Buying past winners and selling past losers — and the crash that rides along

Jegadeesh & Titman ranked stocks on their trailing returns, bought the winners, sold the losers, and reported the whole formation-period × holding-period table rather than one lucky pair. This rebuilds that grid on a synthetic cross-section of 60 assets you control — including the honest control case (no predictability at all) and the failure mode the same literature documents: rare, violent crashes.

The sort

Formation period J6 months

Rank all 60 assets on their trailing J-month return. The documented effect is for 3–12-month winners (Jegadeesh & Titman 1993).

Holding period K6 months

A fresh portfolio is formed every month and held K months, so K overlapping cohorts are open at once and each month only 1/K of the book rolls.

Portfolio size

Concentrating into 3 names sharpens the sort and multiplies the idiosyncratic noise in every monthly reading.

Formation-to-holding gap

Wait one month before holding so the last formation month never overlaps the holding window. A variant you can test here, not a claim about the paper.

Sample & frictions

One-way cost10 bps

Charged on the book's actual traded notional each month as the overlapping cohorts roll on and off.

Momentum signal strength70

Scales the persistence and the spread of an unobserved expected-return state. At 0 that state is identically zero: the cross-section has no predictability whatsoever, so a J/K sort can only pick up sampling noise.

Injected momentum crash

Daniel & Moskowitz (2016) found momentum “suffers rare, violent crashes — losing over 70% in months during 1932 and 2009 rebounds”. This injects one such episode: a market slump, then a sharp rebound in which the prior losers lead violently. An injected episode — not historical data.

Sample#1

Same generator, new random draw. Watch which grid cell wins this time.

Mean monthly winner-minus-loser spread, selected J and K

before costs, the raw sort the 1993 paper tabulated

t-statistic of that spread

Mean monthly turnover of the book

Worst single month, WML book

net of cost

Worst drawdown, WML book

peak to trough of a hypothetical index

WML annualised, net of cost

Equal-weighted market, annualised

same months — the WML book is long-short, so this is not a like-for-like race

Best (J,K) cell in this draw — it moves when you reseed--

Formation J (rows) × holding K (columns) · mean monthly winner − loser spread

Positive spread Negative spread Your selected J and K Best cell in this draw

Cells are percent per month, net of your one-way cost charged on each cell's own turnover — set cost to 0 bps to read the raw sort instead. Short holding periods roll the whole book more often, so they pay the most.

Cumulative index of the winner-minus-loser book, net of cost (100 = start)

Winners minus losers, net of cost Equal-weighted market Injected crash episode

What you are looking at: a hypothetical index of a simulated long-short sort on synthetic data, not an account and not a return anyone earned. Set momentum signal strength to 0 and every cell of the grid becomes noise — there is nothing left in the cross-section to detect. At 0 bps the cells then scatter from one draw to the next, and a whole grid can come up green on luck alone (leave the cost on and they simply all go red). Turn the signal back up and hit “draw a new sample” repeatedly: the grid's best cell jumps somewhere else nearly every time, and the same fixed J and K reads very differently draw to draw. That is exactly why the research claim is about a diversified portfolio at monthly horizons rather than about any single chart.

What the research actually found

Support level: strong. This is the one family of ideas technical trading points at where the evidence is genuinely on its side, and IndicatorEdge says so as plainly as it says the opposite elsewhere.

“Not a chart pattern but a return regularity: assets that performed well keep performing well for months. Documented in stocks by Jegadeesh & Titman (1993) and generalized across 58 futures/FX/bond markets as time-series momentum by Moskowitz, Ooi & Pedersen (2012).”

“This is the best-supported idea 'technical' trading points at. Buying 3–12-month winners and selling losers earned significant excess returns in US stocks (Jegadeesh & Titman 1993); 12-month time-series momentum is positive in nearly every liquid futures market (Moskowitz, Ooi & Pedersen 2012); and a reconstruction back to 1880 finds positive trend-following returns in every decade since (Hurst, Ooi & Pedersen 2017).”

And the limits, from the same literature:

“It is not a free lunch and not a precise entry signal. Momentum suffers rare, violent crashes — losing over 70% in months during 1932 and 2009 rebounds (Daniel & Moskowitz 2016) — and the academic effect is a diversified portfolio phenomenon at monthly horizons, not a promise that any single chart's trend will continue.”

Jegadeesh, Narasimhan, & Titman, Sheridan (1993). “Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency.” Journal of Finance, 48(1), 65–91.

Moskowitz, Tobias J., Ooi, Yao Hua, & Pedersen, Lasse Heje (2012). “Time Series Momentum.” Journal of Financial Economics, 104(2), 228–250.

Hurst, Brian, Ooi, Yao Hua, & Pedersen, Lasse Heje (2017). “A Century of Evidence on Trend-Following Investing.” Journal of Portfolio Management, 44(1), 15–29.

Daniel, Kent, & Moskowitz, Tobias J. (2016). “Momentum Crashes.” Journal of Financial Economics, 122(2), 221–247.