November 18, 2026 – Welcome back to our Safe Withdrawal Rates Series! You can find the landing page with all SWR Posts here. Today’s post serves two purposes. First, I’d like to introduce a new feature in the Google Sheet I use to simulate retirement withdrawal strategies (see Part 28 for more info). I finally added an international stock market series going back to 1871. If you recall, I used non-US stock index data only starting in 1970, and filled in the S&P 500 index returns for the 1871-1969 period. With the new series, I can finally study how a more internationally diversified portfolio would have performed in the earlier historical period. So, what’s the effect of diversification with international stocks? Would it help raise the safe withdrawal rate in the early period? Should we invest in 100% stocks now, as recommended in the infamous Anarkulova, Cederburg, and O’Doherty paper “Beyond the Status Quo: A Critical Assessment of Lifecycle Investment Advice”? All interesting questions! Let’s take a look…
The new International Stock data: 16 markets, going back to 1871
So, what’s in the new series? A reader (who wanted to stay anonymous) reached out and offered to supply a back-filled international non-US stock index total return series. In the back of my mind, I had always planned to construct this non-US index myself. But it is a lot of messy work, so I always procrastinated. But I was happy when a reader offered me this series on a silver platter. I confirmed that the construction is exactly what I would have done if I had had the time, focus, and motivation to do it myself. Specifically, the new series is a market-cap-weighted index of 16 developed markets, in U.S. dollars, with gross dividends reinvested. For the World War II years, it’s blended toward GDP weights (more on that below). Here’s the international stock index replication procedure in more detail:
- Markets. Australia, Belgium, Canada, Denmark, Finland, France, Germany, Italy, Japan, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland and the UK. Nine are in from 1871; the rest join when their data start (Denmark in 1872, Norway in 1880, Japan in 1885, Finland in 1895, Spain, the Netherlands and Switzerland in 1900).
- Local returns. Annual total returns (price plus gross dividends) come from the Kuvshinov-Zimmermann long-run stock market dataset, with gaps filled from the Jordà-Schularick-Taylor Macrohistory Database.
- Currencies. We convert everything to U.S. dollars at year-end exchange rates. Where only annual averages exist, the year-end rates are estimated so that currency moves land in the right year: gold-standard cross rates before 1914, interwar year-end estimates, and December rates through the German hyperinflation. After 1945, we use published year-end rates, with official rates for Germany in 1948-49 and Japan in 1949-51.
- Weights. Each country’s dollar market cap at the prior year-end. Weights drift with prices, just like in the MSCI indexes. Before 1960, the UK usually made up 40-50% of the index.
- The World War II fix. Pure cap weights have a funny quirk in wartime: as the German, French, Japanese, and Italian markets collapse, their weights shrink too, so the war looks milder than it really was (-21% real for 1939-48, vs. -41% in the DMS data). So for 1939-48 and 1949-59 only, each year blends the cap-weighted return with a GDP-weighted one (blend weights 0.49 and 0.43). That hits the DMS figures of -41% and +345% exactly.
- Monthly returns. This step is really the crux of the series construction. Because most countries’ data series are annual, we can only generate an annual series that we then need to “monthify.” Each year’s return is spread over its 12 months in the shape of the UK market’s monthly dollar return (Bank of England share prices and the dollar-sterling rate), scaled so the months compound exactly to the annual number. This isn’t 100% ideal, but it’s the best we can do with the limited data available. This step designs the new series to have a 100% correlation with the UK intra-year, while maintaining the “correct” annual correlations. It’s not too much of a stretch either, considering that during much of the simulation period, the UK was the largest non-US stock index. It’s certainly close enough for a withdrawal rate analysis!
- The splice. From January 1970 on, the existing MSCI World ex USA returns are untouched. The index still starts at 100 in January 1871, so only the post-1970 level is rescaled.
Here’s what the new data look like next to the S&P 500:

International stocks went nowhere in real terms for 36 years, from the end of 1913 to the end of 1949. At monthly frequency, this drought period actually lasted from 5/31/1913 to 12/31/1950, or about 37.5 years. Ouch! Two world wars, a hyperinflation, several currency reforms, and a few outright confiscations will do that to you. The only slight diversification benefit from non-US stocks was that the drawdown 1929-1932 wasn’t as extreme as in the U.S.; about -32% abroad vs. -79% in the U.S.!
Actual results for investors back then may look even worse due to potential survivorship bias, i.e., the USSR and all eventual Warsaw Pact countries probably had a return of -100%, which we now know in hindsight, but an investor in 1913 didn’t know that yet and might have potentially invested in places like Russia, Romania, Poland, etc.
How good is the new International Stock data?
Before I put anything into the Google Sheet, I wanted to make sure this isn’t garbage in, garbage out. Let’s perform two checks:
1: Rerun the replication concurrently with the MSCI World ex-USA.
Same markets, same data sources, same weights and currency conversions, but applied to 1970-2016, where we have the actual MSCI World ex-USA data (gross dividends, nominal U.S. dollars). The test stops in 2016 because the Kuvshinov-Zimmermann data end there. Not bad at all:
| Period | Correlation of annual returns | Annualized Tracking Error | Replicated Series CAGR | MSCI World ex-USA CAGR |
| 1970-2016 | 0.991 | 2.85% | 9.24% | 9.31% |
| 1970-1989 | 0.988 | 3.53% | 15.77% | 16.08% |
| 1990-2016 | 0.993 | 2.26% | 4.65% | 4.55% |
2: Compare with the DMS data before 1970.
Dimson, Marsh and Staunton publish real returns for a 22-country World ex-USA index over selected historical episodes (Credit Suisse Global Investment Returns Yearbook 2018, Table 1, p. 16). Two of those episodes (1939-48 and 1949-59) were used for the World War II fix, so they match by construction. The other three pre-1970 episodes were not used in the calibration:

That’s honestly better than anyone could have expected. For perspective, even my own S&P 500 series (with its predecessor indexes and historical reconstruction) differs from the DMS USA index by up to 3.4 percentage points a year over 1929-31. Yet the two agree in the long run: 6.49% vs. 6.5% real per year over 1900-2017. The new international series compounds at 4.09% vs. 4.5% in DMS over the same window (Yearbook 2018, Figure 5, p. 11), a 0.4-point gap that sits in years outside the episodes above.
How to use the new International Stock data in the Google Sheet
The new series now lives in the “International Stocks” column of the Asset Returns tab, and the input is also relabeled “International Stocks”. If you’re working off an old copy of the sheet, make a fresh copy (File > Make a copy). As always, if you’re new to the toolbox, see Part 7 for the introduction and Part 28 for earlier updates. Then:
- Set your weights in the “Parameters & Main Results” tab: e.g., U.S. stocks in B8, 10-year Treasuries in B9, and non-US stocks in B12.
- Pick which international stock series you prefer pre-1970: Either “Non-US” (the new series) or “S&P 500” to use the old version.
- Set the horizon in B27 and the final value target in B28.
- Check the results on the same tab.

So, in summary, if you don’t like the new international series and prefer to work under the previous assumption, i.e., pre-1969, a U.S. investor would have invested their entire equity allocation in the S&P 500 (and precursor indices), you can still do so. I like my Google Sheet to stay as backward-compatible as possible.
How do International Stocks affect Safe Withdrawal Rates?
As always, I like to simulate two retirement scenarios:
- The stereotypical Trinity/Bengen scenario of a 30-year retirement with asset depletion.
- The more FIRE-style scenario with a 50-year horizon and a final value target equal to 25% of the initial portfolio (of course, adjusted for CPI inflation)
In both cases, I start with my typical 75/25 baseline allocation, i.e., 75% U.S. large-cap stocks (VOO, IVV, SPY, or similar) and 25% U.S. intermediate Treasury Bonds (IEF, VGIT, or similar).
For the start dates from 1970 onward, I use only post-1970 data, so nothing changes there. Even the 1960s will have only mild differences in the withdrawal simulations. But we expect major differences in the earlier cohorts, i.e., around the Great Depression.
30-year horizon simulations
Let’s start with the fail-safe withdrawal rate as a function of the international share. I report results for both all historical cohorts and the Post-1926 period. Across all cohorts, international equities aren’t very useful. I simulate all historical cohorts, shifting the 75% equity weight from 75%/0% domestic/international to 30%/45% international in 7.5% steps. Intermediate bonds will stay at 25%.
The highest failsafe withdrawal rate rises from 3.82% to 3.91% if we allocate one-tenth of the equity share to international stocks (i.e., 67.5% domestic stocks, 7.5% international stocks, and 25% bonds). Not a huge difference. If we consider only the post-1926 era, we’d ideally allocate 22.5% to international (i.e., 52.5%/22.5%/25%), and we’d achieve a pretty impressive improvement in our safe withdrawal rate, i.e., 4.33% vs. 3.82%, or a roughly 13.3% boost in the retirement budget.

Let’s shed some more light on the data and determine the cause of the underwhelming diversification performance when including all cohorts. In the chart below, I plot the failsafe across all cohorts again, along with the decade-specific failsafe rates for the relevant decades (1900s, 1910s, and 1960s). Interestingly, the 1960s would have massively benefited from more international diversification. However, once we add more international stocks, the 1900s become the worst-case scenario (between 7.5% and 22.5% international allocation), after which the 1910s become the worst-case cohort for international equity allocations of 30%+.

If we ignore the 1900s and 1910s, diversification looks more promising. Here’s the same chart for the post-1926 era. I again plot the post-1926 failsafe and the relevant decades that cause those worst-case scenarios in the simulations, i.e., the 1920s, 1930s, and 1960s. Both the 1920s and 1960s would have benefited from a higher international equity share. But there is a limit! Raising the international equity allocation too much makes the 1930s the worst-case scenario, which was previously quite safe (i.e., well above 4%) but then gets worse quickly as the poor performance of international stocks during WW II takes over. The sweet spot appears to be 22.5% international stocks (and 52.5% domestic to reach 75% equities). If you rely on linear interpolation, you’d likely go to 25% international stocks and 50% domestic, where the red and green lines intersect, with a 4.35% withdrawal rate.

Just for completeness, I also include a bar chart with all failsafes by decade. Notice that for the 1990s and 2000s, we don’t have a complete 30-year simulation window, so I “pad” the last few years with calibrated data (explained more in Part 1 of the series). Safe withdrawal rate simulations rely heavily on the first few years of returns (Sequence Risk!) and the second half of returns (whether real, historical, or average padded returns) usually has little impact on the safe withdrawal rate. Still, we take the withdrawal rates based on the calibrated return data with a grain of salt.

… and a table with the same results, plus the failure rates of a 4% Withdrawal Rate.

50-year horizon simulations
Next, let’s check how international stocks improve your FIRE simulations, i.e., we require safety for 50 years, with an additional cushion of 25% of the initial portfolio (adjusted for CPI, of course) at the end. So, let’s go through the same set of charts and tables. I start with the overall failsafe withdrawal rates, both for all cohorts and the post-1926 era; see the chart below. If we include the 1900s and 1910s, no level of international stocks will enhance your retirement. Your failsafe starts at 3.23% and only gets worse as you add more international stocks. However, for the post-1926 era, there is a sweet spot again at 22.5% international stocks (thus 52.5% domestic, i.e., a 70/30 split within the equity portion), which affords you a 3.57% failsafe. Adding more international stocks will then lower the failsafe again.

Let’s look again at what causes this shape of failsafe withdrawals. It turns out, the 1900s are the culprit and provide the constraining cohort for all levels of international stock exposure. I included the 1910s, 1920s, and 1960s just for reference. The 1910s had a similar shape to the 1900s, though just shifted up by a few basis points. In contrast, both the 1920s and 1960s would have significantly benefited from more international diversification.

Next, let’s look at the post-1926 failsafes as a function of the international stock allocation. As before, maximizing the fail-safe withdrawal rate through international diversification faces a tradeoff between the 1920s (diversification helps) and the 1930s (diversification hurts), with a sweet spot at 22.5% international and 52.5% domestic equities. Also noteworthy, during the 2000s, one cohort (3/31/2000, right before the Dot-Com Bust) acts as a spoiler and generates an even slightly smaller SWR (3.57%) than the intersection of the red and blue lines for the 1920s and 1930s. Of course, the 2000s cohort relies on calibrated data post-2026, so we might not rely 100% on that one cohort’s failsafe withdrawal rate.

Here’s a summary chart of all failsafe stats in the same format as above. Intriguingly, over the longer horizon, international stocks didn’t help any cohort in the 1980s, 1990s, or 2000s.

Finally, as before, a more detailed summary table. If we want to maximize the failsafe, we’d pick 7.5% (all cohorts) or 22.5% (post-1926). If we want to minimize the failure rate of the 3.5% or 4% Rule (a very different objective function, clearly), we might add a little bit more international exposure.

With International Stocks, Can We Use 100% Equities Now?
You may remember my post from two years ago, critiquing the paper “Beyond the Status Quo: A Critical Assessment of Lifecycle Investment Advice,” written by Aizhan Anarkulova, Scott Cederburg, and Michael S. O’Doherty. The paper recommends a 100% equity allocation split equally between domestic and international stocks as the simplest and best overall lifetime portfolio. I tend to agree that on your path to retirement, a 100% equity allocation is likely the way to go, probably enhanced slightly by an eventual glidepath into bonds as you get close to retirement (see Part 43 on the optimized pre-retirement glidepath logic). However, once you’re in retirement, a 100% equity portfolio seems too risky, specifically, too exposed to Sequence of Return Risk. But I want to keep an open mind: maybe adding international stocks could have helped hedge away sequence risk in retirement. So, let me simulate that scenario. This time, let’s look at 100% equity portfolios and shift the international share from 0% to 80% in 10% steps. Side note: I did the simulations also for 90% and 100% international stocks, but will display the results only for 0-80%, as there is no gain post-80%.
Let’s start with the 30-year horizon with asset depletion. I plot the blue and black lines again for the “All Cohorts” and “Post-1926 cohorts as a function of the international share (0% to 80% in 10% steps). My results confirm that the Cederberg paper is spot on: If you insist on a 100% equity allocation, you’re far better off with international diversification. A 100% U.S. equities portfolio would have had a gnarly 2.85% safe withdrawal rate (but obviously fantastic upside potential in the non-worst-case scenarios!). Replacing between 40% and 50% of your domestic allocation with non-US stocks would have raised your failsafe substantially to 3.71% for all cohorts and 3.86% for the post-1926 cohorts. So, I’m glad that I confirmed their results, albeit with a different methodology.

But I still maintain my criticism: once you retire, you should revisit the 100% equity allocation that got you there and reoptimize. For example, across all cohorts, even the optimal all-equity portfolio (60%/40% domestic/international) has a worse failsafe (3.71%) than the best 75/25 allocation with international stocks (3.91%). Likewise, in the post-1926 era, the best 75/25 portfolio with international allocation handily beats the 50% domestic, 50% international stocks allocation: 4.33% vs. 3.86%.
I also include a detailed table with more results, in the same format as above:

I can do the same exercise for the FIRE scenarios, i.e., for the 50-year horizon with a 25% final value target. Now the peak performance of the 100% equity portfolio (measured as the failsafe withdrawal rate) occurs at 30% international stocks (all cohorts) and 40% international for the post-1926 cohorts. One difference we notice is that, for all cohorts, the best 75/25 portfolio with international stocks only matches the best 100% equity portfolio. In that case, you’d probably be better off going with 100% equities. However, for the post-1926 era, you’re significantly better off with the 52.5%/22.5%/25% optimized portfolio with only 75% equities.

As before, here are the detailed results in a table:

A Caveat!
How representative is this new international equity return series for today’s retirees? International stocks trailed U.S. stocks by between 2 and 3 percentage points per year; see the table below. Early on, at least, you had a very low correlation, i.e., only 0.30 in the simulated period from 1871 to 1969. But since then, economies and financial markets have become more connected. The correlation between the MSCI World ex-US has been 0.69 since its inception in 1970, and it’s even 0.86 since the turn of the millennium. Withdrawal-rate simulations with such a low early correlation may artificially increase the attractiveness of non-US stocks, when we likely won’t see such low correlations again anytime soon (notwithstanding some crazy theories that the U.S. will lose its position as the world’s economic and financial top dog). So, we must use caution when extrapolating the diversification potential of international stocks and not get too excited about past diversification performance when correlation was at a rock-bottom 0.30.

This discussion reminds me of the small-cap, value, and small-cap value stock discussion: The Fama-French SMB (small over large cap) and HML (high book value over low book value) factors were historically vastly profitable, especially between 1926 and about the early 1980s; see Part 62 for more details. Today it appears that rational market participants have arbitraged away that previously fantastic profit opportunity. So, I always warn people not to put too much faith in historical simulations that rely on rosy small-cap value performance. Here it’s a similar flavor: international stocks have offered much less diversification recently than they had pre-1969. Add to that the lower average returns, likely due to more conflicts, less capitalism, and weaker property rights protections, and international stocks may not offer that much retirement safety.
Conclusion
The new international equity total return series (16 markets, cap-weighted, calibrated to DMS for the World War II years) matches MSCI closely when rerun on 1970-2016 (correlation 0.991). Before 1970, it stays within 1.2 percentage points a year of DMS in the episodes it wasn’t calibrated to.
Before this update, the toolbox’s non-US column was just the S&P 500 before 1970, so an international weight barely changed the worst historical cohorts. Was that a huge problem? Probably not because the typical U.S. investor would have found it challenging to allocate to non-US equities before 1970. Nevertheless, I like to leave the simulation design decision to you and give you the option to run your retirement simulations with “true” international data. Recall that my Google SWR Simulation Sheet results have (at least) two interpretations: 1) how would historical cohorts have fared had they retired between 1871 and 2015, and 2) the thought experiment of how my retirement today and in the future would look if the past return patterns were to repeat? So, even if you don’t believe retirees in 1926 would have easily invested 50% of their portfolio in non-US stocks, we can still use that second approach.
Of course, the sheet is backward compatible, so if you prefer the old convention (U.S. index data for 1871-1969), you can switch back to that as well.
My simulation results indicate that a slight international allocation would have improved your retirement experience, at least for the post-1926 cohorts. The 1900s and 1910s were more of a mixed bag. For a 30-year retirement, we still improved the results somewhat. However, the 36-year drought in non-US stock returns between 1913 and 1949 would have wreaked havoc on some historical retirement cohorts in 1901, 1907, and 1911 if they had invested even a small percentage in non-US stocks. Also noteworthy: international stocks now have a far higher correlation with the U.S. than in the historical simulations before 1970. I wouldn’t blame anyone for keeping a mostly domestic equity allocation.
Please leave your comments and suggestions below! Additionally, be sure to explore the other parts of the series; see here for a guide to the different parts so far!
Title Picture Credit: WordPress AI + ERN edits
Data Sources for the new International Stocks series:
- Kuvshinov, D. and Zimmermann, K. (2022), “The big bang: Stock market capitalization in the long run”, Journal of Financial Economics 145(2); dataset BBdatasetR1, CC BY-NC-SA 4.0: dkuvshinov.com
- Jordà, Ò., Schularick, M. and Taylor, A. M., Jordà-Schularick-Taylor Macrohistory Database, Release 6
- Bank of England, “A millennium of macroeconomic data”, via FRED: UK monthly share prices (MSPIUKM) and U.S. dollars per pound (USUKFXUKM)
- Dimson, E., Marsh, P. and Staunton, M. (2018), Credit Suisse Global Investment Returns Yearbook 2018: Summary Edition: Table 1 (p. 16) and Figure 5 (p. 11)
- Dimson, E., Marsh, P. and Staunton, M. (2006), “The Worldwide Equity Premium: A Smaller Puzzle”, SSRN 891620: Table 2 (p. 12)
Do you plan to update the web-based tool?
Yes, it’s on my to-do list! 🙂
An article on value vs growth would be interesting. I’ve always heard value beats growth… for 20 years. Has this statement been flipped or updated though with last 20yrs of growth beating value? I know these terms appear to be ambiguous
Value has underperformed growth for the last 20 years.
I have published several:
https://earlyretirementnow.com/2019/06/12/my-thoughts-on-small-cap-and-value-stocks/
https://earlyretirementnow.com/2024/12/02/small-cap-value-stocks-diversification-or-diworsefication/
https://earlyretirementnow.com/2025/06/02/small-cap-value-swr-series-part-62/
As usual you’ve provided a very thought-provoking, rigorously data-driven take on this issue. I was wondering whether you might eventually extend the analysis to compare different types of international allocation (e.g. heavier weight on emerging markets) and how they might offer more or less advantage than the baseline you looked at for this article?
I will likely not include EM because it wasn’t feasible to invest in before 1970. If you like to see how an EM-heavy portfolio could be replicated with the current asset returns, please refer to the ETF factor exposure database, where I have data on EEM, VW,O DGS, and DFEVX
They are essentially non-US stocks with additional gold and SMB exposure, plus short 10-year bond exposure.
As you know, all the major brokerage firms recommend 20-40% of non-US stocks in the equity portion of the portfolio. It would be good to summarize how their thinking is different from yours. Let me give it a try, and please correct me if I misrepresent your views. 1. They (major brokerages) follow the global market cap. You are betting on the US economy. 2. They seek sector diversification away from the tech heavy US (not sure what you think about this). 3. They see lower valuations abroad. You think the valuations abroad are lower for a reason. 4. They keep large non-US exposure to protect from the long term tracking error (such as another Japan). While the short term correlations between US and non-US are high this century, the long term correlations are still low (because non-US stocks mostly suck?). You again believe in the US economy, and think we are not headed for “another Japan”.
They might be right; they might be wrong.
The 20-40% number comes from an efficient frontier approach where you assume similar returns for the US and non-US, and a correlation of around 0.8-0.85. Then you’d certainly want non-US exposure.
My bet is that the US is indeed stronger, especially compared to Europe, where the next debt crisis is brewing (France this time), which will be worse than the Greece/Italy/Spain/Portugal issue 16 years ago.
From my foxhole, I’d argue Karsten’s approach is basically spot on – from the decision tree logic, we’re either in an instance where it will outperform relative to US (probably due to US equities being priced on the assumption of non-achievable revenue multiples), or underperform (most likely due to structural debt attached to anemic GDP growth stifling virtuous investment feedback loops)…
So the suggested uncertainty choice would be to ‘sin a little’, and in this case that number being around 20% or higher makes a lot of intuitive sense.
For my part I basically always assume that the professional finance wonks with billions of dollars of toolsets and information at their disposable are going to out-alpha-arbitrage me, and that pricing can only stay wrong for so long, so I’m inclined to suspect that the relative pricing isn’t really that wrong, so my bet would probably align with Karsten’s. Even though my initial exposure target to international was around 25-30% – investing haphazardly without actively rebalancing existing balances (just doing that contribution-side with targets in that range depending on the relative costs based on 401k brokerage houses) I’ve wound up floating between 20-22.5% in terms of actual balance, and I suspect that’s about where I’ll leave it for now.
Migrated over to the new sheet and then confirmed the toggle to “SP500” basically matches the old sheet. The non-US toggle changes CAPE>20 and SPX at all-time high at 0% failure rate from 3.71% to 3.64%, resulting in about $4k less per year at 0% failure rate. I liked having that extra $4k :). It changes All from 3.71% to 3.53%, which gets even steeper at 5% failure rate ($13k less per year).
Maybe I should just stick my head in the sand and act like I never saw this…. 😉
Nice! Thanks for confirming!
+$4k/yr on $208k is too precise for me. With a $5.7M portfolio (what’s required for the delta on 3.71% vs 3.64% to generate $4k/yr), and $208k/yr, life throws far more noise at me than $4k/yr. That is to say, such a conclusion is more of a non-event than a change in $4k/yr. For those who do need that $4k/yr, and I’m not saying that’s insignificant to everyone, most would be better off (mentally and financially) delaying retirement a year to get that extra 0.07% margin than taking their person SWR to 4 decimal points in precision! I’m personally unsuccessful in tax planning to +/-$4k per year!
I can’t emphasize enough how awesome this is as advice – by that point, I’d argue the correct measurement is no longer in $k/year, but in ‘additional WEEKS until the SWR multiple based on current/expected CAEP lines up’. It will probably matter more if employment stretches into a new calendar/fiscal/tax year than any of this stuff, which takes a bit of a mental and emotional load off to me.
Also worth noting the extreme statistical divergence where we’re trying to chase down small single digit percentile performance over time – most of the time that risk doesn’t materialize, and the worry becomes how to expend an additional $40k/yr after waiting out that SORR-wait-and-see delay period.
Very true, I was being a bit tongue and cheek. Although, if history repeats itself, this does call into question the value of having an international/ex-US allocation. I end up with a lesser withdrawal, plus I’m dealing with the additional overhead (albeit small) of rebalancing another fund (VXUS in my case).
Nice work. I’d be glad to study internationally diversified portfolios by myself. You documented your data sources for stock returns per country, but you didn’t document the data source for market weights. Which is quite critical as market weights changed quite a lot over history (according to DMS). Mind documenting the market weights data sources and possibly share some thoughts about its credibility?
Oh, market weights probably came from the Kuvshinov and Zimmermann paper. I still would like to hear some thoughts about the credibility of this paper. Thank you.
Yes, that’s about as credible as it gets.
Well… last year, I used the same data source (afaik, it’s the only one available) and I assembled the same aggregate graph as DMS (where we see historical market caps vary widely over time). And the alignment wasn’t terribly satisfying, to say the least. I don’t know if you ran a similar sanity test. So… If you could please elaborate on WHY you think this is a credible data source, it would be appreciated.
It’s definitely not ideal data, but it’s quite good, likely the best we’ll ever have, and probably more importantly I’d be among those arguing that the structural alpha of making international investment available to both US investors writ large, combined with carry trade shenanigans of the late 20th century, are much larger effects than any market-cap weighting noise that is happening outside of WW2 data (and those data are inherently a bit of a bodge – but it all still passes the sanity test to me as somebody who gets paid to contemplate what some aspects of WW3 would look like for a day job), so there is enough salt grain-iness baked into all of this data that choosing what side of that 15-40% distribution of where the optimal international equity exposure probably is would be the most defensible position based on the data is up to interpretation, but nobody would be surprised if it’s in there, and that the central limit theorem hints at that being closer to the middle of that range than not.
He’s a finance professor at a world-famous university, so I will take his word for it.
Here’s the fit with the DMS data, again:

Market weights are actually documented in my post; see the reference in the last section:
Kuvshinov, D. and Zimmermann, K. (2022), “The big bang: Stock market capitalization in the long
run”, Journal of Financial Economics 145(2); dataset BBdatasetR1, CC BY-NC-SA 4.0:
dkuvshinov.com
See the link to the BBdataset. Column E (“mcap”) in that Excel Sheet.
And, yes, they are from a credible source.
Great addition to the series!
I’d love your thoughts on the current wide gap between the S&P 500 CAPE (high 30s) and non-US CAPE (low 20s). Does your new series have enough earnings data to calculate success rates based on starting ex-US CAPE, like you’ve shown for the S&P 500 in Part 3, Part 43 and elsewhere? I’m curious whether a big US valuation premium historically predicted a bigger benefit from international diversification.
The US has a lot of earnings growth factored in that isn’t visible in the 10-year rolling data (yet). Non-US, especially Europe, is stagnant in both GDP and earnings growth. So, be careful about shopping for “bargains” abroad: these might be cheap for a reason.
(edit) And to address your question: no, I don’t have enough data to calculate the non-US CAPE, especially not before 1970.
This is another way of saying US valuations already require strong future execution just to deliver average returns. If US growth expectations slow even slightly below lofty expectations, a high-30s CAPE leaves virtually zero margin of safety. However, if equities aren’t offering a compelling risk-adjusted return (whether overpriced US mega-caps or stagnant ex-US value traps with horrible demographic trends), the rational move is to let risk-free TIPS do the heavy lifting.
I think you naled it with that – it’s always a question of relative performance, and the ‘bet’ such as it is consists of whether one thinks the US equities will be able to meet those lofty expectations over how much stagnation and debt will limit ex-US equity growth over time. For my part, the demographics (watch some Peter Zeihan content), natural resources (US intercoastal waterways are still cheat codes), and technological innovation (continuing to be enhanced by brain drain effects) edge the US has probably justifies the price offset… and most situations the performance of equities as a whole are probably too coupled to really pick out a ‘winner’ anyway (semiconductors are a great example of outcomes being only everybody wins or everybody suffers).
Yes, that’s the crux! Admittedly, earnings have increased spectacularly since 2009. I’m not sure this trend can continue for that much longer. If it stops, the price reversal would be brutal. But as long as earnings grow like this, the party continues.
Cheers [Anonymous ERN-Reading Legend] !
At this point, I’m more impressed that Karsten was able to keep it under 4,000 words here, but this is just a first installment of a multi-part series for sure.
Can you elaborate on this WWII fix? Because we use backtesting to construct a portfolio for the future, we should model how modern index funds operate—not the realities of 80–90 years ago. Given that today’s funds track market weights, meaning war-affected countries would naturally shrink in weight, why is this adjustment necessary?
Yeah, I’m debating whether I should just revert back to the cap-weighted index. We don’t need to match the DMS series. It’s been on my mind, too.