In many developing countries, the increasing public interest for economic inequality and mobility runs into the scarce availability of longitudinal data. Synthetic panels based on matching individuals with the same time-invariant characteristics in consecutive cross-sections have been proposed as a substitute to such data - see Dang and Lanjouw (2014). The present paper improves on the calibration methodology of such synthetic panels in several directions by: a) explicitly assuming the unobserved or time variant determinants of (log) income are AR(1) and relying on pseudo-panel procedures to estimate the corresponding auto-regressive coefficient; b) abstracting from (log) normality assumptions; c) generating a close to perfect match of the terminal year income distribution and d) considering the whole mobility matrix rather than mobility in and out of poverty. We exploit the cross-sectional dimension of a national-representative Mexican panel survey to evaluate the validity of this approach. With the median estimate of the AR coefficient, the income mobility matrix in the synthetic panel turns out to be close to the genuine matrix observed in the panel. However, this need not be true for extreme values of the AR coefficient in the confidence interval of its estimation.