Applied Macroeconomics
Consumption Theory
Panel Data
Applied Time Series

Macroeconomics
Econometrics

May 2000 (Orals) Macroeconomics (Consumption Theory) and Econometrics (Time Series)
May 1999 (Written) Microeconomic and Macroeconomic Theory (with distinction)

Aggregate Consumption: What U.S. States Have to Say

Professor Peter C. B. Phillips
Professor Robert Shiller
Professor Alexander Michaelides

Summer 2003

M.Phil., Economics, Yale University, May 2002
M.A., Economics, Yale University, May 2001
B.Sc., Economics (valedictorian, University of Cyprus, June 1998

Cowles Prize, Cowles Foundation for Economic Research, Yale University, May 2002 and 2000
Yale University Graduate Fellowship, 1998–2002
Presidential Prize of the Republic of Cyprus for Academic Excellence, June 1998
Awards of academic excellence, Department of Economics, University of Cyprus, 1994–1998
Silver World Medal of London Chamber of Commerce & Industry for Bookkeeping and Accounts, Spring 1991

Teaching Assistant, Theory of Income Determination and Fiscal Policy, Yale University, Spring 2002 and Fall 2000.
Teaching Assistant, Introductory Microeconomics, Yale University, Fall 2001.
Teaching Assistant, Graduate Macroeconomic Theory, Yale University, Spring 2001.

Summer 2001. Intern at the International Monetary Fund, IMF Institute. Research assistant to Era Dabla Norris for two projects: estimation of age-earnings profiles for Ghana, and collection of data for counties liberalizing their fiscal sector.
Summer 2000. Research assistant to Professor Robert Shiller, Yale University, for an econometric paper investigating stock market and home equity wealth effects for U.S. state consumption.
Summer 1998. Research assistant to Professor Panos Pashardes, University of Cyprus, for a project concerning economic development and the role of government.
Summer 1997. Research assistant to Professors Panos Pashardes and Theofanis Mamouneas, University of Cyprus for a project entailing an empirical cost analysis of the production sector of the Cyprus Tourist Industry.
Summer 1996. Research assistant to Professor Aris Spanos, University of Cyprus, collecting historical data to study the impact of the banking sector on the economy of Cyprus.

Korniotis G., 2002, “Catching Up with the Texans and U.S. State Consumption,” Working Paper, Department of Economics, Yale University

Korniotis G., 2002, “Differentiated Propensities to Consume: Evidence from the U.S. States,” Working Paper, Department of Economics, Yale

Korniotis G., 2000, “Where does it come from? A Simple Model of Differentiated Propensities to Consume”, Working Paper, Department of Economics, Yale University

Korniotis G., K. Kalli and C. Christofides, 2000, “Solar-Collector Industry in Cyprus: Technico-Economic Analysis and Future Perspectives,” Energy Studies Review, Volume 9, Issue 2.

Kalaitzidakis P. and G. Korniotis, 2000, “The Solow Growth Model: Vector Autoregression (VAR) and Cross-Section Time-Series Analysis,” Applied Economics 32, 739–747.

Chapter 1

Recently external habits have attracted a lot of attention. For example, Campbell and Cochrane (1999) use external habits to explain U.S. stock market data. However, there is no empirical evidence to support external habits at the aggregate U.S level. The focus of this chapter is to empirically determine whether the effects of external habits are strong enough to influence aggregate consumption. Using U.S. state level data, I find that indeed this is the case. I also show that Campbell and Cochrane (1999) made the right choices by using a difference model for their utility function and by including "keeping up" habits in their model.

My analysis utilizes a panel of U.S. state consumption data. State data are sufficiently aggregated that they can be considered "macroeconomic" data. At the same time, they have enough cross-sectional variation to allow us to define different measures of external habits. For instance, a measure of "catching up" external habit formation for the state of New Jersey could be the average of past consumption choices of all the states contiguous to New Jersey. The ability to differentiate past consumption choices of a state from its external habit measure is what allows us to differentiate between internal and external habits. Such differentiation is lost in U.S. aggregate data, where external and internal habits are observationally equivalent. This has been one of the major hurdles obstructing the application of aggregate U.S. data to test the "catching up" external habits hypothesis.

The economic model assumes that in a given state a representative agent chooses levels of non-durable consumption by optimizing an external habit formation model. External habits can take one of two forms. "Catching up" external habits are defined as a geometric average of past consumption choices of states with a common boundary to state in question. "Keeping up" external habits are defined as a geometric average of current consumption choices of states contiguous to the state in question. Both measures of external habits depend on choices made by economies "external" to the state in question. I also include local "catching up" habits measured by lagged state consumption. I do not include local "keeping up" habits, because they require data that are less aggregated than state data. Such data are not available. The model therefore generates two Euler equations that link the growth rate of consumption of each state either with the average growth rate of past consumption levels of boundary states ("catching up") or with the average growth rate of "current" consumption levels of boundary states ("keeping up"). I pool the state Euler equations to form the regressions to be estimated.

The resulting econometric models include fixed effects, a time-lagged dependent variable, and either a time-and-space lagged dependent variable ("catching up" habits) or a spatial lag ("keeping up" habits). In order to estimate my econometric models, the researcher is facing two challenges. On top of the bias that comes from the lagged dependent variable, the researcher needs to tackle the bias coming from external habits. The current econometric literature provides little guidance on how to address both sources of bias simultaneously.

Ignoring the habit variables, Hahn and Kuersteiner (2002) propose a bias-corrected least-square estimator that is consistent and asymptotically normal. The estimator is also efficient under homoscedasticity and normality of the cross-sectional error terms. Phillips and Sul (2002) conduct an extensive simulation study to compare different estimators of the dynamic panel, fixed effect model. They find that the estimator of Hahn and Kuersteiner has better properties than the GMM and the linear IV estimator. In light of these results, I extend the framework of Hahn and Kuersteiner to accommodate external habits. The new estimator corrects not only for the bias induced by the time-lagged dependent variable, but also for the bias induced by external habits. The bias of external habits is approximated and it is subtracted from the least-square estimator. The new estimator then becomes consistent and asymptotically normal. The current extension can be applied to any macroeconomic panel data set in which the cross-sectional units are expected to influence each other.

The pooled U.S. state regressions are then estimated by the new technique with quarterly and annual state data. In both data sets, external habits are significant, and the data lend credence to the "keeping up" habit model. Using the estimates of the "keeping up" model under quarterly data, I find that a 1% increase in U.S. GDP growth will induce a 0.47% average increase in state consumption growth. Half of this increase is due to the direct effect of U.S. GDP growth on state consumption growth and half is due to a state multiplier effect. Also, I find that 1% increase in the consumption growth of California (New York) will induce a 0.04% (0.07%) average increase in state consumption growth. Thus, ignoring the feedback between states, we will underestimate the aggregate effects of a policy that initially affects either the whole of the U.S. or a particular state. In all the regressions the estimate of the elasticity of intertemporal substitution is also found to be statistically different from zero. In addition, the data support a "keeping up" difference model, where the functional form of the utility function is close to that of Campbell and Cochrane (1999).

Chapter 2

In its original form, the Life-Cycle/Permanent Income model predicts that consumption is determined by the net present value of total wealth. The second chapter takes up this issue by estimating the elasticities of total consumption to different forms of wealth and testing whether they are equal. Such elasticities could be important for policy makers. For example, if the federal government intends to boost economic activity, then the effects of different programs on aggregate consumption need to be determined. Thus, it is valuable to identify whether a tax refund has the same outcome as a tax credit on capital gains or a tax credit on housing equity loans. The relevant consumption elasticities to address such a question are the ones with respect to income, stock market holdings, and housing-equity, which are estimated in this chapter.

I analyze quarterly U.S. state data for the period 1982-1999. Given the empirical evidence for external habits that I find in the previous chapter, I include a time-and-space lagged dependent variable in the regressions I estimate. The time-and-space lagged dependent variable refers to the average lagged consumption growth of the U.S. excluding the state in question. Then, I estimate the model using the extended Hahn and Kuersteiner procedure I developed in chapter 1.

Pooling all the states together, stock market wealth is found to be insignificant in explaining the growth rate of state consumption, whereas total income and home-equity turn out to be statistically significant. The elasticity for total income is estimated to be 9.4% while the one for home-equity is 1%. The results confirm the findings by Case, Shiller and Quigley (2001) in the presence of external habits.

Chapter 3

The empirical results in the previous chapter suggest that state-level consumption growth reacts differently to housing and stock market wealth. Such a result might arise under differentiated transaction costs for withdrawing funds from different forms of wealth to finance consumption. The literature on behavioral economics complements this explanation by introducing mental accounts. Mental accounts refer to a psychological system of "recording" transactions where the individual does not treat wealth as fungible. For example, a series of three basic mental accounts could include current disposable income, current assets, and future income. In this case, the marginal propensity to consume is taken to be account specific, since the individual is subject to a differentiated temptation to spend.

The third chapter introduces a model that gives rise to mental accounts by a fully optimizing individual. The model draws on the Behavioral Life-Cycle Hypothesis of Shefrin and Thaler. I modify the utility function to capture utility gains/losses from saving/dissaving. Such a utility function induces a consumption function with differentiated marginal propensities to consume out of different forms of wealth. The consumption function is almost identical to the "excess-sensitivity" model of Flavin. I show that Flavin’s marginal propensity to consume out of transitory income measures the relative importance of savings to consumption with respect to marginal utility.