Uptrend filter
A stock qualifies only when its price is above the 200-day average and the 50-day average is above the 200-day. The trend is a gate, not a source of points.
RESEARCH LIBRARY
RESEARCH INVESTIGATION 003
63-Day Momentum Ranking
CLOSED · NOT ROBUST
THE QUESTION
Among stocks already in an uptrend, do the ones that rose most over the last three months keep doing better?
After the 100-point score failed, the next test isolated a single idea: rank uptrending S&P 500 stocks by their 63-trading-day price change and hold the leaders.
The leading portfolios averaged a gain over SPY. But the ranking itself pointed the wrong way, half the periods lost to SPY, and almost all of the gain came from a handful of periods after 2022. That is not a reliable effect, so the hypothesis was closed.
Hypothesis Frozen
COMPLETERank Test
COMPLETEPortfolio Test
COMPLETEClosed
ARCHIVEDTHE HYPOTHESIS
One signal.
Written down first.
Among point-in-time S&P 500 stocks in an established uptrend, those with stronger 63-trading-day momentum than their peers on the same date will tend to rank higher in the following 20 trading days.
Momentum ranking
Qualifying stocks are ranked by price today divided by price 63 trading days ago, minus one. Highest first; ticker order breaks ties.
No relative-strength factor
Subtracting SPY's 63-day return from every stock on the same date shifts all values equally, so it cannot change the ranking. It was left out for that reason.
FROZEN TEST DESIGN
What could not change after the results were seen.
Signal
63-trading-day price change
Evaluated stock-dates
36,695 across 144 rebalance dates
Primary test
Within-date rank correlation with the next 20-day return
Portfolios
Equal-weight Top 3 / Top 5 / Top 10
Holding period
20 trading days
Benchmark
SPY over the same 20 trading days
Trading costs
0, 5, 10 and 25 basis points per trade
Not allowed afterwards
Changing the lookback, the trend rule or the horizon; adding volume or near-high factors; picking a favorable period
PRIMARY TEST
The ranking pointed
the wrong way.
If momentum helped, stocks ranked higher should have earned more over the next 20 trading days, giving a positive rank correlation. The average correlation was slightly negative, at −0.021, and it was positive on only 47% of dates.
The effect is small enough that it may be indistinguishable from zero, and no significance test was run. What it does show is that the predicted positive relationship did not appear.
EARLY2015–2018
−0.013Mean rank correlation
Positive on 43% of dates.MIDDLE2019–2021
−0.026Mean rank correlation
Positive on 47% of dates.LATER2022–2026
−0.025Mean rank correlation
Positive on 51% of dates; still negative on average.PORTFOLIO RESULTS
Yet the top portfolios averaged a gain over SPY.
Average excess return per 20-day period before trading costs. The typical period tells a different story: the Top 3 median was almost exactly zero.
Average excess return per period
Median +0.00% · beat SPY in 50% of periodsAverage excess return per period
Median +0.31% · beat SPY in 53% of periodsAverage excess return per period
Median +0.14% · beat SPY in 51% of periods- Beat SPY, 2015–2021
- Beat SPY, 2022–2026
- Trailed SPY
Each bar is one 20-trading-day holding period from January 2015 to May 2026, before trading costs. Large gains cluster in 2022–2026; earlier periods were negative on average.
!
WHY THE AVERAGE MISLEADS
A few big periods did all the work.
Removing just the five best periods out of 144 turns the Top 3 average negative. The 2022–2026 periods contributed 127% of the Top 3 total; the earlier periods subtracted from it.
An average built on a handful of outliers can disappear the moment those outliers stop happening.
Average return versus SPY per 20-day period, before trading costs, after dropping the strongest periods. With 25 basis points of trading costs per trade, the Top 10 portfolio trailed SPY even with every period included (−0.03%).
TEMPORAL STABILITY
Top 3 excess return by period.
Average excess return per 20-day period, before costs.
EARLY2015–2018
−0.24%Average excess return
Beat SPY in 45% of periods.MIDDLE2019–2021
−0.55%Average excess return
Beat SPY in 45% of periods.LATER2022–2026
+2.78%Average excess return
Beat SPY in 58% of periods. The entire gain came from here.THE DECISION
Close it.
Do not flip it.
The frozen hypothesis predicted a positive ranking relationship. It did not appear, and the portfolio gains were unstable and concentrated in one stretch of time.
Reversing the ranking, keeping only the later years or picking the best portfolio size after the fact would all be new hypotheses chosen by looking at the answer. None of those were allowed. The next idea, Trend Quality, was written down as a separate hypothesis before it was tested.
WHAT THIS DOES NOT SHOW
Momentum is not
proven useless.
Academic momentum research typically uses longer lookbacks, broader universes and long-short portfolios. This test covered one definition, one horizon and the largest U.S. stocks only.
It also does not show that the 2022–2026 behavior will continue, and the later period was not an untouched out-of-sample test.
LIMITATIONS
Recorded with the result.
Universe
Point-in-time S&P 500 membership through July 13, 2026
Unverified observations
About 5% of stock-dates excluded; the gaps may not be random
Prices
Adjusted daily prices; not actual trade fills
Costs
Fixed sensitivity assumptions, not realized trading costs
Inference
No significance test was run on this hypothesis
Data sources
Daily prices from Yahoo Finance, with Tiingo for some delisted stocks