One of the most stubbornly profitable patterns in finance is also one of the simplest to state: winners keep winning, for a while. Here’s how quants turn “relative strength” into a ranked, market-neutral strategy — why it works, the brutal way it occasionally blows up, and what an Indian trader can actually do with it.

Picture a school sports day. You don’t need to know whether 

 students are faster or slower than last year’s. To win your bet, you just line everyone up by their times in the last few races, put your money on the kids at the front, and bet against the kids at the back. You’re not forecasting the weather — you’re ranking the runners.

That is cross-sectional momentum. Take a universe of stocks, rank them by how they’ve performed against 

, and rebalance periodically. You don’t care if the whole market is up or down — only about the spread between the strong and the weak. It’s the strategy academics call the “momentum factor,” and it has been documented, argued over, and traded for thirty years.

 by nature. You rank a basket, go long the top slice (say the top 20% — the “winners”) and short the bottom slice (the “losers”), holding both in equal rupee amounts so the market exposure roughly cancels. What’s left is a bet purely on the 

: that recent winners will keep beating recent losers.

The landmark study is Jegadeesh and Titman (1993): stocks that outperformed over the past 3–12 months continued to outperform over the next several months, and the reverse for losers — a pattern strong enough that it later got its own letter in the quant alphabet (the momentum or “UMD” factor, bolted onto the Fama-French model by Carhart). Asness, Moskowitz and Pedersen later showed the same relative-strength effect shows up in nearly every asset class at once. It is one of the most replicated anomalies in all of finance.

Rank a group of stocks by recent performance, buy the strongest, short the weakest, and bet the winners keep beating the losers. You’re trading the 

past 12-month return, skipping the most recent month

 — quants write it “12-1” (twelve months of formation, minus one). You rank every stock by its return from one year ago up to one month ago, and that ranking decides the portfolio.

Why throw away the most recent month? Because of a separate, opposite effect: 

. Over a single month, last month’s biggest winners tend to give a little back and last month’s biggest losers tend to bounce — partly real, partly a microstructure artefact (the bid-ask bounce, one-off liquidity moves). If you don’t skip that month, the reversal pollutes your momentum signal and quietly bleeds returns. Skipping it is one of those tiny details that separates a textbook strategy from a working one.

 period (how far back you look — 12, 6, even 3 months) and the 

 period (how long you hold before re-ranking — usually a month). Shorter look-backs react faster but trade more; longer ones are calmer but slower to catch turns.

Rank stocks by their last year’s return but ignore the most recent month — because over very short windows, hot stocks tend to cool off briefly. That “skip-a-month” trick is the standard recipe.

Nobody has a single airtight answer, which is part of why it persists. There are two camps.

 Information spreads slowly and people react in stages. When good news hits, investors 

 at first — they anchor to old beliefs and update grudgingly — so the price drifts up over weeks instead of jumping (Hong and Stein’s “gradual information diffusion”). The disposition effect adds fuel: people rush to sell winners and cling to losers, which holds winners 

 fair value for a while, leaving room to keep rising. Later, trend-chasers pile in and push it further. Under-reaction creates the drift; herding extends it.

 The other camp argues momentum simply pays you for bearing a nasty risk — specifically the risk of the occasional violent crash (more on that below). On this view the return isn’t a free lunch; it’s a premium for holding something that blows up rarely but badly. Both stories are probably partly true.

Either people are slow to react to news and then pile in (so trends persist), or you’re being paid to hold something that crashes hard once in a while. Likely a bit of both.

“Momentum” gets used for two different things, so let’s be precise. 

: it ranks names against each other and trades the spread, and it’s market-neutral by construction (long and short cancel). There’s also 

 (absolute) momentum, which judges each market against 

 past — long if it’s been rising, short if it’s been falling — and can end up net long or net short everything at once.

The difference matters in a falling market. A cross-sectional book stays roughly market-neutral — it can profit even in a crash 

 winners fall less than losers, but it doesn’t take a directional view. The absolute version can flip fully short and ride the whole decline. This post is strictly about the relative, ranked version.

Cross-sectional momentum trades the gap between strong and weak stocks and stays market-neutral. (Its cousin, absolute/time-series momentum, instead bets on each market’s own up-or-down direction.)

The skeleton is “rank, slice, hold, repeat,” but every choice along the way is a real decision:

Pick a liquid, tradeable set — say the Nifty 200. Momentum is strongest in smaller stocks, but those are exactly where costs and shorting constraints bite hardest, so there’s a tension between where the signal lives and where you can actually trade.

Raw 12-1 return rewards whatever was most volatile. Many serious implementations rank by 

 momentum instead — past return divided by its own volatility — so you’re buying steady climbers, not lottery tickets. (This is exactly what the NSE’s own momentum index does; see the India section.)

Quintiles (20% buckets) or deciles (10%); equal-weight within a bucket, or tilt weights toward the strongest scores, or toward larger caps to keep it tradeable. Finer slices (deciles) give a stronger signal but smaller, costlier positions.

Monthly is classic but churns heavily. Quarterly or semi-annual rebalancing cuts turnover (and tax) at the cost of reacting slower — a trade-off that matters enormously once real costs enter.

Plain momentum quietly becomes a bet on whichever 

 ran hardest (all the hot names cluster in one industry). Pros often neutralise by sector, or by beta, or run 

 — ranking on the stock-specific return left after stripping out market and sector moves — which historically has a smoother ride and a gentler crash.

Choose your stock universe, ideally rank by return-per-unit-of-risk (not just raw return), slice into winner/loser buckets, decide how often to rebalance, and strip out sector bets so you’re trading genuine stock-level strength.

The thing nobody warns you about: momentum crashes

This is the defining risk of the strategy, so it gets its own section.

: long stretches of steady, pleasant gains punctuated by rare, savage losses. The classic trigger is a market that has crashed hard and then violently rebounds. During the crash, the worst-hit names (usually high-beta, beaten-down stocks) sink to the bottom of the ranking — so momentum 

 them. Then the market snaps back, those same junk names rocket the most, and your short book detonates. This is exactly what happened in 2009: as markets bottomed and reversed, momentum strategies suffered one of their worst drawdowns in history.

Daniel and Moskowitz, in their paper literally titled “Momentum Crashes,” showed the payoff looks like you’ve 

 (sold) a call option on the market: you collect a steady premium most of the time, but in a sharp panic-then-rebound you take a brutal, convex loss. That is the price of the momentum premium — you’re being paid, in calm times, to absorb that occasional explosion.

Momentum makes money slowly and loses it suddenly. Its worst days come when a crashed market rebounds violently — the beaten-down stocks it’s shorting bounce hardest and blow up the short side. Plan for it.

Forget formulas for a second — this is the whole mechanic. Take six stocks, compute each one’s 12-1 momentum (last year’s return, skipping the most recent month), rank them, and trade the ends:

You go long C and A (the top third), short F and D (the bottom third), in equal rupee amounts — say ₹5 lakh on each leg, so the book is roughly market-neutral. Next month you recompute everyone’s 12-1 score and reshuffle. Note what you are 

 doing: no forecast of the index, no view on earnings, no story. Just a ranking, mechanically traded.

The catch hides in that word “reshuffle.” If E and B climb the ranking next month while A slips, you’re buying and selling again — and every one of those trades pays STT, charges, and slippage, and triggers a short-term capital-gains event. Cross-sectional momentum is 

, and that is where its real-world edge gets eaten.

Rank your stocks, buy the top bucket, short the bottom bucket in equal amounts, and re-rank every month. The mechanic is trivial; the cost of constantly reshuffling is the hard part.

Cross-sectional momentum runs into Indian market plumbing harder than almost any other factor. Five things to know.

1. The easy route: long-only, off-the-shelf

You don’t have to build this yourself. The NSE runs the 

 index — it picks the 30 highest-momentum names from the Nifty 200 using a “normalised momentum score” built from each stock’s 

6- and 12-month returns adjusted for volatility

 (i.e. risk-adjusted momentum), tilt-weights them, and rebalances semi-annually with a buffer to limit churn. Several AMCs offer index funds and ETFs tracking it. That’s a genuine, low-cost, long-only momentum strategy you can SIP into — no shorting, no monthly grind. For most retail traders, this is the sensible version.

, but shorting cash equity in India is intraday-only unless you borrow stock through SLB — and SLB is thin and pricey for many names. So a true long-short momentum book is genuinely hard for retail. The workarounds: run both legs in 

 (but only ~180–220 stocks have futures, which chops out the small/mid-caps where momentum is strongest), or go 

long a momentum basket and short index futures

 to cancel the market — an approximate, cheaper version of market-neutral.

Monthly reshuffling means most of your gains are realised inside a year — so they’re 

 (raised from 15% in July 2024), versus just 12.5% on long-term gains. Add STT (0.1% each side on delivery), stamp duty, GST and slippage on a high-churn book, and the gap between the gross backtest and your after-tax, after-cost result is enormous. This single fact is why the slow, semi-annual NSE index often beats a hyperactive home-brewed momentum system in practice.

If you implement via futures: limited universe, client- and market-wide position limits, stocks dropping into ban periods when open interest spikes, and lot-size constraints all add friction and cap how cleanly you can hold a ranked basket.

A 30-name momentum basket drawn from the Nifty 200 can quietly become a bet on two or three hot sectors. In a sharp rotation, “diversified” momentum names tumble together. Watch your sector concentration — it’s usually higher than you think.

The clean retail route is a long-only Nifty200 Momentum 30 index fund/ETF — no shorting needed. The full long-short version is hard here because shorting is restricted, monthly churn gets taxed at 20% STCG, and the basket often hides a big sector bet.

Beyond the crash risk (section 6, and worth re-reading), four more failure modes:

 Momentum is famous and crowded. It can underperform for years at a stretch — plenty of investors capitulated on the factor during its weak patches in the 2010s, right before it worked again. A strategy you can’t sit through is a strategy you don’t really own.

 of trading costs. The academic long-short looks great on paper; net of turnover, STT, slippage and 20% STCG, a big chunk of it evaporates — especially in India.

 Why 12-1 and quintiles? Why not 6-1 and deciles? If you tried dozens of look-backs and bucket sizes and kept the prettiest, you’ve curve-fitted. Pick defensible parameters, check that neighbours also work, and expect live results to be worse.

 In a true liquidity panic, the crowded winners you’re long can get dumped alongside everything else, so the “market-neutral” hedge fails exactly when you need it.

It can go cold for years, its real edge is far smaller after costs and tax than the brochures show, it’s easy to over-tune, and its market-neutrality can desert you in a genuine panic.

Below is a complete, runnable long-short momentum backtest on a panel of stocks — ranked by 12-1, quintile long-short, monthly, with costs and (deliberately) a crash-and-rebound episode baked into the data so the famous left tail actually appears. Point the panel at your own wide table of daily stock returns to test it for real.

Two lines do the heavy lifting on honesty. 

 builds the 12-1 signal by summing twelve months and subtracting the most recent one — the skip-month. And 

 makes positions chosen at the end of a month earn the 

 month’s return, so the test can never trade on a price it hadn’t seen yet. Drop either and your backtest turns into fiction.

# Cross-sectional momentum — honest long-short backtest (12-1, look-ahead-free, costed)
import numpy as np, pandas as pd

# 1. PANEL: swap in your own wide DataFrame of daily stock returns (dates x tickers).
#    Synthetic here: market beta + persistent drift + noise, with a crash-then-
#    rebound baked in so the 'momentum crash' actually shows up.
def make_panel(N=60, days=2600, seed=12):
    rng = np.random.default_rng(seed)
    dates = pd.bdate_range("2015-01-01", periods=days)
    betas = np.clip(rng.normal(1.0, 0.45, N), 0.3, None)
    mkt = rng.normal(0.0004, 0.010, days)
    mkt[1500:1545] += -0.0075          # crash (~2 months)
    mkt[1565:1625] += 0.0072           # rebound -> breaks the short book
    alpha = np.zeros((days, N)); phi, sd = 0.94, 0.0011
    for t in range(1, days):
        alpha[t] = phi*alpha[t-1] + rng.normal(0, sd, N)
    idio = rng.normal(0, 0.016, (days, N))
    ret = betas[None,:]*mkt[:,None] + alpha + idio
    return pd.DataFrame(ret, index=dates, columns=[f"S{i:02d}" for i in range(N)])

R = make_panel()
m_ret = (1 + R).resample("ME").prod() - 1            # monthly returns
logm  = np.log1p(m_ret)

# 2. SIGNAL: 12-1 momentum = months t-11..t-1 (SKIP the most recent month t)
mom = logm.rolling(12).sum() - logm

# 3. PORTFOLIO: long top quintile, short bottom quintile, equal-weight, $-neutral
ranks = mom.rank(axis=1); n = ranks.max(axis=1)
top = ranks.ge((n*0.8).values[:,None]); bot = ranks.le((n*0.2).values[:,None])
w = (top.div(top.sum(axis=1),axis=0) - bot.div(bot.sum(axis=1),axis=0)).fillna(0.0)

# 4. RETURNS: *** w.shift(1): positions set at month t earn t+1 -> no look-ahead ***
COST = 0.0020                                         # 20 bps per unit weight traded
turn = (w - w.shift(1)).abs().sum(axis=1).shift(1).fillna(0)
ls_net = (w.shift(1) * m_ret).sum(axis=1) - turn*COST

# 5. METRICS (annualised) + skewness + worst month
def stats(r):
    r = r.dropna(); m = len(r)
    cagr = (1+r).prod()**(12/m) - 1; vol = r.std()*np.sqrt(12)
    skew = ((r-r.mean())**3).mean() / r.std()**3
    return cagr, vol, cagr/vol, skew, r.min()

c,v,s,sk,wm = stats(ls_net)
print(f"long-short  CAGR {c:.1%} | vol {v:.1%} | Sharpe {s:.2f} | skew {sk:+.2f} | worst month {wm:.1%}")

long-short  CAGR 4.9% | vol 13.0% | Sharpe 0.38 | skew -0.29 | worst month -12.0%

Synthetic panel, one representative run. The full version (equal-weight market 14.4% / Sharpe 0.72; long-only top quintile 13.7% / 0.66) shows the long leg roughly tracks the market — the factor lives in the long-short spread.

Read those numbers honestly. The long-short premium here is 

 — 4.9% a year, Sharpe 0.38 — nothing like the eye-watering returns a careless backtest would invent. And look at the last two figures: 

negative skew (−0.29) and a single −12% month.

 That worst month is the engineered crash — the market rebounding and the short book of beaten-down names exploding upward. That fat left tail is not noise to be smoothed away; it 

 momentum’s character. A backtest that shows momentum with smooth, positively-skewed returns has a bug or a look-ahead leak.

This little test proves the mechanics are sound and unbiased. The case 

 momentum as a durable edge comes from the long-horizon, many-market evidence (Jegadeesh-Titman and the “everywhere” studies), not from one synthetic run — and even there, it’s a premium you earn by surviving the crashes.

A trustworthy momentum backtest skips the recent month, lags its trades so it can’t peek, and subtracts costs — and when you do, the edge is modest and comes with an ugly occasional loss. If your backtest looks smooth and spectacular, distrust it.

Cross-sectional momentum ranks a universe of stocks by their past year’s return (skipping the most recent month to dodge short-term reversal), buys the top slice and shorts the bottom slice in equal amounts, and rebalances regularly — a market-neutral bet that recent winners keep beating recent losers. It works because investors are slow to react and then crowd in, and/or because you’re paid to endure its signature risk: rare, brutal “momentum crashes” when a beaten-down market rebounds and the short book explodes, giving the strategy an ugly negative skew. For an Indian trader the clean version is a long-only Nifty200 Momentum 30 index fund; the full long-short is constrained by India’s shorting rules, eaten by high turnover and 20% short-term capital-gains tax, and prone to hidden sector bets. Backtest it honestly — skip the month, lag the signal, subtract the costs — and you’ll find a modest, lumpy premium, not a money machine.

Educational content, not investment advice. The strategies described involve real risk of loss, and shorting/derivatives can lose more than you invest. Verify all current index methodologies, charges, and tax rules with your broker and a qualified CA before trading. Code and figures are illustrative; test on your own data before risking capital.