Criticism

The objections to this strategy are a different matter from the mistakes that happen when it is carried out; those are in the chapter Staying the Course under Typical Mistakes. This chapter is about the rule itself: what is publicly disputed about its derivation, its backtest and its running costs.

Each entry names the objection, this page's counter-position and where it can be read.

The window was selected on the past

The widely used value for the window length has no reason outside the numbers. In the discussion of the German-language series it is called overfitted to the past, because neighbouring values do markedly worse and nobody can say why it should be this one. Later backtests from the scene found a flat zone around the usual lengths instead. On top of that comes the objection that the sensitivity to window length and index choice has hardly been tested in public in any systematic way. Anyone adopting the value adopts a choice that was made on the same market history.

This page's counter-position: it names the zones from the scene's investigations instead of a single value, carries the preset as a starting point and offers no parameter search.

The paper's backtest left costs out

The 2016 version the rule rests on computes the leveraged side without the internal cost of leverage of a real product, without fees and taxes and on gross rather than net total return, and in nominal rather than inflation-adjusted terms. Each of these items lifts the result. The objection is not against the idea but against the size of the effect it is circulated with.

The 2026 revision concedes half of the objection: it includes financing costs, fee and trading costs, and the lead shrinks to a size the paper itself rates as not statistically established. Taxes and inflation are still missing, and it says nothing about the price basis.

This page's counter-position: its backtest models financing costs and fees, names the price basis it computes on, and carries tax as an assumption of its own. On the US versions the gross index is only the signal; the return comes from the prices of the products (before their launch from a reconstruction on the index of the product). In US funds dividends flow into the fund without withholding tax, so gross is defensible there as the signal basis.

Switching costs are a running friction, not a one-off item

The rule lives on switching, and every switch pays the bid-ask spread. In sideways markets, switches that are reversed shortly afterwards pile up, and such phases occur often. As the main weakness of the pure price filter this has been public since 2016 in Growth and Trend, together with the proposal to tie the signal to a second filter. A calculation that shows the benefit in the declines without carrying the friction of the remaining years compares two different things.

This page's counter-position: the histogram in the chapter Staying the Course counts the switches per year.

In a German taxable account the tax on every realised gain comes on top, and it comes at every exit rather than at the end. This case has been computed in public in part 10 of the series ZahlGrafs Exzellente Abenteuer; the result points in the direction that ongoing taxation eats part of the advantage the same rule keeps inside a tax-free account.

This page's counter-position: that is exactly why the tax assumption at the chart can be switched off rather than being built in, and why the chapter Implementation and Taxes spells the German case out. The tax classification there is rough, states its date and is no substitute for tax advice.

Synthetic series overstate the result when costs are missing

For the time before the products existed, every long backtest of this rule stands on a reconstructed series. If financing costs, fees or the gap between a product and its index are missing from it, the result is a leveraged product that never existed in that form, and the rule looks better because its underlying looks better. How large these items are for a real leveraged product has been described for years, among others at Double-Digit Numerics. The objection applies not only to other people's calculations but to every one of them, including the one on this page.

This page's counter-position: its leverage model includes financing costs and fees, the chapter Sources names its assumptions, and only derived quantities are shown, as a distribution rather than as single numbers.

The price history was favourable for the rule

Every backtest of this rule shows the single path the markets took. Evaluations from the scene recombine sections of that history and recompute the rule on the paths that arise from it. They show that the actual price history turned out unusually favourable for the rule: only a small part of the recombined paths ran better. And favourable zones an earlier analysis had found turned out to be partly a consequence of that one return sequence; their author retracts that analysis himself in a later episode. The approach is recomputed in public in Principia Amumbo, part V with several resampling methods side by side.

This page's counter-position: it shows the distribution of the annual return across rolling periods in the chapter Backtest instead of a single outcome, and it describes its backtest as what the rule would have done, not as what it will do.