Anticipating a moving-average crossover before confirmation

Rewrite the crossover as a threshold close

A moving-average-crossover is a rule that fires when a shorter simple moving average crosses a longer one. In this workflow that cross is the event to anticipate, not the entry itself.

A 20-day and 30-day simple moving average pair can be turned into a next-bar threshold close: the next-session close that would make the shorter and longer simple moving averages meet on the following bar. A cross prediction is a same-day signal that the current close has already crossed that threshold close, which implies a confirmed average cross is expected on the next bar.

Keep only probable cross days

A negative or otherwise unreachable threshold close can be rejected by requiring the next bar’s return to stay inside twice the extreme one-day return observed over a long lookback. That filter leaves only a small share of bars as probable cross days, meaning the implied threshold close still lies inside a realistic next-bar return band so a crossover is possible rather than arithmetically excluded.

On a single-name example using an 800-bar lookback, that realistic-return band implied a next-bar daily-return range of +48% to -26%, and most candles were classified as having no practical chance of a 20/30-day average cross.

A one-day lead on names and breadth

When predicted and confirmed crossovers are counted across the same 100-stock NASDAQ 100 universe, the prediction series leads the confirmed-cross series by one day at the market-breadth level as well as on individual names. A composite breadth ticker is that daily count, across a defined stock universe, of how many names show a predicted or confirmed ascending or descending crossover.

Across the NASDAQ 100 sample, four simple-average pairs produced hundreds to thousands of descending and ascending predictions, with the majority later confirmed and only a small percentage failing in each direction. The 20/30-day pair generated 1772 descending and 1766 ascending predictions, of which 85.79% and 88.09% respectively fell into the leading confirmation bucket reported for that pair. Lengthening the slower average reduced prediction counts, for example to 562 descending and 621 ascending predictions on the 20/100-day pair, while the 50/100-day pair showed the highest leading-confirmation shares at 92.50% descending and 91.76% ascending.

The same test on an indicator

The same threshold-close prediction can be applied after replacing raw close with an indicator series, so a slow and fast moving average of that indicator can be anticipated in the same way as a price-average cross.

Place the entry only in a persistent trend

Knowing the prediction at the close of day zero creates a defined execution window through the next session’s range and the following open, so a rule-based-entry can be placed before the lagging confirmed cross would otherwise fire. That rule-based-entry is an entry, delay, or abstention decision defined by a closed-form next-day close, a realistic-return filter, and whether the predicted cross is later confirmed.

The procedure is presented as a short-horizon trend-following test. Trend-following here only treats predicted and confirmed crossovers as actionable when price is already moving in a persistent direction, and treats congestion as a reason to stand aside. Predicted crossovers can be traded during persistent trends, including with partial orders, but are not described as effective during congestion.