Wealth dynamics: reducing noise in panel data

Although the asset data from the Health and Retirement Study (HRS) is of very high quality, there is sufficient noise to frustrate attempts to study saving behaviour by examining wave-to-wave change in wealth. In this research, we attempt to reduce noise by means of reactive-dependent interviewing in which respondents with large inexplicable changes in assets between 1998 and 2000 are called back by HRS interviewers, presented with their prior reports and asked to reconcile the data. We achieved reconciliation for 1255 households (2479 net-worth components) and, as a result, the variance in measured change for the entire sample of 11,583 households with the same financial respondents in both waves was cut in half. The empirical validity of the data also appears to have been improved. The correlation of gross change in net worth and income, for instance, increased from an insignificant negative to a highly significant positive value. Although reconciliation of large asset changes marginally improves the goodness of fit of multivariate models, there remains sufficient noise in the asset-change data to require analysts to employ additional methods to reduce the influence of outliers. Copyright © 2006 John Wiley & Sons, Ltd.