Financial Analysts Journal

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ufaj20

Stocks for the Long Run? Sometimes Yes,

Sometimes No

Edward F. McQuarrie

To cite this article: Edward F. McQuarrie (2024) Stocks for the Long Run? Sometimes Yes,
Sometimes No, Financial Analysts Journal, 80:1, 12-28, DOI: 10.1080/0015198X.2023.2268556

To link to this article: https://doi.org/10.1080/0015198X.2023.2268556

View supplementary material

Published online: 14 Jan 2024.

Submit your article to this journal

Article views: 18616

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ufaj20
Perspectives Financial Analysts Journal | A Publication of CFA Institute
https://doi.org/10.1080/0015198X.2023.2268556

Stocks for the Long Run?

Sometimes Yes,
Sometimes No
Edward F. McQuarrie
Edward F. McQuarrie is a Professor Emeritus at Santa Clara University, Santa Clara, California. Send correspondence to Edward F.
McQuarrie at emcquarrie@scu.edu.

When Jeremy Siegel published his ust stocks beat bonds over the long run? Should investors
“Stocks for the Long Run” thesis, little
was known about 19th-century stock
and bond returns. Digital archives
have made it possible to compute real
M expect a stock portfolio to compound in real terms at 6% to
7% per year?

Over the last three decades, Jeremy Siegel has defended both claims
total return on US stock and bond under the rubric of “Stocks for the Long Run.” He built on foundations
indexes from 1792. The new historical laid by Ibbotson and Sinquefield (1976), who showed substantial out-
record shows that over multi-decade
performance for stocks over bonds dating back to 1926, and on
periods, sometimes stocks outper-
Shiller (2015), who found strong stock performance dating back to
formed bonds, sometimes bonds out-
performed stocks and sometimes they 1871. By the 1980s, the degree to which stocks had outperformed
performed about the same. New bonds, as shown in the historical record, had begun to puzzle theorists
international data confirm this pattern. (Mehra and Prescott 1985).
Asset returns in the US in the 20th
century do not generalize. Regimes of Siegel’s contribution was to push the beginning of the stock record back
asset outperformance come and go; from 1871 to 1802, drawing on the much earlier data collection efforts of
sometimes there is an equity pre- Smith and Cole (1935),1 and to push the bond record back from 1925 to
mium, sometimes not. 1802, drawing on the compilation in Homer (1963). In later book editions,
Siegel (2014) was able to update portions of the 19th-century stock
Keywords: asset allocation; equity record using data from Goetzmann, Ibbotson, and Peng (2001), without
premium; historical stock and bond
returns; international returns; long-term disturbing the thesis of a persistent, strong equity premium.
asset returns
However, in the 30 years since Siegel (1992a, 1992b) first advanced
his thesis, new data have emerged on 19th-century stock and bond
returns. The new findings substantially diverge from the record avail-
able to Siegel and are at variance with the 20th-century record in the
annual compilation maintained by Ibbotson (2020) in the Stocks,
Bonds, Bills and Inflation yearbook. This paper provides a summary of
the new findings and considers the implications for investment
I would like to acknowledge the pathbreaking work of Richard Sylla, head of the
research team whose data compilation made this new market history possible. I
thank him, Rob Arnott, Jason Zweig, James Grant, Evan Lorenz, Jeremy Siegel, Mark
Disclosure: No potential conflict of Hulbert, Allan Roth, Mike Staunton, William Bernstein, and Bryan Taylor for helpful
interest was reported by the author(s). comments on earlier iterations of this work. Special thanks to Bryan for providing
access to GFD data and to the Editorial Team for indispensable guidance.
This article has been corrected with minor changes. These changes do not impact
PL Credits: 0.75 the academic content of the article.

12 © 2023 CFA Institute. All rights reserved. Volume 80, Number 1

 Stocks for the Long Run? Sometimes Yes, Sometimes No

practice.2 The post-1926 period in the US emerges yields. Later, when Goetzmann, Ibbotson, and Peng
as highly distinct within a broader historical view, (2001) became available, Siegel used their post-1824
with outcomes that do not generalize to other times data on dividends. The Goetzmann et al. data were
and places. limited to stocks in New York and, as data collection
proceeded, were found to contain numerous omis-
sions but also unique information, which supple-
The New Historical Record mented the efforts described next.
The online Appendix links to working papers which
The same digitized sources tapped for share counts
detail the methods used to collect the new data. The
also provided many dividends. Nonetheless, before
Appendix also links to files containing the raw secu-
1865 the record of semi-annual dividends remained
rity-level data, hosted by Financial Analysts Journal
spotty. I then discovered that early firms would publish
for use by future researchers and to allow auditing of
a notice in the local newspaper when a dividend was
the findings.
declared. Digitized records were searched for the term
“dividend” year by year, city by city. That search, sup-
Prices. A team led by Richard Sylla examined
plemented by the Goetzmann et al. data, provided
hundreds of newspapers published before 1860 and
what appeared to be 95% or better coverage of the
compiled prices on a much larger set of securities
dividends paid by stocks in the new database.
than had previously been available (Sylla, Wilson,
and Wright 2006). After 1860, I hand-gathered After 1865, the CFC, annual Poor’s editions, and
prices from the American Railroad Journal, Banker’s Martin provided the dividend record. These sources
Magazine, the Commercial and Financial Chronicle were also used to track share splits, stock dividends,
(CFC), and other contemporary publications, and rights issues.
through 1897 (stocks) and 1926 (bonds), supple-
mented by the collections of Hall, Payne, and Summary. The new record improves on the old as
Sargent (2018), Macaulay (1938), and Martin follows.
(1898). The raw price data were condensed into
an annual series of average January prices, as 1. Includes securities trading outside of New York, in
Cowles (1939) had done, the better to link to his Boston, Philadelphia, Baltimore, and southern and
post-1871 series (Shiller 2015). western cities. The new record covers three to five
times more stocks and five to ten times more
The new stock and bond series both begin in January bonds.
1793, the first month where more than three stocks 2. The expanded coverage captures more failures,
were found trading regularly.3 The new stock series reducing survivorship bias. Outside of New York,
ends January 1897; returns did not change much banks failed, turnpikes succumbed to canals,
when survivorship bias in the Cowles (1939) series canals lost to railroads, and new railroads fell on
from 1871 to 1897 was corrected (McQuarrie 2020). hard times and never paid a dividend. States
Accordingly, following 1896 the stock series and fol- defaulted in the Panic of 1837. Corporate bonds
lowing 1925 the bond series used by Siegel are were downgraded or defaulted.
appended to complete the record.4 3. Includes federal, municipal, and corporate
bonds, and large numbers of each, as opposed
Share Counts. Share counts were sought to get to the one bond used each year by Siegel prior
capitalization-weighted stock returns. The develop- to 1862. The new bond record observes price
ment of digital archives of magazines and newspa- changes, where Siegel had to infer price change
pers and the digitization of out-of-copyright books from successive yields.
made this possible. An example is Goddard (1831), 4. Calculates capitalization-weighted total return
which contains share counts for most of the larger for stocks. The old stock record was either price-
stocks of that era. Corporate biographies and state weighted (Goetzmann, Ibbotson, and Peng
legislative reports were also accessed, along with 2001) or equal-weighted (Smith and Cole 1935)
early compilers such as Poor (1860).5 and lacked information on dividends.

Dividends and Distributions. Information on

dividends was gathered to construct an index of the Shortcomings remain. These include: (1) annual fre-
total return on stocks. Siegel’s initial sources did not quency; (2) time-averaged data (Schwert 1990); (3)
observe dividends; he had to estimate dividend exclusion of stocks that traded over the counter6;

Volume 80, Number 1 13

Financial Analysts Journal | A Publication of CFA Institute

and (4) inability to observe ex-dividend and Corrected History of US Stock and Bond
ex-coupon dates and prices. Performance. Figure 1 shows the new record of
US stock and bond performance from 1792. It can
be compared to Figure 5-4 in Siegel (2014, 82). Two
recent periods are marked out in Figure 1, with the
Key Changes. Stock returns before 1871 look dif-
bond performance line reset at the beginning of the
ferent now because of the reduction in survivorship
right panel to facilitate comparison.
bias. The old record omitted the largest stock that
traded before the Panic of 1837, the 2nd Bank of During the decades following 1981 (right panel),
the United States. At the peak before the Panic hit, stocks and bonds performed about the same.
the 2nd BUS accounted for almost 30% of total US Occasionally, stocks soared above bonds, as during
market capitalization. It failed as the Panic pro- the dotcom boom, but as late as December 2008,
ceeded, with shares dropping in price from $120 to stocks had fallen behind bonds in what by then had
$1.50, and never recovered. been a 27-year horse race.

Bond returns look more positive now, due to inclu- The right panel of Figure 1 offers little to support
sion of corporate bonds, inclusion of a broader selec- “Stocks for the Long Run.” Stocks did very well in
tion of federal and municipal bonds, and an the boom that began in 1982—but so did bonds.
adjustment for the greenback price of interest paid in Periods where stocks soared ahead of bonds were
gold coin between 1862 and 1879.7 offset by plunges that took stocks back down to the

Figure 1. Stock and Bond Performance from 1792

Note. Performance through December 2019 (pre-pandemic). The bond performance line (dark brown) is reset equal to stock wealth
at the end of 1981 to facilitate comparison in the years that follow.

14
 Stocks for the Long Run? Sometimes Yes, Sometimes No

bond line. Most of this portion of Figure 1 postdates returns. Counts of the rolls in which each asset pre-
Siegel’s earliest efforts and can be approached as a vailed can treated as the odds that stocks will beat
failure to replicate. bonds over such a holding period. If multiple intervals
are examined, a pattern across roll length can be
The middle panel of Figure 1 shows the prior four interpreted as a change in odds as the holding period
decades. These data dominated the record when
grows longer.
Siegel fashioned his thesis. This portion shows poor
long-term performance by bonds with sustained neg- Siegel (2014) includes such a rolls analysis,9 and his
ative real returns. Stocks rose far above bonds fol- data indicated that the longer the holding period, the
lowing World War II. greater the odds that stocks will outperform bonds.
Specifically, the odds in favor of stocks rose from 3-
The middle and right panels of Figure 1 capture two
in-5 over any single year to 9-in-10 over 30-year
different regimes in the relative performance of
rolls, offering strong support for “Stocks for the
stocks and bonds: a recent era of near-parity perfor-
Long Run.”
mance and an earlier era in which stocks did well and
bonds did poorly. Each was about four decades in The new historical record tells a different story.
length, the long run by any standard measure. And Table 1 provides overall results with a breakout by
yet the relationship between stock and bond returns, sub-period, using 1862 and 1942 as the break points.
and the absolute level of return on bonds, were strik- Recall that the new stock record shows the greatest
ingly different across these two adjacent periods. deviation from the old record before the Civil War
and that World War II marked the beginning of the
The remainder of Figure 1 shows stock and bond
worst bond bear market in US history.
performance from 1792 to 1941. Performance
before 1942 looks more like the right panel and not Results for the entire 227 years in Table 1 are weakly
so much like the middle panel: It shows rough parity supportive of “Stocks for the Long Run”: The odds
in performance, with stocks and bonds producing that stocks will outperform bonds do increase as the
about the same wealth accumulation by 1942. The holding period lengthens from 1 to 50 years. But
extended historical record shows short spans of using the new historical data, these odds never get
extreme stock outperformance, most notably in the much higher than 2-in-3 and increase only slowly as
1920s. But there is no period, going back to 1792, the holding period stretches out from 5 to 50 years.
that looks anything like the sustained postwar surge
of stocks over bonds seen in the middle panel of However, the subperiod results in Table 1 do not
Figure 1. Prior to that point stocks and bonds had support “Stocks for the Long Run.” Prior to the Civil
more typically run neck and neck.8 War, the pattern was the reverse: The longer the
holding period, the greater the odds that stocks
Rolling Returns. Figure 1 used carefully selected would underperform bonds. In fact, for holding peri-
start dates to make a point about regime change. ods of 30 and 50 years, in that era stocks always lost
When return series are found to be sensitive to the to bonds. Conversely, the most recent period is
point of beginning, it is useful to compute rolling strongly supportive: From 1942 the odds that stocks

Table 1. Odds that Stocks Beat Bonds over Rolls of Increasing Length
Holding Period in Years
1 5 10 20 30 50

All data: 55.1% 62.3% 63.8% 64.4% 67.2% 68.0%

Counts: 227 223 218 208 198 178
Sub-periods:
Rolls ending 1793–1862 47.8% 41.5% 36.7% 12.0% 0.0% 0.0%
Rolls ending 1863–1941 52.5% 63.8% 65.0% 70.0% 70.0% 53.8%
Rolls ending 1942–2019 64.1% 78.2% 83.3% 92.3% 98.7% 100.0%

Note. Annualized real returns were calculated for stocks and bonds over each holding period and then a count was taken of the
number of rolls where the stock return exceeded the bond return. The percentage stated is that count divided by the total num-
ber of rolls.

Volume 80, Number 1 15

Financial Analysts Journal | A Publication of CFA Institute

outperform start high and increase with holding regime following World War II when bonds fell into
period, until by 30 years the odds converge on an abyss.
100%.10
Figure 2 charts the stock–bond advantage over 10-
No consistent relationship between asset outper- year rolls to portray the magnitude and durability of
formance and length of holding period can be the stock advantage at different points in market his-
extracted from Table 1. The results are better inter- tory.11 It shows that the stock advantage has regu-
preted as showing changes in regime. Prior to 1942, larly reversed, even in the postwar period.
a regime of parity performance held sway: Unsurprisingly, in view of Table 1, two of the worst
Sometimes stocks outperformed, sometimes bonds. runs occurred before the Civil War. Figure 2 also
After World War II, a new regime of extreme stock shows how unusual the decades following World
outperformance took hold (Figure 1). War II were when viewed through a longer lens.

The rolls analysis broken out by sub-period calls into The pattern visible in Figure 2 is best described as
question the stationarity of stock and bond returns. alternation without periodicity and with few con-
It supports instead the idea of time-varying regimes straints on magnitude. Sometimes stocks win, and
in asset returns. Sometimes stocks outperform bonds, sometimes they lose. Sometimes stocks fall behind
and sometimes bonds outperform stocks. What mat- bonds only briefly, at the depths of a bear market,
ters is not the length of the holding period, but while at other times the disadvantage is sustained
which regime is in operation: the regime that held well past that bottom. Most important: Figure 2 does
before the Civil War, when stocks languished, or the not show a stock advantage that waxes and wanes,

Figure 2. Variability of the Stock (Dis)Advantage over 10-Year Holding Periods

16
 Stocks for the Long Run? Sometimes Yes, Sometimes No

but a stock advantage that repeatedly reverses into a must quickly revert. Rather, whether there is a stock
deficit. advantage at all depends on the regime in place.

Next, Figure 3 superimposes the stock and bond Setting relative performance aside, the next section
return series themselves, also as 10-year rolls. It probes the belief that stock investors have always
shows that a stock (dis)advantage can be created done well when the holding period stretches out
through diverse combinations and permutations. The over decades.
decades following World War II are the only instance
where extreme positive returns on stocks coincided Worst Stock Returns. Figure 4 charts the real
with sustained negative returns on bonds to produce total return on stocks for 20-, 30-, and 50-year rolls;
a large and enduring stock advantage. This unique the top panel of Table 2 calls out the worst cases.
and relatively recent event, fatefully combined with Even with the new history in hand, the record shows
the old, faulty 19th-century sources, led financial his-
no instances of a negative 20-year return,12 much
torians to extrapolate that in the long run, stocks
less a negative 30- or 50-year return. And Table 2
would always beat bonds by a substantial amount.
confirms that the worst 30-year return was stronger
It seems unlikely that two centuries suffice to field all than the worst 20-year return, while the worst 50-
the permutations of stock and bond performance year return was better still and not bad in absolute
that may occur. But two centuries are sufficient to terms, about 4% real. Nonetheless, over multi-decade
undermine the idea that an advantage of stocks over holding periods, there have been repeated instances
bonds, of þ2% to þ4% annualized, represents any where the real total return on stocks was 2% or less
kind of lawful regularity or any range to which values (Table 2).

Figure 3. How Stock and Bond Performance Combine to Augment or Reverse a Stock
Advantage

Volume 80, Number 1 17

Financial Analysts Journal | A Publication of CFA Institute

Figure 4. Best- and Worst-Case Scenarios for Stocks over 20-, 30- and 50-year Holding
Periods

Table 2. How Bad Can It Get for Stocks?

20 years 30 years 50 years
Endpoint Annualized return Endpoint Annualized return Endpoint Annualized return

A. Worst equity returns

1948 þ0.75% 1861 þ2.00% 1859 þ3.80%
1855 þ1.09% 1932 þ2.75% 1931 þ4.29%
1920 þ1.22% 1920 þ3.64% 1978 þ4.48%
1981 þ1.29% 1994 þ4.27% 2008 þ4.72%
B. Worst equity deficits
1858 −5.13% 1859 −3.37% 1859 −2.43%
1833 −1.84% 1843 −2.26% 1884 −1.14%
1893 −1.76% 1877 −1.20%
2008 −1.64% 1896 −1.10%
1932 −1.45%
2019 −1.10%

18
 Stocks for the Long Run? Sometimes Yes, Sometimes No

Worst Bond Returns. Figure 5 shows 20-, 30-, are concentrated in the 19th-century and in the
and 50-year rolls for bonds. The worst rolls all come newly collected data. The decades before the Civil
after World War II and prior to the bull market in War emerge as the very worst; on this metric, the
bonds that began late in 1981. Unfortunately, these 1930s barely make the cut. This information was not
are the bond returns that dominated the first few available 30 years ago.
Ibbotson yearbooks, making investors that much
more receptive to “Stocks for the Long Run.” Before
Failure Out-of-Sample. The new data indicate
that “Stocks for the Long Run” was built on a faulty
the war, US bond investors could have reassured
premise: that the strong returns on stocks seen after
themselves the way US stock investors can still do:
World War II, combined with the poor returns on
that there had never been a negative return on
bonds through 1981, reflected a stationary process.
bonds over a 30- or 50-year holding period. The old historical record appeared to show that
The worst 50-year return before the war, of 3.78% stocks had always returned 6% to 7% real and there
real annualized, was not bad in absolute terms and had always been a substantial equity premium. The
not much different from the worst 50-year return for new data show that the 19th century, particularly
the antebellum era, saw quite different returns than
stocks (Table 2). Then the regime changed.
the 20th century, with repeated equity deficits.

Negative Equity Premia. The bottom panel of Unfortunately, the new data remain vulnerable to
Table 2 lists the worst equity deficits for each inter- dismissal by practical investors, who may deem the
val. Unlike the top panel, the worst equity deficits results too old to be relevant. The next section

Figure 5. Best- and Worst-Case Scenarios for Bonds over 20-, 30-, and 50-year Holding
Periods

Volume 80, Number 1 19

Financial Analysts Journal | A Publication of CFA Institute

addresses that weakness by demonstrating that However, there are problems with testing “Stocks for
“Stocks for the Long Run” also fails out-of-sample the Long Run” using international data. World ex-US
using 20th-century international returns. returns include nations that suffered defeat in a
major war (Germany, Japan), endured a civil war
(Spain, Portugal), or were invaded and occupied
(Belgium). It could be argued that their inclusion vio-
The Evolving International Record lates ceteris paribus conditions. To construct a more
By the fifth edition, Siegel (2014) had what appeared fair test these problematic nations and periods were
to be out-of-sample support for his thesis in the find- excluded as described in the online Appendix.
ings of Dimson, Marsh, and Staunton (2002).13 Table 3 shows worst case returns over 20-, 30-, and
However, as with the US, the international record 50-year rolls within intact foreign markets, revealing
has been steadily expanded, with further reductions a dozen instances of substantially negative 20-year
in survivorship bias (online Appendix), making it pos- returns, and half a dozen cases of negative 30-year
sible to reexamine “Stocks for the Long Run” outside returns.
the US across a broader sample of markets than had
been available even 20 years ago. “Stocks for the Long Run” does not contemplate such
poor investment outcomes. The new international
record shows the thesis not to be supported when
World Ex-US Stock Returns. The 120-year tested out of sample across intact nations outside
annualized real return on stocks ex-US is now esti- the US.
mated by the Dimson team as 4.4%—not the 6.6%
estimated by Siegel (2014) as the true long-term Non-Stationary Equity Premium. Next, the
expected value. That difference does not sound like 2020 Credit Suisse yearbook calls out returns over
much but as the 2020 Credit Suisse yearbook notes: the trailing 50 years. For the World ex-US, for 1970
A dollar invested in US equities in 1900 resulted in a through 2019, the authors now report parity perfor-
terminal value of USD 1937. … An equivalent mance for equities versus government bonds, at 5.1%
investment in stocks from the rest of the world gave a and 5.0% annualized. Outside the US, the equity pre-
terminal value of USD 179 … less than a tenth of the mium has been just that small in recent decades.
US value.
At the country level, the new record shows multiple
The favorable experience of US stock investors in instances where the equity premium was a deficit for
the 20th century does not generalize. Stock wealth decades, as seen in the new 19th-century US data.
accumulated by the long-term international investor Japan since 1989 is a well- known counterexample
has fallen 90% short of expectation. to Siegel’s thesis. But it is not necessary to set the

Table 3. Worst Multi-decade International Stock Returns Excluding War Losses

Nation 20 years Ending in Nation 30 years Ending in Nation 50 years Ending in

Italy −7.34 1979 Norway −4.40 1978 Italy −0.54 2011

Norway −5.92 1977 Italy −2.35 1991 Norway 0.43 1995
Sweden −5.17 1932 Portugal −1.64 1949 Austria 1.10 1996
Japan −5.02 2009 Sweden −1.10 1932 Sweden 1.61 1948
Switzerland −4.39 1981 Austria −1.02 1976 Belgium 2.04 1908
Austria −4.26 1981 Switzerland −0.78 1991 Spain 2.34 2020
Spain −3.36 1983 Japan −0.78 2019 Switzerland 2.41 2011
France −2.98 1981
Portugal −2.34 1939
SouthAfrica −1.40 1920
UK −1.27 1920
Belgium −1.27 1976
Germany −1.20 1980

Note. Table shows all negative 20- and 30-year returns found and all 50-year returns less than 2.5%. Annualized real percentage
returns.

20
 Stocks for the Long Run? Sometimes Yes, Sometimes No

start date at the culmination of a great bubble to intervals, stocks have performed weakly and/or
unearth a multi-decade equity deficit. bonds have outperformed stocks. World outcomes
and 19th-century US outcomes expose a lack of sta-
Table 4 shows that every country had experienced tionarity that undermines the hypothesis of “Stocks
an equity deficit across two decades, and 18 of 19 for the Long Run.”
had experienced an equity deficit across three deca-
des. There is no obvious clustering of dates in
Table 4 nor any concentration in one century or Interpreting the New Historical
another. Some of the worst instances occurred com-
paratively recently.
Record
The old record suggested that the equity premium
The table reveals that the new 19th-century data was stationary across centuries in the US and
from the US are not exceptional. Worldwide, there across domestic and foreign markets and, likewise,
have been numerous instances where stocks under- that the returns-generating process underlying
performed bonds over multi-decade intervals. There the strong 20th-century US equity returns
is no law that guarantees an equity premium will be chronicled in the Ibbotson Stocks, Bonds, Bills &
received if an investor buys and holds for decades. Inflation yearbook was stationary. The emerging his-
Once the historical lens is widened to include the tory suggests an alternative perspective in which
19th century in the US or swung out to include mar- asset returns fluctuate subject to the regime in
kets outside the US, it is straightforward to find place.
instances of bonds for the long run.
The Regime Thesis. In conventional financial
Summing up, the larger sample of markets outside history, greatest weight would be placed on values
the US shows diverse outcomes for stocks and computed over the longest interval available, on the
bonds. In some periods in some markets for some theory that sampling returns over time follows the
length of time, stocks have enjoyed strong returns same logic as sampling people from groups. Longer
consistent with “Stocks for the Long Run.” In time samples ought to provide more precise esti-
other periods in other markets over other mates of the true expected return on an asset, just

Table 4. Lowest Equity Premia Observed Internationally

20 years Ending in 30 years Ending in 50 years Ending in

Australia −1.98 2008 −0.64 2016 1.48 2019

Austria −8.50 1943 −5.05 1953 −2.46 2011
Belgium −4.21 1948 −1.73 1886 −1.69 1886
Canada −5.41 1886 −3.06 1884 −0.56 1902
Denmark −5.28 1932 −3.95 1946 −2.80 1932
Finland −3.26 2019 1.64 2018 3.57 1967
France −3.03 1821 −2.93 1831 0.27 2011
Germany −3.90 1980 −1.47 2002 −0.99 2011
Italy −5.10 1979 −4.56 2016 −2.99 2011
Japan −9.10 2009 −3.85 2019 −1.41 2011
Netherlands −6.10 1932 −2.26 1934 1.09 1950
New Zealand −6.19 2006 −4.35 2016 0.37 1952
Norway −10.03 1938 −7.46 1947 −4.49 1967
Portugal −8.18 1993 −6.26 2003 −2.26 2014
South Africa −3.03 1985 −1.43 1985 0.61 1932
Spain −5.63 1920 −4.25 1915 −3.38 1936
Sweden −8.08 1932 −3.29 1932 −0.74 1932
Switzerland −1.18 1974 −0.66 1991 0.78 2011
UK −1.52 1939 −1.14 1849 −0.16 1759
Deficit in: 19/19 18/19 12/19

Note. Data from GFD except Portugal from Jord a et al. (2019). Shaded cells include years where the nation was defeated in war,
suffered civil war, or was invaded and occupied.

Volume 80, Number 1 21

Financial Analysts Journal | A Publication of CFA Institute

Figure 6. Changes in the Stock and Bond Correlation over Time

as a larger sample of people would produce a better Stationarity prevails within regimes but not across
estimate of any differences in height or weight them. The idea of temporary stationarity distin-
across groups. guishes the regime thesis from Pastor and
Stambaugh’s (2012) idea that the parameters of the
The regime thesis denies that longer time intervals
return distribution are unknown. Under their thesis,
are like larger samples. It does not assume a popula-
the new 19th-century US data and the country-level
tion of asset returns existing outside of time from
international results can be characterized as an
which larger samples can be drawn by lengthening
expanded sample relative to Ibbotson (2020). The
the series. There are only the asset returns that
much larger sample better captures the true volatil-
have been recorded in history thus far. These do
ity of stock and bond returns by allowing more
not predict future asset returns because their pat-
opportunities for extremes to emerge—in particular,
tern is specific to the regime that prevailed at the
extremes over multi-decade intervals, which are
time. No analysis of US bond returns from 1792 to
1941 could have predicted the bond returns seen necessarily few in a single-market, single-century
following the war—such a bond abyss had never sample (Boudoukh, Israel, and Richardson 2019).
occurred before. The regime thesis differs in expecting sustained but
temporary stationarity. Stocks and bonds ran neck
A regime is a temporary pattern of asset returns that and neck for over a century (Figure 1). Then after
may persist for decades. The regime thesis entails no the war, that regime gave way to one where stocks
periodicity and requires no reversion. Rather, it beat bonds, year after year, decade after decade.
expects to find ceaseless variation, as the permuta-
tions and combinations of individual asset returns Variance and Covariance. More examples of
play out (Figure 3). non-stationarity can be found by examining two

22
 Stocks for the Long Run? Sometimes Yes, Sometimes No

Figure 7. Change in the Volatility of Stocks and Bonds over Time

quantities central to asset allocation and portfolio to the short-term fluctuations studied by financial
management: the correlation between stocks and economists, there may be extended regimes in which
bonds and their respective standard deviations. Many correlations are elevated or depressed. The diversifi-
analysts expect the stock–bond correlation to be cation potential from adding bonds to a stock portfo-
low; over the trailing nine decades the 2020 Stocks, lio has not been stationary across the two centuries
Bonds, Bills & Inflation yearbook has it as 0.17 for charted.
corporate bonds and 0.01 for long government
Figure 7 shows rolling 20-year standard deviations
bonds.14 But those values, measured over the arbi-
on stocks and bonds. The augmented history shows
trary interval represented by the modern post-1926
much lower volatility on stocks in the distant past as
era, prove a poor guide to what a longer historical
compared to the 20th century. Bond variability has
record might reveal.
also not been constant. Because bond variability has
fluctuated through a wide range, and because stock
Figure 6 shows that the correlation has been highly
volatility in the new data is much lower, the aug-
variable over 20-year intervals, ranging all the way
mented record shows the bond standard deviation
from about −0.70 to 0.90, consistent with research
sometimes to be on a par with that of the stock mar-
in financial economics (Aslanidis and Christiansen
ket. Relative volatility has not been stationary.
2012; Li 2002; Yang, Zhou, and Wang 2009).
However, correlations in the modern era did reach Neither returns nor variance nor covariance show
low points never seen before 1926. The overall cor- stable values for stocks and bonds. The new record
relation for the modern era of .01, per the SBBI, is supports concerns about stationarity advanced in
much lower than the 0.61 recorded for the preceding Pedersen, Babu, and Levine (2021) and Goetzmann
134 years.15 The new data suggest that in addition (2020).

Volume 80, Number 1 23

Financial Analysts Journal | A Publication of CFA Institute

Figure 8. Real Price Appreciation without Dividends

Contribution of Dividends. It is not hard to In the earliest decades, dividends accounted for all
find assertions in the literature that dividends have the total return on stocks; in recent decades, their
accounted for some stated percentage of the total contribution has substantially diminished. The break
return on stocks.16 The assertion assumes a station- becomes visible with the new dividend record, which
ary relationship between the two components of shows that the composition of total return has not
total return, with a long enough sample allowing for been stationary.
the relative contribution of price appreciation and
dividends to be estimated with precision. Magnitude of Regime Variation. Under the
regime thesis, history is not a sample; hence, use of
But this is not what the new history shows. Figure 8 sample statistics to compare time periods is question-
shows real price appreciation on US stocks in the able. Nonetheless, if taken primarily as descriptive,
absence of dividends. It looks very different than the statistical tests can help to illuminate the magnitude
real wealth index with dividends reinvested that was of the difference between regimes and the incapacity
charted in Figure 1. The differences are concentrated of longer time samples to improve precision of esti-
in the new data. By 1871 and the beginning of the mate. To that end, Table 5 compares the 19th cen-
Cowles (1939) record, one dollar invested in stocks, tury to the 20th century.
with dividends spent rather than reinvested, had pro-
duced a real loss of $0.50. The price had still not The contrast is stark for bonds: 20th-century bond
recovered by 1942, with a real loss of $0.37. Then returns are significantly lower than 19th-century
the regime changed. Real price appreciation began to returns and significantly lower than 20th-century
mount up. The contribution of dividends drops, as stock returns. By contrast, the 19th century in the
price appreciation begins to account for more and US saw a 100-year equity deficit, with bonds slightly
more of total return. outperforming stocks.17

24
 Stocks for the Long Run? Sometimes Yes, Sometimes No

Table 5. Outcomes by Century

Stocks Bonds
Equity premium Inflation
Mean Wealth Mean Wealth Mean Mean

19th century: 6.68% $322 6.98%a $594 −0.29%a −0.27%a

1800–1899 (12.71%) (9.21%) (10.37%) (5.17%)
20th century: 8.85%b $837 2.32%a,b $6 6.54%a 3.17%a
1900–1999 (19.65%) (10.35%) (18.79%) (5.04%)

Note. Arithmetic mean real total returns. Wealth is the value of $1.00 invested for 100 years (compounded returns can be
extracted by taking the 100th root). Equity premium is the mean of the annual subtractions. Standard deviations are in parenthe-
ses. Means with superscript a are different across periods and those with superscript b are different within period (t tests with het-
erogenous variance, all p values < .01).

Stock returns are not significantly different across But again, if taken in isolation, the international
the two centuries, but the 19th-century stock inves- record, which has been available for some years
tor only accumulated 40% of the wealth accumulated (Dimson, Marsh, and Staunton 2004), could also be
by the 20th-century investor—and just a bit more dismissed by US investors on grounds of American
than half what the 19th-century bond investor got. exceptionalism. It is the joint demonstration of recur-
ring equity deficits in the augmented US and interna-
The equity premium for the 20th century, at 654 bp, tional records that drives home the point: Stocks are
is far above theoretical expectations (Mehra and risky.
Prescott 1985); conversely, the 19th century saw an
equity deficit of 29 bp. If one takes the average The new historical record addresses a conundrum
across the two centuries, it comes close to conven- that gradually emerged as the work of Roger
tional expectations, a bit over 300 bp, and similar to Ibbotson and Jeremy Siegel diffused to a broader
the Siegel (2014) estimate, but that value is no more public. If stocks are risky, investors will demand a
illuminating than to speak of the average rate of premium to invest. But if stocks cease to be risky
inflation, combining gold standard and fiat currency once held for a long enough period—if stocks are cer-
regimes. tain to have strong returns after 20 years and certain
to outperform bonds—then investors have no reason
to expect a premium over these longer periods, given
Conclusion that no shortfall risk had to be assumed. The
expanded historical record shows that stocks can
Summary of Findings. The augmented historical
perform poorly in absolute terms and underperform
record reveals substantial variation in values central
bonds, whether the holding period is 20, 30, 50, or
to financial analysis. Stock returns, bond returns, the 100 years. That documentation of risk resolves the
correlation between them, their standard deviation, conundrum. There can be no equity premium unless
and the contribution of dividends have not been sta- stocks are risky, but with that risk documented—even
tionary. Measurement spanning even 100 years need if infrequent in occurrence—investors can, based on
not bring convergence. Moreover, the 20th century the entirety of the historical record, probabilistically
produced phenomena never seen before, like the vol- expect stocks to outperform bonds over other inter-
atility of stocks after 1929 or the bond abyss after vals, including the interval that corresponds to their
the war. own investment horizon. Conversely, if stocks had
never done poorly outside of war and disaster, there
Implications for Investors. US investors must
would be no reason to expect stocks to continue to
accept that stocks will not always beat bonds, no
do well in peacetime, because no shortfall risk would
matter the holding period. However, presented in
have been demonstrated. The fact of a persistent
isolation, the new 19th-century US data could readily
equity premium, and not just its level, would be the
be dismissed as too old and obtained under far dif-
puzzle (Mehra and Prescott 1985).
ferent macroeconomic conditions. The expanded
international record provides crucial support, showing Once investors accept that stocks can disappoint, no
equity deficits and poor stock returns in recent times. matter the interval, diversified portfolios become

Volume 80, Number 1 25

Financial Analysts Journal | A Publication of CFA Institute

more attractive. In an Ibbotson/Siegel world, any can be supported with reference to Pastor and
portfolio not 100% in stocks must underperform Stambaugh (2012) and Mandelbrot and Hudson
over the long term. Only investors with a low toler- (2010).
ance for short-term volatility can justify mixing in
bonds. But if sometimes stocks do well and some- If the parameters of the distribution of stock returns
times not, even an investor with a high tolerance for are unknown, and not ascertainable using compara-
risk can justify diversifying away from a 100% stock tively brief samples of one century drawn from a sin-
position. Dimson, Marsh, and Staunton show that gle market, then the new historical record should not
since 1900 the 60/40 portfolio has offered a higher surprise. Doubling the sample length in the US, and
Sharpe ratio and smaller drawdowns relative to adding dozens of international samples, was inher-
stocks alone (cf. Asness 1996). Similar considerations ently likely to produce patterns not seen in the post-
caused Bernstein (2002) to recommend it as the 1926 US record of Ibbotson (2020). What appear to
default portfolio. This recommendation becomes eas- be regimes might only be further information about
ier to follow once it is accepted that a 100% stock the parameters of the distribution of stock returns
portfolio will not necessarily outperform a stock– over multi-year and multi-decade periods.
bond blend.
On a Mandelbrot and Hudson view, stock and bond
Implications for Financial Analysts. Now returns are fractal, and patterns of return seen at
that the volatility of the equity premium has been short intervals can be expected to recur at longer
established, including reversals into negative terri- intervals of 10, 20, or more years, once the sample
tory over intervals of arbitrary length, the question size expands sufficiently (Boudoukh, Israel, and
for analysts is whether regimes unfavorable to equi- Richardson 2019). No analyst expects stocks to out-
ties can be predicted. The research on stock–bond perform bonds every year, but if return patterns are
correlations provides an analogy: Here, parameter- fractal, then there should also be no expectation
izations of inflation (level, change) have shown that stocks will outperform bonds over each decade
promise in predicting whether correlations will be or over every two-decade period, and so on. For a
high or low. Future research on the occurrence of time, the data compiled by Ibbotson and Siegel
equity deficits might proceed similarly, seeking out seem to suggest that stock returns were not frac-
macroeconomic factors associated with stronger and tal—that the odds of underperformance became
weaker performance of equities relative to fixed vanishingly small as the holding period lengthened
income. For this endeavor, the expanded record out across decades. The expanded historical record
provides test cases spread across two centuries and shows otherwise.
dozens of markets.
If regimes are real, then it may be possible to pre-
The fate of such efforts depends in part on whether dict their occurrence and even the timing of regime
regimes are real, as glossed by the phrase “temporary shifts. But if the new historical record primarily
stationarity.” Regimes need not be real phenomena; serves to expand the sample of fractally distributed
apparent stationarity may be randomness misper- randomness, then successful prediction becomes
ceived as pattern. This challenge to the regime thesis unlikely.

Editor's Note
Submitted 13 February 2023
Accepted 5 October 2023 by William N. Goetzmann

Notes
1. Siegel drew on Schwert (1990), who had as in Arnott and Bernstein (2002), or to use it as an
assembled multiple series compiled by Smith and input for projecting future returns as in Bernstein (2013)
Cole (1935). or Ilmanen (2022). The present paper makes
no predictions; it is focused on accurate
2. Prior attempts to critique the historical record assembled retrodiction.
by Siegel have been few. Arnott (2009) and Zweig (2009)
are notable exceptions. It has been more customary to 3. Per Hall, Payne, and Sargent (2018), the (federal) bond
accept Siegel’s dataset and perform operations upon it, series could be extended back to late 1790.

26
 Stocks for the Long Run? Sometimes Yes, Sometimes No

4. Siegel used the Shiller/Cowles stock series through 10. Mixed results for the middle period reflect the fact that
1925 and then the total market index from the longer rolls ending 1863 and after may include decades of
Center for Research into Security Prices; for bonds data from before 1863, i.e., from the old regime.
after 1925 he used the long government bond record in
Ibbotson (2020). In splicing to the post-1925 series, 11. This chart refers to “stock advantage” rather “equity
which are keyed to December, January 1926 is premium,” because in technical and theoretical work, the
double-counted. equity premium is defined against the risk-free rate rather
than bonds. Because of the duration mismatch between
5. When available from these sources, issue size was used stocks and the bills that provide the risk-free rate, it is
to weight bond returns. common also to express the equity premium against a
portfolio of long government bonds (see Dimson et al.
6. Banks ceased to trade on exchanges in the mid- 2020 for an example of the joint calculation of the EP).
1840s, after which quotes dwindle in the sources But portions of this chart include corporate bonds as well
consulted. Insurance companies also moved off-exchange as government bonds, hence the substitution of “stock
at that point, with railroads beginning their multi- advantage.”
decade domination of exchange trading. However,
insurance companies, which had been almost as 12. Siegel (2014) makes several claims about stock returns
common as banks in the 1820s and 1830s, had over 20 years (pp. 94-95). See also Exhibit 2.9 in the 2020
already been removed from the index once I learned SBBI, which supports a similar inference. Conversely,
that most had gone to zero and then been recapitalized Dimson, Marsh, and Staunton (2004, 2021) show that this
after the fire of 1835 (English 1880). I did not think it proposition cannot be sustained internationally.
would help the credibility of the new record to assert
13. Earlier editions had used older compilations of data for
that over ten million dollars of capitalization had gone
Germany, Japan, and the UK.
to zero in early 1836. But consequently, prior to 1845
an upward bias remains in the new record of stock 14. For real returns the values are .23 and .09, respectively.
returns.
15. Part of the difference stems from the use of an aggregate
7. Greenbacks were legal tender and could be reinvested in bond index for parts of the period before 1926 (corporate
bonds paying gold, requiring this adjustment to capture bonds have a higher correlation with stocks). To reinforce
total return. the overall point about the volatility of correlations, if the
chart were taken out through 2022 the 20-year
8. The scale of this log chart is both misleading and correlation would already have risen to minus 0.17.
illuminating. A doubling or tripling over shorter intervals
reads as a squiggle, obscuring the relative outperformance 16. For example, John Bogle at https://www.etf.com/
that did occur from time to time, as in the 2.5 times publications/journalofindexes/joi-articles/3869-the-
outperformance of bonds by the late 1850s. But the scale importance-of-investment-income.html. For a similar
and the reset bond line also make clear that the estimate and procedure, see Siegel (2005, 126).
magnitude of the outperformance of stocks in the
decades after World War II had no precedent and has not 17. All 100-year rolls from the first, ending 1892, through the
been seen since. one ending 1917 saw equity deficits of dozens of basis
points; there was also a tiny deficit for the 100 years
9. See, e.g., his Table 6-1, p. 96. ending 1932.

References
Arnott, R. 2009. “Bonds: Why Bother?” Journal of Indexes 12 Boudoukh, J., R. Israel, and M. Richardson. 2019.
(3): 6–17. “Long-Horizon Predictability: A Cautionary Tale.” Financial
Analysts Journal 75 (1): 17–30. doi:10.1080/0015198X.2018.
Arnott, R. D., and P. L. Bernstein. 2002. “What Risk Premium Is
1547056.
“Normal”?” Financial Analysts Journal 58 (2): 64–85. doi:10.
2469/faj.v58.n2.2524. Cowles, A. 1939. Common-Stock Indexes. Bloomington, IN:
Principia Press.
Aslanidis, N., and C. Christiansen. 2012. “Smooth Transition
Patterns in the Realized Stock–Bond Correlation.” Journal of Dimson, E., P. Marsh, and M. Staunton. 2002. Triumph of the
Empirical Finance 19 (4): 454–464. doi:10.1016/j.jempfin.2012.
Optimists: 101 Years of Global Investment Returns. Princeton,
04.005. NJ: Princeton University Press.
Asness, C. S. 1996. “Why Not 100% Equities.” The Journal of
Portfolio Management 22 (2): 29–34. doi:10.3905/jpm.1996.29. Dimson, E., P. Marsh, and M. Staunton. 2004. “Irrational
Optimism.” Financial Analysts Journal 60 (1): 15–25. doi:10.
Bernstein, P. L. 2002. “The 60/40 Solution.” Bloomberg Personal 2469/faj.v60.n1.2589.
Finance, 28–30.
Dimson, E., P. Marsh, and M. Staunton. 2020. Credit Suisse
Bernstein, W. 2013. “The Paradox of Wealth.” Financial Global Investment Returns Yearbook 2020. Credit Suisse
Analysts Journal 69 (5): 18–25. doi:10.2469/faj.v69.n5.1. Research Institute, Zurich, Switzerland.

Volume 80, Number 1 27

Financial Analysts Journal | A Publication of CFA Institute

Dimson, E., P. Marsh, and M. Staunton. 2021. “American Martin, J. G. 1898. A Century of Finance. Published by the
Exceptionalism: The Long-Term Evidence.” The Journal of author, Boston, MA.
Portfolio Management 47 (7): 14–26. doi:10.3905/jpm.2021.1.
257. McQuarrie, E. F. 2020. “A Re-Examination of Stock Market
Returns from 1871-1897: Did Cowles Get It Right?” SSRN
Durand, D. 1942. Basic Yields of Corporate Bonds, 1900-1942. #3564496.
New York, NY: National Bureau of Economic Research.
Mehra, R., and E. C. Prescott. 1985. “The Equity Premium: A
English, S. 1880. “Chronological History of the Fire and Marine Puzzle.” Journal of Monetary Economics 15 (2): 145–161. doi:10.
Insurance Companies of the City of New York.” Insurance 1016/0304-3932(85)90061-3.
Times, November, 771–778.
P
astor, L., and R. F. Stambaugh. 2012. “Are Stocks Really Less
Goddard, T. H. 1831. A General History of the Most Prominent Volatile in the Long Run?” The Journal of Finance 67 (2): 431–
Banks in Europe [with] a Statistical and Comparative View of the 478. doi:10.1111/j.1540-6261.2012.01722.x.
Moneyed Institutions of New York. New York, NY: H. O. Sleight.
Pedersen, L. H., A. Babu, and A. Levine. 2021. “Enhanced
Goetzmann, W. N. 2020. “The Financial Analysts Journal and Portfolio Optimization.” Financial Analysts Journal 77 (2): 124–
Investment Management.” Financial Analysts Journal 76 (3): 5– 151. doi:10.1080/0015198X.2020.1854543.
21. doi:10.1080/0015198X.2020.1766287.
Poor, H. V. 1860. History of the Railroads and Canals of the
Goetzmann, W. N., R. G. Ibbotson, and L. Peng. 2001. “A New United States of America. New York, NY: John H. Schultz.
Historical Database for the NYSE 1815 to 1925: Performance
and Predictability.” Journal of Financial Markets 4 (1): 1–32. doi: Schwert, G. W. 1990. “Indexes of US Stock Prices from 1802
10.1016/S1386-4181(00)00013-6. to 1987.” The Journal of Business 63 (3): 399–426. doi:10.1086/
296513.
Hall, G. J., J. Payne, and T. J. Sargent. 2018. US Federal Debt
1776-1960: Quantities and Prices. NYU Stern, Department of Shiller, R. J. 2015. Irrational Exuberance. Revised and expanded
Economics, New York, NY. third edition. Princeton, NJ: Princeton University Press.

Homer, S. 1963. A History of Interest Rates: 2000 B.C. to the Siegel, J. J. 1992a. “The Equity Premium: Stock and Bond
Present. New Brunswick, NJ: Rutgers University Press. Returns since 1802.” Financial Analysts Journal 48 (1): 28–38.
doi:10.2469/faj.v48.n1.28.
Ibbotson, R. G. 2020. Stocks, Bonds, Bills & Inflation. Chicago, IL:
Wiley. Siegel, J. J. 1992b. “The Real Rate of Interest from 1800–1990:
A Study of the US and the UK.” Journal of Monetary Economics
Ibbotson, R. G., and R. A. Sinquefield. 1976. “Stocks, Bonds, 29 (2): 227–252. doi:10.1016/0304-3932(92)90014-S.
Bills, and Inflation: Year-by-Year Historical Returns (1926-
1974).” The Journal of Business 49 (1): 11–47. doi:10.1086/ Siegel, J. J. 2005. The Future for Investors. New York, NY:
295803. Random House.

Ilmanen, A. 2022. Investing amid Low Expected Returns: Making Siegel, J. J. 2014. Stocks for the Long Run. 5th ed. New York,
the Most When Markets Offer the Least. Hoboken, NJ: Wiley. NY: McGraw-Hill.
 K. Knoll, D. Kuvshinov, M. Schularick, and A. M.
Jorda, O., Smith, W. B., and A. H. Cole. 1935. Fluctuations in American
Taylor. 2019. “The Rate of Return on Everything, 1870–2015.” Business 1790-1860. New York, NY: Russell & Russell.
The Quarterly Journal of Economics 134 (3): 1225–1298.
Sylla, R. E., J. W. Wilson, and R. E. Wright. 2006. Price
Li, L. 2002. “Macroeconomic Factors and the Correlation of Quotations in Early United States Securities Markets, 1790–1860.
Stock and Bond Returns.” SSRN #363641. Inter-University Consortium for Political and Social Research,
Ann Arbor, MI.
Macaulay, F. R. 1938. Some Theoretical Problems Suggested by
the Movements of Interest Rates, Bond Yields, and Stock Prices in Yang, J., Y. Zhou, and Z. Wang. 2009. “The Stock–Bond
the United since 1856. New York, NY: National Bureau of Correlation and Macroeconomic Conditions: One and a Half
Economic Research. Centuries of Evidence.” Journal of Banking & Finance 33 (4):
670–680. doi:10.1016/j.jbankfin.2008.11.010.
Mandelbrot, B. B., and R. L. Hudson. 2010. The (Mis) Behaviour
of Markets: A Fractal View of Risk, Ruin and Reward. London: Zweig, J. 2009. “Does Stock Market Data Really Go Back 200
Profile Books. Years?” Wall Street Journal, July 11th.

28