Getting Out of a Hedge Fund: How Four Liquidity Terms Bite

Getting out of Nilgiri Absolute Return Fund, invented, means passing four tests in order. A lock-up decides whether the units are eligible at all. A dealing window decides when. A gate decides how much of what was validly requested is actually paid. A side pocket decides which part of the holding cannot leave at any price. Each test reduces what the one before it allowed.

An offering document prints these four terms as a list: four paragraphs, one after another, each perfectly clear on its own. Read as a list they look like four separate obstacles. A reader who has understood all four still ends up with a wrong answer to the only question that matters: when the money arrives, and how much of it. The four terms are not a list at all. A request meets them in a fixed order, each one working on whatever the one before it left standing. The change of shape, from list to order, decides every answer that follows.

The fund used throughout is Nilgiri Absolute Return Fund, invented, managed by Nilgiri Alternatives Advisors Private Limited, invented. The fund is open-ended: money is subscribed rather than committed, there is no capital call, there is no fixed term and there is no wind-up date on the calendar. In a closed-end vehicle the term itself says when an investor gets its money back. Nilgiri has no term, so four contracted liquidity provisions stand in its place, and those provisions are what a holder has instead of a maturity date.

Why can a fund that holds perfectly liquid things still be hard to leave?

Start with something ordinary. Twelve households share the cost of a wedding hall and pay in over a year. The money is in a bank account. Nothing about that account is illiquid: any rupee in it could be moved this afternoon. And yet if four of the twelve want their share back on the same Tuesday, and the hall has already been booked and half paid for, the account being liquid does not help anybody. The nature of the assets does not decide who gets what. The arrangement the twelve wrote down at the start decides it: who may ask, when they may ask, and what happens when several of them ask at once.

A hedge fund is that situation with a contract behind it. Nilgiri Absolute Return Fund holds shares that trade on an exchange, and a share that trades on an exchange can be sold on any day the exchange is open. So the assets are not the constraint. The constraint is the arrangement between the holders of the fund and the fund itself, and that arrangement exists because of a structural fact that has nothing to do with how quickly a share can be sold.

Here is the fact. When one holder leaves, the fund does not hand over a slice of its positions. The fund cancels that holder's units and pays cash. To find the cash it must sell something, and what it sells is a share of everybody's positions rather than a share of the leaver's. So a redemptionAsking the fund to cancel units and pay out their value. is not one person taking their own property out of a cupboard. A redemption is one person requiring the group to trade. Every rupee paid out to somebody leaving is a rupee raised by selling positions that belong to the people who stayed. A fund that must sell a fifth of its book in a week to meet requests is a different fund the following Monday from the one it was the previous Friday, and the people who did not ask for anything are the ones holding whatever it has become.

The four terms exist to manage that mechanism. Each term does a particular job, and the order in which those jobs happen decides the outcome. Ang, Asset Management, 2014, treats illiquidity as something an investor is compensated for rather than as a nuisance to be complained about, and that framing is his; whether any particular compensation is adequate is a separate question.

What are the four terms, and what does each one actually decide?

All four follow, each in a single sentence, exactly as the fund's own documents arrange them. Three of the four are worked in full separately.

Four sentences, one for each. The lock-upA period from each subscription during which no redemption is accepted at all. is a period. The dealing dateThe date on which units are actually issued or cancelled. is a date. The gateA cap on how much may be redeemed at one dealing date, with the excess scaled back. is two ceilings. The side pocketA separate class holding one position that cannot reliably be valued, in which no dealing happens. is a separate class. All four are this one fund's own contracted terms. The conditions under which a vehicle of this kind is registered in India are set by the Securities and Exchange Board of India.

A list hides the ordering. Each of the four is measured against a different thing. The lock-up is measured against each subscription separately, so one holder can have some units free and some locked on the same morning. The window is measured against a calendar and a notice deadline and never against a rupee amount. The gate is measured against two bases at once, one of them a single holder's own holding and the other the whole fund. The pocket is measured against the holding itself, and because it splits the holding before anything else runs, it changes the base that the other three are struck on. Every reduction in this sequence has a denominator, the denominators are not the same, and almost every wrong answer about hedge fund liquidity comes from carrying one test's denominator into the next test's arithmetic.

Test one: is any of this holding eligible to be asked for?

The first test does not reduce a request. The lock-up decides whether particular units are allowed to be in one. The fund's lock-up runs twelve months from each subscription, and the phrase from each subscription is doing all the work in that sentence. The clock does not run from the day the holder first invested, and it does not run from the day the fund launched. Every time money goes in, a fresh twelve months starts on that money and on nothing else.

Picture a household that has been putting money into a recurring deposit every few months. Each deposit has its own maturity, so on any given day some of the money is free and some is not, and the household does not have one status but several at once. The lock-up works exactly like that. A holder who subscribed three times over two years has three separate clocks running, and on the morning it decides to leave, only the units whose clock has finished are eligibleThe part of a holding that is past its lock-up and therefore allowed to be requested..

The holder followed all the way through holds Rs 25,00,00,000 of Nilgiri Absolute Return Fund and Rs 5,00,00,000 of that was subscribed inside the last twelve months. The holder asks to redeem everything. The subtraction is the whole of test one: Rs 25,00,00,000 less Rs 5,00,00,000 is Rs 20,00,00,000, and Rs 20,00,00,000 goes forward to the next test. The Rs 5,00,00,000 has not been cut, refused or reduced; it has simply never entered the request, and no later test can restore it.

The distinction between an amount cut and an amount never asked for sounds pedantic, and it is not. A cut amount is still asked for and still waits somewhere. An ineligible amount was never asked for, so nothing about it is pending, nothing about it queues, and the only thing that changes its status is the passage of time. Test one is the only one of the four that works on eligibility. The other three all work on amounts.

Test two: is there a date, and did the notice reach it in time?

The second test's mechanics are worked to the day separately: when notice must be given, where the dealing date falls, when payment must be made, and what the audit holdback is for. In the sequence it does one job, and the job is a yes or a no.

A request that has survived test one arrives at a calendar. Either it was received before the notice deadline for the coming dealing date, in which case it joins that date, or it was not, in which case it joins the next one. There is no partial answer. Nothing is scaled, nothing is proportioned, nothing is halved. A request either catches this date or it catches the following one, and the difference between those two outcomes is a full quarter of waiting for an amount that was otherwise entirely in order.

For the holder in this worked case, notice was received before the deadline, so the request catches this dealing date rather than the next one. Catching the date is the entire contribution of test two. A date test does not change an amount, so Rs 20,00,00,000 went in and Rs 20,00,00,000 came out. Test two is the only one of the four that can cost a holder three months without costing it a single rupee of the amount requested.

The dealing calendar is a genuine subject with real arithmetic in it, and it is covered separately. The test sits after eligibility and before any question of amount. There is no sense in asking how much of a request will be paid before knowing which date is paying it.

Test three: how much of a valid request is actually paid?

Now the amount changes. The gate is the only one of the four tests that takes a number that was entirely valid and returns a smaller number, and it does that through two separate ceilings that are usually printed as one paragraph. Both apply, both apply at the same dealing date, and they are measured against completely different things.

The first ceiling is drawn on one investor's own holding: no more than 20.0 per cent of it may be redeemed at a single dealing date. For this holder, 20.0 per cent of Rs 25,00,00,000 is Rs 5,00,00,000. The base that ceiling is struck on decides the answer. The ceiling is struck on the holding of Rs 25,00,00,000, not on the eligible Rs 20,00,00,000 that survived test one, and the difference between those two bases is Rs 1,00,00,000 of allowance. A reader who applies the 20.0 per cent to the eligible amount gets Rs 4,00,00,000 and is wrong by a fifth.

The second ceiling is drawn on the fund: no more than 25.0 per cent of net assets may go out at a single dealing date. Net assets are Rs 5,00,00,00,000, so that ceiling is Rs 1,25,00,00,000. The second ceiling is not a ceiling on any one request. The cap sits on the total of everybody's requests together, and that structural difference produces the most counter-intuitive behaviour in the whole sequence.

Below the total of Rs 1,25,00,00,000 the second ceiling does nothing whatsoever. Not a little; nothing. Every rupee validly asked for is paid, and a holder reading the fund's documents would see a term that has never once affected it. The moment the total crosses that figure, the ceiling divides itself across everybody, and each request is scaled back pro rataCutting every request by the same proportion so the total fits the cap., meaning cut by the same proportion so that the sum of what is paid fits inside the cap. The proportion is the cap divided by the total requested, and because the total sits in the denominator, the proportion falls away steeply once it starts moving at all.

Put both ceilings on the same request and the order inside test three is fixed too. The per-investor ceiling runs first, on each holder separately, and produces the amount that holder brings to the dealing date. Only then are all those amounts added up and compared with the fund-level cap. The fund-level cap is applied to a total that every holder's own ceiling has already narrowed. The total in the arithmetic below is never the total anybody originally wanted.

Which limb of the gate binds, and how would anybody know?

The gate is printed as one term and behaves as two, and the two do not bind in the same circumstances. The bases are the answer, so keep them in view. The first limb is measured against one holder's own holding of Rs 25,00,00,000. The second is measured against the fund's net assets of Rs 5,00,00,00,000. The two bases differ by twenty times, and a term whose two halves are struck on bases twenty times apart cannot bind in the same conditions.

The two extremes make the behaviour obvious. Suppose this holder asks for everything and nobody else asks for a rupee. The total at that dealing date is Rs 5,00,00,000, or 4.0 per cent of the Rs 1,25,00,00,000 cap. The second limb is nowhere near touched. The holder is stopped entirely by the first limb, at Rs 5,00,00,000 out of an eligible Rs 20,00,00,000. Now suppose a great many holders, each asking modestly and each comfortably inside its own 20.0 per cent, together bring Rs 1,60,00,00,000 to the date. Not one of them has breached the first limb. Every single one of them is cut by the second.

The first limb is struck on one holder acting alone, the second on every holder acting at once, and no holder can tell from its own request which of the two is going to bind. That last clause is the part worth carrying away. A holder knows its own holding exactly and therefore knows its own first limb exactly. The holder does not know what anybody else is asking for. The second limb depends entirely on a number the holder cannot see until the dealing date has passed. Every term is perfectly clear in advance, and the amount paid is still knowable only after the fact.

What happens when a lock-up and a gate act on the very same holding?

How Gating and Lock-Ups Affect Fund Liquidity

Two terms, one holding, one morning. Both end with the holder receiving less than it asked for, so it is tempting to treat them as the same kind of thing applied twice. The lock-up and the gate are not the same kind of thing at all, and the figure below draws the difference on a single rupee scale.

The lock-up removes units from the request. Rs 5,00,00,000 of the holding was subscribed inside twelve months, so it is not in the request, was never in the request, and cannot be affected by anything that happens later at the dealing date. If the gate had not existed at all, that Rs 5,00,00,000 would still not have been paid.

The gate cuts the request that survives. The gate takes the Rs 20,00,00,000 that did enter, allows Rs 5,00,00,000 of it under the first limb, and then reduces that to Rs 3,90,62,500 under the second. The first limb is struck on the holding of Rs 25,00,00,000 and is entirely indifferent to how much of it happens to be eligible. If the lock-up had not existed at all, the first limb would still have allowed only Rs 5,00,00,000.

The interaction is the interesting part, and it is not additive, so hold both facts at once. In this worked case the lock-up removed Rs 5,00,00,000 and the first limb of the gate allowed Rs 5,00,00,000. The two figures are identical and they are not the same number. One is what was subscribed inside twelve months. The other is a fifth of the holding. The two coincide only because the recent subscription happens to be exactly a fifth of the holding. Change either figure and they part company immediately. Two equal rupee amounts struck on two different bases are two different numbers that happen to have the same value, and treating them as one is the commonest arithmetic error in this whole subject.

There is a second, quieter interaction. Because the first limb is measured against the whole holding and not against the eligible part, a lock-up does not shrink the gate's allowance. A lock-up shrinks the pool the allowance is drawn from. If a holder had Rs 25,00,00,000 of which only Rs 4,00,00,000 were eligible, the first limb would still allow Rs 5,00,00,000, and the eligible amount would then be the binding number instead. Which of the three figures binds depends entirely on their relative sizes, and there is no general rule that says one of them always wins.

What does one request actually look like as it passes through all four tests?

Everything so far has been one test at a time. Now run a single request through the whole sequence and keep the rupees visible at every stage. The rupees are what a reader remembers. The denominators are what a reader gets wrong. The holder asks to redeem its entire holding of Rs 25,00,00,000 in Nilgiri Absolute Return Fund at one dealing date.

Read the last column downwards and the arithmetic is plain. Rs 25,00,00,000 less Rs 5,00,00,000 of locked money is Rs 20,00,00,000. Twenty per cent of the holding, being Rs 5,00,00,000, is less than the eligible Rs 20,00,00,000, so the first limb binds and Rs 5,00,00,000 goes forward. The cap of Rs 1,25,00,00,000 divided by the total of Rs 1,60,00,00,000 is 0.78125, and Rs 5,00,00,000 multiplied by 0.78125 is Rs 3,90,62,500. Rs 20,00,00,000 asked for, less Rs 3,90,62,500 paid, leaves Rs 16,09,37,500 standing.

Now the two ratios that make the point, and both need their denominator said out loud or they mean nothing. Rs 3,90,62,500 against the eligible Rs 20,00,00,000 is 19.5 per cent. Rs 3,90,62,500 against the whole holding of Rs 25,00,00,000 is 15.6 per cent. The holder asked for everything and received 19.5 per cent of what was even eligible and 15.6 per cent of what it held, and not one contracted term was breached at any point in that sequence. Nobody exercised discretion against it. Nobody was slow. Nobody was unfair to it. Four ordinary provisions ran in their ordinary order and that is what came out.

One coincidence in these numbers deserves flagging so nobody builds a rule on it. The locked amount is Rs 5,00,00,000 and the first limb of the gate also allows Rs 5,00,00,000. The amounts are equal here only because the recent subscription happens to be exactly a fifth of the holding. One of them is a subscription made inside twelve months; the other is a fifth of Rs 25,00,00,000. Move the subscription date, or subscribe a different amount, and the two figures separate at once.

What happens to the part of the request that was not paid?

A reader who has followed the arithmetic can still finish with the wrong picture here. The Rs 16,09,37,500 does not evaporate, and the holder does not have to write in again. The unpaid amount is carried forwardThe unpaid part of a request, which waits for the next dealing date., meaning it stands at the next dealing date as a request already made.

Follow the consequence at the level of the fund rather than the holder. At this dealing date, Rs 1,60,00,00,000 of requests had already survived every holder's own first limb, and the cap allowed Rs 1,25,00,00,000 of it. So Rs 35,00,00,000 of already-narrowed requests was not paid, and that Rs 35,00,00,000 is standing at the next dealing date before a single new request is made. Whatever anybody asks for at that date arrives on top of it.

A gate does not reduce how much people want out of a fund. A gate changes when they can have it, and the demand it did not satisfy is still there at the next date. The difference is between a gate as a valve and a gate as a tap that has been turned down. A valve that closes reduces flow permanently. A tap turned down moves the same water through a narrower opening over a longer time, and everything that has not yet come through is still behind it.

Two honest limits on that Rs 35,00,00,000 before anybody quotes it. First, it counts only the fund-level scale-back. Each holder's own first limb held back more, and those amounts are not in it. Second, the total at the next dealing date depends on what people ask for then, and nobody knows that total in advance.

Test four: what part of the holding cannot leave at any price?

The fourth test is different in kind from the other three. A holding the manager designates as not reliably valuable may be moved into a separate class in which no subscription or redemption is accepted until it is realised. The mechanism fits in that one sentence. The trigger, the unit split, the valuation and the release are all covered separately.

The pocket's position in the order is odd: it acts before the other three rather than after them. Suppose the manager had designated a pocket before this dealing date. Every holder's units split in the same proportion, so a pocket of 8.0 per cent of net assets takes 8.0 per cent of this holder's Rs 25,00,00,000, being Rs 2,00,00,000, and leaves Rs 23,00,00,000 in the dealing classThe units that can still be issued and cancelled, as opposed to those in a pocket..

Everything above is then struck on Rs 23,00,00,000 rather than on Rs 25,00,00,000. The locked money splits too, so Rs 4,60,00,000 of recent subscription sits in the dealing class and the eligible amount becomes Rs 18,40,00,000. The first limb becomes 20.0 per cent of Rs 23,00,00,000, being Rs 4,60,00,000. And the Rs 2,00,00,000 in the pocket is not part of any of that: it does not queue, it is not carried, and it leaves only when the pocket is realised, whatever the other three tests say.

The pocket does not reduce a request. The pocket reduces the holding a request can be made against, so it belongs at the front of the sequence even though the documents number it last. The fund-level proportion is not restruck for the pocketed case. The total everybody else asks for would move as well, and nobody knows that total in advance.

What has a fund actually said when it says quarterly liquidity?

The four tests together explain why one very common sentence carries far less information than it appears to. A vehicle described as offering quarterly liquidity has told a reader precisely one thing: there is a dealing date every three months. The word quarterly carries nothing beyond that one fact. Quarterly describes test two, and only the date part of test two.

Everything else is untouched by it. Quarterly says nothing about the notice period, so a request may need to be in hand well before the date it is aiming at. Quarterly says nothing about a lock-up, so a subscription may not be eligible for that date at all. Quarterly says nothing about a gate, so a valid request may be paid in part. Quarterly says nothing about a pocket, so a share of the holding may not be dealable at any price. And payment follows the dealing date rather than falling on it, so quarterly says nothing about when the cash actually arrives.

Quarterly describes the date and nothing else, and a fund that deals quarterly is not the same thing as a fund an investor can leave quarterly. In the worked case above the fund dealt on schedule, the request was in time and every term operated exactly as written, and the holder still received 15.6 per cent of its holding at that date. Both of those sentences describe the same dealing date.

How does somebody whose job is reading these documents use the order?

Somebody who reads these documents for a living works with the order rather than with the terms, and the work has a recognisable shape.

An analyst covering a vehicle of this kind is usually trying to answer one question for somebody else: if a holding of a stated size were asked for at the next date, what would come back and when. The four terms in a list will not answer that. The order will. The answer is the sequence run forwards, and the analyst's real work is finding the four bases rather than the four percentages. Which subscriptions are inside their lock-up, so what is eligible. Which date the notice can still catch. The size of the whole holding matters, and the first limb of the gate is struck on it rather than on the eligible part. And the fund's net assets, the base of the second limb.

The same order is what makes a stress question answerable at all. An institution's risk function does not ask what happens if it redeems. The question is what happens if everybody redeems, and that question lives entirely in the second limb of test three. The second limb is the only one of the four whose value depends on other people's behaviour. The first limb, the lock-up and the pocket are all knowable from a single holder's own records. The fund-level proportion is not knowable in advance by anybody, including the manager, until the notices are in.

A person running a household's own money meets the same structure in a far smaller way, and the shape is identical. Money placed in an arrangement with a notice period and a queue is money whose availability is decided by a sequence of conditions rather than by a single date, and the useful question is never how liquid is this but which condition stops me first. The order is the answer to that question, and the order is the same whether the amount is Rs 25,00,00,000 or a tenth of a month's salary.