U.S. consumers and the war in Iraq: the non-economics of consumer confidence

Business Economics, April, 2004 by Robert Keyfitz

Consumer confidence receives close attention as a driver of present and future consumer spending prospects. Understandably, concerns about job security and bonuses have an effect on behavior at the shopping mall, but attitudes also seem to reflect factors less immediately connected to the economy such as terrorist threats and 'war jitters.' For instance, the Conference Board's Consumer Confidence Index (1) fell by 25 percent after the terrorist attacks of September 11, 2001 and later soared by 32 percent with a quick end to major combat operations in Iraq (Figure 1). Such magnitudes are far in excess of the economic impacts that might reasonably have been anticipated and indicate the existence of an important psychological component.

It is useful to separate out economic and non-economic influences from each other. In all likelihood, these give rise to very different dynamics since economic conditions evolve slowly while terrorist threats, war jitters, accounting scandals, financial crises, and the like can appear and reverse quickly. Knowing whether confidence is consistent with underlying economic conditions should help to forecast future movements in confidence and hence consumer spending, or to highlight irrationally exuberant or despondent behavior.

Accordingly, this note has two aims. Both should be of independent interest to applied macroeconomists and forecasters. The first is to present a rigorous statistical methodology for decomposing volatility in consumer confidence into economic and non-economic components. Evidently, the task is not straightforward. While both types of influences likely played a role in the movements shown in Figure 1, it is not obvious how to disentangle their relative contributions. At least for the economic component there are indicators pertaining to conditions in the household sector, such as income growth, inflation, stock prices, and unemployment. But, for the non-economic component no such indicators exist. Unforeseen shocks become evident only after the fact, and even then it may not be clear how to measure them objectively for econometric analysis. In short, though the evidence suggests non-economic factors are important, modeling them explicitly must take an indirect route. Fortunately, techniques well suited to this sort of problem are available. To be specific, modeling consumer confidence as a state-space system--which, in a sense, defines the non-economic component residually in a sophisticated way--permits indirect inferences to be made. The next section provides a brief introduction to state-space modeling and then applies the technique to the problem at hand.

The second aim of the paper, hopefully interesting even to those not concerned about the technical details, is to assess the impact of the war in Iraq on consumer spending. Indeed, it was trying to understand the economic impacts of the war that initially led to the approach described here. Consistent with many other studies, a consumption function estimated in the third section finds a small, though statistically significant, link between consumers' confidence and consumption spending. This translates the shock to consumer confidence associated with the war into a cumulative fall in consumption spending of $40.5 billion (at 2000 prices) or 0.3 percent in 2002-03.

Measuring the Non-Economic Component of Confidence

The state-space approach has been developed over the past half century to deal with what is a relatively common statistical problem: the need to draw inferences about things that cannot be measured directly. Many statistical techniques including OLS, Bayesian estimation, and time series models such as ARIMAs and VARs can be viewed as special cases, but the approach is much richer. An example of this richness is the potential for allowing time-varying coefficients. State-space models are also the basis for the Kalman filter, a computationally efficient procedure for sequentially updating a forecast.

A simple state-space model consists of a state equation that describes the way some variable of interest moves through time, and a signal or observation equation that describes its relationship to some measurable occurrence. The observation equation may also depend on other observable inputs, in which case the state variable is sometimes referred to as an "unobservable component." In the present case, there is an observable signal (consumer confidence) which is the combination of an economic component and an unobservable, non-economic component.

The earliest applications of state-space models were to interpret radar signals and guide rockets, and the terminology hints at these origins. Suppose some object is moving through space. In this case, the state equation consists of the laws of physics that govern its movement, while the signal equation describes a radar signal that gives the current location subject to a random error. Over the past several decades, economists have adapted the technique to problems with latent variables such as expectations, trend growth, or the non-accelerating-inflation rate of unemployment (NAIRU), which, though central to theoretical or empirical models, cannot be observed directly. For instance, an early application by Fama and Gibbons (1982) used ex post, observed interest rates and inflation to make inferences about ex ante expected inflation. For a recent survey of other applications in macroeconomics, see Basdevant (2003). In more technical applications, state-space models have been used to fill in missing data, extract higher frequency series from lower frequency ones (e.g., monthly from quarterly), and adjust for seasonality or measurement error (Harvey, Koopman and Penzer, 2003).