What the U.S. Business Cycle Looks Like

In brief: Investment spending (think new factories, housing, or durable goods) takes the biggest swings. Business and household investment moves about 2.6 times as much as output, while everyday consumption moves about two-thirds as much. Hours worked fluctuate more than output, but real hourly compensation barely moves with the cycle. Does that mean wages are rigid? Not necessarily, as we’ll see below. And since 1984, output’s fluctuations have been roughly 38% smaller in this sample.

1. What Do We Mean by the Business Cycle?

Modern macroeconomics studies two broad phenomena: long-run growth and short-run fluctuations in aggregate economic activity around its long-run trend. This post is about the second one: the business cycle.

By the early twentieth century, economists had documented recurring fluctuations in industrial economies, with striking similarities across countries and episodes. As Robert Lucas put it, “one is led by the facts to conclude that, with respect to the qualitative behavior of co-movements among series, business cycles are all alike.”¹ We’ll look at more recent U.S. data, drawing on Eric Sims’s Notre Dame course notes, to review some of the cycle’s best-known stylized facts. In a follow-up post, we’ll use these facts to guide the choices in a Real Business Cycle model and check its predictions.

It’s worth keeping in mind that modern macroeconomics shares much of its foundation with microeconomics. A familiar methodological starting point is that households and firms make choices given their economic environment and adjust their behavior when that environment changes. But their choices also change the environment everyone else faces. If households cut spending, for example, firms may produce less and hire fewer workers, which in turn affects household income and spending. We need to account for those feedback loops, not just the initial decision.

This is where general equilibrium comes in: we work out prices and quantities across the markets considered in the analysis, so that households’ and firms’ decisions are consistent with one another. In a partial-equilibrium analysis, we instead focus on a particular market while holding conditions elsewhere fixed. This is not a strict divide between macro and micro; microeconomics uses general equilibrium too. The point is that economy-wide questions make those connections hard to ignore. We’ll see this more concretely when we set up our first model, but it’s useful to keep it in mind from the get-go. Why? Because the prices and quantities we’ll look at are not independent objects: they emerge together from the decisions of households and firms.

2. The Long Run: Great Ratios and Factor Shares

Let’s start with the long-run facts. They may seem far removed from business-cycle fluctuations, but they will help us choose the model’s parameters later on.

• The Great Ratios (Kaldor facts): Expenditure shares of GDP remain relatively stable over long horizons. Here consumption excludes consumer durables, which we count as investment alongside gross private domestic investment.
  • Consumption accounts for roughly 50% to 65% of GDP across the postwar era.
  • Investment oscillates between 20% and 30%. Together, $C$ and $I$ account for roughly 80% of all output ($Y = C + I + G + NX$).

• Productivity and Real Wages: Average labor productivity (output per hour) and real compensation per hour both rose substantially after 1948, tracking each other closely until the mid-1970s. Both continued to rise afterward, but labor productivity grew faster, creating a widely debated gap between productivity and compensation.

• Hours per capita: Despite the rise in real wages, hours worked per adult (nonfarm business hours divided by civilian adult population) show no long-run upward trend. Higher wages make working more attractive (the substitution effect), but higher income also makes leisure more affordable (the income effect: why keep working without enjoying life?). This gives us a hint about how to model household choices: rising living standards should not make hours trend upward or downward forever. King-Plosser-Rebelo preferences² capture this by making these effects offset each other when wages and consumption grow proportionally, leaving hours unchanged.

• Labor Share: The labor share is the share of national income that goes to workers rather than capital owners. The figure compares two measures: the BLS-published nonfarm business labor share and a measure we construct as compensation divided by GDP. The nonfarm business labor share has fluctuated around 60%–64%, but has trended downward since the early 2000s. BLS reports a share of 52.8% in 2026Q2, the lowest recorded since the series began in 1947.

3. Extracting the Cycle: Long-Run Trend vs. Short-Run Fluctuations

To study the business cycle, we need to separate short-run fluctuations from long-run growth. There are several ways to do this, such as removing a linear or log-linear trend or using a band-pass filter. Here I’ll use the familiar Hodrick-Prescott (HP) filter, with the standard quarterly smoothing parameter $\lambda = 1600$.

• The data: Quarterly U.S. data, 1948Q1–2026Q1 (313 quarters).
• Quantities are measured per capita (using the civilian noninstitutional population age 16+) and expressed in logs.
• The HP filter separates each series into a smooth trend and a cyclical component. Because the series are logged, the cyclical component is approximately the percentage deviation from trend.

4. Business Cycle Stylized Facts

Fact 1: Consumption is smoother than output

Consumption is about 34% less volatile than output, with a relative cyclical standard deviation of 0.66. Here I follow Sims and count durable goods as investment, since they provide services over time like physical capital; consumption therefore includes only nondurables and services. The real-consumption measure here is constructed by deflating their nominal spending with the GDP deflator and dividing by civilian adult population; it is not the official BEA real-consumption series. This is consistent with intertemporal smoothing: people try to more or less stabilize their consumption rather than have it follow every rise and fall in income.

Fact 2: Investment fluctuates much more than output

By contrast, investment is considerably more volatile than output: 2.62 times as volatile, to be precise. In good times, firms and households can devote more resources to capital and durable goods; in downturns, those purchases tend to fall sharply. Consumption is relatively smooth, while investment takes much bigger swings.

Fact 3: Hours move strongly with output

Hours worked move closely with output, with a correlation of 0.87, and are 1.26 times as volatile. Notice the difference from the long-run picture: hours have no sustained upward trend, but they fluctuate considerably around it. Our model will therefore need to allow people to change how much they work over the cycle, without making them work more and more as the economy grows.

Fact 4: Real compensation moves less than hours

Real compensation per hour has a relative standard deviation of 0.71, compared with 1.26 for hours, and its correlation with output is just -0.05. So people work considerably more during booms, but their real hourly compensation does not systematically rise with output. This is something we’ll need to think about when we model the labor market: what makes hours change so much if real wages move relatively little? Sticky wages are one possible explanation, but we cannot infer them from this correlation alone. Different shocks and changes in who is employed can also produce this pattern.

Fact 5: Real interest rates barely move with current output

The ex-post real interest rate has a correlation of just 0.04 with output in the same quarter. This might seem surprising, since interest rates affect households’ decisions about consuming today versus saving for tomorrow. But those decisions concern the future, so looking only at the same quarter can miss part of the relationship. Here, the correlation between the real rate and output four quarters later is about -0.24. This doesn’t establish causality, but it gives us a reason to look at timing, not just whether two variables move together today.

Fact 6: Measured productivity is not necessarily a technology shock

The RBC model we’ll study starts from the idea that changes in technology drive fluctuations. At first sight, the data seem encouraging: measured total factor productivity (TFP) has a correlation of 0.76 with output. But measured TFP is a residual, the part of output that measured capital and labor inputs don’t explain. If firms run their machines longer or use workers more intensively during a boom, that can also show up as higher TFP. Fernald’s utilization-adjusted measure has a correlation of -0.20 and lower relative volatility (0.58, versus 0.77). So we should be careful about treating measured productivity as a direct measure of the technology shocks in our model.

Fact 7: The cycle became less volatile after 1984

Finally, these numbers depend on the period we look at. Output’s cyclical standard deviation falls from about 1.94% before 1984 to 1.20% afterward, a decline of roughly 38%. The later sample is less volatile even with the financial crisis and the pandemic included. This decline is associated with what economists call the Great Moderation. For our purposes, it means we shouldn’t assume that a model calibrated to one period will match another equally well. You can change the sample in the interactive figures to see how much the other facts change too.

5. Conclusion: The Benchmark for Macro Models

We now have a clearer idea of what our model needs to explain. Consumption is smoother than output, investment is much more volatile, and hours move strongly with the cycle while real compensation does not. We also have some long-run restrictions: expenditure shares are relatively stable, and rising living standards have not led to a sustained rise in hours worked per adult.

In the next post, we’ll set up an RBC model and see how far it gets. The long-run facts will help us choose its parameters; the cyclical facts will help us judge the model’s results. Matching one correlation isn’t enough if the same model gets consumption, investment, or hours badly wrong. And given what we saw with utilization-adjusted TFP, we’ll need to distinguish the model’s assumed technology shocks from what we actually measure in the data.

If you’d like to look more closely at the series or try a different sample, the figures are available in the EconLab U.S. Business Cycles Explorer.

Footnotes

1. Lucas, Robert E. (1977). “Understanding Business Cycles.” Carnegie-Rochester Conference Series on Public Policy, 5(1), 7–29. ↩

2. King, Robert G., Charles I. Plosser, and Sergio T. Rebelo (1988). “Production, Growth and Business Cycles: I. The Basic Neoclassical Model.” Journal of Monetary Economics, 21(2–3), 195–232. This is the reference for the balanced-growth preferences discussed above; the empirical results here use the updated EconLab data. ↩