Measuring savers' preferences how and why?

Luc Arrondel (1, 2, 3) , André Masson (1, 2)

1 PSE - Paris-Jourdan Sciences Economiques
2 PSE - Paris School of Economics
3 Banque de france

Measuring individual preferences of savers have two main motivations: to reduce the share of non-observed heterogeneity in explaining households' wealth behaviors, and to construct more accurate tests of the theories of savings and portfolio choices. For France, we have constructed a unique data set, the PAT€R surveys (PATrimony and Preferences vis-à-vis Time and Risk), with the first survey made in 1998 by interview with Insee, and the following surveys conducted by Tns-Sofres via postal questionnaires: PAT€R-2002 focused on the transmission of preferences between parents and children; the three subsequent waves, PAT€R-2007-2009-2011, were conducted in May 2007 (before the crisis), June 2009 and November 2011, on more than 3600 households representative of the French population, with an important panel dimension - about 2000 households were interviewed at least twice, and nearly 1100 households were present in all three waves. Almost each survey replicates direct measures of risk attitudes as well as time preference found in the literature, such as the hypothetical lottery on income of Barsky et al. (1997), selfreported 0-to-10 scales concerning the willingness to take risks in various contexts (Dohmen et al., 2011), or even experimental measures. Yet these measures, based upon a limited number of questions, suffer from serious drawbacks that are analyzed in the paper. This is why we have developed an alternative method of scoring in order to measure three types of preferences, towards risk, time, and offspring (altruism) : synthetic and ordinal scores of preferences are derived from the responses given to a large number of (mainly) simple real-life questions covering different life domains of life (consumption, leisure, health, investments, work, retirement, family). No single question is satisfactory by itself, but the idea is that if a number of questions possess a common dimension, towards risk for instance, this dimension will be isolated by aggregating the replies, beyond framing effects, measurement errors, etc.: the score obtained will then show enough internal consistency, as measured by psychometric tests. Ultimately, the data will thus determine how many indicators - if any - should be introduced for each type of preference. On the five surveys, this scoring method has proved very consistent and robust. Derived from some 60 questions, one single score of preference towards risk was sufficient in each wave (with Cronbach's alpha near 0.7). As regards time depreciation, two scores were each time necessary, marking the contrast between the short term and impulsive choices ("short-term impatience") on the one hand, and the long term and more reasoned decisions "time preference" over the life cycle) on the other. One single score of altruism was retained. he contents of these four scores are very similar from one wave to the next, as well as their individual determinants, their individual correlation over time, the degree of correlation between scores, etc. Finally, the effect of these scores on the level and composition of wealth are similar across surveys, in line with theoretical predictions, and more significant than those obtained for alternate measures of preferences which suffer, moreover, from endogeneity biases.

Mots clés: Preferences; Saving; Wealth; Inequality

Domaines: Economies et finances

Luc Arrondel, André Masson. Measuring savers' preferences how and why?. 2013. ⟨halshs-00834203⟩