The Momentum Factor, Tested on 16 Years. It Was Just Bitcoin and Nvidia.

Every strategy I've torn apart on this channel, a real quant would roll their eyes at. So this time I tested the opposite: the one strategy actual finance professors — tenure, Nobel-adjacent citations — will tell you is genuinely real. It's called cross-sectional momentum. Rank a basket of ~30 things by their trailing 6–12 month return, buy the top third (winners), short the bottom third (losers), rebalance every month. Decades of papers back it; it shows up in stocks, bonds, commodities and currencies going back 200 years. It's the closest thing trading has to a peer-reviewed edge. I ran it on 16 years of data. It passed my bar. Barely. And then I found out the whole thing was Bitcoin and Nvidia wearing a lab coat. The real version the papers describe — long winners, short losers — fails clearly: about 6% a year with a 50% drawdown (a MAR near 0.1), because shorting the weakest names in a 15-year up-market just bleeds, and most people can't even run a multi-year 30-name short book in a normal account. Drop the short side and just buy the winners, long-only — the version a regular person could do — and now it "passes" (MAR ~0.55, Sortino ~1.1). But the drawdown was 68%. At its worst it cut your account by more than two-thirds. A strategy you'd actually deploy keeps its drawdown in the single digits; a 68% hole isn't a strategy you run, it's a strategy that runs you. And it barely passes anyway: strip the 5 best months and it drops back under the bar (~0.44). Then the test that ends it. The top 5 names were 42% of all the profit — Bitcoin, Nvidia, Amazon, Apple, Ethereum (Bitcoin alone was up over 8,000% across the test). Take crypto out — just the two coins — and every single version fails, all of them back under the bar. The professor-approved edge, on this data, wasn't momentum finding a hidden signal. It was being long Bitcoin and Nvidia during the biggest crypto and AI run in history. The factor didn't pick the winners. It just happened to be standing next to them when they went up. To be fair: momentum is real. Over long windows across many markets, winners tending to keep winning is one of the most robust facts we have. But the honest retail-runnable version of it reduces to "own strength, don't sell your winners too early." If you already hold a broad index, the megacap tech names, or a little Bitcoin, you're already running the only part of this that worked — without the 68% drawdown and without the monthly paperwork. You don't need a rebalanced factor model to own Nvidia. You can just own Nvidia. Verdict: for a regular person, no. The academic effect is real; what's actually runnable for you is mostly concentrated beta in whatever already went up, written in the language of a factor. Momentum is one of the truest things in finance and one of the easiest to accidentally reinvent as just owning what's hot. Not investment advice, and I'm not licensed to give it. One guy running 16 years of real data to see whether the respectable version survives an honest look. AI tools help me make these videos. The experiments and the opinions are mine. ▶ Check my work (write-up + code): https://github.com/backtest-graveyard/backtest-graveyard/tree/main/ep15-momentum-factor 0:00 The one strategy the professors respect 1:22 What cross-sectional momentum actually is 2:17 Why it's respected (200 years of papers) 3:05 The real version (long/short) fails 4:13 Drop the short side, and it "passes" — with a 68% drawdown 5:12 And it barely passes: strip the 5 best months 5:36 The test that ends it: take out two coins 6:44 But momentum is genuinely real 7:54 Verdict: the effect is real, what's sold is owning what's hot 9:18 Fifteen strategies later