Holdco / Dual-Class Spread Strategy

The thesis

Many US-listed companies have multiple share classes that trade simultaneously: BRK.A vs BRK.B, GOOG vs GOOGL, FOX vs FOXA, NWS vs NWSA, BF.A vs BF.B, the various Liberty trackers. The classes differ in voting rights and sometimes in conversion privileges, but they are claims on identical underlying cash flows. The price ratio between any two classes should therefore be stable — deviations are pure supply-demand noise.

In practice, that ratio drifts. An institutional fund that gets a large redemption may dump one class disproportionately. An index inclusion may force buying of one class but not the other. Tax-loss selling at year-end may target the more liquid class. Each event creates a temporary divergence that mean-reverts within days to weeks. Buying the cheap class captures the reversion without the directional exposure of the underlying business — the cleanest “no-leg-uncertainty” arbitrage available without access to shorting.

Academic basis

Nenova (2003) computed the implicit value of voting rights across 18 countries by comparing dual-class share spreads, establishing that the spread is meaningful and varies systematically with country-level governance protections. In US samples, the voting-rights premium is typically 1–5%, with significant short-run variation around the long-run mean.

Doidge (2004) extended the analysis to US-cross-listed dual-class firms, finding that the spread compresses when private benefits of control are constrained by US disclosure regimes. Schultz and Shive (2010) directly tested the arbitrage opportunity using transaction-level data, documenting that dual-class spreads exhibit short-horizon mean reversion exploitable by patient capital, with realized returns net of trading costs in the range expected for a low-vol arbitrage strategy.

The reason the spread doesn’t fully arb away is friction: the two classes are imperfect substitutes (voting rights matter for some holders), the absolute spread is small (often 50–200 bps), and the timing of mean reversion is uncertain (the spread can drift further before reverting). These frictions sustain a small but persistent inefficiency.

How Alpha Suite implements it

• Curated 17-pair list — includes BRK.A/B, GOOG/GOOGL, FOX/FOXA, NWS/NWSA, BF.A/B, HEI.A/B, MOG.A/B, LEN.B/LEN, TAP.A/TAP, CWEN.A/CWEN, and the major Liberty trackers (LBRDA/K, LSXMA/K, BATRA/K, FWONA/K, LLYVA/K, LBTYA/K, LILA/K). Pairs that yfinance can’t price are skipped at runtime.
• 6-month rolling z-score — for each pair, compute the daily price ratio (cheap_default / sister_default), then compute the mean and standard deviation over the available window. The current z-score is (current_ratio − mean_ratio) / std_ratio.
• Threshold |z| > 1.5 — only pairs with current z-score outside the ±1.5σ band trigger a signal. The strategy is asymmetric in scoring (|z| ≥ 2.5 gets +15, 2.0–2.5 gets +10, 1.5–2.0 gets +5).
• Long the cheap class — if z is negative, the original “cheap” class is now anomalously cheap and gets the long. If z is positive, the sister class is now anomalously cheap and gets the long. Direction is always BULLISH on whichever class is cheap.
• Take-profit at the mean — the take-profit percentage equals the expected reversion percentage (mean_ratio − current_ratio) / current_ratio. Signals with expected reversion under 50 bps are skipped (not worth the friction).
• Stop-loss wider than TP — sl_pct = max(2%, 1.5 * tp_pct). Holdco spreads can drift further before reverting (timing risk); the wider stop accommodates that.
• 21-day horizon — most spreads revert within a month based on the empirical literature. Longer-horizon signals are rare and typically mean a structural change has happened (one of the legs has different liquidity or coverage profile permanently).
• No regime sensitivity — holdco spreads are macro-insensitive (no equity beta on the long-only leg, since both classes track the same business). A small -2 score adjustment in DEFENSIVE regime accounts for liquidity drying up but the effect is minor.

When it fires

Signals appear most often during (a) index-rebalance flow events that affect one class but not the other, (b) tax-loss-selling pressure at year-end on the more liquid class, (c) single-class-specific news (e.g., a board action that affects voting rights), or (d) pure illiquidity events where one class trades thinly for a few days and the spread blows out. The cleanest fires are on the Berkshire (BRK.A/B), Alphabet (GOOG/GOOGL), and Liberty trackers, where the underlying businesses are well-understood and the spread mechanics are visible.

Realistic signal volume: 0–5 active signals at any time across the 17-pair universe. Most of the time the spreads sit within their normal band; the strategy is designed to fire only on genuine outliers.