Independent blend vs a shared-capital book: does your portfolio actually compete for capital?
Most portfolio backtests quietly assume every holding has its own money. A shared-capital book drops that assumption and runs one account where your sleeves compete for a single cash balance — bar by bar. Here's the difference, why it changes the numbers, and when to use it.
The short answer
A "blend" mixes independent equity curves — each sleeve behaves as if it were fully funded on its own. A shared-capital book runs one cash balance that every sleeve draws from, so a strategy sitting in cash frees capital for the ones that are deployed. The blend is optimistic and great for ranking; the book is what actually happens when real dollars have to be in one place at a time.
The hidden assumption in a blend
Blend five holdings at 20% each and the engine grows five separate curves, then averages them. It's clean — but it silently assumes all five are fully invested at once. You can't deploy 100% of the same dollars in five names simultaneously, and if two of those five are systematic strategies that step to cash half the time, the blend still credits them their full 20% slot sitting idle. The result flatters a book that was never that invested.
What a shared-capital book does differently
One account, one cash balance, resolved every bar. The rule underneath is a single sentence: the account funds positions, not portfolios. What a holding is holding in the market is its own; every euro it is not holding sits in the book's cash, where any other holding can use it. Four things decide who gets the money:
Read left to right and the whole book is one sentence: you may claim your share, you draw only what you are holding, what nobody is holding is lent out — and no one may absorb the account. Each box below is that step in full.
1. Mandate — the allocation policy
Each sleeve's fair share of the book. By default it is simply the weights you set; a rule can take over instead — equal-weight, risk-parity or min-variance — recomputed on a schedule from trailing risk, with no look-ahead. Choosing one is not a setting you tick: it is a verification, run in Review, that measures every rule against a plain equal-weight book out-of-sample and net of cost. On most books the honest verdict is that nothing beats 1/N beyond the noise, and that is the answer.
2. Demand — what each sleeve actually holds
A sleeve draws only what its own position needs, bar by bar. A buy-and-hold sleeve always draws its whole share. A systematic strategy draws nothing while it is flat — and, just as importantly, draws only part of its share when it is partly invested: after a partial take-profit, or when its position sizing, Kelly or Adaptive Exposure deliberately deploys less than the full wallet. That freed capital is available on the same bar it is freed, not at the next rebalance.
3. Redistribution — no idle cash, and no override
Whatever nobody is holding is offered to the sleeves that can use it, in mandate proportion and water-filled up to each one's Max %. One rule keeps this honest: a sleeve is never funded out of its own unused cash. A strategy that chose to deploy half its wallet made a decision, and handing that half straight back would silently run it at double size — so the capital goes to the other sleeves, or it stays in cash. A book holding a single strategy therefore has nobody to borrow from, and reproduces that strategy exactly.
3a. The book allocates capital — the strategy still decides its exposure
That rule protects a sleeve from its own idle cash, and for a long time we assumed it protected the decision itself. It did not. With three sleeves each one may not draw its own unused half, but it may draw the other two's while they draw its — so a book where every strategy had cut its size in half ended up holding exactly what it held at full size. Measured on a real three-sleeve book, a modulator cutting every position to a quarter moved the account's deployed capital by half a percentage point.
So the account now asks two questions instead of one. How much would you hold at full conviction? decides your share of the book — that is the allocator's job, and lending is settled on it exactly as before. What fraction of that conviction are you applying right now? is yours, and what you hold back stays in cash: it is not offered to anyone, because it is capital your own rule declared must not be at risk. Halve your exposure and the account halves with you.
The restraint is measured, not assumed: your strategy is run a second time with its Adaptive Exposure switched off, and the difference between the two is what it was holding back. Two things follow, and both are deliberate. A partial take-profit is not restraint — you closed part of a trade and that money is genuinely free, so it is lent exactly as it always was. And volatility-scaled (ATR) or Kelly sizing has no full-size version to compare against — the sizing rule is the appetite — so the account may still redeploy what those hold back. A book that contains one says so in its warnings rather than leaving you to assume otherwise.
3b. The recall — what happens when a buy signal arrives and the account is full
Lending is the half everyone likes. Here is the other half, because it is the one that surprises people: your weight is a right, and anything above it is a loan. When a strategy signals an entry and the account is already fully invested, the book does not wait and does not skip the trade. It takes a slice from every holding that is sitting above its own line — never all of it from one — and funds the new position to at least its own weight, out of capital the others were only lent while this one sat in cash.
Concretely: every holding that is in the market is held to the same multiple of its own weight — a holding whose strategy has put only part of its share to work draws proportionally less, and lends the remainder. So opening a fourth position simply adds a name to that split and all three incumbents shrink together, each by its own distance to the new line. What each one gives up is part loan — proportional to its weight — and part drift, proportional to how much it has gained since the last re-allocation. So the holdings that have run furthest give back the most, without that ever becoming a rule to always sell the winner.
Two things follow, and they are what make the rule fair rather than merely decisive. Nobody is ever cut below their own weight: a holding sitting exactly on its share is already where the re-cut would put it, so it contributes nothing and is left alone. And nothing is borrowed from a broker — the shares are a partition of the account's own value, so the book cannot deploy more than it has. Measured on a real six-holding book: 125 entries onto an account already 100% deployed, 254 checks on the holdings that were in the market at the time, and not one case of a holding pushed below its own weight. Peak deployment over the whole run: 100.0000%.
This matters more than it sounds, because on a book of strategies this is what moves your capital — not the rebalancing schedule. The same book re-allocated 125 times against 46 scheduled quarterly boundaries. Review now reports it: how often the account recalled capital, and how much of the book changed hands doing it. The dial that governs how much recalling happens is Max %: set it equal to a holding's weight and nobody ever holds more than their share, so there is never anything to call back.
4. The two hard limits
Max % caps how much of the account any one holding can command — the answer to “if everything else is flat, does gold really take 100%?”. Once a strategy trades the book it defaults to twice that holding’s weight, and that default is doing real work: with no ceiling at all, one sleeve on a real six-holding book held more than half the account on 76% of the bars it was invested, and the weights its owner had set essentially never ran. Set it to 100 and you are back to no ceiling; set it equal to the weight and the holdings never share at all. It is brought back inside at the next re-allocation, not on every bar: policing a limit continuously would mean re-cutting the book almost every day, and paying the trading cost for it, to defend it by a fraction of a percent. A winner can therefore sit above its ceiling in between — and the run reports the peak share it reached, so the overshoot is stated rather than hidden. And the total funded can never exceed the account: the shares are a partition of what you have, so the book has no way to borrow. There is no leverage to switch on and no margin to model.
What it reveals that a blend can't
- Deployed capital over time — how full the account actually was, bar by bar, instead of an implied always-100%.
- Reused capital — how much of the account is running on capital borrowed from holdings that were not using theirs. Measured at the moment capital changes hands, so it reads exactly zero on a book where everything is always fully invested and there is nothing to lend.
- Capital recalled — how often the account funded a new position by calling back what it had lent, and how much of the book changed hands doing it. On a book of strategies this, not the rebalancing schedule, is what actually moves your capital.
- Cash drag and buffer — the return you gave up, or the risk you avoided, by not being fully invested.
- An honest CAGR — measured on a book that was sometimes only partly invested, not on five phantom fully-funded curves.
When does the book actually trade?
A shared account is only honest if you can say exactly when it moves capital — otherwise “rebalanced quarterly” can quietly mean “rebalanced every bar”, and the setting you chose is a label rather than a rule. There are three moments, and nothing else moves a cent:
- A sleeve's own position changes. It enters, exits, scales in, or takes a partial profit. The capital that frees up is offered to whoever can use it, on that bar.
- The roster changes. A holding's history begins (or ends) inside your window, so the account has one more — or one fewer — claim on it.
- The scheduled rebalance. On the first bar of each quarter (or year) the book re-cuts straight back to your target weights. It does not wait for a signal, and it does not ask a strategy's permission — but it never closes anyone's trade either: re-cutting resizes the stake behind a position, not the decision to hold it.
Being one of those bars is still not enough on its own. The account compares what each sleeve is funded with against what its mandate calls for, and only trades when the gap is worth trading on — so a book already sitting where it should be is left alone.
That includes the Max % ceiling. A cap is brought back inside at the next of those three moments, not on every bar — because enforcing it bar by bar would mean re-cutting the book almost every day, and paying the trading cost for it, to defend a limit by a fraction of a percent. So a winner can sit above its ceiling in between, and the run says so: each holding reports the peak share of the account it actually reached, and a peak above its own cap raises a warning. The cap still does the job it exists for — the moment the others step aside, nobody is handed the whole account.
One consequence surprises people, so it is worth stating plainly: because capital is re-cut whenever it changes hands, the quarterly rebalance often has little left to correct. That is not the setting failing — it is drift never getting the chance to build up. The Lab measures it rather than assuming it, and tells you when a cadence moved capital on almost none of its dates. Choosing a rule-based allocation instead of fixed weights forces a cadence too: a rule needs a beat on which to re-read risk, so “buy & hold” becomes quarterly, and the app says so instead of doing it silently.
Why there is no “blend” mode to choose
A blend is the optimistic ceiling: fast, and every sleeve gets a fair, isolated hearing — which is exactly why it cannot tell you what one real account would have done. A flat strategy that frees capital for a deployed one is an advantage the blend can't see, and its always-invested assumption is a distortion it can't avoid. So the Lab doesn't ship the blend as a mode you pick: a portfolio is one account, and when nothing ever steps to cash the book reproduces the blend to the byte anyway. You get the ranking and the reality check from one run.
The same anti-overfitting discipline
Choosing an allocation is still a search, and a search overfits. DeMiguel, Garlappi and Uppal (2009) showed that plain 1/N (equal-weight) beats most "optimized" allocations out-of-sample, because optimization fits the covariance matrix to one history. So the honest default is equal-weight or risk-parity, and a fancier policy has to clear a significance bar over 1/N — and hold up out-of-sample — before it's worth trusting. Treat a "perfect" allocation with the same suspicion you'd give an overfit strategy. More on overfitting.
The shared-capital checklist
- Do any of your sleeves step to cash? If so, a blend is over-crediting them.
- How full was the account over time — and what did the cash drag cost you?
- Did capital actually get reused, or was every sleeve fully invested all the time anyway?
- Is your allocation policy beating plain 1/N out-of-sample, or only in-sample?
- Is any single holding commanding more of the account than you want — and have you set its Max %?
How QUANTHEON Lab does this for you
In the Portfolio Lab, build your holdings and simulate as usual — this is simply how the Portfolio Lab simulates. There is no second engine to switch on and no “blend” mode to pick, because a portfolio is one account. You set the weights, an allocation policy, a rebalance cadence and a Max % per holding; the book reports its real equity, how deployed the account was over time, how much capital was reused, the longest stretch it sat idle, and a per-sleeve attribution of who earned the return. Everything is deterministic and strictly no-look-ahead.
FAQ
What is a shared-capital portfolio backtest?
One that runs all your holdings on a single cash balance where they compete for capital bar by bar, rather than blending each holding's independent equity curve as if it had its own money. The account funds positions, not portfolios: a strategy sitting in cash — or one that has just taken partial profits, or that sizes below its full wallet — frees that capital for the holdings that are invested.
How is it different from a normal portfolio blend?
A blend grows each holding's curve separately and averages them, silently assuming all are fully invested at once. A shared-capital book has one account, so total deployment, cash drag and reused capital become visible — and the CAGR reflects a book that was only sometimes fully invested.
Can I combine my own strategies in a shared-capital book?
Yes. Saved strategies and passive assets — stocks, ETFs, indices — can be sleeves in the same book, competing for one cash pool under an allocation policy, each held to its own Max %.
Is Max % enforced on every bar?
No. It is brought back inside at the next re-allocation — a sleeve moving, the roster changing, a scheduled rebalance. Enforcing it bar by bar would turn a limit into a mandate to trade every day, and the book pays real costs for its own re-allocations. The drift above the ceiling in between is reported: each holding shows the peak share of the account it actually reached.
What if my holdings are in different currencies?
Today they are blended as they are, each in its own currency, and so is the benchmark — so a book mixing EUR and USD, or judged against a benchmark in another currency, carries some exchange-rate movement inside its return and its alpha. Treat those two numbers as indicative on a multi-currency book. Converting everything into one unit is the correct answer and is being built; shipping it half-right would be worse than saying this plainly.
Why did my quarterly rebalance move almost nothing?
Because capital is re-cut every time it changes hands, drift rarely accumulates between quarters. That is the cadence having little left to correct, not failing — and the app measures it, telling you on how many of its dates the rebalance actually moved money.
Does optimizing the allocation overfit?
It can — allocation is a search. Plain 1/N is a famously hard out-of-sample benchmark (DeMiguel 2009), so a policy has to beat equal-weight with statistical significance and hold up out-of-sample before it's worth trusting.
Related: Portfolio Lab walkthrough · Adaptive position sizing · The Strategy Score · What is overfitting? · The technical whitepaper