nyasekai

Graveyard / Basis and spreads / US stocks

Dual-class share arbitrage

Can you trade the price gap between two share classes of the same company?

Family
Basis and spreads
Verdict
Dead
Cause
Overfit
Buried
3 October 2026

The idea

Two share classes of one company have the same cash flows but different votes or liquidity. The gap should mean-revert, and liquidity-driven dislocations should close.

How we tested it

Data8 tradeable US pairs where both legs trade at least $5M a day and can be shorted: GOOG/GOOGL, FOX/FOXA, NWS/NWSA, HEI/HEI.A, ZG/Z, LBTYK/LBTYA, UA/UAA and FWONK/FWONA. Z-score of the log spread against a trailing mean; enter when it is large, exit when it returns, 3-sigma stop and 40-day maximum hold, one eighth of capital per pair.
PeriodJanuary 2016 to October 2026.
CostsClosing-auction cost: half the measured closing spread plus 2 bp per leg. A second, harsher assumption of 0.13% per leg.
HoldoutIn-sample pick tested out of sample.
ControlNone stated.

Result

Configurations passing0 of 12 (both cost models)
Best Sharpe0.42
Profit factor1.29 or lower
Max drawdown−2% to −7%
Out-of-sample Sharpe / profit factor of the in-sample pick0.07 / 1.03
Sharpe in 2023-24−1.1 / −1.4
Net per pairZG/Z +33 bp, FWONK/A +29 bp, HEI +20 bp, other five about 0 or negative

Why it failed

A first look at all 16 pairs, including untradeable ones, showed +29.5 bp per trade (t 5.0), but that result had a one-day lookahead: the exit was filled at the same close the z-score was read. With the lookahead removed and only tradeable pairs, the edge shrinks to about the cost of trading.

What would make it worth retesting

Only per pair, for the two names with a structural reason for wider dislocations (ZG/Z and FWONK/FWONA), as a fresh hypothesis with that reason stated in advance, not by dropping the losers afterwards.

Past results of a backtest say nothing certain about the future.