Causality in macroeconometrics: some considerations about reductionism and realism

Abstract

This paper investigates the varieties of reductionism and realism about causal relations in macroeconometrics. There are two issues, which are kept distinct in the analysis but which are interrelated in the development of econometrics. The first one is the question of the reducibility of causal relations to regularities, measured in statistics by correlations. The second one is the question of the reducibility of causes among macroeconomic aggregates to microeconomic behaviour. It is argued that there is a continuum of possible positions between realism and reductionism for both the questions, but, as far as the second question is concerned, the dominant position of mainstream macroeconometrics is strongly reductionist. The paper defends an integrative approach that emphasizes the gradual nature of many real world cases.

Keywords:

• causality
• realism
• reductionism
• econometric models
• identification problem

Notes

1. There are also important ontological questions as to what sort of things the relata of causality exactly are – whether they are events, facts, states of affairs, propositions, etc. – and questions about the relationship between the relata of causality and the relata of regularities, but we do not face this issue here, and we assume that the relata are just macroeconomic variables, without going more deeply into this issue; something more about this will be said in section 4.

2. This section is built upon Moneta (2004: section 1.3).

3. The illustration of the identification problem in this terms and notation follows King et al. (1991: 822).

4. In that case, Γ₀ can be easily obtained via the Choleski factorization of the covariance matrix of the estimated residuals Σ_ε.

5. In fact there is a somewhat circular reasoning in some Structural VAR econometric practices, as argued by Uhlig (1999): to obtain such stylized facts several alternative theoretical restrictions are used and the criterion of choice among them turns out often to be the conforming of the empirical results to the accepted background theoretical knowledge.

6. The Causal Markov Condition can also be seen as a graphical interpretation of the principle of the common cause, formulated by Reichenbach (1956). The principle of common cause says that if X_i and X_j are probabilistically dependent, then either X_i causes X_j , or X_j causes X_i , or X_i and X_j are effects of some common cause X_h .