Page 74 - Invited Paper Session (IPS) - Volume 2

IPS184 Sanvi Avouyi-Dovi et al.
popular demand functions are the generalised Leontief functions (Diewert,
1971), the translog function (Christensen et al., 1975), the Rotterdam model
(Theil, 1965 and 1975a and b) and the Almost Ideal Demand System (AIDS,
Deaton and Muellbauer, 1980 a and b).
One of the advantages of both the AIDS and Rotterdam models is that
they have recourse to theoretical restrictions that are statistically testable
rather than impose them on the functional form. Barnett and Seck (2008) show
that there is no definite evidence of the predominance of the AIDS models
over the Rotterdam models in terms of explanatory power or forecasting
performances. For example, the two classes of models perform well when
substitution among goods is low. Also, according to their findings, the full
non-linear AIDS models by Deaton and Muellbauer (1980 a and b) perform
better than the original Rotterdam models in certain cases, whereas the
Rotterdam ones clearly exceed the AIDS models in other cases. Furthermore,
by comparing the AIDS and Rotterdam models using a model selection
procedure (Amemiya, 1980, 1985), Erkan (2006) finds that the AIDS model
displays better performances than the Rotterdam model.
As the aim of the paper is not to compare different demand systems, our
selection of the demand system is guided by the best practices in this research
area. Thus, the main guideline of our study is given by the robustness and
stability of the empirical results from previous studies. As mentioned before,
the AIDS models deliver relatively good performances under some conditions
in many empirical studies. The appropriate framework should be clearly micro-
founded in order to avoid some spurious relations. Finally, it should be
empirically tractable and intuitively understandable. The AIDS model meets
these requirements and is a good tradeoff between the different classes of
demand systems.
In this paper, following Blake (2004), we apply a financial AIDS (FAIDS)
framework to French households’ portfolio choices. This framework is an
extension of the seminal AIDS model of Deaton and Muellbauer (1980) to the
financial portfolio. Such a model is compatible with the analysis of households’
portfolio choices based on the neoclassical demand theory. In particular, the
analysis of cross-interest rate elasticities enables us to assess substitution or
complementarity effects between the different asset shares in households’
portfolios.

2. Methodology
We aggregate the individual financial assets to distinguish six categories
of assets within French households’ total financial wealth (Banque de France,
2005; Bachellerie et al., 2016). Thus, the share of asset “i, i=1,..., 6” represents
the relative weight of this asset in total financial wealth (M1, M2M1, M3M2,
PEL, ASSETS, LIFEBONDS). The six shares are the following (Chart 1):

61 | I S I W S C 2 0 1 9