The Style Box: Where a Portfolio Sits on Two Axes

A style box places a portfolio on two axes at once, usually the size of what it holds and how expensively those holdings are priced. The placement is a weighted average across every holding, so it describes a centre of gravity that no single holding need occupy. Move the boundaries or the weighting basis and the same portfolio lands in a different cell.

A portfolio landing in a different cell with nothing bought and nothing sold is the hardest of those three claims to believe, and arithmetic settles it rather than assertion. One thing has to be settled first, and then held to throughout. A style boxA grid that files a portfolio into one cell according to two measured characteristics of what it holds. The cell is a filing position, not a score. is a classification, not a rating: no cell on the grid is a better place to sit than any other, and reading the grid as a scoreboard is the single most common way it gets misused. A large filing cabinet with nine drawers shows which drawer a document went into. The drawer does not show that the document is good.

The running example throughout is the Anantara Multi-Asset Portfolio, an invented discretionary mandate of Rs 500 crore run by Faiz Ahmad Ansari for an invented charitable endowment whose investment committee is chaired by Rukmini Deshpande. Its shape is equity 60.0 per cent at Rs 300 crore, fixed income 30.0 per cent at Rs 150 crore and cash 10.0 per cent at Rs 50 crore. The equity sleeve holds 28 names, the largest of them Rs 23 crore, and the mandate caps any single holding at 5 per cent of the portfolio.

The individual sorts and where they came from are covered separately: the value and size sorts attributed to Eugene Fama and Kenneth French in 1993, the momentum sort to Narasimhan Jegadeesh and Sheridan Titman in 1993, and the quality and low volatility sorts carried without a single originator. A style box takes two of those measured characteristics, uses them as the two arms of a map, and leaves a committee to work out what the resulting cell can and cannot support.

What is a style box actually measuring?

Two continuous quantities, one on each arm. One arm carries the size of the businesses behind the holdings. The other carries how expensively those holdings are priced relative to what the businesses report. Both are measurements on a scale, the way height and weight are measurements on a scale, and neither one is naturally a category. A map exists before any label does: the grid lines and the words in the cells are something a person drew afterwards on top of two continuous measurements.

Think about a school noticeboard that plots every pupil by height and by age. Both quantities are measured, so nobody argues about where a pupil sits. The argument starts the moment somebody draws two lines across the chart and writes TALL FOR AGE in one corner. The measurement was arithmetic; the corner name was a decision. A style map works exactly the same way, and almost every disagreement about one is a disagreement about the second step disguised as a disagreement about the first.

The word cell only enters once somebody cuts each arm. Two cuts across a scale produce three sections. Two cuts on each of two arms therefore produce three times three, or nine cells. Three cuts on each arm would produce four times four, or sixteen. The famous nine cell picture is not a fact about portfolios at all: it is the count that follows when whoever drew the map chose two cuts per arm rather than one or three.

None of this makes a style map useless. It makes it a map, and a map is a genuinely useful object as long as everybody in the room knows they are reading one. The trouble starts when a picture of nine cells with words in them stops being read as a filing position and starts being read as a verdict.

How does a portfolio of 28 holdings get one position?

Both arms are measurements and every holding can be measured, so every holding has its own place on both arms. So a sleeve of 28 names is 28 points scattered across the map, not one point. The portfolio's single position is the weighted averageAn average in which each item counts in proportion to its share rather than counting once each. Doubling an item's share doubles how much it pulls the answer. of those 28 points, with each holding counting in proportion to how much of the sleeve it is.

The weighted average produces the single most useful fact about a placement: the portfolio sits at a centre of gravityThe balance point of a set of positions once each one is weighted. The arithmetic lands there, and nothing need actually sit there. that none of its holdings need occupy at all. A weighted average is a balance point, and a balance point is not a description of anything sitting on the beam. The average number of legs in a room full of people and one dog is not two, and nothing in the room has that many legs.

Here is the same idea drawn. Take a constructed sleeve of 28 names in which four sit low on an arm, eight a little higher, ten higher again and six near the top. The balance point falls at 50.71 on that arm, and it falls in a gap: the nearest holdings sit at 40.00 and 55.00, so nothing in the sleeve is anywhere near the point the map would print.

Everybody already knows this shape from somewhere else. A household with two earners on very different incomes has an average income that neither person earns. A street vendor whose cart sells one expensive item and forty cheap ones has an average sale price that matches nothing on the cart. A wedding caterer averaging the spice level across sixteen dishes describes a dish nobody is served. The average is real arithmetic; it is simply not a description of the things it was computed from.

Does the weighting basis change where the same portfolio lands?

Yes, and by more than most readers expect. There are two obvious ways to weight the 28 points, and they are not close to each other. Under equal weightTreating every holding as counting the same amount regardless of how much money sits in it. Each of 28 holdings counts one twenty eighth. every holding counts once, so each of the 28 names counts one twenty eighth of the sleeve, which is 3.57 per cent of the Rs 300 crore equity sleeve. In money that is Rs 10,71,42,857/- a name, and the whole rupee arithmetic leaves Rs 4/- over that cannot be split 28 ways.

Under value weightLetting each holding count in proportion to the money sitting in it, so a larger position pulls the answer harder than a smaller one. every holding counts in proportion to the money in it. The record gives three groups whose sizes it locks. The largest holding of Rs 23 crore is 7.67 per cent of the Rs 300 crore equity sleeve and 4.6 per cent of the Rs 500 crore portfolio. The top ten together are Rs 155 crore, so holdings two through ten come to Rs 132 crore between them, or 44.00 per cent of the sleeve. The remaining 18 holdings are Rs 145 crore, or 48.33 per cent of the sleeve. The three shares sum to 100.00 per cent exactly.

Notice what just happened to the base. Every share in that paragraph is struck against the sleeve except one, and that one says so in the same breath. The largest holding is 7.67 per cent of the equity sleeve and 4.6 per cent of the portfolio, both figures are correct, and moving between them without saying which is which is how a report quietly tells a committee the portfolio is less concentrated than it is.

Drawing that trap is easier than describing it. The Rs 23 crore block is exactly the same block of money in both readings below. Only the ruler underneath it changes, and the percentage changes with the ruler rather than with the holding.

Now put the two bases beside each other for the single largest holding. Equal weighting gives it 3.57 per cent of the pull. Value weighting gives it 7.67 per cent. The largest holding is therefore 2.15 times as influential on a value weighted placement as on an equal weighted one, and a report showing a marker on a map without saying which basis produced it has left out the choice that decides where the marker went.

The same switch does something tidier at the group level, and the arithmetic there is exact. Under equal weighting the ten largest names carry ten twenty eighths of the pull, or 35.71 per cent. Under value weighting they carry Rs 155 crore of Rs 300 crore, or 51.67 per cent. The ten largest therefore carry 1.45 times as much. The other eighteen names move the opposite way, from eighteen twenty eighths at 64.29 per cent down to Rs 145 crore at 48.33 per cent, or 0.75 times as much.

The reason the transfer happens at all sits in the average holding sizes, and the record locks both. The ten largest average Rs 15.50 crore each, or 5.17 per cent of the sleeve apiece. The other eighteen average Rs 8.06 crore, or 2.69 per cent apiece. The average name inside the top ten is 1.92 times the average name outside it, and equal weighting is precisely the choice to ignore that.

The two movements are the same movement seen twice. The total pull is always 100 per cent and nothing else is available for it to come from, so the ten largest gain 15.95 percentage points of pull and the other eighteen lose exactly 15.95 percentage points of pull. Switching the basis is a transfer, not a creation.

The mandate itself puts a ceiling on all of this, so one more figure belongs beside the others. No single holding may exceed 5 per cent of the Rs 500 crore portfolio, or Rs 25 crore, and Rs 25 crore is 8.33 per cent of the Rs 300 crore equity sleeve. So the most any one name in this sleeve could ever pull on a value weighted placement is 8.33 per cent, or 2.33 times the equal weighted 3.57 per cent. A single holding limit written for concentration turns out to be a limit on map influence as well. Nobody wrote the limit for that consequence and almost nobody notices it.

Where do the lines between the cells come from?

Somebody drew them. The choice is the whole answer, and it is not a criticism. A boundaryA chosen threshold on a continuous scale that decides which side of a line something falls. A boundary is a decision by whoever built the map, not a property of what is being measured. on a continuous arm has to be chosen, because the arm itself carries no natural break. Height does not come with a line above which a person is tall. Whoever builds the map picks the thresholds, states them, and everybody using that map inherits the choice.

Put the two placements from the previous section on a map and the consequence becomes visible. Take a constructed sleeve in which the other 27 holdings sit at 50.00 on the size arm and 55.00 on the price arm, and the largest sits at 90.00 and 35.00. The value weighted placement lands at 53.07 and 53.47. The equal weighted placement lands at 51.43 and 54.29. The two placements are 1.83 constructed units apart, and a line drawn anywhere between them puts the same portfolio in two different cells on the same day.

The mirror image is worse still, and it needs no second basis at all. Keep the placement fixed and move the line. A portfolio can change cells with nothing bought, nothing sold and no measurement changing, purely because a boundary moved or a different map was used, and the picture will show a change that never happened in the portfolio.

There is a third route to the same surprise that involves nobody at all. Under value weighting the weights are money, and money moves with prices. If the largest holding rises faster than the rest of the sleeve, its share of the sleeve rises with it, its pull on the placement rises, and the marker slides. Nobody decided anything, no line moved, and the picture still changed.

How far can one holding move the placement?

Not as far as people fear, and the bound is exact rather than a matter of judgement. A weighted average moves by a holding's weight multiplied by that holding's distance from where the rest of the set sits. Nothing else enters it. So the largest holding at 7.67 per cent of the sleeve can shift the sleeve's placement by at most 7.67 per cent of its own distance from the others, however extreme its own position happens to be.

Work it on the constructed sleeve from before. The other 27 sit at 50.00 on the size arm and the largest sits at 90.00, a distance of 40.00 units. Value weighted, the shift is 7.67 per cent of 40.00, or 3.07 units, landing the placement at 53.07. Equal weighted, the shift is 3.57 per cent of 40.00, or 1.43 units, landing it at 51.43. Even at the mandate ceiling of 8.33 per cent the shift would be 3.33 units. Influence is capped by weight. A placement sitting in the middle of a cell is therefore robust, and a placement sitting near a line is not.

The bound has a practical edge to it. A shift of 3.07 units is harmless or decisive depending entirely on where the marker started. Take a constructed cell running from 45.00 to 60.00 on the arm. A placement sitting at 50.00 is 10.00 units clear of the upper line, so the same 3.07 unit shift leaves it exactly where it was. A placement at 58.00 is 2.00 units clear, and the identical shift carries it across.

The everyday version of this bound is a class average. One pupil scoring far above everybody else moves the class average by their share of the class multiplied by how far above they are, and in a class of thirty that share is small. A small share is why a single extraordinary result barely moves a class average, and why a small class is far easier to swing. The map behaves identically, and it is why a marker deep inside a cell can be trusted to stay there and a marker hugging a line cannot.

What does a style box actually establish?

Something real, and it is worth stating fairly before the criticism arrives. A map compresses two continuous measurements across dozens of holdings into one position that a person can take in at a glance and, more importantly, can set beside another position measured the same way. Setting one placement beside another is the point. A cell name on its own is a label. Two placements measured on one map with one basis is a movement, and a movement is evidence about something.

The compression costs more than it feels, and the count is worth doing. Twenty eight holdings measured on two arms is 56 numbers. The weighted average turns those 56 into 2 coordinates. The cuts then turn the 2 coordinates into 1 cell name. Nothing about that is dishonest, and none of it is reversible: the coordinates cannot be recovered from the cell name, and the 56 cannot be recovered from the coordinates.

One reading earns its keep: the same portfolio at two dates, on the same map, with the same weighting basis. Only then is the difference between the two markers attributable to the portfolio rather than to the map. That comparison is also the entire setup for style drift, which takes exactly this comparison and asks what it means when the movement is large.

What does a style box hide?

The dispersionHow far apart a set of values sits from each other, measured for example by a standard deviation. A centre and a spread are two different statistics and neither implies the other. of the holdings, completely and by construction. A weighted average is a centre, and a centre carries no information about spread. The silence about spread is not a flaw in style maps specifically. Every average is silent in the same way. But it becomes a flaw the moment a placement is read as a description of what is inside.

Here is the demonstration. Take three constructed sleeves of 28 holdings each. The first is a cluster, all 28 sitting at 50.00. The second is a skew, 24 holdings at 45.00 and 4 at 80.00. The third is a barbellA set split between two extremes with nothing in the middle. The name comes from the shape of a weight bar, heavy at both ends., 14 holdings at 20.00 and 14 at 80.00. All three place at exactly 50.00 on the arm. Their standard deviations are 0.00, 12.25 and 30.00.

A box gives no information whatsoever about which of those three a portfolio is, so anybody reading a placement as a risk description has quietly swapped a centre for a spread. The barbell in that picture contains nothing in the middle at all, and yet a map places it precisely in the middle and files it in the middle cell.

The household version is a grocery trolley. Twenty two middling items and a trolley holding two luxuries beside twenty basics can come to the same average price per item. The average is honest arithmetic in both cases. The average simply does not distinguish one trolley from the other. If what matters is how the household would cope with one price shock, the average was never going to answer that.

Can the Anantara portfolio be placed on a map?

No, and the reason is a fact about the record rather than an oversight. Placing a portfolio on two arms requires a measurement per holding on each arm. For 28 names that is 28 size measurements and 28 price measurements, being 56 numbers, and a value weighted placement needs the 28 individual weights on top, taking the requirement to 84 numbers.

The record for the mandate carries one of those 84. The record gives the sleeve at Rs 300 crore, the count at 28, the largest holding at Rs 23 crore, the ten largest at Rs 155 crore as a group and the other eighteen at Rs 145 crore as a group. Only the largest holding is pinned individually. Not one size measurement and not one price measurement exists for any name.

Placing the Anantara portfolio would mean inventing 56 measurements. Nothing else in the record could ever be reconciled with them. A picture built that way looks precisely as convincing as one built on real measurements, which is the reason not to build it: it would teach a reader to trust a confident marker with nothing underneath it.

How does a committee actually use one of these in a room?

By asking three questions before reading the picture, and by refusing to read it until all three have answers. Rukmini Deshpande's investment committee receives a monitoring pack, and a style map inside it is worth about as much as the three answers beside it. The questions are the same three whether the pack covers Rs 500 crore or a household's savings.

First, which weighting basis produced the marker? Without it, the largest holding's pull is unknown between 3.57 per cent and 7.67 per cent on this sleeve, a 2.15 times difference. Second, where are the lines and who chose them? Without that, a cell change carries no information about the portfolio. Third, what is the spread behind the marker? Without it, a barbell and a cluster arrive looking identical. All three questions are about the map rather than about the portfolio, and that is exactly the point: a picture of a portfolio is mostly a picture of the choices made in drawing it.

An analyst comparing two mandates does the same work in a harsher form. The two packs almost never share a map. A lender assessing collateral concentration does it too, and a household comparing two savings arrangements described by their labels is doing the identical thing with smaller numbers. In every case the useful move is to ask what was measured, on what base, with which lines, and then to treat the label as the last step rather than the first.

What is left to cover?

Everything about movement. Holding a portfolio still and moving the map around it is the only way to see how much of a placement belongs to the mapmaker. Style drift holds the map still and lets the portfolio move, and asks what it means when a mandate stops sitting where it said it would sit. The two arguments only work in that order. A reader who has not yet seen a cell change with nothing traded will read every movement as a decision somebody took.

One route to a placement is covered separately. Instead of measuring every holding, a portfolio's effective style can be inferred from the pattern of its returns, an approach William Sharpe set out in 1992. The returns based route answers a related question with entirely different inputs and entirely different failure modes, and it is not what a holdings based style box does.