How to Monitor Portfolio Risk Through the Year

Monitoring risk runs as a pass repeated on a schedule, not as a reply to bad news. Write down every limit with its base, refresh the exposures the portfolio currently carries, refresh the spread figures on one window, check that those four tie to each other, measure the worst fall on its own, work out the real room left, label any crossing by its cause, and record what could not be measured.

A portfolio does not sit still between reporting dates. Prices move, so the weights move with them, and a set of limits that were satisfied on the first of the month can be uncomfortable by the twentieth without anybody having placed a single order. So somebody has to look, on a schedule, in the same order every time.

Think about how a landlord checks a building. Not by waiting for a tenant to phone about a leak. By then the ceiling is already down. The check runs on a round: the same rooms in the same order, so March can be read against February. The fixed order and the written output are what make the round a process rather than a habit, and a risk pass is the same kind of object.

The order runs to eight steps, worked end to end on the Anantara Multi-Asset Portfolio, an invented Rs 500 crore discretionary mandate run for a charitable endowment, whose investment committee is chaired by Rukmini Deshpande and whose day to day management sits with Faiz Ahmad Ansari.

The inside of a step is covered separately. Where a step needs a mechanism, that mechanism is named and the sequence moves on.

Why is monitoring a repeating pass rather than a reaction?

Because the alternative is to look only when something feels wrong, and feelings arrive late. A monitoring cycle is a schedule plus an order plus an output. Take any one of the three away and what is left is somebody looking at screens.

The schedule matters because risk in a portfolio is not an event. Nothing announces the day the largest holding gets uncomfortably large. Risk creeps in through a run of good weeks, and it is largest precisely when everyone is pleased. A pass that fires only on bad news will never once look at that.

The order matters for a different reason. A figure computed before its base was fixed gets compared against a different thing, so two passes run in different orders can reach different answers from the same portfolio on the same day. An order that never changes is what lets this month's pass be read against last month's without anybody reconstructing how either was done.

And the output matters because a pass that produces only a feeling of having checked cannot be handed to anybody. A monitoring pass produces eight written outputs, and the eighth is what most processes leave out.

What gets written down before any number is pulled?

Step one. Before a single figure is refreshed, every constraintA written limit on what the portfolio may hold or do, set out in the mandate rather than chosen by whoever is running the check. is written out again, each one beside the baseThe denominator a limit is measured against, chosen when the mandate was written. it is measured against.

Restating the limits looks like paperwork and it is the cheapest minute in the pass. A limit is a ratio, and somebody chose its denominator when the mandate was written, possibly years ago. A pass that opens by pulling numbers has already accepted whatever denominator the system happens to use. Fixing the base first is what stops the rest of the pass from answering a question nobody asked.

Try it on something domestic. A household agrees that no more than a fifth of the money goes on rent. A fifth of what: the salary before deductions, what actually reaches the account, or what is left after the loan payment? One sentence, three bases, three different rent figures.

For the Anantara Multi-Asset Portfolio the four written lines and their bases come out as below. Note that two of them share a base, one is asked separately of every holding, and one is measured against a sleeve rather than against the whole.

Getting the base wrong on purpose is the quickest way to see what it buys. Hold the Rs 23 crore holding still, change nothing about the portfolio, and read it against the two totals a careless pass might reach for.

Which numbers are exposures, and which are not?

Step two refreshes the exposureWhat the portfolio is currently open to: the weights it presently carries, the names it presently holds, and the characteristics of those holdings, as opposed to what any of it has already earned. list, and the discipline is in what the list leaves out.

The list holds the weight of each bucket, the largest single holding, how much sits in the ten largest together, the credit standing of the bond sleeve, and whether anything the mandate excludes has appeared. Five items, all describing the portfolio as it stands right now.

Returns are not on it: not last quarter's, not the year's excess over the composite benchmark, not the worst fall already taken. The return figures are measured elsewhere in the pass, and they are left off step two on purpose. A return describes something that has finished happening. An exposure describes what the portfolio currently carries, and only the second is something a decision can act on.

The everyday version is a shopkeeper who supplies four office canteens, three of them in one building. Last month's takings are last month's takings. The exposure is that three quarters of the trade rides on one lease being renewed, and that is the half anybody can act on this week.

Two of those five items are worth drawing out. A later step compares them against a limit.

What has to be fixed before the spread figures are refreshed?

Step three refreshes four dispersion figureA number describing how much a series moved about, such as a volatility, or how much two series moved together, such as a beta. All of them are computed from a run of returns rather than from a single one. outputs at once: the portfolio's volatility, the composite benchmark's volatility, the beta of one against the other, and the tracking error between them.

The window and the reading frequency are fixed before any of the four, decided once, written at the top of the step, and applied to every one of them. The stated twelve months, on the same series, at the same frequency, for all four.

The reason this is a step rather than a footnote is that a mixture of windows is completely invisible in the output. Four numbers printed in one table look like four numbers describing one portfolio at one moment. Nothing in the printing says the first came from three years of history and the third from twelve months. The only defence is to fix the window before the figures exist.

For the Anantara Multi-Asset Portfolio the stated year gives portfolio volatility of 11.8 per cent, composite benchmark volatility of 10.4 per cent, a beta of 1.08 against that benchmark, and a tracking error of 3.7 per cent. All four from the same series over the same twelve months. Those four figures are the step three output, and they are not yet trusted.

How do the four figures check each other?

Step four is the check, and it exists because those four figures are not four independent readings. Three of them determine the fourth. The set carries three degrees of freedomThe number of values in a set that could be chosen freely. Where a relationship ties the set together, fixing all but one of them leaves the last with no choice at all., not four, so the last one can be computed and compared with what was measured.

Run it on the stated year. The relationship, drawn in full below, leaves a remainder of 13.7744, whose square root is 3.71 per cent. Why it holds at all is covered under tracking error as active risk.

The measured tracking error was 3.7 per cent. The computed one is 3.71 per cent. The two figures land on each other to within the rounding, so all four came from one sample over one window and the pass moves on. Had the computed figure come out at, say, 3.2 per cent against a measured 3.7, no single set of returns could have produced both, and the correct response is to stop the pass rather than to report the numbers with a note.

Stopping is the part people skip. A check whose only permitted outcome is a footnote is not a check, and this one has teeth because a mismatch ends the pass and sends somebody back to find which figure came from a different window.

Why does the worst fall get its own line?

Step five measures the path, and it is separated from step three because the two are different kinds of object. Steps three and four describe how much a series of period returns moved about. Step five describes one distance, from the highest point the portfolio reached to the lowest point after it, inside a stated window.

Inside the stated twelve months that distance was 9.7 per cent for the Anantara Multi-Asset Portfolio against 8.1 per cent for the composite benchmark. Both carry that window every time they are printed. A fall quoted without its window is not a measurement of anything.

Now the specific trap in this record. The step three figures also say how much of the portfolio's movement was shared with the composite benchmark and how much was not, and on the stated year the unshared part is 9.4 per cent of variance. The step is to record it, not to explain it.

So the pack now holds a 9.4 and a 9.7. The 9.4 per cent is a share of variance across the whole year and the 9.7 per cent is a single distance from a peak to a trough, and no arithmetic anywhere connects them. They are close in size, they will land on the same sheet of the pack, and somebody will eventually write a sentence treating one as a check on the other. Keeping them on separate lines with their units written out is a step five duty.

Step five also produces a short list of things it could not produce, and that list is worth seeing beside the figures it did.

How much room is actually left under a limit?

Step six compares each refreshed figure with its limit and works out the headroomHow far a figure can travel before it reaches its limit. Where the limit is a share of a total that itself moves, the room left is not simply the gap between today's reading and the limit.. Step six is where a spreadsheet quietly produces the wrong answer.

Take the largest holding. The holding is 4.6 per cent of the Anantara Multi-Asset Portfolio, so on Rs 500 crore it is Rs 23 crore, being Rs 23,00,00,000/-. The cap is 5 per cent of the portfolio, or Rs 25,00,00,000/-. Room of Rs 2 crore, so the obvious answer to how far the price can rise is 2 divided by 23, or 8.70 per cent.

The 8.70 per cent answer is wrong, and wrong in a specific direction. The holding is part of the portfolio, so as the holding rises the portfolio rises with it. At an 8.70 per cent gain the holding is worth Rs 25.00 crore but the portfolio is now Rs 502.00 crore, so the holding is 4.98 per cent and the cap has not been reached.

Solved properly, with the portfolio growing underneath, the rise is 9.15 per cent. At that point the holding is Rs 25.11 crore of a Rs 502.11 crore portfolio, or exactly 5.00 per cent. The naive figure understates the room by nearly half a percentage point of price, so a pass built on it raises an alarm before there is anything to be alarmed about.

Which limit gives way first?

The second half of step six is the question the pass has to answer and the document cannot. Several limits apply at once, and only one of them is the binding constraintThe limit that is reached first as circumstances move, so that it, rather than any of the others, is what actually restricts the portfolio.. Finding out which takes arithmetic.

Try the case where the whole equity sleeve rises together, prices moving and nobody trading. Two limits are in play: the 50 to 70 per cent equity band and the 5 per cent cap on any one holding, both measured against the whole portfolio.

For the cap: the holding and the portfolio both rise with the sleeve, and the cap is reached on a 25.0 per cent sleeve gain. Check it: the sleeve is Rs 375 crore, the portfolio Rs 575 crore and the holding Rs 28.75 crore, or exactly 5.00 per cent.

For the band, the same movement takes equity to its 70 per cent ceiling only at a 55.6 per cent gain, and at the moment the cap is reached equity is still 65.2 per cent. The cap gives way at a 25.0 per cent sleeve rise and the band survives to 55.6, so the close limit is the one listed second in the mandate and the pass has to say so.

One thing worth separating. The two crossing figures above are for a rise in prices. If instead somebody moves money into equity by selling bonds, the portfolio total does not change and the band is reached by shifting Rs 50 crore. Seventy per cent of Rs 500 crore is Rs 350 crore. Same limit, two very different headroom figures, depending on whether a trade or the market brings the portfolio there. The distinction between a trade and the market is step seven.

There is a third crossing in the same mandate and it sits on the other side of today. Run the sleeve downwards and the 50 per cent floor is reached on a fall of 33.3 per cent, where the sleeve is Rs 200 crore of a Rs 400 crore portfolio. All three belong on one scale.

What has to travel with the word breach?

Step seven. If any figure has crossed its limit, the pass records what caused the crossing, and it does that before anybody discusses what to do about it.

There are two causes and only two. Either somebody placed an order that put the portfolio outside, or nobody did anything and prices moved the weights across the line while the pass was between runs. The first is an active breachA limit crossed because of a decision taken inside the portfolio, such as an order that should not have been accepted in the size it was. and the second is a passive breachA limit crossed without anybody acting, because price movement changed the weights on its own between one check and the next..

The two need different repairs, so a report carrying only the word breach has destroyed the one piece of information that says what to fix. If a trade caused it, something that was supposed to test the order before it was placed did not. If prices caused it, no control failed at all, and the questions are how often the pass runs and how much room the internal trigger leaves before the mandate's own line.

The household version is a car that fails its inspection. Failed because somebody drove it into a wall is a different repair from failed because the tyres wore down over eighteen months, and writing only failed on the form loses exactly that.

What goes in the last field of the pass?

Step eight records what could not be measured, and why. Step eight is the field most processes leave off the form, and it decides whether a monitoring process improves or merely repeats.

Everything above it describes what was seen. Step eight describes the edges of what was seen: the item that had no data, the figure the record could not produce, the test that was not run at all. Written down, with a reason each time.

At this date the field carries five entries. Listing status and credit standing name by name are not in the record, so the third and fourth mandate lines were never tested. The dates of the high and low points are not recorded, so nothing about how long the fall took or how long recovery took can be produced. The return basis and reading frequency behind the dispersion figures are not stated. The average weights across the month are absent, so only this date's weights are known. And there is no holding by holding comparison against the composite benchmark.

Notice what the first of those does to the earlier steps. Two of the four written limits could not be tested, so a report saying all limits satisfied would be a false statement. Every real check has edges, and the edges are the only thing that tells the next pass where to look, so a pass whose eighth field is empty has not looked hard enough.

What does one full pass produce, end to end?

Eight steps, eight written outputs. Here is the whole pass on the Anantara Multi-Asset Portfolio at one stated date, inside the stated twelve month period.

Look at the seventh row. There is no crossing, and the field is still filled in, with the words not applicable rather than with nothing. A blank field is read by the next person as a clean result, so an untested item is marked rather than omitted. That habit costs one word and separates a report saying nothing happened from one saying nobody looked.

What must never be a step in this procedure?

Three moves look like sensible responses to an uncomfortable reading, and none of them belongs anywhere in the pass.

Never change the measure because the reading was uncomfortable. A volatility that came back higher than expected is a result, not a defect in the measure. Swap it for one that reads lower and the only reading anybody had is gone.

Never move the window so that a crossing disappears. A fall of 9.7 per cent over the stated twelve months becomes something else over ten, and a pass whose window is chosen after the figures are seen cannot be compared with the pass before it.

And never report a number the record cannot support. Two of the four limits went untested this pass, so the sentence all limits are satisfied is not available, and no careful wording makes it available.