Generations and Business Cycles - Part I
ABSTRACT: Prior to industrialization the Kondratiev cycle reflected biological generations. It's economic effect was at least partially induced by the generational war cycle, with its impact on government finances. Shorter economic cycles did not reflect the longer K-cycle, but rather operated on their own schedule, possibly also generational in nature. Hence financial panics followed a real estate cycle of about 18 years, denoted as the Kuznets cycle. Slumps occurring in between the Kuznets cycle defined a half-cycle that was of the same length as the business cycles first identified by Clement Juglar. Ordinary business cycles were of apparently random length up to a full Juglar. With the coming of industrialization, the ordinary business cycle became associated with a new phenomenon, the inventory cycle, also called the Kitchen cycle. It decreased to a lower, more uniform length of about 40 months. The Kuznets and Kondratiev cycles continued on more or less as before. From these changes induced by industrialization, the SMECT structure emerged, in which sixteen 40 month Kitchen cycles "fit" into a standard Kondratiev cycle and the K-cycle is subdivided into 1/2, 1/4 and 1/8-length sub-cycles. Standing aside from this was the old Kuznets real-estate cycle. After 1933, the Kondratiev cycle remained, but its length gradually increased to about 72 years today. The Kuznets real estate cycle continued, but was much weaker for about 40 years until the 1970's when something like the old cycle was again active. Although it has been 22 years since the last real estate peak in 1980, valuations have not yet reached past peak levels and a real estate bust is not expected now (Nov 2002). Instead the boom spawned by the Fed's rate cuts should continue to produce increasing real estate valuations for a couple more years--much as happened with stocks after the 1998 crisis.
In my recent book on the Kondratiev cycle, I developed the idea that the K-cycle is fundamentally generational in nature. From late Medieval times up to the early 19th century, the K-cycle was equal to two biological generations in length, or about 50 years. Two Kondratiev cycles in turn form one saeculum, a generational cycle described by the American authors William Strauss and Neil Howe. The K-cycle was closely aligned with wars, and a possible mechanism for the cycle was alternating periods (of generational length) of government debt growth and decline associated with war finance. After the economy became industrialized in the late 19th century the relation between the cycles changed. Instead of two K-cycles per saeculum, there was now only one. The K-cycle lengthened and the saeculum shortened to the point where both are about 72 years long today. This new pattern has only developed after 1929. In between 1860 and 1929 the situation is unclear. Strauss and Howe note the anomalous behavior of their cycle at this time and call it the Civil War anomaly. They attribute it to a skipped generation caused by an unusually bad outcome of the Civil War. My thesis is that it was caused by a shift in cycle dynamics around this time caused by industrialization.
Table 1. Structure of socioeconomic periodizations
|Cycle||Pre-industrial (before 1860)||Modern (after 1929)|
|Saeculum||Two K-cycles (90-115 years)||One K-cycle (72 years)|
|K-cycle||45-60 years||72 years|
|K-wave/Stock Cycle||Biological gen (20-30 years)||36 years|
|K-season/Secular Trend||10-15 years||Psychological gen (18 years)|
Each Kondratiev cycle is itself split into two Kondratiev waves, each associated with what I call the Stock Cycle. The Kondratiev wave is subdivided into two Kondratiev seasons, each associated with a secular market trend. Table 1 shows how these cycles were related before and after industrialization. Still shorter economic cycles exist, such as the Kuznets cycle of 15-20 years (related to building/real estate valuation cycles), the Juglar cycle of 7-11 years (related to fixed investment) and the Kitchen cycle of about 40 months (related to inventories). The Juglar cycle was first noted by Clement Juglar in 1862, and can be thought of as existing in the pre-industrial economy. The other two cycles were identified much later (Kitchen in 1923; Kuznets in 1930) and might be thought of as cycles in the industrialized economy.
In this paper I present information on shorter economic and stock cycles to look for evidence suggestive of a change in structure among the shorter cycles that is associated with industrialization. This would provide further evidence for the idea that the Civil War anomaly of Strauss and Howe actually reflects a shift in socioeconomic cycle dynamics associated with industrialization and not a "bad" outcome of the Civil War conflict. I will also employ market researcher Robert Bronson's SMECT system of nested economic cycles as a means to help characterize all these cycles. The shorter stock cycles I will use is the collection of major stock bear markets shown in Table 2. The methods used to select these bear markets are given in the Appendix A. The shorter economic cycles I will consider are ordinary business cycles as defined by the National Bureau of Economic Research (NBER) plus those I have identified before 1854 when the NBER cycle dating begins. The methods used are given in Appendix B and the cycles themselves appear in Table 3.
Economic cycles up to 1933
The first thing to look at is how the spacing of bear markets and economic recessions have changed over time. Figure 1 shows the spacing of bear markets (which defines a bull/bear market cycle) and that of recessions (which defines the business cycle) from the end of the 18th century to the present. Several features are immediately apparent from this figure. The first is that the stock market and the economy showed cycles of similar length up until around 1930, after which the economic cycles have averaged longer than the stock cycles. The second is that prior to 1885 and after 1930 business cycles were longer (about 5-6 years on average) than during the 1885-1930 period, when they averaged about 40 months. Thirdly, the stock cycle, which shortened along with the business cycle in the 1880's, remained at this shorter length until the present. That is, it did not rise again in length in recent decades as the business cycle has.
Finally, note that the timing of the downward shift in cycle length in the 1880's corresponds to the period during which the economy was transforming from being mostly agricultural to mostly industrial. That is, the business cycle shortened to a "Kitchen-like" length of about 40 months right at the time the fraction of the workforce engaged in agriculture fell below 50%. Since the Kitchen cycle reflects fluctuations in inventories of manufactures, one might expect it to have appeared only as agriculture faded in importance. This interpretation would have the pre-industrial business cycle operate on some cycle other than the Kitchen. These shifts in business and bull/bear cycle lengths define three eras of interest: the pre-industrial (before 1885) the industrial (1885-1933) and the regulated (after 1933).
One of the salient features of U.S. financial history in the 19th and early 20th centuries was the fairly regular occurrence of financial panics/major depressions at roughly 20 year intervals: 1798, 1819, 1837, 1857, 1873, 1893, 1907 and 1929. One can use the major bear market bottoms associated with these panics and major lows in between them to define a cycle of 7 to 13 years as shown in Figure 2. In this way two cycles can be defined; a 20 year "panic cycle" with bottoms shown in bold italic and a ten year half-cycle. The "panic cycle" corresponds reasonably well to the pre-depression U.S. real estate cycle. Land sales by the US government before the Civil War showed massive surges in activity in 1818, 1836 and 1854 followed by sharp drops in 1819-20, 1837-41 and 1855-60. Chicago land values showed peaks in 1836, 1854, 1872, 1891, and 1925 and troughs in 1840, 1859, 1875, 1894 and 1933. The number of new buildings constructed in Chicago showed major peaks in 1892, 1910 and 1925 and troughs in 1900, 1918 and 1933. This relation to real estate cycles suggests that the panic cycle is a Kuznets cycle. The ten-year half-cycle is of the same length as the cycles identified by Juglar and so I will call them Juglar cycles. Table 4 below summarizes these cycles and compares them to the Kondratiev and minor cycles (e.g. NBER business cycles and bear markets).
Table 4. Relation between Kondratiev, Kuznets, Juglar and minor cycles before 1933
|Kuznets Troughs||Juglar Troughs||Minor Cycles|
|1787|| 1867 P (con't)
Table 4 shows how the four cycles discussed so far are aligned. We have already seen that the ordinary or minor business cycles shortened upon the transition from the pre-industrial to the industrial era. Prior to 1885 business cycles averaged 64±23 months in length, between 1885 and 1933 they averaged 41±9 months. This difference is statistically significant. Using the data in Table 4 one can calculate the average length of the other cycles for these two periods. The Kondratiev cycle was unchanged at 53-54 years for both periods. The Kuznets cycle length was unchanged as well: about 18 years (peak-to-peak) or 20 years (trough-to-trough) in length for the pre-industrial era versus about 18 years in the industrial era. The pre-industrial Juglar cycle (9.7 years) was a bit shorter than the industrial cycle (12 years), but this difference is not statistically significant.
During the pre-industrial era, anywhere from 1 to 4 business cycles could fall within a Juglar. That is, the lengths of business cycles could vary from 0.25 Juglar to 1.0 Juglar with an average of 0.56 Juglar. The shortest actual business cycle observed over this time was 2.5 years, about 0.25 of a standard 10 year Juglar. The longest was 9 years, which occupied a full Juglar. If business cycles were of random length between 0.25 and 1.0 Juglar, they would average around 0.62 Juglars, not too different from the observed average of 0.56. The standard deviation of the lengths of such random cycles would be equal to 35% of their average length. In actuality, the standard deviation of cycle length was 35% of average length. What this suggests is that before industrialization the shortest natural cycle was the Juglar at about 10 years. Shorter fluctuations were essentially random in nature, except for the requirement that they be bound by a Juglar cycle.
Compare this to the situation after 1885 in which business cycles fell into a narrow range between 0.25 and 0.33 (average 0.29) Juglars in length. The standard deviation of these cycle lengths was only 22% of the average length. After industrialization, the impact of inventory fluctuations on the economy came to dominate short term economic fluctuations, producing the 40-month Kitchen cycle. The shortest natural cycle was now the Kitchen, which fixed business cycle length into a narrow range around 40 months. The Juglar cycle still operated more or less at its old length, but there was now an interaction between it and the shorter Kitchen cycle. Since the Juglar was required to be an integral number of Kitchen cycles, which had their own natural length, the result was a Juglar cycle that was comprised of either three longer-than average Kitchen cycles or four shorter-than-average Kitchen cycles. The result was a cycle comprised (on average) of 3.5 Kitchen cycles that ran 11-12 years in length (3.5 x 40 months), just a bit longer than it had before industrialization.
Since the (trough-defined) Kuznets cycle must contain a whole number of Kitchens, the same issues can arise. Five or 6 Kitchen cycles would be necessary to produce a trough-defined cycle length in the 18-20 year window observed for the pre-industrial Kuznets cycle. Table 4 shows that the industrial Kuznets ran five Kitchens in length. This suggests that the natural length for the Kuznets is closer to the 18 year spacing shown by the real estate peaks and not the 20 year spacing shown by the troughs. Table 4 also shows a 2:1 relation between Juglar cycles and Kuznets cycles that ran throughout the entire 19th century. In the early 20th century the two cycles became uncoupled. The Panic of 1907, which produced a severe recession that marked a Juglar bottom, was the first financial panic that was not immediately followed by a major decline in real estate activity or valuation. There was instead a gradual drop in values and activity which bottomed during WW I, well before the next Juglar bottom in 1921.
Because of the great length of the Kondratiev cycle, it is not possible to assess how industrialization might have changed the relationship between this cycle and the shorter ones. No relation between the Kuznets cycle and the Kondratiev cycle is apparent from Table 4. A clear relation between the Juglar and the K-cycle is also not apparent. Table 4 shows 5 or 6 Juglars per Kondratiev during the 19th century. There is some evidence for a relation between the Juglar and the K-season after 1897. The 1897-1908, 1908-1921 and 1921-33 Juglars roughly correspond to Kondratiev spring, summer and fall, suggesting that four Juglars per Kondratiev might have been an emergent pattern after industrialization, but such a K-cycle would be only 48 years long, a bit on the short side. Table 5 summarizes all these observations.
|Saeculum||~100 years (two K-cycles)||80-90 years (uncoupled with K-cycle)|
|K-cycle||1/2 saeculum; 5-6 Juglar||4 Juglar?|
|Generation||1/2 K-cycle; 1/4 saeculum||20-22 years (uncoupled from K-cycle)|
|K-wave/Stock Cycle||1/2 K-cycle||2 Juglar?|
|Kuznets cycle||~2 Juglar (18-20 years)||1-2 Juglar (~18 years)|
|Juglar cycle||1-4 bus cycles (7-11 yrs)||3-4 Kitchen (~12 yrs)|
|Kitchen cycle||N/A||1 bus cycle (~40 mo)|
Economic cycles after 1933
We now turn to the period after 1933, during which governmental economic policies of various types have been in force. I call this era the regulated era to reflect these policies. Several independent lines of evidence suggest that the Kondratiev cycle became considerably longer after 1933. The 35 year rise in both interest rates and reduced prices over 1946-1981 are two direct measures of post-1933 Kondratiev upwave length. This can be compared to the average upwave of 24.4 years as defined by interest rates and prices before 1933. The four secular market trends since 1929 have lasted from 16 to 20 years, with an average length of nearly 18 years, suggesting a 36 year Stock Cycle, compared to 25.4 years for the five Stock Cycles before 1929. These data imply a 70-72 year length for the K-cycle after 1933, compared to about 50 years before--a 40% increase in length.
*Housing starts after 1960. New non-farm dwelling units  for 1900-1960, and building construction in Chicago  before 1900. All values in thousands of units per year. Housing starts and new non-farm dwelling units appear to be the same series. The Chicago series was used to extrapolate the housing start series back from 1900 to 1885.
Since the Depression there have been no financial panics so the method used to find Juglar cycles in Figure 2 cannot be applied. Figure 3 presents a graph of building activity from 1885 to the present. Peaks in building activity can be seen in 1892, 1909 and 1925, which reflect the 18 year industrial Kuznets cycle peaks. Ignoring WW II, after the trough in 1933 there was a smoothly rising trend in building activity that reached a peak in 1950. This peak was not followed by a conventional bust, however, but rather, a plateau. It was not until the late 1960's that something like an old-fashioned real estate boom developed. This boom peaked in 1972 and was followed by a bust which bottomed in 1975. This boom/bust cycle was immediately followed by two more which showed peaks in 1979 and 1986, and troughs in 1982 and 1991. Since 1991 building activity has shown a more leisurely rise and has yet to show any sort of peak.
Figure 4 shows a housing price index obtained by dividing the GDP implicit residential deflator by the total GDP implicit deflator as described by McFadden This McFadden index reflects the trend in real housing values without land included. A series of peaks in 1873, 1887, 1907 and 1929 can be seen which roughly correlate with peaks in Chicago land values in 1872, 1891 and 1925, and in building activity in 1892, 1909, and 1925. Thus, figures 3 and 4 describe a pre-1933 real estate cycle that is reasonably consistent with the idea of 18 year cycles in real estate activity and valuation as described earlier. Both figures show a major rise in both housing construction and price after the 1933 cycle bottom that lasted until around 1950. Both show no significant decline in values or activity commensurate with a bust after this date. A boom in housing prices developed only after the 1970 recession and lasted the entire decade. There was no intermediate peak in 1972 as for building activity. Likewise there is no price peak in the 1980's as was seen with building activity. Both series show troughs in the early 1990's and a rise afterward. If we interpret the price peak in 1969 as a separate peak, and then assign troughs at the price minima in 1958 and 1971, we can see what appears to be a continuation of the 18 year cycle after 1933. But this pattern is not supported by the building activity data.
Figure 4. Real housing price index (1982=1.0)
Further insight may be gained by examining median housing prices, which are available after 1963 for new houses and after 1970 for existing houses. Figure 5 shows a plot of these prices and household income in constant 2002 dollars. Income of the 65th percentile of households (the middle of the top 70% of households) was chosen as reflective of the income of the "typical" home buyer, and thus most comparable to the median home price. Also shown are ratios of the housing price to income, which might be thought of as a type of "valuation" for housing, sort of like P/E for stocks. Prices for existing homes have tended to trend with income; that is, the ratio between price and income is range bound. This ratio defines a cycle with a peak in 1978-9 and trough around 1990, followed by a slow recovery. Since 2000 a steeper uptrend appears to have gotten underway, suggestive of the beginning of another boom. This data is consistent with the McFadden index in Figure 4, which suggests only one peak in the 1970-1990 period and not three as suggested by housing starts in Figure 3.
Prices for new homes have shown a consistent rise over time with noticeable peaks in 1969, 1979 and 1989. The ratio between new house prices and income also displays a rising trend over time and three clear peaks corresponding to the price peaks. The first two of these peaks are consistent with McFadden peaks in Figure 4, the third is not. The different trends for new versus existing houses after the early 1980's in Figure 5 suggests that much of the higher price for new houses since then may reflect increased quality and not stretched valuations indicative of a bubble. The McFadden index can be thought of as a quality-adjusted real price index for residential construction. Thus, the correspondence between the existing house price index (which is inherently largely quality-adjusted) and the McFadden index makes sense, even though the latter is based on new housing. The strong possibility that the post-1980 trends in the new housing price/income ratios reflect rising quality and not inherent valuation suggests that the story told by the existing-house price data and the McFadden index is the correct one.
Figure 5. Median housing prices in
2002 dollars and the ratio of price* to 65th percentile household
*Prices for new homes were obtained from census data. Data for existing homes come from the National Association of Realtors and from statistical abstract of the US before 1989.
The evidence from prices suggest that an attenuated real estate cycle continued on after 1933 that grew in strength after 1970. The building activity data since 1933 don't show this. The minor fluctuations in building activity between 1950 and 1972 closely follow the ordinary business cycle and not some longer cycle. The fluctuations grow larger after 1970 but they continue to reflect the ordinary business cycle and not a longer cycle like prices do. This increase in intensity after 1970 is consistent with the growth in strength of the price cycle after this date. Thus, it appears that building activity no longer follows a Kuznets real estate cycle in the post-1933 era, whereas prices still do.
It has been 22 years since the last McFadden price peak in 1980, so another one is due now. The recent rise in existing housing prices in Figure 5 suggests that we may have entered the boom stage the marks the final approach to a Kuznets peak. The data in Figure 5 is not consistent with the idea that housing is particularly overvalued right now. For both existing and new houses, prices relative to income are well below the levels reached at previous real estate peaks in the late 1970's and 1980's, respectively. Existing home prices would have to rise another 10% relative to income for valuations to reach late 1970's levels. This level would be reached in a couple of years if current trends continue, implying a Kuznets real estate peak in 2004. At 24 years this cycle is on the long side, but the cycle before was unusually short, so the average of about 18 years is preserved. Applying the 18 year standard spacing to the last trough in 1992 projects the next Kuznets trough for around 2010.
Table 6 summarizes the post-1933 cycles using the price data to define the Kuznets cycle, and obtaining business and bull/bear market cycles from Table 2 and 3 respectively. For this presentation, I assume that the 1980 and 1982 recessions are really a single "double dip" recession. Table 6 shows that the tendency for one bull/bear stock market cycle to coincide with a business cycle persisted until the 1960's, after which business cycles started to reflect two or even three bull/bear market cycles. This extension in business cycle length seems to have occurred just before the time that larger-scale real estate cycles re-appeared.
|1933- 1938||1932||1966 (con't) (K-summer)
|1945||1946||1982 P (K-fall)
1The bear markets in 1934, 1960, 1980 and 1984 are skipped for
reasons given in the next section on SMECT
2The 1980 and 1982 recessions are treated as one double-dip recession
The 12 regular business cycles since 1933 average 68 months in length, much longer than the 41 month average before 1933. The standard deviation is very large (47% of the length) compared to that for the industrial period (22% of length) or even the pre-industrial era (35% of length). This degree of dispersion is even greater than what one would expect for cycles of random length. Closer examination shows that they appear to fall into two categories: eight short cycles, 28 to 64 months in length (average 49) and four long cycles, 88 to 125 months in length (average 108). A t-test on these two sets of cycles shows the hypothesis that there exist two populations of modern business cycles of two lengths is significant at only the 80% level (95% is usually considered significant). If we consider the twin recessions in 1980 and 1982 as a single "double-dip" recession, we would then have 11 business cycles since 1933 that fall into two categories, long cycles of 105±16 months and short cycles of 50±9 months. These two populations are statistically different at the >96% confidence level.
These two classes of cycles have rather small standard deviations relative to length (15% and 18%), similar to the tight dispersion of the industrial cycles. This suggests that each cycle has its own mechanism, just like industrial era business cycles were linked to inventory cycles. Table 6 shows that business cycles have been either one ~4 yr bull/bear market cycle long or 2-3 (2.25 average) 4-year bull/bear cycles long. This suggests that the mechanism associated with the shorter 4-year business cycle is also associated with the 4-year periodicity of bear markets. The 4-year length of this cycle is similar to that of the old Kitchen cycles, suggesting that this cycle has persisted after 1933, but now is slightly longer than it used to be. The 9-year length of the longer cycle is similar to that of the old Juglar cycle. No clear-cut relation between Juglar and Kuznets cycles is evident from Table 6. To explore cycle structure further I present the SMECT system of nested binary cycles in the following section.
The SMECT system of nested cycles
The market researcher Robert Bronson recently presented an intriguing scheme for connecting the K-cycle with shorter business cycles which he calls the Stock-Market and Economic Cycles Template or SMECT (see Table 7). SMECT is primarily focused on the stock market rather than the economy and Kitchen cycles (the shortest cycles) are defined by the periodicity of bear markets rather than by NBER business cycles. Thus, one can count three Kitchen cycles, 1896-1900, 1900-1903, 1903-1907 in the secular bull market era (K-spring) from 1896-1907. Similarly there are four Kitchens in the 1907-1920 K-summer period, four Kitchens in the 1920-1933 K-fall period and four Kitchens in the 1933-1949 K-winter period, for a total of 15 Kitchen cycles in the 1896-1949 K-cycle. (The 1934 bear market was not counted for reasons that will be discussed later). Fifteen Kitchen cycles per Kondratiev is close to the 16 expected from SMECT (Table 7). In SMECT, the Kondratiev seasons/secular trends are called Kuznets cycles, which are further subdivided into two Juglar cycles. Note that the 18-year real estate (Kuznets) cycle or the panic-defined Juglar cycle are not the same as the stock market cycles called Kuznets or Juglar in SMECT. Bronson points out that the idea that 16 Kitchen cycles make up one Kondratiev is consistent with the traditional length of the K-cycle of 53 years and the standard length of the Kitchen cycle of 40 months. That is, sixteen 40-month cycles span about 53 years.
Table 7. The SMECT scheme of nested cycles.
|Cycle||Typical length today|
|Juglar||2 Kitchens or 8 years|
|Kuznets||2 Juglars, 4 Kitchens, or 16 years|
|BAAC Supercycle (Stock Cycle)||2 Kuznets, 4 Juglars, 8 Kitchens, or 32 years|
|Kondratiev cycle||2 Supercycles, 4 Kuznets, 8 Juglars, 16 Kitchens, or 64 years|
Although the larger cycles in SMECT are formally defined as geometric series multiples of Kitchen cycles, the larger cycles are not without auxiliary meanings. For example, the recessions of 1908, 1914, and 1921 (respective unemployment of 8.0%, 8.5% & 11.7%) were more severe than the neighboring recessions in 1904, 1911 and 1919 (respective unemployment 5.4%, 6.7% and 1.4%). The latter involve only Kitchen troughs, while the former denote both Juglar and Kitchen troughs. The 1908 recession is even more special, as it was associated one of the recurrent financial panics and marked the end of a secular bull market. This recession was also a Kuznets bottom. Finally, the 1921 recession was the most severe of them all and saw a long-term interest rate peak in the prior year. This recession marked an even higher-order change: the beginning of the Kondratiev downwave. Similarly, the 1933 and 1921 recessions were by far the worst of the 1920's era and marked Kuznets bottoms. The 1942 bear market was larger than either the 1938 and 1946 events showing the effect of the Juglar cycle. Recall the hypothesis that a relationship between the old Juglar cycle and the K-season emerged at the same time the Juglar cycle was becoming uncoupled from the real-estate cycle. The 1907 Panic was not real-estate related. It seems to have been produced by the old Juglar financial cycle. It would appear that the SMECT Kuznets cycle is the same as the financial Juglar cycle. On the other hand, the real-estate cycle continued to operate independently of SMECT.
Bronson advances the idea that with the onset of government economic management, the Kitchen cycle expanded from 40 months to four years in length, reflecting an emergent alignment between Kitchen-type business cycles and the four year electoral cycle. This increase in Kitchen length then gave rise to a corresponding increase in length of the higher-order cycles. Specifically, the Kondratiev cycle length increases from about 53 to 64 years. Support for the idea of a link between the electoral cycle and periodicity of bear markets is the fact that 14 of the 20 bear market troughs since 1933 (Table 2) fall in non-Presidential election years. The probability of this arising from chance is 0.0016%, providing strong evidence for such a four year cycle in the stock market. Thus, the SMECT model presents an explanation for Kondratiev cycle lengthening as well as providing an elegant categorization of the cycle structure of the 1896-1949 K-cycle. It also provides a mechanism for the 4 year post-1933 business cycles described earlier.
Proceeding beyond the 1949 Kondratiev trough one can count the number of bear markets that occurred during the 1949-1966 secular bull market (also a Kuznets cycle in SMECT) for the post 1949 Kondratiev cycle. Consulting Table 2, one counts five bear markets instead of the four expected from SMECT. However, using the strong correlation between the electoral cycle and the stock bull and bear markets, a good case can be made for mid-term elections (like this year) acting as "bear market attractors". Bear markets that occur in Presidential election years should be ignored as spurious, in which case the bear market bottoms in 1932, 1960, 1980 and 1984 should be ignored. Of course, the bear market bottom in 1932 is simply too large to be ignored, and since it happened before the rise of regulation, it should be included, but then the 1934 bottom should be skipped in its place because it occurred too close to the 1932 bottom to be part of a four year cycle. In this case there are only 4 bull/bear cycles in the 1949-1966 Kuznets cycle and also 4 bear markets in the 1966-1982 Kuznets cycle. Thus, the first half of the current Kondratiev cycle also conforms to SMECT.
Here the power of the SMECT model is shown by its ability to bridge the transition for the industrial era to the regulated era. The explanatory pattern seen in the earlier cycles remained as well. The Juglar bottoms in 1958, 1966, 1974 and 1982 showed more severe bear markets (average severity 3.0) than the Kitchen bottoms in 1954, 1962, 1970, and 1978 (average severity 1.7). Similarly, the recessions associated with the Juglar bottoms showed greater severity (average unemployment 8.4%) than those associated with only a Kitchen bottom (average unemployment 5.9%).
However, when we come to the post-1982 period there is a problem. Counting "election" bear markets for the 1982-2002 secular bull market period (Kondratiev fall) gives five rather than four as expected from SMECT. The enormous length of this Kondratiev fall season (20 years for 1982-2002 as compared to 12 for the previous one over 1920-1932) is the problem. Bronson resolves this problem by setting the end of the recent BAAC Supercycle bull market in 1997 when the equally-weighted stock index began its secular bear market, instead of 2000, the nominal peak in the major capitalization-weighted indices. In this way 1998 becomes the Kuznets cycle bottom, equivalent to1932 in the last K-cycle. But this assignment makes the current bear market (the worst of the current K-cycle) simply a reflection of a Kitchen cycle by itself. It destroys the explanatory power of SMECT in which more severe events generally were associated with the bottoms of higher-order cycles. Yet if we preserve the explanatory power of SMECT by setting 2002 as a Kuznets and Juglar bottom and 1990 as a Juglar bottom, then we have three Kitchens in the most recent Juglar cycle--in violation of the SMECT structure.
Continue to Generations and Business Cycles Part II