Research

Kondratiev Waves, Stocks and GDP

Published

The Kondratiev long wave (康波) is the hypothesis of a 45–60-year economic cycle, famous in China through the slogan "fortunes are made on the K-wave". We reviewed its theory, a century of statistical tests, its actual relationship with stocks, rates and GDP, and the forecasting record of its proponents, and re-ran the tests ourselves on Maddison 2023 GDP for five countries and 150 years of Shiller US stock data. Verdict: the K-wave has no usable predictive power in GDP or stock data; by the usual demarcation criteria it is not strictly pseudoscience either, but an unconfirmed, weakly falsifiable conjecture. The pictures tell most of the story.

1. What the K-wave is: the textbook picture and the datings

The idealised K-wave and four dating schemes
Kondratiev read about two and a half long waves out of British, French and US prices and interest rates in 1925; later authors (Schumpeter, Mandel, Perez, Korotayev) attached a technological revolution to each wave. The 19th-century turning points are agreed; after 1940 the datings scatter: wave 5 starts anywhere from 1971 to 1992, the 5→6 transition anywhere from 2000 to 2030. "Which phase are we in" has no answer until you pick an author.

2. The K-wave and GDP

20-year moving average of GDP per capita growth with K-wave upswings
If the wave were real, the conventional upswings (shaded) should sit systematically above the downswings. The 20-year smoothed growth of the US, the UK and the five-country aggregate shows no such alternation; the 1921–1944 "downswing" is one of the fastest-growing pre-1992 phases for the US and the UK.
GDP per capita growth in K-wave A and B phases
Averaging all A years against all B years: US 1.41% vs 1.41%, five countries 1.77% vs 1.76%. No country's gap is significant under permutation tests (France's comes from the two world wars and halves without them).
Power spectrum of GDP per capita growth
The power spectrum looks straight at the 40–60-year band: for the US and the five-country series the band holds less than the 1.7% of variance white noise would give it, with Monte-Carlo p-values above 0.75 against AR(1) red noise. The one "significant" result the literature keeps citing (Korotayev & Tsirel 2010, p≈0.045) needs 1914–1946 replaced by a constant and the 52-year and 17-year peaks pooled.
Slutsky–Yule demonstration
Why does the eye keep seeing waves? Slutsky showed in 1927 that moving averages manufacture smooth pseudo-cycles out of noise. Smoothing AR(1) noise of the same length and variance as US growth the same way, about four in ten paths grow 2–5 "long waves" 35–75 years apart with at least the real data's amplitude. Kondratiev himself used 9-year moving averages on polynomial-detrended series.

3. The K-wave, stocks and interest rates

150 years of US real total return with secular phases
US secular bull and bear markets (1901–21 bear, 1921–29 bull, 1929–49 bear, 1949–66 bull, 1966–82 bear, 1982–2000 bull, 2000–09 bear, 2009– bull) run about 17–20 years each, two per K-wave, and track CAPE valuation and the inflation regime rather than a 50-year wave. By the investor "seasons" the deflationary "Kondratiev winter" began in 2000, yet the real return since 2009 is among the highest in the record.
10-year forward US real return: CAPE vs K-wave sinusoid
Used as a predictor of 10-year forward real returns, log CAPE scores R² = 0.29; the best 45–60-year sinusoid scores 0.09, and reshuffled returns beat it 82% of the time. On top of CAPE the wave adds nothing; a dummy for the conventional A phases even has the wrong sign.
US long bond yield since 1871
The "interest-rate long wave" proponents like to cite (trough 1941 → peak 1981 → trough 2020) is one hump in data since 1871, matching neither the 1873 peak nor the 1896, 1945 and 1992 troughs. Schmelzing's 700-year real-rate series calls the 1950–85 rise "the really striking divergence from trend" and what followed a reversion to trend, not a cycle.

4. The forecasting record (dated calls, not hindsight)

Forecaster / dateCallOutcome
Zhou Jintao, lecture of 2016-03-16Commodity rebound in 2016; A-share decline in 2018Hit (inventory-cycle-scale calls)
Zhou Jintao, same lectureChinese and US assets fall across the board after Q3 2017; 2018–19 the "years of no return", final low in 2019Miss: US stocks kept rising after 2017; 2019 was no low
Zhou Jintao, same lectureWave-5 depression lasts to 2025; dollar bull ends 2017Miss: US real equity return about +9.5%/yr since the lecture; global stocks at records in 2025
Ian Gordon (Longwave Group), July 2010Dow to 1,000 at a minimum, gold $4,000, Dow/gold 0.25Miss: gold reached $4,000 only in 2025, with the Dow above 47,000 and Dow/gold near 12
Harry Dent, books 1999–2025Opposite-direction calls through 1999–2015; crashes and deflation in 2016, 2023, 2024, 2025All missed
Korotayev & Tsirel, 2010 paperTwo incompatible post-2008 readings, one with a K-wave peak in 2018–2020Unfalsifiable: one of the two readings is always "right"

No peer-reviewed paper published a dated K-wave forecast before 2008 or 2020 that was later confirmed, and the 2022 inflation ran against the "winter = deflation" rule. The hits are 1–3-year inventory or valuation calls, or a permanent bear being right eventually.

5. Pseudoscience, or an unconfirmed conjecture?

  • Not pseudoscience in the strict sense. By multi-criteria demarcation (Hansson), pseudoscience relies on authority, cherry-picks, ignores refutation and never revises; K-wave researchers publish in refereed journals, report negative results and have revised or dropped their claims (Metz retracted his own 1992 findings; Perez explicitly abandoned fixed periodicity).
  • But weakly falsifiable by Popper's standard. Turning points are confirmed only in hindsight, every author re-dates the waves, and a "sixth wave" has been declared to start anywhere from 2000 to 2030. The academic critics (Kuznets, Garvy, Solomou, Maddison, Metz, Diebolt) say "not demonstrated" and "not a useful analytical tool".
  • Keep the sequence, drop the calendar. Perez's "big bang → finance-led installation → bubble and crash → institutional recomposition → golden-age deployment" is a mechanism you can reason with; the parts of the K-wave story that do have data behind them have shorter, measurable stand-ins: valuation and inflation regime (CAPE, 17–20 years), credit and house prices (the BIS financial cycle, about 16 years), industrial demand (commodity super-cycles, 30–40 years).

Data and method: Maddison Project Database 2023 (GDP per capita, US/UK/FR/DE/JP, 1800–2022), Bank of England "millennium of macroeconomic data" (UK GDP 1700–2016), Shiller monthly US data (1871–2023, datahub mirror). Spectra are 10%-tapered periodograms with Daniell smoothing, tested with 5,000 Monte-Carlo draws against white and fitted AR(1) noise; forward-return regressions use Newey–West errors for the overlapping windows. Every figure is reproducible with tools/research/kwave/make_figures.py; the full report and the five source notes are in the repository under docs/research/.

Open this guide in Wave Radar

More guides