From Haver.com:
Michigan Consumer Sentiment Down Yet Again
April 11, 2008
By Tom Moeller
- The preliminary reading of April consumer sentiment from the University of Michigan fell another 9.1% m/m to 63.2. Consensus expectations had been for a lesser decline to 69.0. The decline dropped sentiment to near its lowest level since 1982.
- The expectations component accounted for the largest part of the decline in April sentiment with an 11.1% m/m drop. The index is at its lowest since 1990. Expectations for personal finances fell out of bed with a 13.4% m/m drop (-23.6% y/y). Expectations for business conditions during the next year also fell a hard 8.7% (-51.7% y/y). Expectations for conditions over the next five years fell 11.1% m/m (-20.0% y/y).
- Opinions about government policy, which apparently influence economic expectations, fell 4.4% m/m (-28.6% y/y). The percentage of those surveyed who indicated that they thought government was doing a good job fell to the lowest level since 1992 and 43% had a poor opinion.
- The expectations component accounted for the largest part of the decline in April sentiment with an 11.1% m/m drop. The index is at its lowest since 1990. Expectations for personal finances fell out of bed with a 13.4% m/m drop (-23.6% y/y). Expectations for business conditions during the next year also fell a hard 8.7% (-51.7% y/y). Expectations for conditions over the next five years fell 11.1% m/m (-20.0% y/y).
- Opinions about government policy, which apparently influence economic expectations, fell 4.4% m/m (-28.6% y/y). The percentage of those surveyed who indicated that they thought government was doing a good job fell to the lowest level since 1992 and 43% had a poor opinion.
- The mean expectation for inflation during the next twelve months rose again last month to 5.6%, nearly the highest level since 1990.
- The current conditions index fell 6.9% m/m after a slight rise during March. The view of current conditions for buying large household goods fell hard (-24.3% y/y) and the view of current personal finances also fell sharply (-26.9% y/y).
- The University of Michigan survey is not seasonally adjusted. The reading is based on telephone interviews with about 500 households at month-end; the mid-month results are based on about 300 interviews….
Here’s the series over the past ten years (more at the Haver website).
Figure 1: University of Michigan: Consumer Sentiment
Index, not seasonally adjusted, Index 1st Quarter 1966=100. Source: St. Louis Fed FREDII.
See additional commentary by Jim Hamilton at the time of the announcement, and his post on the Conference Board indicator.
Given the startling nature of this release, one is driven to ask, do sentiment indices matter? From S. Ludvigson, 2004, “Consumer Confidence and Consumer Spending,” Journal of Economic Perspectives 18(2):
Despite the widespread attention given to surveys of consumer confidence, the
mechanisms by which household attitudes influence the real economy are less well
understood. Do consumer confidence surveys contain meaningful independent
information about the economy, or do they simply repackage information already
captured in other economic indicators? Do the surveys provide information about
the future path of household spending, or do they reflect current or past events?
Finally, do the surveys correspond neatly to any well-defined economic concept, or
do they furnish only a nebulous barometer of household disposition?
This paper begins with an overview of how consumer confidence is measured
and reported. It then evaluates what is known about the relationship between
consumer attitudes and the real economy. The evidence suggests that the most
popular survey measures do contain some information about the future path of aggregate consumer expenditure growth. However, much of that information can
be found in other popular economic and financial indicators, and the independent
information provided by consumer confidence predicts a relatively modest amount
of additional variation in future consumer spending. Moreover, there is some
evidence that consumer confidence surveys reflect expectations of income and
non-stock market wealth growth, but evidence on the connection between these
surveys and precautionary saving motives is mixed.
A more adamant conclusion regarding consumer surveys and future consumption comes from Dean Croushore:
The conjecture that began this article seemed sensible: The use of real-time data might have a better chance of showing that measures of consumer confidence could prove useful in forecasting. After all, the measures of consumer confidence could reflect what people know that has not yet been captured by government statistical agencies. However, in trying to predict consumer spending, evidently the measures of consumer confidence reflect other events affecting the economy and do not sufficiently tell us what people know that government statistical agencies do not know.
The bottom line: If you are forecasting consumer spending for the next quarter, you should use data on past consumer spending and stock prices and ignore data on consumer confidence.
So does this mean one should ignore these indices? No, none of the studies argues that particularly extreme point. Rather, the consensus seems to be that the consumer sentiment indices summarize extant data. To the extent that this is an accurate assessment, the following bit of news is not surprising. From Macroeconomic Advisers (as noted by RealTime Economics), here’s the latest estimate on February real GDP:
Figure 2: Log real GDP, in billions of Ch.2000$. Source: Macroeconomic Advisers, accessed 15 April 2008.
RealTime Economics summarizes it the best:
Forecasting firm Macroeconomic Advisers updated its monthly GDP estimate for February, showing a 1.2% decline in February. It was the second largest one-month decline in the nearly 16-year history of the index, behind a 1.6% drop in September 2001.
Technorati Tags: consumer sentiment indices,
consumer confidence,
consumption, forecasts.
With respect to the “expectation for inflation,” I submit that the respondents have now arrived at a clear understanding of the intentions of Benjamin Bernanke.
This is the moment when this con man’s con has been fully busted.
The evidence of inflation is all around us, the man (and woman) in the street sees it and feels it, and there simply is no lie sufficiently sophisticated to obscure the reality.
The problem is that it is our society (and civilization) that is being degraded,
and we have nine more months to wait before the new Obama Administration (now advised by Paul Volcker) can demand an immediate and ignominious exit.
One would be wise to take note of the unprecedented and not-so-polite “trial-balloons” emanating from Volcker regarding Bernanke.
BizLinks | 4.17.08
Continental posts loss, Southwest profit shrinks Texas coast misses out on insurance drop Did Ending Regulation Help Fliers? Cash for ‘Bumped’ Fliers Doubled DAL/NWA: Proposed Merger So Large, Yet It’s Worth Less Than Yahoo Kryptonite Caused GE’s Miss…
Menzie, has anyone constructed a correlation table? Looking at Croushare’s piece, his baseline model used past consumer spending, income, changes in the interest rate, and changes in the real value of the stock market for prediction and evaluated the gain to RMS forecast error.
If there is multicollinearity, that might not be a good way to do it. A correlation table helps to identify multicollinearity and decide what the best combination of fundamental variables is.
Charles: I’m sure he did along the way. See the technical version of the paper, here. Note, for the key question of whether the consumer sentiment index helps in reducing the RMSFE relative to a set of macro variables, multi-collinearity between the macro variables does not necessarily diminish forecasting accuracy.
Greenspan sometimes told Congress that his concern wasn’t reported consumer confidence, but actual consumer confidence. That was pretty obscure, which allowed me to given it any interpretation I wanted. Mine was this – Greenspan was saying that when consumers bring their wallet into alignment with their mood, their mood matters. Do we have any evidence on the circumstances under which US consumers make sudden (1 to 2 quarter) changes in spending behavior? Being debt-strapped and loosing access to mortgage equity withdrawal sound nice, but I’m not sure we’ve ever had this set of circumstances before. All this consumer debt is a relatively new thing, at least since the days of debt slavery and Jubilee.
It seems to me there is a real problem with these surveys if they are simply random samples of the population, for any such survey will ignore the effect of income inequality.
In the USA the top 5% recieve 31% of all income, the bottom 60% 26% of all income. Hence any random survey will overweight the views of those with a far lower purchasing power than the very rich. This would seem to be represented in the Michigan survey which shows a pretty continuous downward trend since the turn of the millennium, a period in which the incomes of the average person have stagnated, in stark contrast to those of the very rich.
Haver’s posting on April 11 of the high correlation between consumer confidence and personal consumption expenditures — especially over the last year — sure convinced me to pay attention to the series.
Menzie says, “multi-collinearity between the macro variables does not necessarily forecasting accuracy.”
Could you spot me a verb, Menzie? “improve,” say, or “diminish”?
For whatever reason, the SSRN server is being obnoxious, so I’m not able to download Croushore’s paper. But I’m not at all certain that he did the analysis. I find so many people using ad hoc approaches to selecting and rejecting variables.
Constructing correlation tables between variables to maximize orthogonality ahead of the analysis is a hobbyhorse of mine. I think that the best analysis is the one that uses the fewest variables while providing maximum explanatory power/variable, and this is a perfect way to achieve that before running the analysis.
The index will have more or less importance depending on the industry looking at it. I would not be anxious to increase production of automobiles or start new housing developments in the second quarter since those are easily deferred purchases and major commitments.
Charles: Sorry. “diminish” was missing, and I’ve inserted it now.
Thanks, Menzie.
I’ll see if I can get Croushore’s piece to download tomorrow.
Ok, Menzie, having looked at Croushore’s paper, it looks as if he is using ad hoc methods. Basically, he re-ran the model using fewer parameters (and as a linear regression) to see if it fits better. While the effort to simplify the model is commendable, it’s neither systematic nor a priori.
Constructing a correlation table allows one to qualitatively break down the explanatory power of a model before even running it. For example, we might find that incomes explain 30% of consumer spending, but that they also explain 90% of consumer confidence. In that case, adding in consumer confidence is obviously a case multicollinearity.
Multicollinearity is dangerous not because of its effect on explanatory power, which is often negligible, because it increases the possibility of finding false interpretations. It can reduce the correlation of the variables that are actually important, even to the point where they seem irrelevant.
I’ll drop him a note to suggest the idea of constructing a correlation table.
Charles: If parsimony is in your objective function, then I understand why you want to do what you want to do. If forecasting accuracy is your objective, aside from identifying economic motivations, one might still want to do what Croushore did.
There’s no fundamental conflict between the approaches, Menzie. As I wrote to Croushore, one can add in variables until one achieves the desired explanatory power, which roughly equates to the forecasting accuracy you desire.
I’m sure you’re aware of cases in which modeling has failed. It’s not very comfortable being the author of a paper that fits the data brilliantly today and a couple of years later proves to be completely wrong. (Or worse, using modeling for wrongful ends, like selling one’s talents to the coal industry to rationalize doing nothing about global warming.) How do models fail? Multicollinearity is a central issue.
Being right, rather than achieving parsimony per se, is what should be the objective function of us all.