Drawdown — why these strategies exist

Lesson 1 opened on a number: the S&P 500 fell 50.8% from its peak in the backtest these strategies are measured against, over Dec 1970 – Dec 2016, measured at month-end. This lesson is about what the six do with that problem, and how far you should trust the answer.

The authors set the target themselves

This is not a framing the site applied afterwards. Keller states it in the opening of the VAA paper:

with VAA we aim at moderate but offensive returns above 10% but with defensive drawdowns of less than 20%, preferably less than 15%.

And it is built into the measure the papers optimize. They score candidate rules with K25, a return measure defined to hit zero once maximum drawdown reaches 25% — so a rule that earns beautifully and falls 25% scores nothing at all. The ceiling is inside the authors’ own objective function, not applied to their results by us.

What they published

The papers set out to keep the worst fall under 20%. In the published backtests of the variants this site runs, the worst month-end fall ranged 8.7% to 16.4%, against 50.8% for holding the S&P 500 over the same decades. A 60/40 stock-and-bond portfolio fell 29.4–29.5% over comparable spans.

Here is the deepest of the six, so you meet the ceiling rather than the best case:

Every strategy page carries its own version of that block. The period differs per paper, which is why it is printed beside the number every time rather than once at the top of the site.

Where it does not hold

Those are the headline variants. Stating the rest raises rather than lowers what the figures are worth, because a number with no edges is not a measurement.

• Other variants in the same papers do worse. VAA’s own pre-1945 span shows 24%, and the HAA paper reports a 25.2% fall for one alternative configuration.
• Small implementation choices move the number a lot. An independent replication by AllocateSmartly found VAA’s drawdown going from 16.1% to 25.2% when a single asset (AGG) was dropped from the universe. Nothing about the rule changed.
• The early decades are not tradable history. ETFs did not exist in 1970. The backtests use index proxies for those years, which carry no spread, no commission and no tracking error.
• Month-end is not the floor. Every figure here is measured at the end of a month, because that is how the papers measure. Within a month the fall ran deeper, and your account would have shown it.

Taken together: these are the results of rules applied to the past, by the people proposing the rules, on data that flatters the early years. They are the best evidence available and they are not a forecast. What they do support is the shape of the claim — that these designs were aimed at the depth of the fall rather than the height of the return, and that on the record they were measured against, the falls were shallower.