Reconciling Consumer Confidence and Permanent Income

Kevin H. McIntyre, Western Maryland College

 

Objectives:

The forecasting power of consumer confidence indices for consumption spending is in sharp contrast to the predictions of the permanent income hypothesis (PIH). This paper attempts to reconcile these seemingly divergent propositions by developing a “confidence augmented'' permanent income hypothesis (CAPIH).

 

Background:

Consumer sentiment and its impact on the macroeconomy is a topic receiving increasing attention from economists, policymakers, journalists, and financial analysts alike. The basic argument suggests that consumer sentiment, measured using indices created from household surveys, is an important determinant of not only present, but also future household consumption expenditures. On a cursory level, the evidence supporting this story is quite compelling. Indeed, household consumption has tended to slow following a dip in consumer sentiment and vice versa for decades (Abderrezak, 1997).

           

Evidence also exists on a not-so-cursory level. A recent study by Bram and Ludvigson (1998) shows that, depending on the confidence measure used, consumer confidence has significant predictive power when forecasting household expenditures, particularly consumer durables. Danthine et al. (1998) take this a step further, linking consumer sentiment to agents' stochastic expectations concerning long term labor productivity growth and hence permanent income. Likewise, Matsusaka and Sbordone (1995) have shown that consumer sentiment accounts for approximately 20% of the business cycle innovation in postwar U.S. GDP. Accordingly, consumer sentiment affects not only the intensity of business cycles, but also their duration.

 

As an explicit measure of forward-looking behavior on the part of households, it is very important to understand its relationship to the permanent income hypothesis (henceforth PIH). In many ways, consumer sentiment fits naturally with the PIH. It reflects the macroeconomic conditions such as the jobless rate, equity prices and interest rates (among other things), and is thus a reflection of permanent income. As a coincident indicator, it is highly correlated with the present state of the economy and is good predictor of the future income and the overall strength of the economy. Moreover, the predictive power of consumer sentiment exceeds what is already found in other macroeconomic variables (Acemoglu and Scott, 1994).

 

Along one very important dimension, however, consumer sentiment is not consistent with standard versions of the PIH. Under the PIH, consumer sentiment (and other variables such as current income, inflation and unemployment rates, etc.) should not have any predictive power for consumption. This, however, is not the case empirically. Studies looking at sentiment often reject the PIH on the grounds that consumer confidence is a good predictor of consumption and/or that consumer confidence Granger causes consumption (Acemoglu and Scott; Matsusaka and Sbordone, 1995).

 

This paper attempts to reconcile the predictive power of consumer sentiment with the permanent income hypothesis by developing a “confidence augmented” permanent income model by introducing consumer confidence as a shift term to household preferences.

 

Data:

The consumption data used in this paper is N.I.P.A. consumption is of non-durable goods and services. Income is measured using quarterly chain-weighted N.I.P.A. personal income.  Both consumption and income are converted to per-capita terms. Four alternate measures of consumer confidence are considered: the Conference Board's index of consumer confidence, the Conference Board's index of consumer expectations, and their respective University of Michigan equivalents. The Conference Board's index of consumer confidence is selected as the baseline sentiment index; it has been shown that Conference Board indices outperform Michigan's in terms of both explanatory power and forecasting ability, even though the Michigan indices have a considerably longer history (Bram and Ludvigson, 1998).

           

A variety of interest rates and asset returns are also considered. The primary interest rates are either the yield on 3-month commercial paper, the 3-month Treasury yield, or the 90-day CD rate. Stock returns are measured using the Standard & Poor's 500 index. Since it is reasonable to assume that households formulate consumption plans taking into consideration the returns to all manner of assets they hold, I also construct an “effective household interest rate,” a weighted average of the above yields in addition to the return to real assets, i.e. housing, calculated using the National Association of Realtors' median house price series.  Nominal interest rates and asset returns are converted to (ex-post) real rates using 4-quarter changes in the Consumer Price Index. Finally, the sample period is 1978:Q1-1998:Q4, the former date corresponding to first release of the Conference Board indices at monthly frequencies.

 

Results:

Preliminary evidence suggests that that so including consumer confidence does not significantly alter the predictions of standard PIH models as far as the degree of permanent income spending undertaken by households is concerned. In contrast, including consumer confidence has significant implications for intertemporal substitution: adding consumer confidence typically decreases estimates of the intertemporal elasticity of substitution by up to a factor of 10. Further analysis shows that these results are largely invariant to the measure of consumer confidence used, minor structural alterations to the baseline model, and the choice of instrumental variables.

 

Return to Session E00-1 | Return to Preliminary Program