Cookies on Macro by Mark.We use essential cookies for sign-in, locale, and security. With your permission, we also use optional cookies for preferences, for analytics, and for advertising measurement with the Meta (Facebook) Pixel. Read our cookie policy.
Trend-cycle decomposition for a single macro series, comparing smooth HP trend extraction with Beveridge-Nelson permanent-transitory decomposition.
Graph workspace
The defining comparison is how differently the HP and BN methods decompose the same series. The top panel shows the original data with both trends overlaid. The bottom panel shows the two implied cyclical components. This layout makes the smooth-HP-trend vs. volatile-BN-trend contrast immediately visible.
HP cycle variance share
Fraction of total series variance captured by the HP cyclical component. Larger values indicate more variation is attributed to the business cycle rather than the trend.
0.0064
proportion
BN cycle variance share
Variance of the BN transitory component over the variance of the series. Much smaller than the HP cycle share when growth is only moderately persistent, reflecting the BN tendency to attribute variation to the permanent component; as nears 1 the BN trend overshoots the series, the offsetting transitory component becomes the larger of the two, and the ratio can exceed 1. The BN trend and transitory component are driven by the same shocks, so they can also move against each other: with within a few hundredths of its -0.99 end (-0.99 to -0.97) and no drift, where the BN trend is almost flat, the ratio can edge just above 1 (by up to about 4 percent) whatever is below about 0.9.
0.0006
ratio of variances (can exceed 1)
BN long-run multiplier
The cumulative impulse response of to a unit innovation. When is large, permanent shocks dominate and the BN trend is volatile. When is near zero, transitory shocks dominate.
1.40
multiplier
Trend correlation (HP vs BN)
Correlation between the two trend estimates. High correlation means the methods agree on the broad trend shape despite different smoothness. Low correlation signals a major disagreement about what constitutes the trend.
0.9958
correlation
Penalty on a bumpy trend line: higher settings force a smoother trend.
Increasing makes the blue HP trend smoother and pushes more variation into the blue HP cycle in the bottom panel.
Higher settings make swings in the growth rate carry over for longer.
Increasing makes the red BN trend more volatile and, beyond about 0.7, the red BN transitory component much larger: the BN trend overshoots the data and the transitory component offsets it.
Moving-average coefficient in . Negative reduces the long-run multiplier; near -1 makes the BN transitory component large.
Moving toward -1 shrinks the red BN trend's volatility until the trend is almost a straight line, and the red BN cycle carries the series' wiggles instead. Moving toward +1 makes the BN trend more volatile, and the BN cycle grows with it rather than shrinking, since the BN cycle carries : at the other defaults its standard deviation is about 2.9 at , against about 1.0 at .
Sets the size of the random shocks, scaling all variation in the generated data.
Higher increases the scatter of gray dots and the amplitude of both trends and cycles.
Sets the average growth rate of the series.
Higher drift tilts both trends upward. The cycles are unaffected in amplitude but the level of the top panel changes.