How to Evaluate the Manager Running a Mandate

Evaluating whoever runs a mandate starts with what the mandate asked for, not with the return. Fix the window, the basis and the benchmark, compute the excess, split it and say which split was run, set it against the risk it was taken with, ask what that sample size can carry, read the constraint record, and write down what the evidence cannot settle.

A version of this already happens in most households, and it happens badly. A household hires a tuition teacher in June. In March the child scores well, so the teacher is good. Another child, another year, a weaker score, and that teacher is not. Nobody in the room ever asked what was agreed in June, what else changed in the year, or how many terms it would take before one result meant anything at all. The conclusion arrived first and the evidence was fitted around it.

An investment committee does exactly the same thing with a much larger number attached. The order that stops it can be written down and run. Almost every failure in evaluation is a step taken out of turn rather than a calculation done wrong, so the eight steps run in a fixed order, and the order is the method.

Every step below is worked on the Anantara Multi-Asset Portfolio, an invented discretionary mandate of Rs 500 crore run for a charitable endowment, whose investment committee is chaired by Rukmini Deshpande and whose mandate is run by Faiz Ahmad Ansari. Every figure belongs to one stated twelve month period.

What has to be settled before any return is looked at?

Step one is to read the mandate. Not skim it, not remember it, read it: the mandate objectiveThe written statement of what the money is being asked to do, which the mandate document sets out before any holding is chosen., the constraints, the benchmark named in the document, and the period that was agreed.

Reading the mandate first looks like a formality and it is the opposite of one. The household version shows why. A household that reads the March mark sheet first spends the rest of the conversation deciding whether the June agreement was reasonable, and the agreement has to argue for its own relevance against a number already known. Read in the other order, the agreement comes first and the mark sheet arrives as evidence about something already defined. A manager can only be evaluated against what was actually asked of them, so reading the return first sets an expectation that the mandate then has to argue against.

For the Anantara Multi-Asset Portfolio the reading produces a short list. The mandate asked for a multi-asset portfolio held inside an equity band of 50 to 70 per cent, with no single holding above 5 per cent of the portfolio, no unlisted holdings, and a minimum credit standing on the fixed income sleeve stated as a policy rather than as a rating symbol. The mandate named a composite benchmark. The evaluation is against that list and against nothing else. Anything the committee wishes had been asked for in June is a lesson for the next mandate, not evidence about this one.

Why are the window, the basis and the benchmark one decision?

Step two fixes three things at once: the period the evaluation covers, the return basisWhether a return figure is stated before or after fees, and whether it is weighted by time or by the cash moving in and out., and the benchmark the return is compared with. Three separate conversations produce three separate opportunities to choose whichever version flatters. One decision, taken once, before any comparison, removes all three.

The basis question is the one people skip, and it decides what a return figure even describes. A return before fees and a return after fees are different quantities. A return weighted by time and a return weighted by the cash moving in and out are different quantities. Two people quoting the same portfolio on different bases are not disagreeing; they are describing different things and using one word for both.

The benchmark half of the step has its own duty. Where the benchmark does not match the shape the mandate requires, the difference is structural, it will show up in every comparison, and it belongs in the record now as a known offsetA structural difference between the portfolio and its benchmark, written down at the start so it is subtracted rather than argued about later. rather than in an argument later. A structural difference discovered at the end gets argued about. The same difference recorded at the start gets subtracted before anybody is asked to explain anything.

On the Anantara Multi-Asset Portfolio the window is the one stated twelve month period. The return basis the mandate agreed to be judged on is not recorded anywhere in the material available. The basis is marked unknown and carried as unknown through every step that follows. Carrying an unknown that far is uncomfortable and honest. The benchmark is a composite of 60 per cent a broad equity index and 40 per cent a broad bond index, both left unnamed here. The mandate requires 10 per cent cash and the composite holds none, so ten points of the composite sit in bonds where the mandate holds cash. The cash requirement makes a ten point difference in each of two buckets, and the difference is written down before anything is compared.

One more thing belongs to step two, and it carries the largest single consequence in this whole evaluation. The 14.2 per cent is what the portfolio itself returned in the stated twelve months. The record states separately, in the commercial terms this mandate carries, what delivering that portfolio cost the endowment in the same year: a management charge of 1.25 per cent of the Rs 500 crore, which is Rs 6,25,00,000/-, and a performance charge of 15 per cent of the 4.2 points earned over the 10 per cent hurdle the mandate wrote, which is 15 per cent of Rs 21 crore, or Rs 3,15,00,000/-. Together that is Rs 9,40,00,000/-, and on Rs 500 crore it is 1.88 per cent of assets. So the same twelve months reads two ways. Gross of that cost the portfolio returned 14.2 against 12.6, an excess of plus 1.6 points. Net of it the endowment held 12.32 against the same 12.6, a shortfall of 0.28 points. An excess return quoted with neither the word gross nor the word net beside it is ambiguous between beating the benchmark and missing it. In an evaluation that ambiguity is the entire question. The 1.6 points travel as a gross figure and carry the label wherever they appear. Which of the two bases the mandate agreed to be judged on is the part the record does not state, and step two writes that down as unknown rather than quietly choosing one.

How is the excess split, and how is the split declared?

Step three computes the excess returnThe portfolio return less the benchmark return over the same window, stated in percentage points rather than as a ratio. and then takes it apart. On the Anantara Multi-Asset Portfolio the portfolio returned 14.2 per cent in the stated twelve months against 12.6 per cent for the composite, so the gross excess is plus 1.6 percentage points. The 1.6 point excess is where most reviews stop, and where step three starts.

More than one honest way to take 1.6 points apart exists. The splits are not rival estimates of one quantity. Each split asks a different question on a different basis, and each reconciles to 1.6 on its own. So the step carries an instruction attached to it: name the question before running the arithmetic. Naming the question is what a declared splitA statement, written before the arithmetic, of which question the decomposition is answering, so its outputs are not read as answers to a different one. means, and it costs one sentence.

Run the first one, the exposure question: how much of the excess is simply carrying more market than the benchmark carried? The composite beat the 6.5 per cent risk-free rate by 6.1 points in the stated year. The portfolio ran at a beta of 1.08 against that composite. The extra 0.08 of exposure delivers 0.08 times 6.1, or 0.488 points. The return expected at a beta of 1.08 is 6.5 plus 1.08 times 6.1. Adding 6.5 and 6.588 gives 13.088 per cent. The residual is 14.2 less 13.088, or 1.112 points. The two parts, 0.488 and 1.112, add to 1.600.

Now stop, close that split, and declare the next one. The second question is where the excess came from across the mandate: an allocation effect of plus 0.35 points and a selection effect of plus 1.25 points. Those two also add to 1.60. The 1.11 point residual is not a check on the 1.25 point selection effect, and neither pair is the true split. Both splits are complete, both reconcile to the same 1.6 points, and no term from one may sit beside a term from the other. Cross the terms over and the sums say so at once: 0.35 plus 1.112 is 1.462, and 0.488 plus 1.25 is 1.738, and neither total is the 1.6 gross points or any other quantity anybody measured.

What does the risk beside the return actually show?

Step four puts the excess next to the risk it was taken with, and keeps both in view. A return quoted alone is half a sentence. The half that is missing is what had to be accepted to produce it, and a committee that never sees the second half will reward whoever accepted the most of it.

Four figures go beside the 1.6 points, all from the same stated twelve months. Portfolio volatility of 11.8 per cent against the composite's 10.4 per cent. A beta of 1.08 against that composite. Tracking error of 3.7 per cent. The worst peak to trough fall inside the same window, 9.7 per cent for the portfolio against 8.1 per cent for the composite. A different window produces a different number from the same price history, so that drawdown figure means nothing without the window attached.

One warning belongs here and it is arithmetic rather than judgement. The four figures are not four independent measurements. Given the two volatilities and the beta, the tracking error follows: 11.8 squared is 139.24, 10.4 squared is 108.16, twice 1.08 times 108.16 is 233.6256, and 139.24 plus 108.16 less 233.6256 is 13.7744, whose square root is 3.7114 per cent. Three of the four figures are free and the fourth is determined by them, so quoting all four as separate evidence counts the same information twice.

What can one year of this record support?

Step five asks the question everybody knows and nobody says out loud: is this record long enough to conclude anything from? Ask it as a step, in writing, before the conclusion is drafted, or it will never be asked at all.

The information ratioThe excess return divided by the tracking error over the same window. How the ratio is built is covered separately. for the Anantara Multi-Asset Portfolio in the stated twelve months is 1.6 divided by 3.7, which is 0.43. Read as a signal against the noise around it, a ratio of 0.43 reaches a conventional level of two after a number of years equal to two divided by 0.43, all squared. Two divided by 0.43 is 4.65, and 4.65 squared is 21.6. So roughly twenty two comparable years, assuming the ratio holds steady and the yearly figures are independent of each other, and both of those assumptions are generous. Turn the arithmetic around and it says what a single year would have to look like: one year settles the question on its own only at a ratio of 2.0, and the stated year holds 0.43, about a fifth of that.

The Anantara portfolio has one year. One year at an information ratio of 0.43 supports nothing at all about skill. The honest response is to record the emptiness rather than to hunt for a measure that returns a friendlier answer. Notice how uncomfortable that is: the sample sizeHow many independent observations the record actually contains, which sets a ceiling on what any conclusion drawn from it can claim. question is the only step that reliably produces an answer nobody in the room wanted. The sample size question is a numbered step for precisely that reason, and not a matter of temperament.

What can the constraint record settle that the return record cannot?

Step six reaches for evidence of a completely different kind. The constraint recordThe record of whether the portfolio stayed inside each written limit through the period, and how close it ran to any limit that was near to binding. answers three questions: did the portfolio stay inside what it was allowed, was any breach caused by a decision or by prices moving on their own, and how close did the binding limit actually run.

Read what that record has that the return record does not. The constraint record covers exactly the period in question, is complete the moment the period ends, and settles its questions outright rather than in twenty two years. A clean constraint record is a fact about conduct in the period under review, and unlike a return figure it is available on the first morning of the evaluation.

On the Anantara Multi-Asset Portfolio it reads like this. Equity at 60.0 per cent, inside the 50 to 70 per cent band with ten points of room on either side. The largest holding sits at 4.6 per cent of the portfolio. On Rs 500 crore that is Rs 23 crore against a cap of Rs 25,00,00,000/-, so the holding is inside. How close is that? If that single holding rose in price while nothing else moved, its weight would reach 5 per cent after a rise of about 9.15 per cent. The arithmetic runs like this: 23 crore times one plus the rise, over 500 crore plus 23 crore times the rise, hits 5 per cent at a rise of 2 divided by 21.85. No breach at the date examined. And two constraints, the ban on unlisted holdings and the minimum credit standing, could not be tested at all from the material recorded. An untestable constraint is itself a finding and gets written as one. State the base every time as well: measured against the Rs 300 crore equity sleeve rather than against the portfolio, that same holding is 7.7 per cent, an answer to a different question and a breach of nothing.

How is the stated process examined?

Step seven sets what the manager said in advance beside what the portfolio actually held. Three questions do the work. Does the stated approach match the positions taken. Were the reasons given checkable at the time, or only once the outcome was known. And was the record reported in a way that removed the choice of what to show, or in a way that left it open.

The process evidenceThe comparison of what was described in advance with what was actually held and reported, which unlike a return can still be altered. is the only part of an evaluation that identifies something that can be changed. A return cannot be revised, and a sample cannot be lengthened on demand. That is why a review that spends its whole hour on the return figure leaves the room with nothing to do differently.

On the Anantara Multi-Asset Portfolio the honest answer at this step is short. The equity sleeve is held across 28 names and portfolio turnover was 34 per cent over the stated year, meaning about a third of the portfolio was replaced, and turnover carries a cost that none of the return figures above show. Both of those are computable from what is recorded: 28 names across a Rs 300 crore sleeve averages Rs 10.71 crore each, and 34 per cent of Rs 500 crore is about Rs 170 crore replaced. The material recorded does not contain what Faiz Ahmad Ansari described in advance, so whether 28 names and 34 per cent match it cannot be answered. So the evaluation writes that down as a gap rather than filling it with an inference, and the committee chaired by Rukmini Deshpande now knows one specific thing to require in writing next year.

How is a conclusion written when the evidence stops short?

Step eight writes the conclusion, and the conclusion includes the part the evidence does not reach. The closing step is where an exercise in evaluating somebody is most likely to slip into a verdict, so the rule is flat: an evaluation ends in a statement of what the evidence supports and what it cannot reach, and never in an instruction.

An evaluation that records an unresolved findingA question the evidence in hand cannot answer, written into the conclusion so the next reviewer inherits the question rather than a guess. is completed work, not unfinished work. The alternative is a conclusion that reads as settled and is not. Written out for the Anantara Multi-Asset Portfolio, all eight steps produce this.

Read as a paragraph, the conclusion says this. Every constraint that could be tested was obeyed in the stated twelve months. The gross excess return of 1.6 points reconciles under two separate splits, and about 0.49 points of it is arithmetic on exposure rather than a judgement anybody made. One year at an information ratio of 0.43 cannot separate skill from a draw and will not for decades. The agreed return basis, the drawdown dates, two constraints and everything about what was described in advance are unknown. The evidence carries no verdict on anybody, so the evaluation states none.