Non-response and Weighting Systems in Business Tendency Surveys: Are Expectations Influenced?

Rationality of economic agents belongs to the basic assumptions of neoclassical economic theory, and for decades it has inspired research on whether expectations are indeed formed rationally. Direct data on expectations are available mainly through business tendency surveys which are subject to various types of non-response problems. Inclination of industrial enterprises to respond may be correlated with values of measured variable, introducing response bias. Response bias may also occur as a result of introducing weighting systems to control variable size of respondents. The two key properties of rational expectations, on which the majority of empirical analyses of survey data are focused, are unbiasedness and orthogonality. We analyze several sample balance statistics and expectations series based on quantified survey data, taking into consideration issues of non-response and weighting schemes. Alternative definitions of expectations series aim to account for: 1) influence of arbitrary assumptions concerning weighting of individual data, 2) changing sample structure that results from non-response, 3) response rates varying with degree of optimism / pessimism of respondents. Results of our analysis indicate that expectations concerning relative changes in production are unbiased but not efficient with respect to freely available information, namely, observed relative changes in production (lagged three months) and expectations balance (lagged two months). This result holds for a range of weighting schemes and non-response issues analyzed, including changes introduced to sample structure by non-response, and increased inclination of "optimists" and "pessimists" to supply answers in the business tendency survey, as long as their shares remain constant in time. (original abstract)Rationality of economic agents belongs to the basic assumptions of neoclassical economic theory, and for decades it has inspired research on whether expectations are indeed formed rationally. Direct data on expectations are available mainly through business tendency surveys which are subject to various types of non-response problems. Inclination of industrial enterprises to respond may be correlated with values of measured variable, introducing response bias. Response bias may also occur as a result of introducing weighting systems to control variable size of respondents. The two key properties of rational expectations, on which the majority of empirical analyses of survey data are focused, are unbiasedness and orthogonality. We analyze several sample balance statistics and expectations series based on quantified survey data, taking into consideration issues of non-response and weighting schemes. Alternative definitions of expectations series aim to account for: 1) influence of arbitrary assumptions concerning weighting of individual data, 2) changing sample structure that results from non-response, 3) response rates varying with degree of optimism / pessimism of respondents. Results of our analysis indicate that expectations concerning relative changes in production are unbiased but not efficient with respect to freely available information, namely, observed relative changes in production (lagged three months) and expectations balance (lagged two months). This result holds for a range of weighting schemes and non-response issues analyzed, including changes introduced to sample structure by non-response, and increased inclination of "optimists" and "pessimists" to supply answers in the business tendency survey, as long as their shares remain constant in time. (original abstract