Managed Futures: Systematic Trend Following

Managed futures is approach 2 of the eight hedge fund approaches described in this library. Positions are taken by a written rule applied to prices across many markets at once, with no view about any company, and the rule is applied the same way whether it is working or not. The cost of the rule is the number of times it changes its mind, and that count is arithmetic rather than opinion.

Everything here rests on one plain distinction, and it is worth sitting with before a single rupee of arithmetic arrives. A position can come to exist in exactly two ways. Somebody can look at a situation and decide, or a decision taken earlier can be applied to the situation without anybody looking. Managed futures sits entirely in the second of those, and that is the whole of what the word systematicDecided in advance by a written rule rather than in the moment by a person. conveys. Notice what the second way does not claim. The second way does not claim the earlier decision was a better decision. The claim is only that the decision was already made, and that when today arrived, nobody was asked.

What does managed futures actually mean, and what makes it systematic?

The plainest version of this sits on an ordinary street. A woman runs a small provision shop and keeps a card taped inside the storeroom door. The card says: when the rice sack drops below ten kilos, order another sack. She wrote it three years ago, on an afternoon when she had time to think properly about how long deliveries take and how much floor space a spare sack eats. Now she does not think about rice at all. She looks at the sack, she reads the card, she orders or she does not. The thinking happened once. The daily work is only the application of the card.

Her cousin runs the same kind of shop two lanes away, and he has no card. Every morning he looks at his rice, thinks about the week ahead, remembers that a wedding party bought heavily a few days ago, and decides. Some mornings he orders early. Some mornings he waits. He is not being careless; he is deciding, each time, with everything he happens to know that morning.

The two shops are the whole distinction, and the finance version changes nothing about them except the scale and the subject. A systematic approach is the card. The rule is written down before the situation arises, the situation is then fed into it, and what comes out is the answer. There is no step in that path where a person looks at the answer and asks whether it feels right this time.

The word systematic states when the decision was made and who made it, and it states nothing at all about whether the decision is any good. The card in the storeroom could say ten kilos or it could say two kilos, and it would be exactly as systematic either way. Somebody chose ten. A written rule has no opinion about itself, so the card cannot say whether whoever chose ten kilos was wrong. Mistaking when a decision was made for whether it was a good decision is the single most common error made about the word systematic.

Managed futures is the name this label carries when the instruments used are futures contracts. Systematic trend followingA rule that holds a position while a price stays on one side of a measure taken from its own past. is the name for what the rule inside it does. Eight approaches to running a hedge fund book are set out across this library, and managed futures is the second of the eight, described rather than ranked.

What is a futures contract doing in this approach at all?

Futures are covered as a subject in their own right elsewhere, so only the three structural properties that make them the instrument this approach reaches for are named here. Futures are exchange traded. The contract is standardised, and the exchange stands between the two sides rather than each side relying on the other. Futures are also margined. Cash is posted against the position rather than the full value of the position being paid over. And a long position and a short position are expressed with the same ease: there is no borrowing to arrange and no separate machinery, just the opposite sign on the same contract.

The ease of taking a short position is why a rule can be written without a preferred direction. A rule that can only say hold or hold nothing has to be told which way is up. A rule that can say hold, hold nothing, or hold the other way does not, and the same written sentence then answers a rising series and a falling one without anybody adjusting it. Short selling is covered properly elsewhere. The one structural point that matters here is that a short position's gain is bounded by the price falling to nothing. The loss is not bounded at all.

The notionalThe size of the position carried, as opposed to the cash posted against it. of a position is the size being carried. The cash posted against it is a different and smaller number. The difference between the two is the third of the three risks described below, and the arithmetic connecting them is set out under exposure and leverage.

What shape does a trend following rule have, without being a recipe?

Here is the shape, and reading it carefully is worth more than any number would be. A boundary is computed from the price series' own past. While the price sits on one side of that boundary, the rule holds a position. While it sits on the other side, the rule holds nothing, or holds the other way. That is it. The rule is that whole object and nothing more.

Trend following is the name for a rule of that shape, and the name is a description of the mechanism rather than a promise about it. The rule is not forecasting. The rule is not estimating what a company is worth. The rule holds no view about an industry, a currency or a season. The rule reads a series of numbers, compares them to a boundary drawn from those same numbers, and returns a position.

Three things about the boundary remain unstated: how far back it looks, how far the price must move before it is treated as crossed, and which answer the rule returns on the day the price sits exactly on the line. Each of those is a parameterA number inside a rule., and a description of what a rule is becomes a rule somebody could run as soon as those numbers are named. The subject here is the shape and what operating it costs, and a shape with the numbers left out is still a complete shape.

One consequence of that shape deserves its own sentence. A great deal follows from it. The position is either on or it is off. The position does not lean, it does not scale with how convinced anybody is, and it does not shift because the price moved a lot on one day rather than a little. Between two crossings of the boundary, every day is the same day as far as the rule is concerned.

Why is the same rule pointed at many price series at once?

Because the rule has nothing to say about any particular one of them. Having nothing to say about any one series follows directly from the shape. A rule that reads a series of numbers and compares them with a boundary drawn from those same numbers does not need to know what the numbers are the price of. The rule does not know whether it is reading a metal, a grain, a currency or a rate. The rule could not say which, and nothing inside it would work differently if it were told.

So the natural way to operate a rule of that shape is to point it at many series at the same time. The rule stays one object. The positions become many, one for each series, each answered separately on the same day by the same written sentence. There is nowhere inside the rule for a view about a single market to live, so the approach carries no view about any single market.

A night watchman with one instruction makes the same point: if a door is open after eleven, close it and note it in the book. He walks a building with forty doors. He is not making forty judgements about forty doors. He is making one instruction happen forty times. Asked what he thinks of the third floor store room, he would have no answer, and the absence of an answer is not a gap in his work. The absence is the arrangement working exactly as intended.

Whether pointing a rule at many series improves what the approach returns is a question about outcomes rather than about mechanism. The description stops at the structure: one rule, many positions, no view about any of them.

Why is a trend following rule late at both ends?

Lateness at both ends is the most important structural fact about a rule of this shape, and it is arithmetic rather than criticism. The fact has two halves, each worth taking slowly.

A rule of this shape detects a move by watching a price cross a boundary computed from that price's own past. For the crossing to happen, the move has to have happened first. The rule reacts to the move itself, so no arrangement of that sentence puts the rule in before the move begins. The same argument runs the other way at the far end. The rule stops holding when the price crosses back. For that to happen, the move has to have finished and turned. A rule that follows trends is therefore late getting in and late getting out, always, by construction, and no choice of numbers inside it changes that.

Look at what that means as points on a path. Take a constructed shape, a price that goes from 100 to 120 and settles back at 110. The boundary is crossed on the way up at 106 and crossed back on the way down at 112. The rule was not holding for the first 6 points of the rise. Nothing had been detected yet. The rule held the 6 points between 106 and 112. The end had to arrive before the end could be detected, so the rule did not keep the 8 points between the high of 120 and the crossing back at 112. Six missed at the start, six held in the middle, eight given back at the finish, and those three add to the 20 points of the whole move.

Now change only the path after the crossing, and change nothing whatever about the rule. The price crosses the same boundary at the same 106, rises to 110, turns, and falls to 96. The rule enters at 106 exactly as before and leaves when the price crosses back at 100. The 6 points between the two crossings this time ran the other way. Same rule, same lateness at both ends, and the arithmetic between the crossings landed on the opposite side of nothing.

Both of those paths are shapes drawn to make the arithmetic visible rather than records of any market. How often a path of either kind appears is a question about outcomes, for this approach and for the other seven alike.

What can the rule not see, and what does it do about it?

Nothing, and nothing. The two answers are worth having in that order.

Go back to the storeroom card. The rice sack falls below ten kilos, so the card says order. The card does not know that the wholesaler's lane is flooded, that the shop three doors down closed last week and half its customers now walk in here, or that the shopkeeper's supplier is about to raise prices. The card reads exactly one input, the level in the sack. None of the rest reaches it. A rule responds only to what it was written to read, so anything outside those inputs may as well not have happened.

Blindness of that kind is not a flaw somebody forgot to fix. Blindness is the same property that makes the approach systematic, seen from the other side. The instant something outside the written inputs is allowed to change the answer, a person is deciding again, and the card is no longer a card. The first property cannot be kept while the second is dropped: they are one property.

So the useful question is not whether a rule is good. The useful question is which inputs never reach it. Two rules of identical shape, reading different inputs, are different objects, and the difference between them is entirely in what each one is blind to. The question about inputs is answerable by anybody holding the document that describes the approach, and answering it needs no number from inside the rule.

The comparison below sets the rule beside a person watching the same screen. The comparison holds two different sets of failures rather than a scoreboard. A person can notice the flooded lane, and a person can also talk themselves out of a position that was doing exactly what it was supposed to do, and can do so for reasons they will not remember accurately a month later. The rule can do neither of those things. Neither column is a better version of the other, and neither ranks above the other.

What does it cost every time the rule changes its mind?

Cost is the one part of the approach that can be worked to the rupee without anybody making a claim about outcomes, and cost is where the subject stops being abstract.

When the price crosses the boundary, the rule's answer changes. The rule was holding and now holds nothing, or it was holding nothing and now holds. The moment of that switch has a name worth fixing: a change of stateThe moment a rule switches from holding a position to holding none, or the reverse.. A change of state is not an opinion, a signal or an event in the world. A change of state is the rule returning a different answer than it returned yesterday, and the only way to make that answer real is to trade. Each change of state is therefore one executionA single trade placed to put the rule's current answer into effect., and every execution is charged.

To put a rupee figure on that, the size and one contracted rate are borrowed from Nilgiri Absolute Return Fund, an invented vehicle managed by Nilgiri Alternatives Advisors Private Limited. The fund had net assets of Rs 5,00,00,00,000 at its record date and pays a contracted transaction cost of 0.05 per cent of value on each execution. The rate is the fund's own contracted arrangement rather than anybody's idea of what dealing costs in India. Nilgiri runs approach 6 of the eight and not approach 2, so the arithmetic that follows is a labelled counterfactual that borrows only a size and a rate.

Run it. On a notional of Rs 5,00,00,00,000, one execution at 0.05 per cent of value is Rs 25,00,000. Four crossings in a quarter is four executions, being 0.20 per cent of the notional, being Rs 1,00,00,000. The calculation is finished, and notice how little went into it: a size, a rate, and a count.

Carry it out to a year on the same counterfactual. Sixteen crossings across a year is 0.80 per cent of the notional, being Rs 4,00,00,000. Set that beside the management fee purely for scale and for nothing else: 2.00 per cent a year on net assets of Rs 5,00,00,00,000 is Rs 10,00,00,000, so Rs 4,00,00,000 of execution cost is 40.0 per cent of the fee. Both are costs the assets have to cover, and only one of them appears in a fee schedule. That is the reason this arithmetic is worth a reader's time. Execution cost is a real cost of operating the approach, and it is not written down in the place where a reader goes looking for costs.

And now the sentence that keeps this block on the right side of the line. None of the above says whether the rule made money or lost it. The charge attaches to the execution and not to the outcome, so the Rs 1,00,00,000 is incurred identically whether the rule was right on all four crossings or wrong on all four. The cost of a rule is the number of times it changes its mind, and that sentence is the whole of the worked instance.

What risks does the approach carry, in its own terms?

Three, and each of them falls straight out of the mechanism rather than being bolted on afterwards. Naming them is one thing; how large any of them is, how often any of them bites, and what any of it does to what the approach returns are questions about outcomes.

Risk one: the position is held all the way to its exit condition

The rule leaves when the boundary is crossed and not before. The rule does not leave when the position has moved against the fund, when somebody is uncomfortable, or when the move looks finished to a person watching. All of those are outside the rule's inputs and therefore invisible to it. The consequence is the lateness described earlier, now read as a risk rather than as arithmetic: between the point where a move turns and the point where the boundary is crossed, the position is still on, and the rule is doing exactly what it was written to do the entire time.

The gap is what detection means, so no version of this rule can be written without it. Making the boundary tighter shortens it and produces more crossings, which the section above priced. Making it looser lengthens it and produces fewer. Neither choice removes it, and naming numbers on either side of that trade would amount to handing over a rule.

Risk two: a path that keeps turning is charged on every turn

A reversalA price crossing back over the boundary the rule watches, so the rule changes its mind again. is the price crossing back over the boundary, and every reversal is a change of state, an execution and a charge. Take a constructed path that crosses the boundary twelve times in a quarter. Twelve crossings are twelve executions, twelve charges of Rs 25,00,000 on the same Rs 5,00,00,00,000 notional, and Rs 3,00,00,000 in total, being 0.60 per cent of the notional in a single quarter.

The charge does not read the outcome, and a charge blind to the outcome is a risk rather than an inconvenience. A crossing the path then carried on from and a crossing the path turned straight back over are billed at the identical rate. The bill is written by the shape of the path and by nothing else the rule or anybody in it can influence.

Risk three: the position is margined, so the exposure is not the cash

A margined position is carried against cash posted rather than paid for in full, so the size being carried and the cash standing behind it are two different numbers, and the first is the larger. Sizing an approach by what has been put up therefore understates what is being carried. How margin is set, by whom, and what happens when a position moves are covered under the prime broker and under exposure and leverage.

One pair of numbers is worth naming so that the difference is concrete rather than theoretical. Nilgiri Absolute Return Fund, at its own record date, carried gross exposure of 180.0 per cent of its net assets and net exposure of 80.0 per cent. Gross and net describe the same book on the same day, and the two figures stand 100 percentage points apart. A reader who takes either figure alone as the size of the position has taken the wrong number, and the ladder connecting the two is set out under exposure and leverage. That fund runs approach 6 rather than the approach described here, and the pair is quoted here only to show that the two measures separate.

Why is the rule followed on the days it is losing?

Readers ask this question last and feel most strongly about it, and the answer is duller than the question deserves. The rule is followed because otherwise there is no rule.

Go back to the storeroom card one final time. Suppose the shopkeeper starts overriding it. Some weeks she orders on the card and some weeks she looks at the sack and decides. Ask what she has now. She does not have a card that she sometimes follows. She has a person deciding, who occasionally agrees with a card. A rule that is abandoned when it becomes uncomfortable is not a rule, and the whole reason to write one down in advance is to remove the moment of choosing. Choosing to stop is that moment arriving through a side door.

So following it on a bad stretch is mechanism rather than virtue. Holding to the rule is not fortitude, it is not character and it is not proof of anything about the manager. It is the arrangement continuing to work exactly as described. If it stops, what the reader was told about the approach has stopped being true, and nothing that came before is a reliable description of what happens next.

Which produces the one genuinely practical question in this whole area, and it is a question about a document rather than about a market. Ask what the offering document says about when the rule may be changed, who may change it, and what has to be recorded when it is. A rule that can be set aside quietly on a bad afternoon is a different object from a rule that cannot, and the difference does not show up in any description of the strategy. The difference shows up in the governance paragraph, a long way from the paragraph a reader would think to check.

What is actually worth asking of a document describing an approach like this?

Reading a document is the practical end of the subject. Most people who meet the approach are not running a book. Such readers are reading about one: an analyst at an investment office writing a note, a student working through an offering document for the first time, somebody at a distributor who has to explain a factsheet to a person who will ask a hard question. Four questions do most of the work, and every one of them is answerable from paper.

First, ask what the rule reads. Every document sets out at length what a rule does. The narrower question is which inputs reach it, and the answer establishes where the approach is blind. Blindness is the property that survives every change of market conditions. Second, ask how many changes of state the period contained and at what contracted rate each execution is charged. The count and the rate multiply into a real cost, as the arithmetic above works out, and neither number is usually presented as a cost.

Third, ask for the exposure carried, gross and net, and not for the cash posted. Two numbers, not one. A single figure cannot describe a book that holds positions in both directions. Fourth, ask what the document says about changing the rule: who may, on what authority, and what gets recorded. Changing the rule is the question treated at length above, and it is the one most readers never reach.

The everyday version is a household that sits down once, in a calm week, and decides to move a fixed sum into savings on the day the salary arrives, before anything else is paid. The household's decision is a written rule, and it is systematic in exactly the sense used here. Its inputs are the date and the amount, and it therefore cannot see that the roof needs repairing this month. Whether the household should keep it or override it in a given month is a separate question. The moment the household starts deciding month by month, it has a person deciding, and the rule it wrote for itself in that calm week has quietly stopped being a rule.