A 2004 paper from the Banco de Espana not only proved highly prescient, but it increased our understanding of both consumption behaviour and the techniques to explore it.
Recently, I have covered research that used event analysis and statistical techniques to show how excess credit matters to other macroeconomic indicators. It was highly useful to see this at such a macro level, but what is also needed to understand credit’s role in the economy is a more detailed analysis of how it relates to consumption. This is precisely what was addressed by Carmen Martínez-Carrascal and Ana del Río from the Banco de Espana (hereafter BdE).
What were they looking at?
The BdE paper set out to address the lack of empirical literature on the potential impact that large levels of indebtedness can have on the economy. Not surprisingly, they looked at aggregate data for Spain, citing the rise in indebtedness in Spanish households. They noted that some level of increased credit is due to structural improvements, such as the liberalisation of Spain’s financial system, but they quite rightly suggested that this high level of debt left households vulnerable to shock.
How did they do it?
The authors used a Vector Error Correction model (VECM) to ask whether consumption reacts when household borrowing departs from it’s long run determinants. This is the advantage of VECMs over other types of models, as we are able to examine the long run fundamentals of a macro variable and assess how quickly things return to this long run equilibrium. They authors use a Johansen procedure for the ECM, rather than the Engle and Granger technique, as they are jointly modelling consumption and borrowing, which implies that they are looking for the possibility of more than one cointegrating vector.
What was found?
Consumption was positively related to income and also to lending, while borrowing was positively related to consumption, housing wealth and negatively to nominal interest rates. Interestingly, in the borrowing equation, they found that housing wealth was more important than financial wealth, most likely reflecting the notion that financial assets cannot be used as collateral by individuals.
Highlighting the advantage of the Johansen procedure, the analysis of the loading factors (the alphas) determine the dynamic adjustment towards long run equilibrium. Firstly, when lending is above its long run level, its equilibrium is restored via reductions in lending combined with a contraction in consumption. This result is the most fascinating of the whole paper, as the loading factor on consumption implies that disequilibrium in borrowing and lending can result in a lengthy contractionary effect on consumption. This result ties in very well with the concept of financial recessions, covered here and the evidence of deleveraging found by Mian & Sufi . In plain language it says that credit matters for consumption, especially when it’s above it’s long run level, and it’s corrective phase will have detrimental effects on consumption.
For consumption itself, positive deviations from its long run level are corrected by reductions in consumption as well as increases in future labour income.
The authors conclude by stating that some increase in fundamentals leading up to 2004 would explain the rise in borrowing, but from their analysis, it is likely that the borrowing level is (was in 2004) above that implied by these fundamentals. Given what is happening now in Spain, this looks like the correct assessment. What would be very interesting, if possible, would be to plug the actual values from 2004 into their model and to see by how much the 2004 level of borrowing was above its fundamentals.
I’ve often returned to an IMF paper that looked at Australian house prices which also used a VECM. This study was highly informative as it was able to suggest that Australian house prices were approximately 5-10% above their equilibrium levels. If we can use these VECM procedures to measure the disequilibrium in the credit market, it might be a more rigorous approach than simply using a statistical technique like the Hodrick-Prescott filter.
Personally, I will need to explore in more detail the methodology of the Johansen VECM procedure. For example, can I try and model consumption on its own, or should I model it jointly with borrowing like the BdE study? From my theoretical work , is it the consumption or borrowing function that is more relevant? In any case, it would seem an ideal tool to embed some assumptions from my point of view and to test their worth empirically.
 Tumbarello, P. & Wang, S. (2010). What drives house prices in Australia? A cross-country approach. IMF Working Paper.