity of the expectations assumptions has led to a crisis of credibility. Survey data is a possible solution… “The data I have in mind are self-reports of expectations elicited in the form called for by modern economic theory; that is subjective probabilities” (Mansky, 2004) Survey data Probabilistic questions Juster (1966) showed that elicited purchase probabilities are better predictors of subsequent behaviour Vague concepts like “future economic conditions” may be avoided with questions about personal facts Harmonization of survey across countries is more likely to be complete when based on numeric response scales Numeric probability scales allow the comparability of responses among different people, across situations and over time Survey data Probabilistic questions: Examples The Health and Retirement Study in the USA (subj. prob. of living 75/85, job loss etc.) The Bank of Italy Survey on Household Income and Wealth and The Dutch VSB-Panel Survey (subj. prob. of one year-ahead growth rates in income) The Bank of Italy Survey on Business Investment (one of the few examples of probabilistic questions to firms) The Michigan Survey of Consumers Survey data: Further improvements Survey should follow the evolution of people behaviour Developments of financial markets Aging populations Reforms of the welfare … Imply new forms of uncertainty Survey data : Further improvements Survey data should be released in a more detailed way… on a geographical, sectoral, dimensional basis (but also new classifications as technologically advanced v. traditional sectors) by income, age, employment classes (better match with macroeconomic variables)