How to systematically beat the 60-40 portfolio

Dynamic strategic asset allocation

In the previous dispatch, we documented the results of a controlled experiment where, using Doeswijk’s capital market assumptions, we only varied the expected returns on the asset classes and kept the rest of the inputs invariant. We found that an unexpectedly large allocation to real-estate may be warranted. The exercise involved only unconditional strategic asset allocation. We suggested that conditional strategic asset allocation may yield more promising results. Here we document a premium over the 60/40 benchmark that is freely-available to disciplined investors willing to tactically rebalance between asset classes.

Our investable universe consists of US equities, US investment grade bonds, US MBS, US munis, US high yield bonds, commodities, as well as US-listed private equity and real-estate. These asset classes are proxied by the following ETF tickers. In general, these are the most liquid and oldest ETFs available.

• "U.S. Equity": "IWV",

• "U.S. Investment Grade Credit": "AGG",

• "U.S. MBS": "MBB",

• "U.S. Intermediate Municipal": "MUB",

• "U.S. High Yield": "HYG",

• "Commodities": "DJP",

• "U.S. Private Equity": "PSP",

• "U.S. Core Real Estate": "VNQ"

Our capital market assumptions (CMA) are taken from AQR’s 2022Q1 CMA. AQR does not provide expected return and covariance estimates for MBS and munis. We obtain these from BNY-Mellon’s 2022 CMA. The correlation matrix and expected asset class volatilities are also obtained from BNY-Mellon. These are conservative and reasonable assumptions.

The “core” risk assets (equities, PE, RE, HY) are associated with well-documented and significant ex ante risk premia. We assume, following AQR, forward-looking real expected return of 3.6% on US equities, 5.9% on private equity, 2.6% on real estate, and 0.3% on high-yield bonds. Unfortunately for investors, these core risk assets are highly-correlated. Here’s the BNY-Mellon correlation matrix.

We assume, following BNY-Mellon, that the correlation with equities is 0.85 for PE, 0.78 for real estate, and 0.65 for high-yield bonds. We think that equities-commodities correlation may be a bit lower than 0.40. But we keep the BNY-Mellon correlation matrix because we don’t want to fiddle with any numbers in the detail unless we really have to.

The best market benchmark for multiasset portfolios is the “60/40” portfolio. We construct a monthly-rebalanced 60/40 portfolio with 60% on U.S. Equity (IWV) and U.S. Investment Grade Credit (AGG). “BNY-Mellon” is a monthly-rebalanced portfolio whose target weights are optimal for a 60/40 investor. Recall that a 60/40 investor has a target volatility of 10% per annum. “Unconditional” is a monthly-rebalanced portfolio whose target weights are optimal given our forward-looking assumptions documented above and the 10% target risk. All these portfolios are unconditional, fixed-weight portfolios. These will serve as our benchmarks.

The “Conditional” portfolio is a monthly-rebalanced, longonly portfolio with weights conditional on the past 12 months of performance. Specifically, on each rebalance date, we estimate expected returns by the geometric means of daily returns over the past year. These are then “shrunk” towards the forward-looking real expected returns using a learning rate parameter of 0.75. We also estimate the covariance matrix using all available daily returns up to the rebalance date. We assume a target volatility of 10% so that we can compare the performance with the 60/40 benchmark.

We impose sector constraints on all portfolios as follows. Because the equity risk premium is the principal source of returns, we constrain equity allocation to 40-80%. Bond allocation is constrained to 10-40%; commodities are constrained to 0-20%, and PE, RE, and MBS are constrained to 0-10% each. These choices are governed by the trade-off between the risk that we’re overweight equities in a scenario where equities fall persistently, and the risk that we’re underweight equities in a scenario where equities again surprise on the upside. In our judgement, these bounds are consistent with institutional practice and common sense.

The “long_conditional_short_60/40 (130-30)” portfolio is, as the name suggests, a standard longshort portfolio with $130 long on our conditional portfolio and $30 short on the 60/40. The purpose of the longshort is to document the relative outperformance of our conditional strategy against the 60/40 benchmark.

We test the performance of all these portfolios using daily returns from 2008-10-03 to 2022-06-23. Note that the target volatility for all portfolios (except the longshort) is 10%, so we’re comparing apples-to-apples.

We find that all unconditional portfolios had a realized geometric mean return of 7.8% per annum over the test period. But they differ in risk-adjusted terms. The 60/40 realized a Sharpe ratio of 0.63, compared to 0.44 for our unconditional portfolio. So, the 60/40 is indeed a hard benchmark to beat. However, dynamic strategic asset allocation offers a significant unadjusted and risk-adjusted premium over the 60/40. Our longonly conditional portfolio realized an expected return of 10.2% per annum over the test period, and a risk-adjusted return of 0.71. We thus estimate a premium of 2.4% per annum of our dynamic longonly portfolio over the 60/40 benchmark.

The performance of the longshort portfolio (which is long the conditional and short the 60/40) shows that dynamic strategic asset allocation beats the 60/40 benchmark systematically. The realized expected return on the longshort was 11.0% over the test period, thus offering a premium of 3.2% per annum over the 60/40. Meanwhile, the realized Sharpe ratio of 0.72 on the longshort is the same as that of the conditional longonly portfolio and higher than that of the 60/40. This shows that you can systematically beat the 60/40 using a straightforward dynamic strategy. The machine works!

Year-to-date, the 60/40 is down 16.7%, whereas our conditional longonly is down only 12.8%, and our longshort is down 11.6%. The recent premium for the dynamic portfolios over the 60/40 is thus 3.9% for the longonly and 5.1% for the longshort. These are large numbers! Digging into the dynamic asset allocation reveals that the premium is due to rotating away from equities (from 60% in March to 40% by the May rebalance) and an exposure to commodities at the upper bound of 20%.

The next figure shows the minimum, mean, and maximum allocations to the different asset classes for our dynamic portfolio. Our maximum and minimum allocations are all respectively at the upper and lower bounds, so the constraints do bind — suggesting a loss of efficiency. So, there’s a nontrivial shadow price of the strategic constraints.

Our mean equity allocation is 66%, considerably higher than the unconditional allocation of 46%. The average allocations for real estate and PE are 5.7% and 5.1% respectively, lower than the unconditional allocations of 10% each. The unconditional allocation to high yield bonds is 34% but the average conditional allocation is only 9.4%. The 2.4% premium of the longonly over the 60/40 is not simply due to a broader universe of assets. The conditional portfolio sports the same 2.4% premium over the unconditional portfolio as well. The 2.4% premium, then, is due entirely to tactically moving in and out of asset classes based on recent evidence. We can think of it as a “tactical premium.”

We have shown that it is possible to systematically beat the 60/40 benchmark by dynamically adjusting allocations to various asset classes. We have thus documented the existence of a tactical premium. Like the equity risk premium, the tactical premium is freely-available to disciplined investors. One can uncover additional premia by replacing the broad asset class ETFs with more efficient portfolios within asset classes. These will the building blocks of a multiasset product we hope to bring to market soon. It should serve as a nice complement to our flagship portfolio that harvests the overnight drift signal.