The oldest idea in trading — buy what’s going up, sell what’s going down — turned into something you can write in code, size with maths, and stress-test honestly. Here’s how trend following actually works, where the edge comes from, and where it quietly bleeds you.

A surfer doesn’t forecast the ocean. They don’t sit on the beach with a model of wind and tide trying to predict the perfect wave. They paddle out, wait, and when a wave starts lifting them, they ride it — until it flattens, then they kick off and wait for the next one.

That is the entire philosophy of trend following in one image. You are not trying to be right about where the market is 

. You are trying to notice, mechanically, when a move has 

, climb on, and stay on until it’s clearly over. The forecast is replaced by a reaction.

For decades this was a discretionary art — chartists drawing trendlines, the old Turtle traders following hand-written rules. The quant’s job was to take that intuition and make it precise: define “a trend” as a number, define “ride it” as a position size, and define “honestly” as a backtest that doesn’t cheat. That’s what this post is about.

Strip away the indicators and trend following is a bet on 

: the belief that, on average, something that has been going up is slightly more likely to keep going up than to suddenly reverse — and the same in reverse for things going down. You buy strength, you short weakness, you 

That last asymmetry is the whole game. A trend follower is wrong often — more than half the time, in fact — but the rules force the losers to be small (you exit when the trend turns) and allow the winners to be large (you stay in as long as the trend lasts). It’s a strategy that makes its money from a few big moves and pays a steady toll of small losses in between. The shape of the payoff matters far more than how often you’re “right.”

Don’t guess the future. Wait until a move is clearly underway, get in, hold while it works, and bail the moment it stops working. Many small losses, a few big wins.

If markets were perfectly efficient, prices would jump instantly to fair value and there’d be no trend to follow. They don’t, for two broad reasons.

 to new information at first — they anchor to the old price and only adjust slowly, so good news leaks into the price over days and weeks instead of all at once. Then, later, they 

 — they pile into a move that’s already running (herding, FOMO), pushing it further than fundamentals justify. Under-reaction starts the trend; over-reaction extends it. There’s also the disposition effect: traders sell winners too early and hold losers too long, which dampens prices on the way up and props them on the way down — again, slow adjustment.

 A lot of money in markets doesn’t trade on opinion at all. Risk-management rules force funds to cut positions as losses mount (selling into a falling market, extending the fall). Index funds mechanically buy what’s being added and sell what’s removed. Central banks move rates in slow, telegraphed steps rather than all at once. All of this creates predictable, drawn-out pressure in one direction — exactly what a trend follower feeds on.

And the evidence is not thin. AQR’s Hurst, Ooi and Pedersen tested a simple trend strategy across 67 markets in four asset classes (commodities, equity indices, bonds, currencies) going back to 

, and found it was profitable, net of costs, with remarkable consistency — through the Great Depression, world wars, the 2008 crisis, and every rate cycle in between. The related “time-series momentum” work (Moskowitz, Ooi and Pedersen) showed the same effect holds across nearly every liquid futures market since the 1980s. Trend following is about as close to a universal market regularity as quant finance has found.

Trends exist because humans are slow to update and quick to crowd, and because a lot of big money is forced to trade in the same direction over time. The pattern has shown up in market data for over a century, almost everywhere.

03Two flavours: time-series vs cross-sectional

Quants split “momentum” into two cousins, and it’s worth keeping them straight.

. Did NIFTY go up over the last few months? Then be long NIFTY. Did crude go down? Be short crude. Every instrument is judged on its own. This is the classic “trend following” of managed-futures funds (CTAs), and it can be net long, net short, or flat across everything at once — which is why it can make money in a crash.

 of instruments against each other and goes long the strongest, short the weakest — regardless of whether the whole group is rising or falling. This is the “momentum factor” you see in equity quant (buy the top-decile stocks by past return, short the bottom decile). It’s market-neutral by construction; it cares about 

Time-series momentum asks “is this thing trending, yes or no?” and trades it on its own. Cross-sectional momentum asks “which of these is strongest vs weakest?” and trades the spread. The first is what most people mean by “trend following.”

04How a quant turns “a trend” into a number

“The trend is up” has to become a rule a computer can evaluate at the close every day. There’s a small menu of standard ways, and they mostly agree with each other:

Take a fast average (say the last 20 days) and a slow one (say 100 days). When the fast crosses 

 the slow, the recent path is steeper than the long path — trend up, go long. When it crosses below, go short. Simple, robust, and the one we’ll code below.

Go long when price makes a new N-day high, short on a new N-day low. This is the famous Turtle-trader logic: a fresh 20- or 55-day high 

 the trend signal. Breakouts catch moves early but whipsaw more in choppy ranges.

The most minimal of all: look at the total return over the past, say, 3, 6 or 12 months. Positive? Be long. Negative? Be short. No indicators — just the sign of a past return. Surprisingly, this bare-bones rule is what most of the academic evidence is built on.

The unifying trick: normalise by volatility

A 2% move in a sleepy bond future is a big deal; a 2% move in crude is a Tuesday. So quants rarely use raw price changes — they divide signals (and positions) by each market’s recent volatility, so a “strong trend” means the same thing everywhere. This single step is what lets one system trade NIFTY, gold, and the rupee side by side without crude dominating just because it wiggles more.

There are a few standard trend rules — moving-average crossovers, breakouts to new highs/lows, or just the sign of the last few months’ return. They mostly say the same thing. The key extra step is scaling everything by how volatile each market is, so they’re comparable.

Here’s the part beginners skip and professionals obsess over. The signal (long/short/flat) is maybe a third of the strategy. 

: instead of trading a fixed number of lots, you trade a position sized so your 

 is roughly constant through time. When a market gets calmer, you lever up to keep the same risk; when it gets wild, you shrink. Mechanically, your position is proportional to (target volatility ÷ recent realised volatility). The effect is a smoother equity curve and far fewer ugly surprises than “always trade 1 lot.”

 — and this is the actual “secret” of the big trend funds. One market trending is a coin toss in any given year. Fifty mostly-uncorrelated markets, each running the same rules, vol-targeted and combined, smooth each other out: when equities chop, maybe bonds or commodities are trending. The edge per market is small and unreliable; the edge of the 

 is what pays. (Keep this in mind — it’s exactly the thing an Indian retail trader struggles to reproduce. More on that below.)

Getting the buy/sell call right is only part of it. Sizing each bet so your risk stays steady — and spreading the same rules across many unrelated markets — is what actually turns a flaky signal into a strategy.

Trend following has a payoff shape unlike almost anything else, and it’s the reason institutions pay for it. Plot a diversified trend strategy’s return against the stock market’s return and you get a 

: it tends to do well when stocks fall hard 

 when stocks rise hard, and it bleeds gently in the calm, directionless middle.

 — slow, grinding declines (2008, early 2020 before the bounce). A time-series system simply flips short as those markets roll over and rides the fall, making money exactly when a long-only portfolio is being torched. That’s why trend following is often called “

” or described as having a “long option” profile: like owning insurance that, unlike real options, doesn’t cost a fixed premium — instead it slowly leaks money in quiet, choppy periods and pays off in dislocations.

Trend following tends to shine when markets move violently in either direction — especially crashes, because crashes trend — and to slowly lose during calm, sideways stretches. It behaves like crash insurance you pay for in small instalments.

. Today the 20-day average crosses above the 100-day — signal flips long. You buy one lot.

One NIFTY futures round trip · FY2026 cost stack

NIFTY ≈ 24,000 · lot size 65 · one lot ≈ 

 notional. You go long, and three weeks later the trend turns and you exit.

 (0.08%) of notional for the full in-and-out. For the trade to even break even, NIFTY has to move about 19 points in your favour 

 you make a single rupee. Notice STT (₹780) dwarfs everything except slippage — and it’s charged whether you win or lose.

Here’s the saving grace, though, and it’s specific to trend following: 

 A positional crossover on one market flips only a handful of times a year. So the April-2026 STT hike that gutted the options scalpers barely scratches a slow trend follower — the cost is real but it’s spread across months of holding, not paid forty times a day.

Every entry-and-exit on one NIFTY lot costs you around ₹1,250 all-in, mostly STT and slippage — about 0.08%. It sounds small, but it’s charged on every trade regardless of profit. Trend following survives this only because it trades rarely.

The textbooks describe trend following on a global basket of 50+ futures. An Indian retail trader lives in a narrower, costlier world. Five things matter.

 be able to go short — half its magic is selling falling markets. In India, shorting cash equity is intraday-only without stock lending (SLB), so you can’t hold a short stock overnight. The practical home for positional trend following is therefore 

: NSE index futures (NIFTY, Bank Nifty), liquid stock futures, MCX commodities (gold, silver, crude, natural gas), and currency futures. These let you be long or short, for as long as the trend lasts.

The full stack — STT (now 0.05% sell-side on futures, 0.15% on option premium), exchange and SEBI charges, stamp duty, 18% GST, brokerage, and slippage — punishes frequent trading. As we saw, this is survivable for slow trend systems but lethal for fast ones. The faster your crossover, the more the cost stack quietly eats your edge.

Indian futures expire monthly (NIFTY now on the 

). A positional trend follower holding for months must 

 — close the expiring contract and open the next — every single month, paying spread and a slice of STT each time. Build this into your cost assumptions; it’s a recurring drag the textbooks (which assume continuous contracts) hide.

There are client- and market-wide position limits in F&O; stocks can enter a ban period when open interest gets too high; and the November-2024 derivatives tightening (upfront premium, fewer weekly expiries, bigger contract values, the Jan-2026 lot-size revision) all change the playing field. None of this stops trend following — but it caps size and adds friction.

This is the big one. The reason institutional trend following is smooth is breadth — dozens of unrelated markets. Indian retail typically ends up running the same signal on NIFTY, Bank Nifty, and a few correlated large-caps, which all crash together. Without genuine diversification (adding MCX commodities and currencies helps), your trend strategy will have far lumpier, scarier returns than any glossy CTA tear-sheet. Respect that gap.

F&O income in India is generally treated as 

, taxed at your slab — but losses can be set off and carried forward, and STT itself is a deductible business expense. (Intraday equity is “speculative” with its own rules.) None of this is advice — talk to a CA — but build the after-tax picture, not just the gross one.

For an Indian trader, trend following lives in futures (so you can short), costs you on every roll and every trade, runs into SEBI position limits, and — most importantly — is hard to diversify, because our liquid markets move together. Expect a bumpier ride than the foreign brochures show.

 Trend following’s worst enemy isn’t a crash — it’s a long, choppy, going-nowhere market. Every time price pokes above your average you buy; it falls back, you sell at a loss; it pokes up again, you buy again. Death by a thousand small cuts. Real trend strategies can underperform a simple buy-and-hold for 

 at a stretch, and the psychological pain of bleeding through a flat decade is what makes most people quit at exactly the wrong time.

 A trend can give back weeks of profit in two days when it snaps (a V-bottom, a surprise RBI move, a budget gap). You always exit 

 the turn, never at the top — that giveback is the price of admission. Overnight gaps in Indian futures can blow through your exit level before you can act.

 Why 20 and 100? Why not 15 and 90, or 30 and 150? If you tried hundreds of combinations on past data and picked the prettiest, you’ve fitted noise, and it will decay out-of-sample. The honest fix: pick sensible parameters for good reasons, test that nearby values also work (robustness, not a single lucky peak), and expect live results to be worse than the backtest.

 In a true panic, “uncorrelated” markets can all move together and even trend strategies can get hit at once. Diversification is a fair-weather friend at the extremes.

Trend following’s real risk isn’t a dramatic crash — it’s boredom: long flat stretches that chop you up and tempt you to quit. It also gives back profit on every reversal, punishes over-tuned parameters, and offers less protection than you’d hope in a true panic.

Below is a complete, runnable trend system: a moving-average crossover, volatility-targeted, with realistic costs — and, crucially, 

 at your own NIFTY-futures / stock / MCX CSV (a 

 column) to test it for real. The synthetic series here just makes it reproducible without a data feed.

. A backtest’s most common lie is acting on information it couldn’t have had yet — using today’s close to trade today’s close. Lagging the signal by one bar forces every decision to use only what was known at the 

 close. Miss this and your backtest will look spectacular and your live account will not.

# Trend following — minimal, honest backtest (look-ahead-free, vol-targeted, costed)
import numpy as np, pandas as pd

# 1. PRICES: swap in your own CSV (date, close) for NIFTY-fut / stock / MCX
#    df = pd.read_csv("nifty_fut.csv", parse_dates=["date"], index_col="date"); close = df["close"]
def make_prices(n=2600, seed=3, start=18000.0):
    rng = np.random.default_rng(seed); drift = np.zeros(n); i = 0
    while i < n:
        L = int(rng.integers(30, 130))
        reg = rng.choice(["up","down","range"], p=[0.44,0.20,0.36])
        drift[i:i+L] = {"up":0.0011, "down":-0.0013, "range":0.0}[reg]; i += L
    price = start*np.exp(np.cumsum(drift + rng.normal(0,0.011,n)))
    return pd.Series(price, index=pd.bdate_range("2015-01-01", periods=n))

close = make_prices(); ret = close.pct_change()

# 2. SIGNAL: fast/slow moving-average crossover
FAST, SLOW = 20, 100
raw = np.sign(close.rolling(FAST).mean() - close.rolling(SLOW).mean())

# *** the one line that separates a real backtest from a fantasy ***
# act only on what was known at yesterday's close -> lag the signal by one bar
signal = raw.shift(1)

# 3. SIZE: volatility targeting (constant risk, not constant lots)
TARGET_VOL, ANN = 0.15, np.sqrt(252)
realised = ret.rolling(20).std().shift(1)            # lagged, no peeking
lever    = (TARGET_VOL / (realised*ANN)).clip(upper=3.0)
position = (signal*lever).fillna(0.0)

# 4. COSTS: charged on turnover (every change in position)
#    ~3 bps per unit of notional turned over = STT(0.05% sell) + exch/GST/stamp + slippage
COST_BPS = 3.0
turnover = position.diff().abs().fillna(position.abs())
gross = position*ret
net   = gross - turnover*(COST_BPS/1e4)

# 5. METRICS
def stats(r):
    r = r.dropna(); n = len(r)
    cagr = (1+r).prod()**(252/n) - 1; vol = r.std()*ANN
    eq = (1+r).cumprod(); mdd = (eq/eq.cummax()-1).min()
    return cagr, vol, (cagr/vol if vol else 0), mdd

for name, r in [("Buy & hold", ret), ("Trend gross", gross), ("Trend NET", net)]:
    c, v, s, d = stats(r)
    print(f"{name:12} CAGR {c:6.1%} | vol {v:5.1%} | Sharpe {s:5.2f} | maxDD {d:6.1%}")

Synthetic data, one representative run. Costs cost ~0.4%/yr here because the system trades only 29 times in a decade.

Now read those numbers honestly, because this is the lesson: 

 The market trended cleanly upward, so simply holding won, and the crossover’s occasional shorts and whipsaws cost it both return (8.4% vs 16.6%) and a deeper drawdown. That is not a bug — it’s the truth about trend following in a one-way bull market. Across other simulated decades I ran, net returns ranged from about −3% to +17%, and the trend system’s drawdown was smaller than buy-and-hold in roughly half the runs and larger in the rest. 

Trend following is not a machine that beats the market; it’s a different return stream

 — one that earns its keep in falling and choppy-divergent markets and in the ability to go short, not by out-running a steady rally.

 evidence of an edge — that comes from running these same honest rules across many markets and many decades, which is exactly what the century-long studies did.

A trustworthy backtest lags its signals so it can’t peek at the future, sizes by volatility, and subtracts real costs. When you do that honestly, trend following often 

 a plain buy-and-hold in calm bull markets — its payoff is a different shape, not a bigger number. Be suspicious of any trend backtest that looks too good.

Trend following is the systematic version of “cut your losses and let your profits run”: define a trend as a number (a moving-average crossover, a breakout, or the sign of a past return), go long uptrends and short downtrends, size each bet so your risk stays constant, and spread the same rules across as many unrelated markets as you can. It wins by being wrong often but small and right rarely but big, it has paid off across markets for over a century, and its signature gift is making money in crashes — but it bleeds through long sideways markets, gives back profit on every reversal, and for an Indian retail trader it’s constrained to futures, taxed on every roll, and hard to diversify. Backtest it honestly — lag the signal, subtract the costs — and you’ll find it’s not a market-beating machine but a different, sometimes uncomfortable, return stream.

Educational content, not investment advice. Derivatives carry the risk of losing more than you put in. Verify all current contract specs, charges, and tax rules with your broker and a qualified CA before trading. Code is illustrative; test on your own data before risking capital.