Only weakly: +29.3% a year for the top quintile against +21.0% for holding every name equally weighted, a gap of +6.8% a year over 110 months. But t = 1.7 and it led in just 56% of months, so a sample this size cannot separate that gap from luck. Worst drawdown on the long leg was -17%. What that justifies holding: +2.0% of the portfolio — sized on 0.4%/yr - the lower bound of the measured 6.8% ± 3.9% - at quarter Kelly, capped at 5%. The headline +6.8% is not the number to size on; +0.4% is what survives its own error. The academic long-short spread was +2.7% a year net of costs (t = 0.8). Trading each name on its own trend did not clearly work: +4.2% a year net of costs (t = 1.9), long the names above their own trend and short the rest. Over the same months, simply buying and holding the benchmark returned +14.5% a year.

That tests the composite this table is sorted by. Ranking on 12-1 momentum alone — the literature's factor, and the reference the composite has to earn its extra parts against — gives +6.8% a year (t = 1.7). Where that is the better number, the extra factors are costing you rather than helping.

Long vs universe is the column that matters if you do not short: it is the top quantile's return minus an equal-weight hold of every name in the list, so it asks whether the ranking added anything over owning the whole universe. Spread is the academic long-short factor return, kept because it is the standard measure — but it needs a borrow and survives short squeezes, and in a high-volatility asset class an unlevered short leg is not implementable, which is why its drawdown can reach 100%. Cross-sectional: top quintile against bottom quintile, equal weight, monthly. Time-series: each name long when its own 12-1 momentum is positive, short otherwise. Costs assume 10 basis points per side on every replaced position. t is a mean monthly return over its standard error; below 2 it is not distinguishable from luck. Sample 2017-09-29 to 2026-11-11, run 6h ago.

No, but it did not reliably hurt either: +20.7% a year for the top quintile against +20.8% for holding every name equally weighted over 110 months, a gap of -0.1% a year — and at t = 0.1, with the top of the list ahead in 48% of months, that gap is not distinguishable from noise. The ranking carries no information here in either direction; it is a curiosity, not a signal and not a contrarian one. What that justifies holding: nothing — the ranking lost to holding everything; nothing to size. The academic long-short spread was -1.3% a year net of costs (t = 0.2). Trading each name on its own trend did not clearly work: +2.5% a year net of costs (t = 1.1), long the names above their own trend and short the rest. Over the same months, simply buying and holding the benchmark returned +10.9% a year.

That tests the composite this table is sorted by. Ranking on 12-1 momentum alone — the literature's factor, and the reference the composite has to earn its extra parts against — gives +1.0% a year (t = 0.5). Where that is the better number, the extra factors are costing you rather than helping.

Long vs universe is the column that matters if you do not short: it is the top quantile's return minus an equal-weight hold of every name in the list, so it asks whether the ranking added anything over owning the whole universe. Spread is the academic long-short factor return, kept because it is the standard measure — but it needs a borrow and survives short squeezes, and in a high-volatility asset class an unlevered short leg is not implementable, which is why its drawdown can reach 100%. Cross-sectional: top quintile against bottom quintile, equal weight, monthly. Time-series: each name long when its own 12-1 momentum is positive, short otherwise. Costs assume 10 basis points per side on every replaced position. t is a mean monthly return over its standard error; below 2 it is not distinguishable from luck. Sample 2017-09-29 to 2026-11-12, run 6h ago.

No, but it did not reliably hurt either: +24.7% a year for the top quintile against +154.8% for holding every name equally weighted over 100 months, a gap of -51.1% a year — and at t = -1.1, with the top of the list ahead in 45% of months, that gap is not distinguishable from noise. The ranking carries no information here in either direction; it is a curiosity, not a signal and not a contrarian one. What that justifies holding: nothing — the ranking lost to holding everything; nothing to size. The academic long-short spread was -90.2% a year net of costs (t = -1.3), but with a -100% drawdown it is not implementable unlevered — short legs in this asset class get squeezed. Trading each name on its own trend did not clearly work: +40.6% a year net of costs (t = 1.0), long the names above their own trend and short the rest. Over the same months, simply buying and holding the benchmark returned +31.0% a year.

That tests the composite this table is sorted by. Ranking on 12-1 momentum alone — the literature's factor, and the reference the composite has to earn its extra parts against — gives -39.5% a year (t = 0.0). Where that is the better number, the extra factors are costing you rather than helping.

Long vs universe is the column that matters if you do not short: it is the top quantile's return minus an equal-weight hold of every name in the list, so it asks whether the ranking added anything over owning the whole universe. Spread is the academic long-short factor return, kept because it is the standard measure — but it needs a borrow and survives short squeezes, and in a high-volatility asset class an unlevered short leg is not implementable, which is why its drawdown can reach 100%. Cross-sectional: top quintile against bottom quintile, equal weight, monthly. Time-series: each name long when its own 12-1 momentum is positive, short otherwise. Costs assume 10 basis points per side on every replaced position. t is a mean monthly return over its standard error; below 2 it is not distinguishable from luck. Sample 2018-07-31 to 2026-11-12, run 6h ago.

No, but it did not reliably hurt either: +2.4% a year for the top quintile against +11.0% for holding every name equally weighted over 110 months, a gap of -7.7% a year — and at t = -0.9, with the top of the list ahead in 48% of months, that gap is not distinguishable from noise. The ranking carries no information here in either direction; it is a curiosity, not a signal and not a contrarian one. What that justifies holding: nothing — the ranking lost to holding everything; nothing to size. The academic long-short spread was -21.8% a year net of costs (t = -1.2), but with a -89% drawdown it is not implementable unlevered — short legs in this asset class get squeezed. Trading each name on its own trend did not clearly work: -1.7% a year net of costs (t = -0.1), long the names above their own trend and short the rest. Over the same months, simply buying and holding the benchmark returned +9.9% a year.

That tests the composite this table is sorted by. Ranking on 12-1 momentum alone — the literature's factor, and the reference the composite has to earn its extra parts against — gives -5.6% a year (t = -0.4). Where that is the better number, the extra factors are costing you rather than helping.

Long vs universe is the column that matters if you do not short: it is the top quantile's return minus an equal-weight hold of every name in the list, so it asks whether the ranking added anything over owning the whole universe. Spread is the academic long-short factor return, kept because it is the standard measure — but it needs a borrow and survives short squeezes, and in a high-volatility asset class an unlevered short leg is not implementable, which is why its drawdown can reach 100%. Cross-sectional: top quintile against bottom quintile, equal weight, monthly. Time-series: each name long when its own 12-1 momentum is positive, short otherwise. Costs assume 10 basis points per side on every replaced position. t is a mean monthly return over its standard error; below 2 it is not distinguishable from luck. Sample 2017-09-29 to 2026-11-12, run 6h ago.