An estimation of deep parameters describing Argentine consumer behaviour

Abstract

This paper investigates the microfoundation of consumer decisions in Argentina (1980:1–2001:3). Structural parameters are estimated following the Euler Equation–GMM approach. Attention is focused on parameter instability, an empirical difficulty for applying this method often pointed out in the literature. The rates of return on assets are approximated by the real interest rate and the rate of growth of real exchange rate as they have been considered as the main variables explaining variations of Argentine ‘wealth’. The results show that parameter estimates have the expected values and the correct signs. Overidentifying restrictions are tested and the null hypothesis of validity of instruments is not rejected. Estimates are also robust for different specifications of the weighting matrix. However, parameter constancy is jointly rejected. Recursive estimates show that the risk aversion coefficient appears as more unstable than the impatience parameter, which is also the less uncertain within sample. Observed changes in estimates seem to be an expected response to different macroeconomic frameworks.

Acknowledgements

An earlier version of this work was presented at the XXXVIII Reunión Anual de la Asociación Argentina de Economía Política on 12–14 November 2003. We would like to thank the comments received there. The views expressed herein are solely the responsibility of the authors and should not be interpreted as those of the Central Bank of Argentina.

Notes

 Since the exchange rate was nominally fixed under Convertibility, it was approximated in real terms by the ratio of wholesale to consumer prices given the greater weight of non-tradables in the last index.

 See Muellbauer and Lattimore (1995) and Favero (2001) for the introduction of liquidity constraints in the optimization problem and for the empirical question of assuming the behaviour of an individual for aggregate data.

 EViews was used. See user's guide for a description of each weighting matrix estimator.

 This statistic and the following ones are computed for the results obtained using the Variable Newey–West estimator of the weighting matrix (the less unrestricted form).

 Note that the more precise the estimates, the narrower the forecast interval and the higher the probability of rejecting the null of constancy. It could be a recurrent problem with GMM estimators since usual statistics reject the null (of zero coefficient) too often (see for example, Hayashi, 2000).