SPY, AGG, GLD and cash: 11.6% a year against 8.1% for 60/40. We rebuilt it, delayed the trades, and it mostly held.

The idea worth investigating

Stocks, bonds and gold are the least exotic ingredients in the business. The interesting question in this paper is how much the recipe matters, and I went in expecting the answer to be "less than they say".

This week we tested "Simple Dynamic Stock/Bond/Gold Portfolios" by Nikhil Devanathan, Alexandros E. Tzikas and Stephen P. Boyd, arXiv 2609.07946, September 2026. Long-only, monthly rebalanced, SPY, AGG and GLD plus cash, no leverage, a 7% volatility target, and a constraint that keeps the mix near 50/30/20. The complex version forecasts returns from ETF, macro and factor inputs; the simple version uses an exponentially weighted average of past ETF returns.

What the authors report, 2006 to 2025 after modelled trading costs: 11.6% annualized return with an 18.1% maximum drawdown, against 8.1% and 33.7% for their 60/40 benchmark. Their dynamic portfolios trade monthly; the fixed benchmark trades annually.

The desk question: how much of the advantage survives when the trades happen a session later, the inputs arrive later still, and the benchmark is given the same tools?

Here's the plan:

• 1. Reconstruct the allocation on freshly downloaded, distribution-adjusted ETF prices with the published forecasting rules.
• 2. Move execution from the same close to the following session's close, then add assumed publication delays to the macro and factor inputs.
• 3. Raise the trading charge from 2.5 to 10 basis points per dollar traded.
• 4. Build the fair comparison: a 50/30/20 mix with the same monthly schedule, optional cash and 7% volatility target.

One result we can share here. Our reference reconstruction produced 11.65% annualized return and an 18.07% maximum drawdown, against the paper's 11.6% and 18.1%. Moving execution one session later cost about 0.48 percentage points of annual return. These are house historical simulations on current-vintage data, pretax, with base trading costs of 2.5 basis points per dollar purchased or sold and cash earning the modelled federal funds rate; they do not establish a live edge.

Subscribers get the seven-row comparison table (complex, simple, volatility-controlled 50/30/20, fixed 50/30/20, fixed 60/40, each with return, volatility, drawdown and Sharpe), the input-delay and 10-basis-point results, the answer to "is this just gold and cash", the average allocation and turnover, the paired block-bootstrap interval on the Sharpe difference, and our assessment of what the evidence still lacks.

Let's get started.

AI assists curation and drafting. The research status distinguishes paper reviews from house tests. Read our research approach.