Economics · Data dashboard
Real Business Cycles
Economies generally grow along a long-run trajectory. Output, consumption and investment follow a long-run path, or trend, but fluctuate above and below it in the short run. These short-run fluctuations are known as business cycles.
A business cycle is characterised by broad fluctuations in economic activity, with periods of expansion followed by periods of contraction. A contraction eventually gives way to a recovery, marking the beginning of a new expansion. Although this sequence of phases is recurrent, it is not periodic: business cycles do not occur at regular intervals and vary in duration, typically lasting between one and ten to twelve years.
In studying business cycles, we are particularly interested in two of their features: how volatile each variable is, and how it moves with output.
Volatility
How much a variable moves over the cycle, compared with output.
Output Variable
Co-movement
Whether a variable rises and falls with output.
Output Variable
To measure cycles, each time series is detrended with a statistical filter that separates its long-run trend from its cyclical component: for most variables, the percentage deviation from trend. Measured in this way, volatility and co-movement turn out to be regularities shared by many cycles, not features of a single recession or expansion. This page presents the main business-cycle facts for the United States since 1950.
Filter comparison: each chart shows its first variable under the three filters; the tables show the HP filter.
- Shaded bands mark US recessions, from the NBER-based quarterly recession indicator (USRECQ).
- The series are prepared as described on the Main Economic Indicators page; see Missing data and estimates for details.
Output and its expenditure components
The cyclical component of output, and of each expenditure component compared with it. Investment and durable spending move much more than output, consumption less; imports rise and fall with output.
Business-cycle statistics: output and its components
Consumption
Each component of consumption compared with total consumption. Durable goods can be postponed and are bought in bursts, so their spending is far more volatile than spending on nondurables and services.
Business-cycle statistics: consumption and its components
Investment
Fixed-investment components compared with total investment. Residential investment often turns down before recessions. The change in private inventories is not shown: it can be negative and has no logarithm, although it explains much of the volatility of total investment.
Business-cycle statistics: investment and its components
Capital
Capital input of the business sector from John Fernald’s quarterly TFP data. The capital stock changes slowly relative to its size, so its cyclical component is small and lags output.
Prices
Labour market
The blue line is the unemployment gap, \(100 \times \ln(u_t/u_t^*)\): the percent deviation of the unemployment rate from the Congressional Budget Office’s noncyclical rate (NROU), the trend used on the Main Economic Indicators page; it does not depend on the filter. The green line is the filtered unemployment rate: the cyclical component of log unemployment, its percent deviation from the trend estimated by the selected filter (a value of 10 means unemployment is about 10 percent above its trend, for example 5.5 rather than 5 percent). Both rise when output falls. See “Unemployment” under Definitions and methods.
Total hours, or hours per person aged 16 and over with the per-capita measure (nonfarm business sector).
Productivity
Measured TFP rises and falls with how intensively capital and labour are used; the utilization-adjusted series removes that effect (Fernald, 2014; Basu, Fernald and Kimball, 2006). In the comparison, output is shown in light grey for reference.
Interest rates
Summary statistics
The main business-cycle statistics for all variables (without the components of consumption and investment), then their leads and lags relative to output, and their correlations with one another. All follow the filter selected above; the switches on the right set the measure for the three tables and the basis of inflation and interest rates in the first one.
Volatility and co-movement
Leads and lags: cross-correlations with output
Correlation between each variable \(k\) quarters ahead, \(x_{t+k}\), and output today, \(y_t\). A peak at negative \(k\) means the variable tends to move before output (it leads); at positive \(k\), after output (it lags). The peak is the largest correlation in absolute value, highlighted in each row.
Correlation matrix
Contemporaneous correlations between the cyclical components of the variables selected below (the selection also applies to the leads-and-lags matrix). Blue cells are positive, orange cells negative; darker cells are stronger correlations.
Leads and lags: cross-correlation matrix
For each pair of variables, the correlation between the row variable \(k\) quarters ahead, \(x_{t+k}\), and the column variable today, \(z_t\), for \(k = -4, \dots, +4\). Each cell shows one of these nine correlations and, below it, the \(k\) at which it occurs: the strongest (largest in absolute value, positive or negative), the highest or the lowest, as chosen with the switch. For the \(k\) below each value: negative \(k\) means the row variable tends to move before the column variable (it leads), positive \(k\) after it (it lags), and \(k = 0\) together. The matrix mirrors itself: the cell for column and row swapped has the same correlation with the opposite \(k\). With “Strongest”, the output column repeats the peaks of the leads-and-lags table above.
Definitions and methods
What the cycle measures
Each filter splits a series into a trend and a cyclical component. Series measured in levels (output, its components, capital, hours, wages, productivity, TFP) and the unemployment rate are filtered as \(100 \times \ln x_t\), so a cycle value of 2 means the variable is roughly 2 percent above its trend. Inflation, interest rates and the participation rate are filtered in levels, and their cycles are in percentage points.
Inflation and interest rates. These are filtered in levels rather than in logs, and their cycles are in percentage points (a value of 1 means the rate is one point above its trend), as is usual in the business-cycle literature (Stock and Watson, 1999). Their logarithm is not always defined: inflation was negative in some quarters, nominal rates were close to zero in 2009–15 and 2020–21, and ex-post real rates are often negative. A change in a rate is also read in points, not as a percentage of the rate itself. The participation rate, which moves little, is treated in the same way. Unemployment is the exception (see “Unemployment” below).
The cycle is a statistical estimate, not an observed quantity. Different filters define the trend differently, so their cycles differ in size and timing; comparing them shows which regularities are robust to the method. Every series is filtered on its full published history; charts and statistics start in 1950 Q1.
Filters
Three filters estimate the trend, and therefore the cycle, in different ways. The filter switch at the top of the page applies to every chart and table; “Filter comparison” shows each chart’s first variable under all three.
Hodrick–Prescott (HP) filter. The HP trend is the smooth series that best fits the data while penalising changes in its growth rate (Hodrick and Prescott, 1997). The smoothing parameter is λ = 1600, the convention for quarterly data and the standard choice in the real-business-cycle literature (Kydland and Prescott, 1990; Cooley and Prescott, 1995). The filter is two-sided: each trend value uses past and future observations, so the latest values are revised as new data arrive.
Hamilton filter. Hamilton (2018) regresses each series on its four most recent values known two years earlier (h = 8, p = 4 for quarterly data). The cycle is the part of the series that could not be predicted from that information. The method uses only past data, the first 11 quarters of each series have no value, and its cycles are usually larger than HP cycles because they include two years of unpredictable changes.
Christiano–Fitzgerald (CF) band-pass filter. The CF filter keeps fluctuations that repeat every 6 to 32 quarters (1.5 to 8 years), the usual definition of business-cycle frequencies, and removes slower movements and shorter-lived noise (Christiano and Fitzgerald, 2003). This page uses the asymmetric full-sample version, which assumes each series behaves like a random walk with drift. Its most recent values rely on fewer observations and are revised as new data arrive.
Unemployment
The unemployment rate is filtered in logs, like output, so its cycle is a percent deviation from trend rather than a deviation in percentage points. This is the convention of the labour-market literature. Because unemployment is a small number, a deviation in percentage points understates how much it moves: a rise from 5 to 6 percent is 1 point, but a 20 percent increase in the rate.
Shimer (2005) measures unemployment, vacancies and their ratio (labour-market tightness) as log deviations from a very smooth HP trend (\(\lambda = 10^5\)) in quarterly US data from 1951. The standard deviation of unemployment is about 0.19 (19 percent of trend), roughly ten times that of labour productivity, and the vacancy–unemployment ratio is about twenty times as volatile as productivity. The standard search-and-matching model (Mortensen and Pissarides, 1994), calibrated in the usual way, generates only a small fraction of these fluctuations from productivity shocks: the “unemployment volatility puzzle”.
Hagedorn and Manovskii (2008) use the standard quarterly smoothing parameter, λ = 1600, which assigns more of the slow movement in unemployment to the trend; the standard deviation of log unemployment is then about 0.125 (12.5 percent). They argue that when the value of non-market activity and workers’ bargaining power are calibrated to the cost of posting vacancies and to the cyclicality of wages, the model is consistent with the data. Whether that calibration is plausible remains debated.
The HP cycle on this page uses λ = 1600, as for every other variable, so it is comparable to Hagedorn and Manovskii; the Hamilton and Christiano–Fitzgerald filters are checks. The sample is longer than in both papers and includes the Great Recession and the 2020 pandemic, when unemployment rose very sharply, which raises the standard deviations.
Unemployment gap. The gap is \(100 \times (\ln u_t - \ln u_t^*)\), where \(u^*\) is the Congressional Budget Office’s noncyclical rate of unemployment (NROU). It is the same transformation as the cycle, with the NROU in place of a statistical trend, so the two lines are in the same units and can be compared directly; it does not depend on a filter. For small gaps it is close to the relative gap, \(100 \times (u_t - u_t^*)/u_t^*\); the two differ for large gaps, as for any log deviation (for example, unemployment twice the natural rate gives about 69 rather than 100). The Phillips-curve and monetary-policy literature usually states the gap in percentage points, \(u_t - u_t^*\) (Galí, 2011; Crump, Eusepi, Giannoni and Şahin, 2019); the log version used here expresses the same gap relative to the natural rate, as the labour-market literature does for the cycle.
Business-cycle statistics
- Standard deviation: of the cyclical component, in percent of trend (percentage points for rates and ratios). It measures volatility.
- Relative standard deviation: the standard deviation divided by that of output over the same quarters (and, for the components of consumption and investment, by that of their total).
- First-order autocorrelation: the correlation of the cycle with its value one quarter earlier. It measures persistence.
- Correlation with output: the contemporaneous correlation of the two cycles. It measures co-movement: positive for procyclical variables, negative for countercyclical ones, close to zero for acyclical ones.
- Cross-correlations: \(\operatorname{corr}(x_{t+k}, y_t)\) for \(k = -4, \dots, +4\) quarters, as in Kydland and Prescott (1990) and Stock and Watson (1999). The peak is the \(k\) with the largest correlation in absolute value.
Annual or quarterly rates. Inflation and interest rates are annualised by default. The quarterly basis divides their cycles by four (inflation becomes \(100 \times \Delta \ln P_t\), interest rates the rate earned over one quarter), the units of a quarterly model such as Smets and Wouters (2007). This changes their standard deviations and relative standard deviations by the same factor, but not their autocorrelations, correlations or leads and lags. See the Main Economic Indicators page for which basis to use in estimation.
Each statistic uses all quarters from 1950 Q1 for which both series exist (for example, the federal funds rate starts in 1954 Q3 and Fernald’s data end one quarter earlier than the national accounts). The 2020–21 pandemic quarters are included; they are large and raise the standard deviations of most variables.
Variables and data
The series are those of the Main Economic Indicators page, with the same sources, quarterly averages of monthly series and treatment of missing values; see its Definitions and methods.
- Per capita: output, its components and capital services are divided by the BEA mid-period population; hours by the civilian noninstitutional population aged 16 and over. Rates, ratios, wages, productivity and TFP are the same under both measures.
- Unemployment gap: \(100 \times \ln(u_t/u_t^*)\), where \(u^*\) is the Congressional Budget Office’s noncyclical rate of unemployment (NROU): the percent deviation of the unemployment rate from that rate. It is already a deviation, so it is not filtered and is the same under every filter.
- Capital and TFP: John Fernald’s capital input and TFP series for the business sector (Fernald, 2014).
- Real interest rates: ex-post rates: the nominal rate minus GDP-deflator inflation over the following quarter.
Missing data and estimates
The cycles on this page are computed from the series published on the Main Economic Indicators page, after the same preparation: monthly series are averaged over complete quarters; real components of consumption and investment are built in chained 2017 dollars with BEA’s formula before 2007; per-capita measures divide by BEA’s population (hours by the population aged 16 and over).
When a source did not publish a value for a period (for example, the October 2025 household survey, not collected during the federal government shutdown), the value is estimated by interpolation between the published values on either side and flagged. Such estimates are rare and short, so their effect on cycles and statistics is negligible. The list of estimated values, with the method and reason for each, is in Missing data and estimates on the Main Economic Indicators page.
Recession shading and data revisions
Shaded bands show US recessions from the NBER-based quarterly recession indicator published on FRED (USRECQ).
Official statistics are revised after their first release, and two-sided filters revise their latest values as new data arrive. Each update of this page uses the latest available history, so earlier values can change.
References
- Basu, S., J. G. Fernald and M. S. Kimball (2006). “Are Technology Improvements Contractionary?” American Economic Review 96(5): 1418–1448.
- Burns, A. F. and W. C. Mitchell (1946). Measuring Business Cycles. New York: National Bureau of Economic Research.
- Christiano, L. J. and T. J. Fitzgerald (2003). “The Band Pass Filter.” International Economic Review 44(2): 435–465.
- Cooley, T. F. and E. C. Prescott (1995). “Economic Growth and Business Cycles.” In T. F. Cooley (ed.), Frontiers of Business Cycle Research, Princeton University Press, 1–38.
- Crump, R. K., S. Eusepi, M. Giannoni and A. Şahin (2019). “A Unified Approach to Measuring u*.” Brookings Papers on Economic Activity 50(1) (Spring 2019).
- Fernald, J. G. (2014). “A Quarterly, Utilization-Adjusted Series on Total Factor Productivity.” Federal Reserve Bank of San Francisco Working Paper 2012-19 (updated March 2014).
- Galí, J. (2011). “The Return of the Wage Phillips Curve.” Journal of the European Economic Association 9(3): 436–461.
- Hagedorn, M. and I. Manovskii (2008). “The Cyclical Behavior of Equilibrium Unemployment and Vacancies Revisited.” American Economic Review 98(4): 1692–1706.
- Hamilton, J. D. (2018). “Why You Should Never Use the Hodrick-Prescott Filter.” Review of Economics and Statistics 100(5): 831–843.
- Hodrick, R. J. and E. C. Prescott (1997). “Postwar U.S. Business Cycles: An Empirical Investigation.” Journal of Money, Credit and Banking 29(1): 1–16.
- King, R. G. and S. T. Rebelo (1999). “Resuscitating Real Business Cycles.” In J. B. Taylor and M. Woodford (eds.), Handbook of Macroeconomics, Vol. 1B, Chapter 14: 927–1007. Amsterdam: Elsevier.
- Kydland, F. E. and E. C. Prescott (1990). “Business Cycles: Real Facts and a Monetary Myth.” Federal Reserve Bank of Minneapolis Quarterly Review 14(2): 3–18.
- Mortensen, D. T. and C. A. Pissarides (1994). “Job Creation and Job Destruction in the Theory of Unemployment.” Review of Economic Studies 61(3): 397–415.
- Shimer, R. (2005). “The Cyclical Behavior of Equilibrium Unemployment and Vacancies.” American Economic Review 95(1): 25–49.
- Smets, F. and R. Wouters (2007). “Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach.” American Economic Review 97(3): 586–606.
- Stock, J. H. and M. W. Watson (1999). “Business Cycle Fluctuations in US Macroeconomic Time Series.” In J. B. Taylor and M. Woodford (eds.), Handbook of Macroeconomics, vol. 1A, Elsevier, 3–64.
Page created on 28/10/2026.
© Arthur Galichère.
- Errors: Every effort has been made to ensure the data and calculations are accurate; any remaining errors are mine. If you believe you have found a mistake, or have a comment or suggestion, please let me know.
- Reuse and citation: Data are from the agencies named under each chart: via FRED (Federal Reserve Bank of St. Louis) and, for total factor productivity and capital input, from John Fernald at the Federal Reserve Bank of San Francisco. The cycles, statistics, charts and text are the author’s; the source data remain subject to the terms of their providers. When reusing these figures, please cite the original data providers and this website: Arthur Galichère, Real Business Cycles.