Quantitative Momentum: A Systematic Process to Identify High Momentum Stocks

ELI5/TLDR

Jack Vogel walks through Alpha Architect’s recipe for a long-only momentum portfolio: start with the 1,000 largest US stocks, pick the 100 with the strongest 12-month return (ignore the most recent month), keep the 50 whose climb was smoothest, rebalance every three months timed to land just before quarter-end, and equal weight. The premise is that winners keep winning and losers keep losing — a stubborn anomaly that academics have prodded since 1993 and still cannot kill. The interesting parts are the design choices: why the gap month, why the smoothness filter, why three-month rebalancing instead of one, and why the calendar matters more than you would expect.

The Full Story

The factor that refuses to die

Eugene Fama, the patron saint of efficient markets, once called momentum “the premier anomaly.” That is awkward for a man whose career was spent arguing that anomalies should not exist. The original sin is the 1993 Jegadeesh and Titman paper, Returns to Buying Winners and Selling Losers, which showed that if you sort stocks by their past 12-month return and buy the top decile while shorting the bottom, you collect about 1.3% a month. Annualised, that is roughly 15%. The follow-up literature has been mostly attempts to explain it away. Few have stuck.

Vogel’s framing is simple. Pick a moment in time. Look back at how every stock in the universe did over the last twelve months. Rank them. Buy the winners. That is it. The rest of the talk is variations on how to make that crude idea less crude.

Step one: a sane universe

Alpha Architect starts with the top 1,000 US firms by market cap. The reason is defensive. If you backtest momentum on micro-caps you will find an enormous premium, but it will mostly come from stocks too thin to trade with real money. Vogel calls this the “micro-cap effect” and treats it as a warning sign rather than a feature. The first filter is therefore not a momentum filter at all — it is a tradability filter.

Step two: the 12-2 lookback

From 1,000 down to 100. The screen is the past 12 months of return, but excluding the most recent month. This is the so-called “12-2” measure, and it exists because of an embarrassing fact: short-term momentum runs in reverse.

If you look at one-month winners and one-month losers, it would be the exact opposite. Short-term losers would have positive return. Short-term winners would be negative.

So you cut the last month off the calculation. The 12-2 is highly correlated with the simple 12-month, and Vogel concedes that if you cannot easily compute 12-2 you may as well just use 12 months — but the lag squeezes out a little extra signal by avoiding the short-term reversal. The intermediate-term band — roughly six to fifteen months — is where the effect is cleanest. Go shorter than six months and you start contaminating it with reversal. Go longer than fifteen and you start contaminating it with mean reversion.

Step three: frog in the pan

This is the part of the methodology that does most of the intellectual work. Two stocks can both be up 80% over the year. One got there by grinding up day after day. The other jumped 50% on a single FDA approval and then drifted sideways. The paper Vogel cites — Da, Gurun, and Warachka’s Frog in the Pan — argues that the smooth one keeps going, and the jumpy one does not.

The intuition is behavioural. When a stock pops on a discrete piece of news, the market processes it, prices it in, and there is no edge left. When a stock creeps up on small, dull, continuous improvements, investors underreact — the news arrives in pieces too small to notice. By the time the market catches up, you have already been holding for a year.

The metaphor is grim. Drop a frog in boiling water and it jumps out. Drop it in cold water and slowly raise the heat and it stays put. The slow-cooked stocks are the ones to own.

The measure itself is a count of positive minus negative trading days over the year, scaled and signed by the direction of the momentum. Vogel admits you can use cruder proxies — idiosyncratic volatility, beta, plain old return volatility — and get most of the way there. The point is you want the smooth winners, not the spiky ones. Top 100 becomes top 50.

Step four: timing the calendar

The fourth filter is not a stock filter. It is a date filter. Momentum has a seasonal heartbeat, and you can either work with it or be the patsy on the other side of it.

Two forces drive the seasonality. The first is window-dressing. Institutional investors report holdings at quarter-end. Nobody wants to file a 13F showing they were sitting on losers, so they sell losers and buy winners going into the report. That is, definitionally, momentum trading. The second is the tax calendar. If you are sitting on a winner in December, you wait until January to sell so the capital gain falls into the next tax year. If you are sitting on a loser, you sell in December to harvest the loss. Both forces push winners up and losers down at quarter-ends, especially year-end.

Vogel’s twist is to rebalance just before the quarter ends, not after. He calls this the “smart rebalance” — end of February, May, August, November. The “dumb” version waits until the end of March, June, September, December, which means you only ride the seasonal tailwind on the way out. January, incidentally, is the one month where momentum is historically negative.

Step five: equal weight, hold your nerve

The final 50 stocks get equal weight. No optimisation, no fancy constraints. The resulting portfolio has 95% active share against the S&P 500. That number measures how different a portfolio is from its benchmark, and 95% essentially means the two have almost no overlap. This is the whole point. If you want returns that are different from the index, you have to actually be different from the index.

The art-not-science choices

Two questions came up that the methodology cannot answer cleanly. How many stocks should you hold, and how often should you rebalance?

Vogel showed a grid. As you concentrate (fewer names), returns rise — exactly what you would expect from a real factor, since the tails should be the strongest. As you rebalance less often, returns fall, and they fall sharply: going from monthly to quarterly costs about 2%, and going from monthly to annual costs nearly 3% across every column. So the gross-return optimum is “very concentrated, very frequently rebalanced.” But that is the paper portfolio. Real money pays trading costs and taxes.

The 50 stocks / 3 months choice is a compromise. More concentration would help returns. More turnover would also help, on paper. But turnover compounds into trading costs and tax drag in a taxable account. Vogel’s advice if you want to do this yourself: hold it in an IRA, or use an ETF wrapper, because the moment you let taxes into the equation, the edge thins quickly.

The trading-cost fight

The most honest moment of the talk is when Vogel walks through the academic-vs-practitioner brawl over how much capacity the momentum factor has. The academics, looking at historical bid-ask spreads from databases, say the strategy chokes at around $5 billion. The practitioners — Andrew Ang’s BlackRock paper in particular — argue the academic spreads are off by an order of magnitude, and that with realistic execution you can run $65 billion to $324 billion. Vogel’s own position is that trading costs eat into the anomaly but do not destroy it. He is also clear that he writes academic papers and runs a fund that charges fees, so he has no incentive to settle the debate either way.

The retail version of this debate is simpler. Zero commissions are nice, but the real cost is the bid-ask spread and the small price impact of your own trades. For a 50-stock portfolio rebalanced four times a year, that is manageable. For a 30-stock portfolio rebalanced monthly, it is not.

Why it works (maybe)

Vogel offers the standard behavioural story: investors underreact to information. Good news arrives, the stock goes up, but not enough — so it keeps drifting up as the news is digested. Bad news arrives, the stock falls, but not enough — so it keeps drifting down. Value works because investors overreact and write off bad businesses too aggressively. Momentum works because investors underreact and price in good news too slowly. The two factors are negatively correlated, which is why pairing them gives a smoother ride than either alone.

The other half of the argument is “limits to arbitrage.” Why doesn’t every institution run this? Because no CIO wants to walk into a board meeting and explain that the momentum strategy lost by 40% over five years. The historical data shows that this happens — long stretches where momentum underperforms the index by five to seven percent annualised. Individuals who are willing to live through those drawdowns have an edge precisely because most institutions cannot.

Key Takeaways

Lookback window: 12-2. Use the past 12 months of return, skipping the most recent month. The skip avoids contamination from the short-term reversal effect, where last month’s winners tend to underperform.
Useful band: anywhere from six to fifteen months. Shorter and you hit reversal. Longer and you hit mean reversion.
Frog in the pan smoothness: among high-momentum stocks, the ones with smoother price paths beat the jumpy ones. The cleanest measure counts positive minus negative trading days, signed by momentum direction. Proxies like idiosyncratic volatility or beta work nearly as well.
Premium size: the original Jegadeesh-Titman long-short ran at ~1.3% per month, roughly 15% annualised. Long-only top decile beat the market by a meaningful margin from 1927 to 2014 on Ken French’s data.
Concentration helps, rebalancing matters more. Going from monthly to quarterly costs about 2% in gross return; monthly to annual costs nearly 3%. Going from 300 names to 50 adds return because you actually express the factor.
Frictions: turnover is high — much higher than for value. Trading costs are debated (academic models say $5B capacity, practitioner models say $65-324B), but they are real. Run momentum in tax-sheltered accounts or via ETFs whose creation/redemption mechanics handle the rebalances.
Seasonality: quarter-ending months (Mar, Jun, Sep, Dec) are momentum’s best months because of window-dressing and tax-loss selling. January is the worst — historically the only month where momentum is negative. Rebalance into the quarter, not out of it.
Tax structure: December is when you want to be already positioned, not when you want to be transacting. Anyone running this in a taxable US account loses a chunk of the edge to short-term capital gains.
Active share: the Alpha Architect index runs around 95% active share against the S&P 500. The corollary is multi-year stretches of 5-7% annualised underperformance. The factor only pays if you can sit through them.
Pairing: value and momentum are negatively correlated as long-short factors but both have positive premiums — so the standard recommendation is to combine them.

Claude’s Take

This is a clean exposition of a strategy that has been the subject of about 30 years of argument. Vogel does not oversell. He concedes the 12-2 lag is marginal, the 50-stock choice is “more art than science,” and the trading-cost debate is unresolved. That is more honesty than most factor talks deliver. The frog-in-the-pan filter is the genuinely interesting piece — most people stop at “buy the winners” and never ask whether smooth winners differ from spiky ones.

What is missing — and this is the standard critique — is a serious treatment of post-publication decay and crowding. The Jegadeesh-Titman paper is from 1993. AQR, Alpha Architect, iShares MTUM, and dozens of others now systematically buy whatever passes a momentum screen. If the anomaly was driven by underreaction, you would expect a lot of that underreaction to have been arbitraged away by now. Vogel waves at this with the “limits to arbitrage” argument — institutions cannot stomach the drawdowns — but does not actually show that the post-publication premium is intact. The Ken French chart that anchors the talk runs to 2014, conveniently ending before momentum’s miserable 2015-2020. A real-world question worth asking is whether the premium that survives after fees, taxes, and crowding is large enough to justify the volatility of the strategy.

The other piece that is glossed is the momentum crash. 2009 was a famous one — momentum portfolios got crushed in the rebound off the financial crisis because the screen mechanically loaded into financials that had been beaten down, then reversed sharply. The strategy is not just “underperforms by 5% sometimes.” It is “occasionally has tail events that wipe out years of premium in months.” Risk-adjusted, the picture is less flattering than the long-run chart suggests.

Score is a 7. Useful as a structured walkthrough of how a published factor gets translated into a real portfolio, with most of the design tradeoffs spelled out. Loses a point for the dated data and the soft treatment of crowding, and another for the inevitable pitch-flavoured framing — this is a fund manager talking about his own fund, and the boundary between exposition and marketing is not always clean.