50 Years of Investing in the World (Part 3)
This article is the third part of a study looking at global and domestic investing from the perspective of local investors.
In Part 1 and Part 2, we took the position of a local investor in one of 16 countries of interest and we explored opposite approaches of either investing 100% global or 100% domestic. In Part 2, it became clear that global bonds tend to hurt local investors, while global stocks definitely helped for most scenarios. It is now time to try a middle ground and study portfolios mixing global and domestic stock investments. We will notably look at the mitigation this could bring to the countries having fared the worst, but also consequences for countries having fared better. Of course, it is easy to look at such numbers in hindsight and draw hasty conclusions, so let’s keep in mind that nobody could have predicted winners and losers ahead of time.
Quick Reminder About Data Sources
We will leverage various data sources which were described in Part 1, mostly based on MSCI data series for per-country stock historical returns (in USD and in local currency) and OECD for per-country inflation rates. Using the MSCI World data series and currency exchange rates, we obtained 16 data series expressed in local currency of such world index. In addition, we will leverage the 16 domestic bond data series which were assembled in Part 2, based on OECD/IMF interest rates and Barclays/SIX indices.
Investing Globally, while Tilting Domestically
Many people diversifying internationally tend to use an Asset Allocation (AA) with X% in local stocks and Y% in international stocks, hence a fixed split. For our study, we’re making use of the MSCI World series and every country is of course part of the world. We therefore need to be a bit careful when interpreting numbers, as X% World + Y% Domestic really means a local (home country) exposure higher than Y%.
In other words, we are starting from a global position (World) and adding a tilt towards the Domestic (local) market of interest. Let’s take a few examples:
- 100% World and 0% Domestic means a domestic exposure corresponding to the current market capitalization of the country (which varies over time).
- 100% Domestic is trivial, full home country (local) exposure, nothing else.
- Say a country represents ~50% of the world’s market cap (e.g. US in 2018, Japan in the late 80s). Then a 80% World + 20% Domestic asset allocation means 60% local exposure (and 40% international).
- Say a country represents ~25% of the world’s market cap. Then a 80% World + 20% Domestic asset allocation means 40% local exposure (and 60% international).
- Say a country represents ~5% of the world’s market cap (e.g. UK in 2018). Then a 80% World + 20% Domestic asset allocation means 24% local exposure (and 76% international).
A tilt has actually quite some advantages, as it adjusts itself when things change. A fixed split is, well, fixed. With a tilt, if the country market cap grows or shrinks over time, the actual exposure to the corresponding country will automatically vary as one would expect. This should make ‘sticking with the plan’ easier, avoiding to tweak your asset allocation.
As the graph below (from Credit Suisse) clearly shows, the evolution of per-country market capitalization over the past century varied quite rapidly. We’re all subject to recency bias, and this graph helps reminding us how much the world can change in a decade or two.
For the record, here is where we are as of January 2020 (chart from Statista). Please note that the numbers from the previous graph and from the graph below show percentages relative to the entire world (as opposed to just developed countries; otherwise the weight of the US would be closer to ~60%).
World 80%, Domestic Tilt 20%
Let’s cut to the chase a little bit. After playing with historical numbers for a while, the author hypothesized that splitting stocks 80% global world and 20% domestic would have displayed fairly good properties in the past 50 years (1970-2019), a sweet spot of sorts. It would certainly have helped a local investor in countries with the worst track record (e.g. Italy, Spain, Japan) to mitigate various types of risk. It doesn’t seem to significantly impact the performance of the countries with an average track record. It does dampen the performance of the countries with the best track record (e.g. Sweden, Denmark) compared to a more domestic-oriented approach, but this is a consideration made in hindsight. Fact is, back in the 70s, nobody had a crystal ball allowing to guess that Sweden would do that good.
As a fully diversified ‘Invest in the World’ approach is our general theme and starting point, most charts below will compare side-by-side a mixed 80% global (world) / 20% domestic tilt approach with a 100% global stocks approach, while using domestic bonds on both cases. We will also compare with a mixed 60% global, 40% domestic approach, to better test the ‘sweet spot’ hypothesis.
Let’s check the numbers for a 70/30 asset allocation, where the stocks part (70%) is split in various ways between global (world) and domestic, while the bonds part remains fully domestic. The reader is encouraged to compare with the content of Part 2, which provided similar charts with a fully domestic asset allocation. Most charts in this article will use a different horizontal scale though, to better emphasize the differences between 80% global, 60% global and global-only.
Here is the usual growth (CAGR) vs. volatility (standard deviation) chart for the 1970-2019 time period. Click on the graph for a larger display. As you can see (if you squint a bit!), there is a nice improvement on volatility (horizontal axis) for nearly all countries when setting global exposure to 80% (hence an extra 20% tilt towards domestic exposure), compared to 60% global. The improvement over 100% global is there too, but less marked and more spotty. From a return standpoint (vertical axis), differences are fairly small, higher global exposure mostly reducing the dispersion.
Now let’s look at maximum drawdowns for the same time period. Those charts are more of a mixed bag. When focusing on the worst performers, 80% global or 100% global wins. When looking at averages for the 16 countries, then 60% global wins, although not by much.
Finally, let’s compute the Safe Withdrawal Rate (5% percentile) for 30-years periods, using the same 70/30 portfolio. There is a marked improvement in favor of 80% global against 100% global. The case is more muddled between 60% and 80% global. Note that Spain badly hurt in any case, although more global exposure would have helped.
A More Systematic Comparison
The previous section asserted that broad exposure to global stocks (while still keeping a modest tilt towards the domestic market and without introducing any such globalization on the bonds side) seemed a fairly solid approach, illustrating the point through a few charts and a few metrics of choice. For the inquisitive and number-minded readers, more details would surely be welcome to further investigate such assertion. To do so, we will vary the global exposure (stocks) from 0% to 100%, by increments of 20%, and monitor multiple metrics.
On the emotional/risk side of things, standard deviation and maximum drawdowns are useful and classic metrics, spanning the entire time period being studied (1970-2019). Maximum drawdowns certainly illustrate the gut wrenching feeling of investors during a deep crisis, although it only captures a single point back in history, which is unlikely to repeat itself as is. The Ulcer index is a much improved way of tracking both depth and duration of all combined drawdowns during the time period. We will compute a ranking score combining all three metrics (for which the lower the value is the better).
On the performance/reward side of things, the Compound Annual Growth Rate (CAGR) is a classic, but fact is it primarily captures the trajectory of an initial investment left to grow for a period of time. In practice, people tend to either accumulate (regularly adding money to their portfolio, e.g. workers) or spend (regularly withdrawing money from their savings, e.g. retirees). The Safe Withdrawal Rate (SWR) metric is very significant, quantifying the withdrawal rate not be exceeded to avoid prematurely depleting one’s portfolio. It also has the nice property of accounting for sequence of returns effects (e.g. a deep crisis at the beginning of a retirement period is bad news for retirees). On the accumulation side, the Internal Rate of Return (IRR) allows to account for both regular extra savings (assumed to be constant in inflation-adjusted terms for simplicity’s sake) and sequence of returns effects, hence a much more accurate metric for accumulators.
All three performance/reward metrics (CAGR, SWR, IRR) will be analyzed for all investment cycles of 30 years contained in the time period of interest, while varying the starting year (hence 1970-1999, 1971-2000,…, 1990-2019). Consistently with SWR typical practice, we’ll look at the 5% percentile of the results (basically averaging the two worst cycles). On the other hand, the oil crisis in 1973 is nearly always the worst starting date (Japan being the notable exception) and one could convincingly argue that such a test is centered on a single historical singularity that is unlikely to repeat itself in the exact same shape, and that it mostly ignores the sequence of events which unfolded since 2000. Also it makes little sense to gate all results by the worst possible outcome, as opposed to a broader set of possible outcomes. To make the comparison more statistically significant, one could look at a larger subset of (bad) starting years by computing a percentile on the various cycles. To do so, we can look at the 50% percentile (the ‘not so good’), the 25% percentile (the ‘bad’) and the 5% percentile (the ‘ugly’) of the collection of cycles. We will do so for CAGR, IRR and SWR.
Finally, for each individual test, we will compare the outcomes for the 16 countries of interest, and rank the results based on the average value for the lowest performing countries, i.e. the subset of 4 countries which struggled the most for a given metric. This is the ‘LOW 16’ column in the tables below. This column is then summarized with a combined average ranking. Knowing that the most troublesome countries were Spain, Italy and Japan (in this specific time period, that is), which are all sizable and diversified economies, it seemed more appropriate to focus on mitigating the most severe situations instead of muddying the water by averaging with less troubled outcomes. Any sizable economy may end up in a similar dire situation in the coming decades, in the author’s humble opinion (US investors having doubts about this may want to come back to the per-country market capitalization chart at the beginning of the article and ponder the 1970-1990 time period).
For information purpose, two more columns (‘AVG 16’ and ‘SD 16’) compute the average of the results across all 16 countries as well as the standard deviation of such results (unsurprisingly, the higher the global exposure, the lower the deviation between countries). ‘AVG 16’ is mostly useful to compare asset allocations, not so much results for a given asset allocation.
The table below provides all details for the 16 countries, comparing various levels of stocks global exposure in the asset allocation. Conditional formatting (green is good, red is bad) is applied to vertical groups of numbers as a visual aid. The top-level asset allocation remains constant, 70% stocks and 30% bonds. As you can see, it is a bit of mixed bag, while still validating our assertion that 80% global exposure and a 20% domestic tilt appeared to be a robust strategy. Click on the images to see a bigger version. Yes, it’s a lot of numbers…
Long story short, based on such multi-criteria assessment, the 80% global and 20% domestic approach appears the most optimal from both an emotional/risk standpoint and a performance/reward standpoint. A few more observations:
- Ranking results vary across metrics and their significance may depend on an investor’s specific goals, but overall they seem to display reasonable consistency across metrics.
- The results display very little sensitivity to the exact percentiles applied to the performance/reward metrics.
- The results do display sensibility to the scoring across countries. Instead of focusing on the worst 4 countries, focusing on the bottom half gave the same results for the emotional/risk score (80% global remains the best approach) while the performance/reward score ranked better with 60% global exposure. As previously explained, the bottom 4 averaging seems a better way to assess risk mitigation though.
- The inclusion of global bonds never really helped, always damaging the emotional/risk scores. If one were to focus on the smallest countries, this finding may not stand though (a test the author didn’t perform).
- Varying the asset allocation to 80/20, 70/30, 60/40 or 50/50 didn’t seem to alter the high-level conclusion. Differentiations are of course more pronounced as the stock exposure increases.
- Finally, the author would be happy to share those results in numerical form if readers want to assess other forms of ranking process.
A fully domestic investment can deliver fairly good results, as anybody having invested in the US or Canada knows. It can even deliver very impressive results as Swedish and Danish citizens experienced in the past decades. But it can also put local investors in devastating situations, with decades-long drawdowns for both stocks and bonds (in real terms), and ruin even the most conservative retirement plans, as Spain, Italy and Japan investors went through.
We often say that “past performance is not an indicator of future success”. And yet the past is all we have to try to calibrate ourselves. The excellent Credit Suisse Global Investment Return Yearbook annual publication provides a 100+ years of history of investment returns from numerous countries and always includes the following warning:
There is an obvious danger of placing too much reliance on the excellent long-run past performance of US stocks. The New York Stock Exchange traces its origins back to 1792. At that time, the Dutch and UK stock markets were already nearly 200 and 100 years old, respectively. Thus, in just a little over 200 years, the USA has gone from zero to more than a majority share of the world’s equity markets. Extrapolating from such a successful market can lead to “success” bias. Investors can gain a misleading view of equity returns elsewhere, or of future equity returns for the USA itself. That is why this Yearbook focuses on global investment returns, rather than just US returns.
Instead of overly relying on the past history of the country we live in, we should try to open our minds to what happened to the rest of the world and learn from it. One could always argue if one specific comparison is significant or not, or if a coarse rule of thumb (e.g. 80% global) applies in every case, but at least, this represents a solid foundation of real-life facts and history to reflect upon. The author believes that this study makes a convincing case to seek a high exposure to global (or international) equities, while keeping a tilt towards domestic equities. Some readers might perceive otherwise, but should by now have more factual material to refine their thinking and possible temptations of home country bias.
The author also hopes that this study will inspire various interested parties to perform more research and analysis about international data sets, based on the general idea of putting ourselves in the shoes of a local investor in a large set of diverse countries.