Model Portfolio Results from 2002 to 2017
The world of investment management is filled with backtests — the use of historical data to show how a proposed investment strategy or portfolio would have performed in the past. The problem is that this creates a strong temptation to tweak proposed strategies so that their backtests produce impressive results.
However, in a complex adaptive system like the global economy and financial market there is a strong likelihood that the future will not perfectly resemble the past. Put differently, in such systems it is often the case that the harder you try to fit your model to historical data, the less robust it will be to future uncertainties.
It is therefore highly interesting for investors to see how real, implemented strategies actually play out. Unfortunately, the in this case the problem is "survivorship bias", which is a fancy way of saying that this type of historical analysis can lead to overoptimism and overconfidence because poorly performing strategies and funds are often killed off quickly, and disappear from data sets without leaving a long-term track record to examine.
Our time-out from publishing for a few years deprived us of the opportunity to quickly tweak our model portfolios had they not been working, or had we simply lost confidence in our methodology when financial markets hit hard times. We can therefore look at 15 years of historical data — from December 2002 to December 2017 — to see (warts and all) how they actually performed.
Methodology
Our model portfolios were based on a regime switching model. We assumed that the financial markets could be in one of three regimes, that we termed normal times, high inflation, and high uncertainty. Based on our analysis of historical time series data, we estimated the probability that, conditional on being in one regime in a given year, the system would switch to one of the two others.
For each regime, we then estimated asset class inputs, including the average real return, standard deviation of return, and correlation with the real returns on other asset classes. To do this, we combined historical data with the outputs from our asset pricing model. Within each regime we assumed Gaussian/normal distributions of asset class returns. However, the different distributions within each regime and use of the regime switching model produced an aggregate distribution of asset class returns that was quite close to the distribution features observed in the historical data (i.e., fat tails, clustered volatility, etc.).
We used broad asset class definitions, and only included those classes for which retail index fund products were available at the end of 2002. As the introduction of new products made more asset classes investable, we updated our model again in 2007 to include them, as well as a weighted mix of uncorrelated alpha products that had become available to retail investors (e.g., equity market neutral and global tactical asset allocation).
The 2002 asset classes (and associated funds) include (1) real return bonds (VIPSX); (2) domestic investment grade bonds (both government and credit) (VBMFX); (3) foreign government bonds (without currency hedging) (PFBDX); (4) commodities (PCRDX); (5) domestic commercial property REITs (VGSIX); (6) domestic equity (VTSMX); (7) foreign developed markets equity (VTMGX); and (8) emerging markets equity (EEIEX).
In 2007, we added (9) foreign commercial property (RWX and VNQI), and (10 timber (a wighted mix of timber REITs RYN and PCL, and WY after it bought the latter).
We also established certain constraints on maximum drawdown and allowable asset allocation solutions, to avoid so-called "corner solutions" where almost all of a portfolio is allocated to one asset class.
Our models also sought to optimize rebalancing frequency to minimize risk and gain incremental return. This was accomplished by setting asset class portfolio weight thresholds that would trigger rebalancing (e.g., 10% above target weight) and when those occurred a target for incrementally overweighting the most underweight asset class or asset classes and underweighting the most overweight (to capture incremental returns from markets overshooting fair value in either direction).
For each long-term real return target (7%, 5%, and 3%), we employed simulation optimization to identify robust portfolios that maximized the probability of achieving the target within he constraints we set. These simulations included both the regime switching model and, within each regime, the range of possible return on each asset class.
Our Benchmarks
We have judged the performance of our model portfolios against two benchmarks. The first is whether they have met their compound annual (i.e., geometric average) real return targets. The second is how they compared to the results from an equally weighted mix of the broad asset classes we used.
To be clear, equal weighting implies that an investor has no confidence in their ability to predict meaningful differences in future asset class returns, standard deviations, and correlations. Put differently, it implies an investor has no confidence in their ability to predict future asset class exposures to the factors that drive returns (e.g., macroeconomic and investor behavior variables), and/or in their ability to predict the future returns associated with different degrees of exposure to those factors.
Model Portfolio Results
The following table shows the nominal return and standard deviation for our target return and equally weighted portfolios along with (a) the amount of return per unit of standard deviation; (b) the value at the end of December 2017 of an investment of 100 made at the end of December 2002; (c) the compound nominal return on this investment (i.e., the geometric average), and the compound real return (which can be compared to the portfolio target).
One equally weighted portfolio keeps the same eight asset classes for all 15 years. The second adds two new asset classes in 2007. The 7%, 5%, and 3% portfolios also added allocations to active alpha strategies that were expected to have a low correlation with domestic equity returns.
Please keep some caveats in mind when looking at these results:
Model and Benchmark (Equally Weighted) Portfolios | Average Nominal Return, 2002-2017 | Standard Deviation of Nominal Returns | Average Return/Standard Deviation | Value at Dec 2017 of 100 Invested at Dec 2002 | Compound (Geometric Average) Real Return, 2002-2017 | |
Equally Weighted 8 Asset Class Portfolio for all 15 Years | 8.80% | 14.70% | .60 | 308 | 5.69% | |
Equally Weighted 8 Asset Class Portfolio 2002-2007, then 10 | 9.10% | 14.60% | .62 | 322 | 6.01% | |
7% Target Real Return Portfolio. Up to 8 Asset Classes to 2007; then up to 11 | 10.60% | 17.60% | .60 | 366 | 6.95% | |
5% Target Real Return Portfolio. Up to 8 Asset Classes to 2007; then up to 11 | 8.20% | 13.50% | .61 | 291 | 5.30% | |
3% Target Real Return Portfolio. Up to 8 Asset Classes to 2007; then up to 11 | 6.00% | 8.70% | .69 | 227 | 3.53% | |
7% Target Real Return Portfolio. 2002 Weights for All 15 Years | 9.60% | 15.90% | .60 | 337 | 6.35% | |
5% Target Real Return Portfolio. 2002 Weights for All 15 years | 9.20% | 13.20% | .70 | 336 | 6.32% | |
3% Target Real Return Portfolio. 2002 Weights for All 15 Years | 5.90% | 7.60% | .78 | 229 | 3.59% | |
Asset Class
Value of 100 Invested at 31Dec2007 on 31Dec2017
Compound (Geometric) Average Real Rate of Return
Real Return Bonds
136
1.51%
Domestic Investment Grade Bonds
145
2.20%
Foreign Government Bonds
161
3.24%
Domestic Commercial Property
165
3.51%
Foreign Commercial Property
166
3.55%
Commodities
56
(7.19%)
Timber
174
4.06%
Domestic Equity
228
6.98%
Foreign Developed Market Equity
129
0.95%
Emerging Markets Equity
114
(0.29%)
Uncorrelated Alpha Strategies
156
2.92%