Some strategies bet that a move will keep going. Mean reversion bets the opposite: that a price stretched too far from “normal” will snap back. It’s one of the most intuitive ideas in trading — and one of the easiest to blow up with. Here’s the quant version, in plain language.

Stretch a rubber band and let go. It snaps back. Stretch it a little further — it still snaps back, harder. Stretch it too far, and it breaks — and now it never snaps back at all.

That single image is mean reversion, including its fatal flaw. The strategy is a bet that prices behave like a rubber band: pulled too far from their average, they tend to spring back. The whole game is buying when something has stretched unusually cheap, selling when it’s stretched unusually rich, and collecting the snap-back. The danger is the day the band breaks.

Pick something with a sense of “normal” — a stock around its recent average, or the gap between two related stocks. Measure how far today’s price has stretched away from that normal. If it’s stretched unusually 

, buy (betting it climbs back). If it’s stretched unusually 

, sell or short (betting it falls back). Close the trade when it returns to normal. Repeat.

lots of small, frequent wins, and the occasional big loss

. Most of the time the band snaps back and you pocket a little. Once in a while it doesn’t, and that one trade gives back weeks of gains. Traders describe this as “picking up pennies in front of a steamroller” — pleasant, until it isn’t.

Buy what’s gotten unusually cheap, sell what’s gotten unusually expensive, and bet it returns to average. You win often and small — and lose rarely but big.

 Panic and euphoria push prices past where they should be in the short run. A fund dumps a big position and the price drops more than the news deserves; once the selling is done, it drifts back up. Mean reversion is the bet that the overshoot gets corrected.

, they pay up (sell cheap) to whoever will take the other side. The trader who buys that panic and waits for normal to return is being paid for providing liquidity. A lot of mean reversion is just that service, dressed up.

 Over very short windows — days to a couple of weeks — last period’s biggest losers tend to bounce and the biggest winners tend to cool off. It’s a well-documented effect, partly genuine and partly the mechanical wobble of the bid-ask spread.

Prices overshoot when people panic, someone gets paid to calm the panic by taking the other side, and over short windows extreme moves tend to partly reverse. Mean reversion harvests all three.

Quants turn “stretched too far” into one tidy number: the 

. Take the price, subtract its recent moving average (the “normal”), and divide by how much it normally wobbles (its standard deviation):

z  =  ( price − moving average )  ÷  standard deviation

A z-score of 0 means “right at normal.” A z of −2 means “unusually cheap, two standard deviations below average.” A z of +2 means “unusually rich.” The rule writes itself: 

buy when z drops below about −2, short when it rises above about +2, and exit when z comes back near 0.

 If you’ve seen Bollinger Bands or RSI “oversold/overbought” signals, those are the same idea wearing different clothes — all of them just measure how far price has stretched from normal.

The z-score says how unusual today’s price is. Buy when it’s unusually cheap (z below −2), sell when it’s unusually expensive (z above +2), exit when it’s back to normal (z near 0).

 A single stock or index stretched far below its 20-day average — buy the dip, sell the rip. The most basic version, and the one we’ll code below.

 Rank a basket of stocks by the last week’s return, then 

buy the biggest losers and short the biggest winners

, betting the extremes snap back over the following days.

 Two closely related stocks (say two banks) usually move together. When the gap between them stretches abnormally wide, buy the laggard and short the leader, betting the gap closes. This is the heart of pairs and relative-value trading.

You can fade one asset against its own average, fade last week’s extreme movers against each other, or fade the gap between two related stocks. Same rubber band, different setups.

 Mean reversion’s killer is simple: sometimes a price falls because something is genuinely, permanently wrong — bad results, fraud, a sector in structural decline. To the z-score it just looks “cheap,” so you buy. It gets cheaper. You buy more. It keeps falling. You are now 

 — fading a real trend that has no intention of snapping back. One of these can erase months of small, satisfying wins in a single stretch. The rubber band didn’t stretch; it broke.

This is the mirror image of the strategy’s charm. The high win rate that makes mean reversion feel safe is exactly what makes the rare loss so dangerous — you get lulled into sizing up, right before the trade that doesn’t come back. Every serious mean-reversion system therefore needs a hard 

 or a time limit: a rule that says “if it hasn’t reverted by here, I was wrong — get out,” so a broken band can’t turn into a broken account.

The big risk is buying something cheap that just keeps getting cheaper because something’s truly broken. Always set a stop — a level where you admit the snap-back isn’t coming and walk away.

Say NIFTY’s 20-day average is 24,000 and today it closes at 23,400 — 600 points below normal. If its usual wobble (standard deviation) is about 250 points, that’s a z-score of roughly −2.4: unusually cheap. You buy one NIFTY futures lot.

. But flip the story: if 23,400 was the first step of a real slide to 21,000 on bad news, that same “cheap” signal keeps you buying all the way down. The single trade that doesn’t revert is the one you must plan for before you place the first one.

When the bet works, it’s a fast, tidy profit. The math only stays friendly if you’ve already decided where you’ll bail if it doesn’t.

Mean reversion runs head-first into two Indian frictions.

This is the big one. Mean reversion is naturally 

 — lots of small in-and-out trades. Every one pays STT (0.05% sell-side on futures, 0.025% on intraday equity, 0.15% on option premium after the April 2026 hike), plus brokerage, exchange and stamp charges, GST, and slippage. The edge per trade is small, so costs can quietly eat most of it. The faster you trade, the worse this gets — a system that looks brilliant before costs can be a slow bleed after them.

To short the “too rich” side and hold it more than a day, you need futures — cash-market short selling in India is intraday-only without stock lending. So positional mean reversion lives in NIFTY/Bank Nifty and liquid stock futures. And because you’re deliberately trading 

 the current move, in a thin or illiquid name you can get run over filling your order.

High churn means everything is short-term: intraday equity is speculative business income, F&O is non-speculative business income, both taxed at your slab. None of the gentle long-term 12.5% treatment applies here. (Not advice — check with a CA — but model the after-tax number, not the gross.)

Mean reversion trades a lot, so costs and taxes hit hard in India — and to short-and-hold you need futures. The strategy’s tiny per-trade edge survives only if you keep costs brutally low.

Here’s the whole thing in a few lines: build a price that mostly wiggles around its average, fade the z-score, and — crucially — bake in one regime where the band 

 (a permanent drop that never reverts) so the danger actually shows up. Point 

 at your own NIFTY-futures or stock CSV to test it for real.

: positions are set from yesterday’s close and earn today’s move, so the test can never peek at a price it hadn’t seen. Drop that one line and any mean-reversion backtest will look like a money-printer.

# Mean reversion — simplest honest backtest (z-score fade, look-ahead-free, costed)
import numpy as np, pandas as pd

# 1. PRICE: swap in your own CSV (date, close). Synthetic here: wiggles around a
#    slow fair value, plus ONE permanent 'break' that never reverts (the danger).
def make_prices(n=2600, seed=5, base=20000.0):
    rng = np.random.default_rng(seed)
    fair = np.cumsum(rng.normal(0.0002, 0.004, n))         # slow fair value
    dev  = np.zeros(n); phi, s = 0.90, 0.013               # mean-reverting wiggle
    for t in range(1, n):
        dev[t] = phi*dev[t-1] + rng.normal(0, s)
    brk = np.zeros(n)                                       # the rubber band breaks:
    brk[1420:1455] = np.linspace(0, -0.30, 35)             # 30% drop over 5 weeks...
    brk[1455:]     = -0.30                                  # ...that stays down forever
    return pd.Series(base*np.exp(fair + brk + dev),
                     index=pd.bdate_range("2015-01-01", periods=n))

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

# 2. SIGNAL: z-score of price vs its 20-day average, then FADE it
N = 20
z = (close - close.rolling(N).mean()) / close.rolling(N).std()
signal = -z.clip(-2, 2) / 2          # long when cheap, short when rich, in [-1,1]

# *** lag one bar so we never trade on a price we hadn't seen ***
position = signal.shift(1).fillna(0.0)

# 3. COSTS on turnover (mean reversion trades a LOT)
COST = 5/1e4                          # 5 bps per unit of notional turned over
turn  = position.diff().abs().fillna(position.abs())
gross = position*ret
net   = gross - turn*COST

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

for name, r in [("Gross", gross), ("NET of costs", net)]:
    c,v,s,d = stats(r)
    print(f"{name:13} CAGR {c:5.1%} | vol {v:5.1%} | Sharpe {s:4.2f} | maxDD {d:6.1%}")
brk = net.iloc[1420:1460]
print(f"the 'break': {((1+brk).prod()-1)*100:5.1f}%  | turnover {turn.sum()/(len(turn)/252):.0f}x/yr")

Synthetic data, deliberately generous to the strategy. One representative run.

Now read it honestly. A Sharpe of 1.65 looks 

 — and that’s the trap. This is a synthetic series built to mean-revert, so of course fading it shines; real markets are far less obliging. Three things deflate the fantasy. The single 

 is one regime where the band snapped and the fader kept buying a falling knife — that one episode drives the whole −24% drawdown. The 

 means it trades 64 times its own capital in a year, so even a tiny 5-bps cost drains nearly 

 before tax. And the high win rate that makes the gross line so smooth is exactly what tempts you to over-size into the one trade that breaks you.

The Sharpe is the bait. The rare blow-up and the cost bleed are the hook. A trustworthy mean-reversion backtest always shows both.

It’ll look amazing in a backtest — that’s the warning sign. Check what one un-reverting trend does to it, and how much its constant trading costs you. Those two numbers, not the headline Sharpe, are the real story.

Mean reversion bets that a price stretched unusually far from its average will snap back, so you buy what’s gotten cheap and sell what’s gotten rich — measured cleanly with a z-score (how many standard deviations from the moving average) — and close out when it returns to normal. It works because people overreact, because liquidity providers get paid to take the other side of panics, and because short-term extremes partly reverse, and it wins often and small. Its fatal flaw is the trend that never reverts: a price that’s cheap because something is genuinely broken, where fading it means catching a falling knife — which is why a stop-loss isn’t optional. For an Indian trader the strategy’s constant trading makes costs and slab-rate tax the real enemy, and holding a short needs futures. Backtest it honestly — lag the signal, subtract the costs — and treat a gorgeous Sharpe as a warning, not a win.

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 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.