The Confidence Game
Late last month word came that the Consumer Confidence Index has fallen yet again. This seemed odd, for three reasons. Just a few days earlier, a different gauge, the Index of Consumer Sentiment, had suggested that confidence was beginning to recover. Moreover, actual retail sales had spiked in October — up 7.1% after having fallen 2.2% in September. Finally, this news of deepening consumer doubt came right after we’d all seen the footage on the evening news of American shoppers lining up at 3 a.m. for the annual after-Thanksgiving sprint through department stores to buy up discounted television sets and so on.
Admittedly, cheap auto-financing deals explained a good chunk of that retail sales jump, and those early-morning shoppers were snapping up bargains. But that shouldn’t matter if all you’re trying to measure are what John Maynard Keynes called "animal spirits" — and however you explain it, those consumers looked a good deal more spirited than they had in September. Nevertheless, the CCI stood at 82.2, the lowest point since 1994, so stocks obediently fell.
In the twitchy post-September 11 economy, consumer confidence has become the It Statistic. The idea is that with business investment shrinking and the markets on the fritz, free-spending shoppers had been the last bastion against a disastrous economic slide. With the administration exhorting consumers not to lose their nerve and keep spending for the sake of the commonweal, the Consumer Confidence Index boiled our seemingly unmeasurable animal spirits into an authoritative figure, which news anchors could blithely describe as "all important."
So what does that definitive-sounding 82.2 figure mean exactly? Where does it come from, who cooks it up, and how? Finally, does it really deserve all the attention it’s been getting? When you hear a number as specific as 82.2 being attached to a concept as vague as "consumer confidence," it’s a good idea to start asking questions.
* * *
The Consumer Confidence Index (CCI) is compiled by an arm of a nonprofit business-research organization (annual budget: about $50 million) with the pleasingly vague but officious-sounding name The Conference Board. A marketing-firm subcontractor, NFO Worldwide, conducts the underlying survey by mail. Although it's routinely described as a survey of 5,000 households, only about 3,500 generally return the form. This form essentially asks for a positive, negative, or neutral response to five questions about current and future business conditions.
So it's a poll. Polls have their place, of course, but simply reporting that X percent of Americans surveyed feel "positive" about business conditions doesn't really seem like the kind of news that should be dominating business coverage and roiling the stock market. After all, polls from Harris and Gallup also address basic consumer confidence issues, and they never make the same splash that the confidence indexes do. Which goes to show that when you're trying to numberize a slippery idea like sentiment, an index trumps a poll every time.
How does the Conference Board convert its poll into an index? By combining the responses to its five questions and converting the resulting figure into a composite number "relative" to a benchmark score of 100.0 for 1985. (Why does 1985 equal 100 on this scale? Because it was "a basic, bland year," explains Lynn Franco, director of the Conference Board's Consumer Research Center, offering some insight into the formal science of consumer confidence.) The upshot is a number that can be easily compared over time, and seems conclusive. In August consumer confidence was flying high at 114.3; following the terrorist attacks and three months of bad economic news, it had dropped to 82.2, a decline of 28 percent. See how scientific that was?
The Conference Board's chief rival in this confidence game is the University of Michigan's Index of Consumer Sentiment. In this case, 500 households are surveyed by phone, and the questionnaire is longer and more detailed. Nevertheless, the answers here are also boiled down to five categories and finessed into an index similar to the Conference Board's. (Michigan's "base" year is 1966.) Most of the time the two indexes more or less move in sync, but the Michigan survey, after dipping to 81.8 in September, rose slightly in October and November, and then again this month, to stand at 85.8.
Why do the two indexes show the confidence trend moving in opposite directions? It depends whom you ask. The Conference Board's Franco — after noting that the Michigan survey draws on a smaller sample — suggests the difference might be that two of her survey's five questions deal with employment, compared with only one of Michigan's. (This means, by the way, that when cnbc anchor Tyler Mathison exclaims, "Confidence is all about jobs," in the course of interviewing someone from the Conference Board, as he did recently, what he's articulating is not a fundamental truth the CCI has revealed about the economy, but rather a fundamental truth about the CCI's methodology.)
Meanwhile, Michigan survey director Richard Curtin — after noting that the Conference board's less-nuanced survey is outsourced and conducted by mail — suggests the difference might also stem from the fact that Michigan's future-looking questions have a one-year, not a six-month, time horizon, and that its inquiries about assumed spending-power take inflation, or the apparent lack of it, into account.
Both the University of Michigan and the Conference Board bluntly claim their surveys have predictive value — which is why they get so much attention. But the power of these assertions depends on how narrowly you define your terms.
In 1998 the New York Fed compared the two surveys and generally found the CCI to be more predictive of future consumption growth. (Not surprisingly, Franco pointed me to this study.) Interestingly, though, the study indicates that forecasts of consumption growth are more improved by factoring in only the Conference Board’s "expectations" component, rather then the broader index.
* * *
Although it doesn't get much attention, both the Conference Board and the Michigan indexes are each made up of two sub-indexes, one concerning present conditions and one concerning future conditions. In the case of the Conference Board's survey, respondents are asked to give a positive, negative, or neutral "appraisal of current business conditions" and "appraisal of current employment conditions." Answers to those two questions are numberized, benchmarked to 1985, and reported as the Present Situation Index. Three more questions ask respondents for a positive, negative, or neutral take on "expectations regarding business conditions," "expectations regarding employment conditions," and "expectations regarding their total family income." Those answers become the numerical Expectations Index. The two sub-indexes are combined to create the overall Consumer Confidence Index. (The Michigan survey does more or less the same thing, making one sub-index from its two present-focused categories, and another from its three future-focused ones, and combining the two to come up with its overall figure.)
A good example of how the sub-indexes get glossed over (despite being more predictive in at least some cases) came in the Conference Board's November report. It found that although the all-important CCI number was down, the Expectations Index had actually risen. This latter finding seems to make more sense, given the evidence — and it would jibe with the findings of the Michigan index. But it was largely ignored at the time. (Though it's worth noting that now the Conference Board can claim its findings are vindicated so long as things get either a) better, or b) worse.)
So why do the overall indexes get all the attention? Presumably because they cover more territory, encompassing the way consumers feel about today and tomorrow — never mind that the additional vagueness may make them less useful. Michigan, which has been running its survey since late 1946 (20 years before the CCI was launched), actually didn't begin pouring its data into a single, clean index number until 1952. Curtin explains that "the media didn't want to hear" some complex set of survey answers that were subject to interpretation; "they wanted to know, `Is [consumer confidence] better or worse?'" Curtin himself refers to the resulting index as "a communication device."
So the person doing the daily market wrap-up, hungry for a hook of the day, seizes on the tidbit that the big confidence numbers are at their lowest levels in six or seven years. But no one bothers to stop and ask: What predictive power does that suggest? After all, in 1993 and 1994, the economy was well into an expansion phase that was just about to get a good deal more expansive. Curtin says that the confidence figures didn’t really start its own runup until after unemployment peaked. "It kind of took a while for consumer confidence to rebound," adds Franco. The CCI hit an all-time high of 144.7 in January 2000, more or less in step with the stock markets. It started a decline then bounced up and down between 106.8 (in February of 2001) and 114.3 (in August). It’s hard to see any prescience in that. (The indices defenders say you have to look at the downward trend, but given the choppiness, that trend is a lot clearer in hindsight.)
* * *
Actually, what you learn when you figure out how those confident-sounding index numbers are pulled together is that the parts a lot more interesting than the sum. When Michigan's monthly number is announced, the release also highlights assorted nuggets culled from the phone survey —for example, consumers expect inflation over the next year to be its lowest since the 1950s and anticipate an unemployment rate of 6.5 percent — which are largely ignored in the press. Similarly, in addition to its vague multiple-choice questions, the Conference Board's mail survey also makes several specific queries about recipients' spending plans — whether they intend to buy a car (new or used), a house, or various appliances (TV set, refrigerator, etc.). Despite being more specific and, presumably, more predictive than responses to the general questions — after all, any given respondent probably has a better idea of whether she's going to buy a car in the next six months than she does of overall employment trends — none of this information winds up in the overall indexes.
Of course, none of this means that consumer confidence, as a concept, doesn't matter, or that it's not worth trying to gauge. But the importance the indexes have taken on lately is almost farcical. (Apparently stocks sold off after the most recent Conference Board announcement partly because economists "expected" the index to come in at 86.5; the idea that there are economists actually trying to predict what this figure will be is too dismal to dwell on.) The attraction of the all-in-one composite index numbers is not just that they seem to take everything into account, but that they're expressed so decisively — they sound like facts, like the sorts of figures that deserve a place next to weekly jobless claims, monthly nonfarm payroll reports, or quarterly GDP. But of all the information that Michigan and the Conference Board gather, that official-seeming number is probably the least interesting. Too bad it gets all the attention.
A version of this story appeared in the December 24, 2001 edition of The New Republic.
