A Model Portfolio Is Not a Separately Managed Account

A model portfolio is a specification: a stated list of holdings and weights that any number of accounts can be run against. A separately managed account is a container: one holder's securities, held in that holder's name, run against a strategy. A model and an account are not rival products. One is the design and the other is the thing built to it, and the gap between them is where most of the confusion lives.

Everything below rests on three things settled earlier. Portfolio management manages a whole rather than a collection of separate decisions. A policy weightThe weight a portfolio is designed to hold in an asset class. A policy weight is a stated intention, not a measurement of what is held today. and an actual weight are different quantities, and the actual weight drifts away from the policy weight between one rebalancing and the next. A separately held account carries its own dates and its own amounts. One account's return is that account's return and nobody else's. The working example throughout is the Anantara Multi-Asset Portfolio, an invented discretionary mandate of Rs 500 crore run for an invented charitable endowment. Rukmini Deshpande chairs the endowment's investment committee. Faiz Ahmad Ansari runs the mandate.

What exactly is a model portfolio?

A recipe makes the point. A recipe for a wedding biryani states quantities, proportions and an order of steps. The recipe feeds nobody. A recipe cannot be eaten and cannot be weighed, and it cannot say what the dinner cost. A recipe is a specificationA written statement of what something should be. A specification describes a design rather than recording anything that exists., and its whole usefulness is that it can be handed to any number of kitchens.

A model portfolioA stated list of holdings and weights, maintained by whoever designs it, against which any number of accounts can be run. A model is a document, not a position. is exactly that. The model is a stated list of what to hold and in what weights, maintained and updated by whoever designs it. When the design changes, the list changes, and every account being run against it is expected to move towards the new list. A model holds nothing at all. A model is a document, not a position, and no money is ever invested in a model. There is nothing there to hold and nothing to settle. A model therefore has no bank account, no custodian, no contract note and no tax status.

The Anantara model states equity 60.0 per cent, fixed income 30.0 per cent and cash 10.0 per cent, summing to 100.0 per cent. Those three weights are the entire document as far as asset classes go, with the mandate's stated limits sitting alongside them: equity may run between 50 and 70 per cent, and no single holding may exceed 5 per cent of the portfolio. Notice what is absent. The design was never funded, and there is no date on which it was. There is no quantity of anything. The weights are pure ratios and they add to one hundred whether the portfolio behind them is Rs 5 crore or Rs 5,000 crore.

What exactly is a separately managed account?

A separately managed accountOne holder's securities and cash, held in that holder's name rather than pooled with anyone else's, run against a stated strategy. is the dish that came out of the kitchen. The account is one holder's securities and cash, held in that holder's name rather than pooled with anybody else's, run against a strategy by a manager who has been given the authority to trade it. An account holds everything: it is a position with a value, a set of transactions, a settlement history and a return that belongs to one holder.

The Anantara portfolio is such an account. At its last rebalancing it held Rs 300 crore of equity, Rs 150 crore of fixed income and Rs 50 crore of cash, totalling Rs 500 crore, or Rs 5,00,00,00,000/- in full. Each amount is money that exists somewhere, held with a custodian. The equity sleeve is spread across 28 names. The largest single holding is 4.6 per cent of the portfolio, or Rs 23 crore, and the ten largest together are Rs 155 crore, or 31.0 per cent of the portfolio and 51.7 per cent of the equity sleeve. Every one of those figures is a measurement of something that exists. Not one of them could be asked of a model.

Are these two things to choose between?

No. Nobody picks a recipe instead of dinner. One is the design and the other is what is built to it, so an account is normally run against a model rather than instead of one. Once that is seen, every remaining question here answers itself, and until it is seen the questions do not even have well formed answers.

The reason the confusion is so common is that both words turn up in the same sentence in marketing material, laid out as though a reader were choosing. The two words are not parallel. A model can exist with no accounts behind it at all, in which case it is a design nobody has built. An account can exist with no model behind it, in which case somebody is making each decision on its own without a stated specification to check against. The normal arrangement is one model above many accounts, and the interesting questions all live in the space between the two.

How do the two compare, line by line?

Set against the same seven criteria, a model and an account differ on every one. A document and a position would be expected to differ like that, rather than behave like two competing products. One is a design and the other is a holding. The two differ on every line for that reason, not because one of them is the better arrangement. Read either column on its own and it still makes sense, which is the test of an honest comparison.

Why do two accounts on one model give different results?

The same biryani recipe handed to four kitchens on the same street produces four different dinners. One bought tomatoes on Tuesday when they were cheap and one bought on Saturday when they were not. One had a guest arrive halfway through and stretched the pot. One household has a member who cannot eat groundnut, and the dish came out without it. And one has a smaller pan, so the quantities were rounded to what would fit. None of those four kitchens did anything wrong, and none of the four dinners is the recipe.

The same four causes produce dispersionThe spread of outcomes across several accounts run against the same specification over the same period. Dispersion measures how far apart the containers ended up, not how good the design was. across accounts run against one model, and it is worth naming them separately rather than reaching for one vague word.

First, timing. Accounts funded on different dates buy the same holdings at different prices, and that difference never washes out afterwards. Second, flows. Money arriving in or leaving one account has to be invested or raised in that account alone, at that account's prices on that account's dates, and the other accounts are untouched. Third, restrictions. A holder who cannot hold something the model holds, for any reason at all, carries a permanent gap where that holding should be, and the substitute behaves differently. Fourth, size and rounding. A small account cannot hold a fractional position, so a 4.6 per cent target becomes whatever whole quantity comes nearest, and rounding errors in twenty eight names do not politely cancel each other out.

All four causes are structural, none of them is a mistake, and a manager reporting zero dispersion across many accounts either has one account or has a reporting problem. That last consequence is the useful one. Zero spread across two hundred containers built by four independent mechanisms is not a compliment paid to the implementation; it is a signal that the number being reported describes the model rather than the accounts.

How far can an account drift before a limit binds?

An account matches its model at the instant it is rebalanced and starts to stop matching it immediately afterwards. The reason is arithmetic rather than conduct. A weight is a ratio: a sleeve's value divided by the total. Let any sleeve's value move and both the numerator of that sleeve and the denominator of every sleeve move with it, so all three weights change even though nobody placed a single order. Movement of that kind is driftThe movement of an account's actual weights away from its stated weights caused by price changes alone, with no buying or selling involved., and drift is the ordinary condition of every account between one rebalancingThe act of trading an account back towards its stated weights. Between two rebalancings the weights are left to move with prices. and the next.

Take the Anantara account at its last rebalancing: equity Rs 300 crore, fixed income Rs 150 crore, cash Rs 50 crore, total Rs 500 crore, reading 60.0, 30.0 and 10.0 per cent. The account matches the model exactly. Now let the equity sleeve rise to Rs 330 crore while the other two together stay at Rs 200 crore. The rise is assumed, not recorded. The total becomes Rs 530 crore. Equity is 330 divided by 530, or 62.264 per cent, 62.3 per cent rounded. Fixed income is 150 divided by 530, or 28.3 per cent. Cash is 50 divided by 530, or 9.4 per cent. The three still sum to 100.0 per cent, nothing was bought, nothing was sold, and the actual weightThe weight an account is holding right now, computed from today's values. An actual weight moves with prices and rarely equals the stated weight. in equity is now 2.3 points above the model.

Now for the finding, the part that surprises people. The mandate permits equity between 50 and 70 per cent. A twenty point band sounds tight. Ask instead how far the equity sleeve's value would have to move before either edge binds, holding the other two sleeves at Rs 200 crore. The ceiling binds when E divided by E plus 200 equals 0.70. Multiply out: E equals 0.70E plus 140, so 0.30E equals 140, so E equals Rs 466.67 crore. The rise from Rs 300 crore to Rs 466.67 crore is 166.67 divided by 300, or 55.6 per cent. The floor binds when E divided by E plus 200 equals 0.50, giving 0.50E equals 100 and E equals Rs 200 crore, a fall of 100 divided by 300, or 33.3 per cent.

A range of 50 to 70 per cent sounds narrow. The range still lets the equity sleeve lose a third of its value or gain more than half of it before anything binds, if nothing else is done. That is worth sitting with. A committee reading the mandate sees two numbers twenty points apart and pictures a close leash. The arithmetic says the leash is eighty nine percentage points long once it is measured in the sleeve's value rather than in the weight. The limit is stated and does what it says; it simply does not say what people assume it says. Drift is also not a breach. Nothing has been done wrong, and what a committee should do about a drifted account is covered separately.

One more thing falls out of the same arithmetic, and it is the part a fixed income analyst notices first. Watch what happens to the other two sleeves while equity travels between its two edges. At the floor, with equity at Rs 200 crore, the total is Rs 400 crore, so an untouched Rs 150 crore of fixed income reads 37.5 per cent and an untouched Rs 50 crore of cash reads 12.5 per cent. At the ceiling, equity is Rs 466.67 crore and the total is Rs 666.67 crore, and that same Rs 150 crore now reads 22.5 per cent and that same Rs 50 crore reads 7.5 per cent. Not one bond and not one rupee of cash was traded, and the fixed income weight still moved fifteen points. That is what it means to say a weight is a ratio: every sleeve's weight is at the mercy of every other sleeve's value, which is why a stated weight only ever describes a moment.

Which failures come from treating one as the other?

There are two, and they are opposites. Opposite errors are hard to spot together. The first is reading a model's stated result as though the accounts earned it. A model holds nothing, and so there is no funding date inside it, no money arriving, no money leaving, no holder restrictions and no trading cost. Every one of those five things sits inside a real account and every one of them moves the number. A stated result for a specification is a statement about a design, and a design cannot have a bad Tuesday.

The second failure is made by the same person a month later, when an account is found not to match the model and every gap is treated as a failure of implementation. Timing, flows, restrictions and rounding produce gaps that nobody did anything wrong to create. A manager pushed to close all of them will trade more, and the account pays for that trading. The model never pays for any of it. Both failures come from one idea, that the model and the account are the same object, and they push in exactly opposite directions.

How does a practitioner actually use this distinction?

The base of a weight and the base of a comparison are the same kind of trap. Faiz Ahmad Ansari, running the Anantara mandate, uses the distinction every time he answers a question about concentration. The largest equity holding is 4.6 per cent of the portfolio, or Rs 23 crore, and the stated 5 per cent limit is written against the portfolio, so the holding sits inside it. Measured against the Rs 300 crore equity sleeve instead, the very same holding is 7.7 per cent. Neither figure is wrong and they answer different questions. Moving between them without saying which base is in use misleads the listener about how concentrated the account is. The base is named every single time, exactly as a result is named as belonging to a model or to an account.

The same discipline turns the stated 5 per cent limit into something that can be reasoned with rather than nodded at. Five per cent of the Rs 500 crore portfolio is Rs 25 crore, and the largest holding is Rs 23 crore, so it sits inside. The useful question is what the limit permits rather than what it forbids. The equity sleeve is Rs 300 crore, and at Rs 25 crore a name that sleeve could be filled by as few as twelve holdings without the limit being touched once. A cap written at 5 per cent of the portfolio rules out one very large position and does nothing else. By itself the cap is not a statement about how spread out the sleeve is. This sleeve happens to be held across 28 names, which is a decision somebody took rather than something the limit produced, and telling those two apart is the whole skill.

An analyst reviewing a manager uses it as a two question filter before looking at any number. Is this figure a property of the design or a property of a container? If it is a container, whose container, funded when, with what flowing in and out? A household reading a printed allocation is doing the same thing without the vocabulary: the sheet showing sixty, thirty and ten is a plan for money, and the account statement that arrives at the end of the quarter is the money. The practitioner's habit is simply to refuse to compare a design with a container until the design has been given dates and amounts.

One more use, and it is the quietest. Once it is understood that the account holds everything and the model holds nothing, tracking differenceThe gap between what an account returned and what the specification it is run against would have returned over the same period. stops being a scorecard and becomes a diagnostic. The question is no longer whether the gap is large. The question is which of the four causes produced it, and whether the answer is something anybody could have changed.

The Anantara Multi-Asset Portfolio, the endowment that holds it, Rukmini Deshpande and Faiz Ahmad Ansari are invented.
Educational material. Not advice on any investment, tax, budget or market position.