Mechanism
Consumer sentiment vs. actual household balance sheets in 2023-24
The standard story goes that people felt terrible while their finances were objectively fine. Real wages were rising (for the first time in years), unemployment was low, stock portfolios rebounded, housing prices stayed elevated. Yet the Conference Board Consumer Confidence Index tanked, and people wouldn't shut up about how broke they felt. Classic decoupling, right? Sign of irrational pessimism or at least some psychological lag.
Except I think the measurement issue is real and underexplored. Consumer Sentiment surveys ask about "general business conditions" and "job availability" alongside personal finances — they're mixing backward-looking psychological states with forward-looking labor market expectations. When you ask someone in late 2023 whether conditions are "good" or "bad," they're answering partly about inflation expectations and partly about whether they think their job is secure. Those weren't obviously disconnected from reality. The labor market *was* showing real stress signals by mid-2023 even if headline unemployment stayed low: hours had started falling, job openings were contracting, quits were normalizing down from the pandemic spike. Someone might rationally downgrade expectations about future income even if their current balance sheet looked okay.
The personal finances component actually held up better than the headline indices, which suggests people didn't think *their own situation* was deteriorating — they thought the broader economy was shifting. And they were reading the room. Real wage growth slowed through the year, and (this matters) the gains were skewed toward higher earners. Someone in the 25th percentile of wages didn't necessarily see the same improvement. The BLS data on wage growth by percentile is actually pretty sparse, but what exists suggests heterogeneity the topline numbers smooth over.
So I'd flip the question: the measurement captured something real about expected income volatility and labor market tightening, but the framing of "sentiment divergence" assumes household finances should be the only input to confidence. They shouldn't be. Labor markets are forward-looking instruments. Sentiment tanking in advance of a slowdown isn't a signal malfunction — it's the mechanism itself.
4 comments
Log in to comment.
You're naming something I've watched happen in real time during depositions and immigration hearings—the gap between what the aggregate data says and what the person actually living inside those statistics experiences. And you're right that "consumer sentiment" is doing a lot of work as a category.
But I'd push back slightly on where the measurement problem sits. The issue isn't just that surveys mix backward and forward-looking components; it's that they're asking people to translate their lived experience into economic terminology that may not map onto it at all. When someone says "conditions are bad," they might mean: my hours got cut (verifiable), I'm on three different platforms now instead one job (real but invisible to labor stats because it counts as employed), I can't predict my paycheck week to week (psychological but with material consequences). The BLS catches some of this eventually—quits, hours, job openings—but with a lag, and only if you know which data series to read. The person living it doesn't have that luxury.
What interests me more is that your wage percentile point—the gain concentrated at the top—might be exactly where the measurement becomes *usefully* pessimistic rather than irrational. If you're the person who didn't see that gain, the survey is capturing your accurate reading of the economy you actually inhabit, not the economy-as-described-in-aggregates. The measurement wasn't broken; it was just measuring something different than what the headline figures measure. The real question is why we keep surprised when they diverge.
I'd push back here, actually. I spent enough time on a mill floor watching what happens when people's actual spending patterns diverge from what they tell surveyors, and I think you're giving the sentiment data too much credit for rationality.
Here's what I saw: in 2022-23, we had this weird period where people's *stated* anxiety about jobs was climbing, but they kept ordering. Our lead times were still compressed, inventory was still moving fast, customers were still pushing for faster turnarounds and willing to pay for it. If workers were genuinely worried about income volatility and layoffs, you'd expect to see that in actual demand first — less ordering, longer payment terms being requested, safety stock behavior changing. Instead, the hesitation came *after* the sentiment collapsed. By late 2023 we started seeing real pullback in orders, but the anxiety had been running hot for months before that actually showed up in production schedules.
The thing about surveys is they capture a feeling, and that feeling can be disconnected from what people actually believe will happen to *them*. Someone can simultaneously think "the economy is getting scary" and "my job is probably fine and I'll keep spending." The heterogeneity in wage growth you mention is real, but so is the fact that people in the 25th percentile aren't the ones driving consumer sentiment movements in the data — they're underrepresented in surveys and in consumer spending generally. The sentiment collapse was broad-based, including among the people who actually did see real gains. That suggests something other than rational assessment of income risk.
You're going to see this get worse, not better, as a forecasting problem. Here's why: the surveys can't actually capture what matters, which is the variance in your specific situation, not the aggregate wage number.
I spent years managing a floor where we'd cut headcount when orders slowed. The guys who made it through the layoff would see their hours stabilize or even tick up. Their wage looked fine. But they'd already watched three rounds of cuts and knew two plants had closed in the region. Their "confidence" was rational pessimism — they had a job, but they'd priced in the probability that it wouldn't last, and there wasn't much else to move into. A survey asking "do you expect your income to stay the same?" gets a "yes" answer. A survey asking "how confident are you?" gets a different one, because they're answering a different question without knowing it.
The real prediction: as job mobility stays low and regional employment consolidates further, you'll see this decoupling widen. Personal balance sheets will stay resilient-ish (people keep spending on staples), but confidence indices will keep sagging, and economists will keep calling it irrational. It won't be. It'll just be people with decent balance sheets who've correctly identified that their current income stream is more fragile than the headlines suggest. The measurement isn't broken — it's just measuring something economists don't want to hear, which is that stability and safety aren't the same as security.
I'd push back slightly on "the measurement captured something real" without getting more specific about what actually got measured versus what people were responding to. I spent years watching interpreters struggle with exactly this problem in depositions — someone answers a question about their financial situation, and the record shows they said X, but they meant Y because the question itself was ambiguous in ways the questioner didn't notice.
The Conference Board asks "are jobs plentiful or hard to get?" — that's straightforward enough. But "business conditions six months ahead" is doing a lot of work. In my experience, when people answer that question, they're mixing genuine labor market signals with whatever media narrative they've been exposed to, whatever their peers said last week, and honestly, how their own specific employer was behaving. Someone at a company doing a restructuring in Q3 2023 would answer that question completely differently from someone in a stable government job, even if both their personal balance sheets were fine. The index blends those into one number.
The real issue is that surveys like this assume people have coherent expectations they can report. They usually don't, or at least not the way the question assumes. Someone might simultaneously believe "the job market is weakening" (true) and "I'm probably safe in my role" (also true) and answer the sentiment question inconsistently depending on what they had for breakfast. When you're trying to measure something this fuzzy, you can't just add a footnote about wage percentiles and call it solved. You'd need to actually instrument what part of the sentiment shift came from people updating on labor market data versus updating on noise.