Quant Memo
Foundational

Convergent vs Divergent Strategies

Convergent strategies bet that a price will snap back toward some anchor; divergent strategies bet that once a move starts, it keeps going. The two need opposite risk controls.

Prerequisites: Trend Following, Cointegration

Ask two traders what a strategy "believes" about price, and you get one of two answers: it thinks price has gone too far and will come back, or it thinks price has started moving and will keep going. Nearly every systematic strategy is a bet on one of those two, opposite views, and confusing them is a common way to blow up a book — because the two families need opposite risk management.

Convergent strategies profit when prices return to an anchor and lose when a move runs away from them. Divergent strategies profit when a move runs away and lose when it snaps back. They have opposite skew, so a single risk framework built for one will misprice the other.

Two families, two shapes of P&L

A convergent strategy — pairs trading, statistical arbitrage, most fixed-income relative value — sells a spread when it looks stretched and buys it back when it normalizes. Most trades are small winners as the spread mean-reverts, but occasionally the anchor itself breaks (a merger fails, a peg is abandoned) and the loss on that one trade dwarfs many wins. This is negative skew: frequent small gains, rare large losses.

A divergent strategy — trend-following, breakout systems, most CTA programs — buys strength and sells weakness, cutting losses quickly when a move fails to continue and holding on when it does. Most trades lose a little as false breakouts get stopped out, but the rare trade that turns into a real trend pays for all of them. This is positive skew: frequent small losses, rare large gains.

convergent rare, large divergent rare, large wins →
Convergent P&L is many small wins with a fat left tail; divergent P&L is many small losses with a fat right tail.

Worked example

A convergent spread trade risks 1 unit to make 0.2 units on each of 20 independent trades, winning 90% of the time, but the losing trade loses 8 units instead of 1 (the anchor broke). Expected value: 0.9×0.20.1×8=0.180.8=0.620.9 \times 0.2 - 0.1 \times 8 = 0.18 - 0.8 = -0.62 per trade if that tail loss is underestimated — the strategy looks great for 19 trades and then erases years of gains on the 20th.

A divergent trend trade risks 1 unit per attempt, wins only 35% of the time at 1 unit, but the rare winner runs to 6 units. Expected value: 0.35×60.65×1=2.10.65=1.450.35 \times 6 - 0.65 \times 1 = 2.1 - 0.65 = 1.45 per trade — a losing hit rate that is still profitable because the tail is a gain, not a loss.

What this means in practice

The two families demand different discipline. A convergent trader must size the tail loss correctly — stress-test what happens if the spread never comes back — because the strategy's whole edge rests on that tail being rare and bounded. A divergent trader must be willing to be wrong most of the time and cut losses fast, because the edge rests entirely on not missing the rare large winner. Mixing the two without separating their risk budgets is how a fund with a working mean-reversion book gets blindsided by a regime where everything trends, or vice versa.

A backtest's Sharpe ratio looks similar for both families, but Sharpe ratios treat gains and losses symmetrically and hide skew. Judge a convergent strategy by its worst-case drawdown under a broken anchor, not its average trade.

Related concepts

Practice in interviews

Further reading

  • Chan, Quantitative Trading (ch. 2)
ShareTwitterLinkedIn