When the King Falls: Backtesting the Sovereign Exit Rule

In my last Sovereign post — The Top Three And The Tundra — I named the rigged game and the operational recipe: DCA into the top 3 names by market cap, never sell, let cap-weighted passive flow do the compounding for you. I finished that post confident in most of the argument and hand-waving on one specific question: what do you actually do when a name falls out of the top 3?

My instinct was "nothing." Never sell. Ride every position to death. Let the estate step-up wipe the basis tax-free. That's the Buy-Borrow-Die posture I run on everything else, and I assumed it extended cleanly to Sovereign.

It doesn't. I ran the backtest. The "never sell" rule is mediocre — not catastrophic, but provably leaving money on the table. Tight rotation wins.

Here are the results up front, because you should know where this is going:

1. Strict rotation — sell the moment a holding drops to #4 by company market cap — returned 9.48× on initial capital across four entry cohorts (2006, 2010, 2015, 2020). Never-sell returned 8.13×. SPY buy-and-hold returned 5.29×.
2. The passive-flow concentration premium is massively real. Even the worst SP3 exit rule I tested crushed SPY by ~47%. The structural thesis from The Top Three And The Tundra is validated — the question was never whether to concentrate, only how to manage the rare rotations.
3. Tax drag on tight rotation is smaller than I expected. R1 (strict) paid $59K–$241K of realized capital-gains tax per cohort, and still won on after-tax terminal value. The ranking signal was load-bearing enough to justify the drag.
4. The 2015 cohort is an instructive exception. Tight rotation briefly sold AAPL during the 2015-16 rank shuffle and had to buy it back. Rules that act on fast-moving rank changes can misfire when rankings are noisy rather than informative. Both costs are real, and both are smaller than the baseline win.
5. My production rule is now: sell any existing Sovereign position the quarter it drops to #4 or lower by company market cap. Redeploy proceeds equal-weight across the current top 3 on the next DCA tranche. One quarterly check, one trade when it fires, and that's it.

The rest of this post is the methodology, the honest look at where the rule misfires, and the tax math that quietly kills the "never sell" framework. If you've already bought the conclusion, you can skip to The Production Rule section.

The question Tundra didn't answer

The Top Three And The Tundra made the structural case for SP3. It didn't make the operational case for any specific rotation rule, because I hadn't done the work yet. I just asserted the simplest possible discipline — DCA in, hold forever, let step-up handle the taxes on death — and treated the rotation question as a footnote.

That footnote matters. Here's why.

If you deployed SP3 at 2006-Q1 and held forever, your starting basket was Exxon, GE, and Microsoft. That's literally the top 3 by cap that quarter. Twenty years later, only Microsoft is still in the top 10. Exxon is around #15. GE is no longer a recognizable continuing entity — it split into three separate companies. If you'd held all three through 2026-Q1, the terminal basket would have been dominated by old Microsoft gains, with the GE and Exxon legs dragging sideways for most of the window.

The never-sell rule produced $9.25M on a $1M initial deployment over those twenty years. Not bad. But the strict rotation rule — which swapped each of those names out as they lost their top-3 status — produced $11.75M over the same window, after paying $241K of realized capital-gains tax along the way. The rotation captured the compounding of newer entrants (Apple, Alphabet, Amazon, NVIDIA) that the frozen basket missed.

That's a 27% improvement in terminal wealth for the discipline of actually responding to the signal the mechanism is sending.

I'd been running an elegant theory of inaction. The data says inaction is worse than modest action.

The setup

The backtest is in cfo/sovereign/backtest/ — code, data, and the findings document. If you want to reproduce it, it takes about ten minutes of yfinance downloads and a few seconds of simulation.

Mechanics:

• Universe: top 20 US companies by market cap, with share-class aliasing (Alphabet = GOOG+GOOGL, Berkshire = BRK-A+BRK-B, Meta = FB/META).
• Initial deployment: $1,000,000, equal-weight across the top 3 at entry.
• Evaluation cadence: quarterly (measured at quarter-end close).
• Rotation mechanic: on trigger, sell the demoted name at quarter-end, pay 23.8% combined federal capital gains (20% LTCG + 3.8% NIIT), redeploy proceeds equal-weight into the current top 3.
• Terminal: 2026-Q1, with step-up at death applied to remaining unrealized gains (zero tax on the finish line).
• Entry cohorts: 2006-Q1, 2010-Q1, 2015-Q1, 2020-Q1.

The rules I tested run the full range from "never do anything" to "twitch at every wobble":

• R0 — Never sell. Ride forever, step-up at death.
• R1 — Strict. Sell the moment a held name drops to #4.
• R2 — Sell when out of top 5 for 1 quarter.
• R3 — Sell when out of top 5 for 4 consecutive quarters.
• R4 — Sell when out of top 8 for 1 quarter.
• R5 — Sell when out of top 8 for 4 consecutive quarters.
• R6 — Sell when out of top 10 for 4 consecutive quarters.
• R7 — Sell when out of top 20 for 4 consecutive quarters.
• SPY — Benchmark. Buy and hold the broad index.

The results

Here's the full matrix. Cells show terminal $ on a $1M initial deployment, after step-up at terminal. Numbers in the last column are average multiples across all four cohorts.

R1 wins three of four cohorts. Where it loses (2015), it loses by ~9%. Where it wins (2006, 2010), it wins by 27% and 36% respectively. The average advantage over never-sell is 135 basis points of multiple — about 17% more terminal wealth for the discipline of responding to rank changes.

Two observations jump out once you stare at the table:

Observation 1: every reasonable rotation rule beats SPY by at least 47%. Whether you rotate aggressively or conservatively, the SP3 concentration premium shows up. This is the mechanism thesis landing empirically — cap-weighted passive flow is doing the work, and your return is mostly a function of participating in the top of the cap distribution.

Observation 2: the rules cluster tightly from R0 through R7. The spread between the best rotation rule and the worst rotation rule is smaller than the spread between any rotation rule and SPY. Choosing among the rotation rules is a tuning problem. Choosing whether to run SP3 at all is the big decision — the rotation rule is the fine knob, not the coarse one.

Why "never sell" is weaker than I thought

The whole appeal of R0 was supposed to be the tax shelter. Every rotation realizes capital gains. The longer you hold, the more the estate step-up at death can wipe clean. In principle, minimizing realizations is pure alpha.

In practice, the math doesn't carry as much as I was crediting it. Here's the honest arithmetic.

Suppose a Sovereign position is at $300 today, cost basis $100. I'm about 54 now; step-up realistically kicks in maybe 30 years out. What does the "never sell" bet actually need?

Hold and wait for step-up:
Final value = $300 × (1 + r_hold)^30
where r_hold is the held name's forward return.

Sell and redeploy to current top 3:
Tax = 23.8% × $200 = $48
Net proceeds = $252
Final value = $252 × (1 + r_top3)^30

For "never sell" to beat "sell and redeploy" at 30 years, the held name needs to compound at within ~1% of the forward top-3 return. That's the entire breakeven. If the held name is merely a 1% annual laggard, you've given up more than the step-up saves you.

And the held name is a laggard — that's exactly what the falling rank tells you. The ranking change is the market pricing in an expected return differential. Ignoring it to chase a ~1% tax shelter is trading an 80-cent tax dollar for a dollar-and-a-half of forgone return.

Once I understood the math at 30 years, the strict rule stopped being aggressive and started looking obvious. R0 was trading higher expected return for a tax break that mostly wasn't there.

The 2015 exception, and what it teaches

The 2015-Q1 cohort is where R1 actually underperformed every other rule except SPY. It made $9.18M against R0's $10.09M. The reason is instructive and worth naming.

During 2015-2016, the top of the S&P went through a brief reshuffle. Apple, which had been #1 or #2 for years, slipped to #3 and briefly to #4-#5 as Alphabet, Microsoft, and (temporarily) Exxon's cap dynamics flipped. R1 — strict top-3, one-quarter trigger — fired on AAPL, sold it at a taxable gain, redeployed, and then had to buy AAPL back later in the cohort as it reasserted dominance.

That's the cost of fast triggers on noisy signals: you trade in and out at tax, with no fundamental thesis difference between your buy and your sell. In a period where the rank signal is genuinely informative, R1 extracts the full benefit. In a period where the signal is mostly cap-volatility noise around the same few companies, R1 pays tax for nothing.

The 2015 miss isn't a refutation of tight rules. It's a reminder that rules are rules — they fire on what they're designed to fire on, and sometimes the signal isn't there. You take the average performance, not the path-dependent best case.

Softer variants of the rule (R2 — top 5, one quarter) avoid the 2015 miss while still beating R0 on average. If you're willing to trade some peak upside for fewer whipsaw events, R2 is defensible. I'm choosing R1 because I think the 2015 pattern is less representative of the coming decade than the 2006 and 2010 patterns, where genuine structural rotations were happening at the top. But reasonable people pick R2 and sleep better.

The production rule

Concretely, here is what I now run:

Sovereign Exit Rule (v2):
On quarterly review, check each held position against the current top 3 by company market cap. If any held name has dropped to #4 or lower, sell the entire position at quarter-end close. Redeploy the proceeds equal-weight across the current top 3 on the next monthly DCA tranche. Pay the tax.

That's it. Two minutes of checking per quarter. One trade executed when a rotation fires — which, based on historical frequency, is roughly once every 2-3 years on average.

The full Sovereign discipline now reads:

1. Entry. Monthly DCA, equal weight across the current top 3 by company market cap.
2. New entrants. If a new name joins the top 3, future DCA tranches include it. Existing positions don't rebalance toward the newcomer.
3. Rotation. Existing positions rotate out per the exit rule above. Strict top-3, one-quarter trigger.
4. Exceptions. Tax-loss harvest opportunities, corporate actions, or life events can override the mechanical rule. Everything else is mechanical.

Three rules, and the whole framework runs. The hardest part of following it will be not checking prices daily.

What this tells us about the mechanism

The thing I want to take away from this backtest — and the thing I want the Sovereign series to keep returning to — is how load-bearing the market cap ranking signal actually is inside a heavily-passive market.

In a fundamental-first market, rank changes would be noise. Companies rise and fall in cap for a thousand idiosyncratic reasons, most of which are orthogonal to their long-term return potential. In that market, "never sell" would be the right discipline, because every sell is a taxable event in exchange for essentially random rank information.

We don't live in that market anymore. We live in a market where ~50% of flow is cap-weighted and non-price-seeking. In that market, rank changes aren't noise — they're the output of a feedback loop that can amplify itself if the rank persists. A company falling out of the top 3 isn't just losing ground on fundamentals; it's losing its share of the passive buying pressure that keeps the top of the distribution compounding at 15% when the bottom of the distribution compounds at 6%.

The ranking signal is the mechanism calling the outcome. When a name falls, it's not just reflecting a weaker company — it's getting cut off from a specific kind of structural flow that it had been receiving. The backtest says: listen to that. Don't be sentimental about which of the current top-3 names you own. Be sentimental about owning the top 3.

Postscript: day 2, Cook steps down

Two trading days after deploying $100K into AAPL as part of my Sovereign basket, Apple announced that Tim Cook will step down as CEO on September 1, handing the role to John Ternus — SVP of hardware engineering. After-hours trading shaved 0.77% off AAPL. My position went from $99,919 cost basis to roughly $99,200 market value. A rounding error.

It's an almost perfect stress test of the framework on day 2.

The news confirmed a lot of what I had already written into the Apple bear thesis. Ternus is a hardware engineer, not an AI executive. Srouji's parallel elevation to Chief Hardware Officer doubled down on the hardware focus. The new CEO is not a signal that Apple is pivoting to close its AI deficit — it's a signal that Apple is continuing to invest where it has always been strong (silicon and integrated hardware-software) and hoping that's enough. My existing view was that Apple is becoming the sophisticated rent-collector on top of other people's AI. Nothing in the Cook-to-Ternus handoff changes that read. If anything, it cements it.

My instinct, reading the news after hours Monday, was to sell AAPL immediately. Crystallize the thesis. Avoid the further repricing I suspect is coming.

The framework said no.

The rule doesn't fire on news. It fires on the company market cap ranking. As of Monday close, AAPL was still #3 by company market cap — behind NVIDIA and Alphabet, ahead of Microsoft. The Sovereign Rotation Rule doesn't care that Tim Cook is leaving. It cares whether Apple's cap has dropped behind Microsoft, Amazon, or someone new. Until that happens, I hold.

This is harder than it sounds. The instinct to act on vivid news is enormous — I could write you a perfectly coherent narrative for why AAPL at $271 two days ago is now overvalued at $269 after the CEO news, and that narrative would even be right on the merits. But "right on the merits" is exactly what the framework is designed to protect me from acting on. Backing up one section: I was right about AAPL in 2015 too, and R1 sold it, and then R1 had to buy it back six quarters later at a higher price, having paid tax for the round trip. The merit was right. The trade was wrong.

So: I hold. The cap ranking is the signal the mechanism is actually sending. If the AAPL bear thesis plays out, it will show up in the ranking — Apple will slide below Microsoft, or below the first IPO of OpenAI/SpaceX/Anthropic that enters top 3, and the rule will fire. If the bear thesis is wrong and AAPL continues to compound at market-like returns despite its AI limitations — which is plausible given the installed base, the services cash flow, and the iPhone upgrade cycle — then I get to participate in that compounding without having prematurely bet against it.

Two minutes a quarter. Respond when it fires. Ignore everything else. Day 2 of the discipline is the hardest day to practice it.

The Sovereign series is about running a simple discipline against a rigged game. The Top Three And The Tundra established the game is rigged. This post establishes the discipline is more active than I originally claimed — not constantly active, but responsive. Two minutes a quarter and a willingness to cut a former king when the flow math says the kingship has passed.

The next post in the series is about dollar debasement in real terms. M2 grew 3.3× while CPI grew 1.6× — the gap went somewhere, and that somewhere is disproportionately the top of the cap distribution. Which is another way of saying: the mechanism isn't just rigging returns higher. It's rigging them higher in nominal dollars that are worth less every year. SP3 isn't a return strategy. It's a debasement strategy with a return on top.

More on that in a couple weeks. In the meantime, the discipline: check once a quarter. Respond when it fires. Ignore everything else.

Code and data: cfo/sovereign/backtest/ in the repo. Findings doc at EXIT_RULES_FINDINGS.md. Raw results at results_exit_rules.csv. Re-run with python3 exit_rules.py.

Previous in the Sovereign series: The Top Three And The Tundra