New Micro-foundations for Macroeconomics
"Household Decisions, Credit Markets and the Macro-economy: Modelling Issues"
John Muellbauer (Oxford)
Friday 10 July, 10.30 - 11.15
Abstract
The econometric models used by central banks and the mind-set they represent have contributed to policy failings in dealing with the global economic crisis and also to policy errors in coping with past economic expansions. This theme was taken up Don Kohn in a November 2008 speech:
“The challenge is to improve our understanding of the linkages between the financial sector and real activity. The recent experience indicates that we did not fully appreciate how financial innovation interacted with the channels of credit to affect real economic activity--both as credit and activity expanded and as they have contracted. In this regard, the macroeconomic models that have been used by central banks to inform their monetary policy decisions are clearly inadequate.
These models incorporate few, if any, complex relationships among financial institutions or the financial-accelerator effects and other credit interactions that are now causing stresses in financial markets to spill over to the real economy. Rather, these models abstract from institutional arrangements and focus on a few simple asset-arbitrage relationships, leaving them incapable of explaining recent developments in both credit volumes and risk premiums.
Economists at central banks and in academia will need to devote much effort to overcoming these deficiencies in coming years”.
Charles Goodhart made similar points at the September 2008 BOE Round-table. Even Jordi Gali admitted at the European Area Business Cycle Network conference in November 2008 that the New Keynesian-DSGE models had nothing to say about the current crisis.
Underlying conceptual reasons for model failure include representative agent assumptions, the assumptions about information and efficient markets, the treatment of uncertainty in these supposedly stochastic models, and the linearization techniques used to solve them. Calibration/estimation methods currently in use typically ignore inconvenient truths.
This paper provides empirical evidence/literature discussion, focusing on consumption, credit and asset prices in separate sections. Section 2 revisits a 1990 debate with Mervyn King and Marco Pagano on the causes of the consumption boom of the 1980s. Many, including Attanasio and Weber(1995) and Attanasio et al (2009), have argued that the main cause of the rise in the ratio of consumption to income in the 1980s was an exogenous shift in income growth expectations. I show using a mix of non-parametric and parametric methods that, for any plausible discount rates, neither perfect foresight models nor plausible econometric models for income growth expectations with any data coherence could account for the rise in the ratio of consumption to non-property income in the 1980s. This contrasts with explanations emphasising credit supply shifts, the relaxation of down-payment constraints and their interactions with housing collateral and other variables such as interest rates and income growth expectations as suggested in research with co-authors Janine Aron and Anthony Murphy summarised in Muellbauer (2007).
Section 3 examines research by Attanasio et al (2009). This examines micro data from the Family Expenditure Survey from 1977 to 2002. The authors argue that the expected correlations of consumption with house prices by age (and tenure) implied by housing wealth effects are contradicted by the data. Using the consumption residuals for young, middle aged and old households generated by Attanasio et al, I show that these are entirely consistent with the predictions of consumption differences by age implied by our own research. This implies differences by age in the response of consumption to changes in interest rates, unemployment rates, income and housing collateral. The evidence strongly supports a housing collateral but not housing wealth interpretation and is consistent with a direct credit channel role for interest rates and a role for precautionary saving. The evidence is also consistent with measurement error in consumption data at the micro level.
Section 4 turns to the centre-piece of standard DSGE models, the consumption Euler equation for the representative consumer. The hugely influential paper by Hall (1978) explained the martingale property predicted by the rational expectations permanent income theory of consumption. An inconvenient truth ignored by most of the DSGE literature is the ‘excess sensitivity’ of consumption growth to predictable income growth established by Campbell and Mankiw (1989, 1990) for major OECD countries. I show that using income forecasting models for the UK, US and Japan with plausible economic content (for example, introducing roles for monetary and fiscal policy variables, including a Ricardian element for the latter), that excess sensitivity is strongly confirmed, and cannot be explained by habit persistence.
Section 5 considers a recent ‘fix’ of the DSGE approach by Iacoviello (2005) and Iacoviello and Neri (2008) which introduces a housing market with a simple financial friction into the DSGE framework. In their paper, the budget constraint says that consumption + housing purchase=labour income + expansion of mortgage debt
i.e. consumption = labour income + mortgage equity withdrawal (MEW).
But if the maximum loan to value ratio is significantly less than 1 (calibrated at 0.85 by the authors), then housing purchase is almost always more than rise in debt (unless extraordinary house price appreciation is expected) and so MEW is almost always negative, and so inconsistent with US and UK data.
Other weaknesses, not in order of importance, include the following:
1. A closed economy is assumed so one group of households lends to the rest and to firms and so misses large US current account deficits – lending from the rest of the world via the financial services industry.
2. The papers miss amplification from an explicit financial sector, as discussed by many articles on financial crises e.g. Bernanke (1983 AER) or Brunnermeier (2008).
3. Credit rationing driven by an exogenous taste difference and an unexplained LTV limit misses micro foundations in asymmetric information and misses the shifts over time in credit technology.
4. A fixed fraction of credit constrained households is assumed. This misses time-varying expectations of income, capital appreciation and uncertainty.
5. The model also misses the life-cycle: young households need to borrow but have to save first for their initial housing deposit. Their saving should depend on both the LTV constraint and level of real house prices. There is no role for this.
6. Precautionary saving also missing.
7. No default risk – mortgage foreclosures in US are now at record levels.
8. Housing preference shocks play an implausibly major role in explaining real house prices and housing investment.
9. There is no role for extrapolative expectations of house price appreciation or other housing market inefficiencies, thus ignoring the large literature on housing market inefficiency. Extrapolative expectations are one major amplification mechanism in the financial accelerator. Empirical evidence for their importance will be discussed.
In the light of these empirical findings and discussion, the concluding Section 6 will consider possible ways of making progress towards central bank policy models useful for short to medium term policy analysis. Endogenising the financial sector with some realism is a necessary step towards better models. The Bank of England’s developing RAMSI model is making some progress towards building in the feedbacks to the financial system and back to the real economy. But such approaches need to be incorporated in a larger model encompassing the financial accelerator and the more conventional macroeconomic linkages.