How to Write a Rebalancing Policy That Can Be Checked

A rebalancing policy settles three things in writing: what fires a rebalance, where the portfolio is traded back to, and who may waive the rule. The first is a date, a distance from the policy weight, or both. Write only the first and the policy fires exactly on time, then hands the size of every trade to whoever is at the desk.

A mandate is already in place. The investment policy statement settled the weights and the range the portfolio is allowed to move inside, and it settled that a range can be crossed on a quiet Tuesday with nobody trading anything. A rebalancing policy is the maintenance rule for that range. Writing one is a procedure, run in order, and every step below produces something the next step consumes.

Start with the picture that makes the whole thing obvious. A household buys two sacks of rice and one sack of dal, and puts them in the store cupboard in that proportion because that is what the household eats. Nobody touches the cupboard for six months. The household ate through the two at different rates, so at the end of six months the proportion in the cupboard is not two to one any more. Nothing was decided. The shape changed anyway.

A portfolio does exactly this, and faster. Prices move weights on their own, so a portfolio left alone stops being the portfolio the document authorised, and rebalancing is the maintenance of a decision already taken rather than a fresh decision about what to hold. Maintenance rather than fresh decision is the reason the policy is written down at all, and losing hold of that distinction is how rebalancing policies get quietly abandoned, one reasonable-sounding waiver at a time.

Everything below runs on the Anantara Multi-Asset Portfolio, an invented Rs 500 crore discretionary mandate held by an invented charitable endowment. Its committee is chaired by Rukmini Deshpande and the mandate is run by Faiz Ahmad Ansari. Its policy weights are equity 60.0 per cent, or Rs 300 crore, fixed income 30.0 per cent, or Rs 150 crore, and cash 10.0 per cent, or Rs 50 crore. Its equity range is 50 to 70 per cent. Every figure in this guide is worked from those five numbers and from nothing else.

What does a rebalancing policy actually have to decide?

Three things, and they are genuinely separate. A rebalancing policyThe written rule that says when a portfolio is brought back towards its stated weights, how far back it is brought, and who may set the rule aside. that names only the first of them is the one most often met, and it is the one that produces arguments.

The first decision is the trigger. The second is the destination, meaning where the portfolio is traded back to once the trigger has fired. The third is the waiver route, meaning who is allowed to not do it and what they have to write down when they do not. A policy carrying only a trigger fires at exactly the right moment and then leaves the size of every trade undecided. An undecided trade size is left to whoever happens to be at the desk that morning.

Step one: what is the policy maintaining?

The first line the policy writes is not a rule. The line states what is being maintained, and the statement is copied out of the mandate rather than settled here.

For the Anantara Multi-Asset Portfolio the line reads that the policy maintains the equity policy weightThe weight the governing document states the portfolio is run at, as distinct from the weight it happens to be sitting at on any given day. of 60.0 per cent inside a stated range of 50 to 70 per cent, together with the fixed income and cash weights that follow from it. Step one ends there. The step consumes the mandate and produces a sentence.

Every later step is then plainly in service of a weight that was agreed somewhere else, so writing this line first keeps the rest of the procedure from turning into a fresh argument about the allocation. A committee that starts at step two, with no maintained target named above it, will find that every rebalance reopens whether 60.0 per cent was right in the first place. Reopening the allocation is a different meeting.

Step two: what fires the rebalance, a date or a distance?

Step two picks the trigger and writes it down. There are two kinds and they answer different questions.

A calendar triggerA rule that fires on a stated date or at a stated interval, without reference to where the portfolio's weights have got to. fires on a date, whatever the portfolio has done. The last working day of March, or every quarter end, or the first Monday of the year. A calendar trigger consumes a calendar and produces a list of dates known in advance. A threshold triggerA rule that fires when a weight has moved a stated distance away from the policy weight, without reference to what the date is. fires on a distance from the policy weight, whatever the date. A threshold trigger consumes a daily or monthly weight reading and produces a firing whose date nobody knows in advance.

The two triggers answer different questions, so a policy carrying both must state which one wins on a day when they disagree. They will disagree. A quarter end arrives and the weights have barely moved; or a weight touches its edge in the second week of a quarter. Both are ordinary. A document with no line for the disagreement is not ordinary.

Neither trigger is the better one by construction. Whether one produces a better result than the other over a long period is covered separately. Step two produces a written trigger and, if there are two, a written precedence line, and nothing else.

Step three: how far must the portfolio move to touch an edge?

Now the arithmetic starts, and step three is the one committees skip. Step three takes the range written in the mandate and converts each edge into the market move that would actually reach it.

A range written in percentage points says nothing about how often it will fire. Ten points sounds tight to one person and loose to another, and neither of them is reasoning from anything. A committee that signs a range without converting its edges into the moves that reach them has chosen a trigger frequency without knowing what it chose.

Here is the conversion on the Anantara Multi-Asset Portfolio, with nobody trading anything and only the equity sleeve moving. To sit at 70 per cent, equity has to be seven parts against three parts of everything else. Everything else is Rs 200 crore, so equity must reach Rs 466.67 crore, or Rs 4,66,66,66,667/-. The gain needed is Rs 166.67 crore on a sleeve that started at Rs 300 crore, or 55.6 per cent. To sit at 50 per cent, equity must match the Rs 200 crore of everything else exactly, so it must fall to Rs 200 crore. The fall needed is Rs 100 crore, or 33.3 per cent of the sleeve.

Look at what step three produced. Two rupee figures, Rs 100 crore and Rs 166.67 crore, and two sleeve percentages, 33.3 per cent and 55.6 per cent. The range itself has not changed. The change is that the committee can now read the range as a distance rather than as a phrase.

Step four: is a symmetric band symmetric in reach?

No, and this is the most useful point in the whole procedure. Step four takes the two figures step three produced and sets them beside each other.

The tolerance bandThe stated distance a weight is allowed to sit away from its policy weight before the policy calls for a trade. on the Anantara Multi-Asset Portfolio is ten percentage points either side of 60.0 per cent. The band reads as a symmetric rule, and every committee that signs one reads it that way. The two moves that reach it are Rs 100 crore downward and Rs 166.67 crore upward. The upper edge is 1.67 times as far away as the lower one, so a band that is symmetric in percentage points is not symmetric in the market moves that reach its edges.

The reason is arithmetic rather than anything about markets, and it is the same reason a share of a moving total behaves oddly everywhere else. A weight is a fraction whose denominator moves when the numerator does. When equity falls, it drags the total down with it, so the fraction shrinks on both ends at once and the weight falls faster than the sleeve did. When equity rises, it drags the total up with it, so the fraction is fighting its own denominator and the weight rises more slowly than the sleeve did. Going down is helped; going up is resisted.

The same thing happens on a move nowhere near either edge. Moving the equity sleeve by Rs 30 crore, ten per cent of itself, in each direction is the cleanest proof there is.

There is a second route to the same two edges, and running it is the check that step three was done correctly. Ask instead how far equity has to outrun everything else. To sit at 70 per cent, equity has to have grown by a factor of seven over six against three over four for everything else, a ratio of 1.556 times. To sit at 50 per cent, equity has to have grown by five over six against five over four for everything else, a ratio of 0.667 times. Both routes land on Rs 466.67 crore and Rs 200 crore, so the arithmetic holds.

Step four hands the committee one small grid. The same two edges written three ways, with the ratio between them in the last column, and it is the last column that the minutes need.

Step five: where does the portfolio get traded back to?

The trigger has fired. Step five decides where the portfolio is traded back to, and it is a wholly separate decision. Three destinations are ordinary. Back to the policy weight of 60.0 per cent. Back to the near edge of the band. Part of the way back, at some stated fraction.

Here is the trap sitting inside that list, and it is easy to write by accident. The trigger point and the destination are the same point, so a policy that triggers when the portfolio touches the edge and then trades it back to that same edge executes nothing at all. The rule fires, the desk computes the trade, the trade is zero, and everybody files a report about a rebalance that did not happen.

None of this argues against trading back to an edge. The argument is for noticing that the trigger point and the destination are two separate lines in the document, and that writing them identically produces a policy which is perfectly consistent and completely inert. Where a policy triggers on crossing an edge rather than touching it, the same destination gives a small non-zero trade instead.

Step six: how large is the trade the rule produces?

Step six consumes the destination step five chose and produces a rupee figure. A rule producing a trade the portfolio cannot execute in the time the rule allows is not a policy, so step six decides whether the policy is operable at all. The size is arithmetic rather than judgement.

Take the upper edge first. Equity has run to 70 per cent, so equity is Rs 466.67 crore and the portfolio stands at Rs 666.67 crore. Sixty per cent of Rs 666.67 crore is Rs 400 crore. The trade sells Rs 66.67 crore of equity, or Rs 66,66,66,667/-, and that is 10.0 per cent of the portfolio as it stands at that moment.

Now the lower edge. Equity has fallen to 50 per cent, so equity is Rs 200 crore and the portfolio stands at Rs 400 crore. Sixty per cent of Rs 400 crore is Rs 240 crore. The trade buys Rs 40 crore of equity, or Rs 40,00,00,000/-, and that is also 10.0 per cent of the portfolio as it stands at that moment.

Both trades are the same share of the portfolio even though the moves that produced them were nowhere near the same size. The match is not a coincidence and it is not deep. The trade back to the policy weight is simply the gap in percentage points multiplied by the total at that moment, and at either edge the gap is ten points. The rupee amounts differ because the totals differ.

The general form removes any need for the desk to work anything out, so the general form is worth writing into the policy itself. The trade is the distance from the current weight to the chosen destination, in percentage points, applied to the portfolio total on the day. Trigger at 72 per cent on a Rs 714.29 crore portfolio. The trade back to the 70 per cent edge is two points, or Rs 14.29 crore. The trade back to the 60.0 per cent policy weight is twelve points, or Rs 85.71 crore. Six times the size, from the same trigger, on the same day.

Step seven: who may waive it, and what gets written down?

Step seven writes the waiverA recorded decision not to follow the rule on a particular occasion, together with who took it and why. route. Name who may set the rule aside on a given occasion, and name what they must record when they do: the date, the weights on that date, the reason, and the date by which the position will be looked at again.

A waiver route reads like administration and is not. A rebalancing policy with no waiver route gets waived anyway, and the waiver then leaves no trace at all. The missing trace is worse than the waiver itself. A committee reviewing the year afterwards sees a rebalancing record with a gap in it. A deliberate decision and an oversight look identical from the outside, so the committee has no way of telling them apart.

Think of the household again. A household that decides to skip one month of its recurring savings and says so at dinner has made a decision. A household where the transfer silently fails and nobody notices has had one made for it. The money position is the same in both cases and the two are not remotely the same event. A traced waiver is worth more to a committee than an untraced compliance. The committee can review the first and cannot even see the second.

How often would this policy have fired in the stated year?

Everything so far was drafting. Now run the finished rule against the one twelve month period the record covers, the only period any figure worked above belongs to.

The worst move the Anantara Multi-Asset Portfolio recorded in that stated twelve month period was a fall of 9.7 per cent from its highest point to its lowest. The fall is Rs 48.5 crore, or Rs 48,50,00,000/-. Charge every rupee of that fall to the equity sleeve. Charging it all to equity is the harshest reading the record permits and certainly not what happened. Equity becomes Rs 251.5 crore and the portfolio becomes Rs 451.5 crore. The equity weight becomes 55.7 per cent, cash rises to 11.1 per cent and fixed income to 33.2 per cent, and the three sum to 100 per cent as they must.

Every one of those weights sits inside the mandate, so a threshold rule on this band would have fired zero times over the stated twelve months. A quarterly calendar rule would have fired four. The observed worst move was Rs 48.5 crore against the Rs 100 crore that would have been needed to touch the lower edge, or 48.5 per cent of the distance. The band was never approached.

Two more things follow, and both matter more than the count itself. The first is that a single window says nothing about what the rule would do over a longer one. A drawdown is measured between a peak and a trough inside a stated window, and a different window gives a different figure. The window is quoted every single time the figure is used.

The second is about the record itself. The portfolio recorded turnoverThe share of a portfolio replaced over a stated period, which counts every trade regardless of the reason it was placed. of 34 per cent over the same stated twelve months, or Rs 170 crore of a Rs 500 crore portfolio. A threshold rule on this band fired none of that, and a quarterly rule at these weights would not have produced anything close to it. Whatever the Rs 170 crore came from, on this evidence it did not come from rebalancing to this band.

One caution about the drawing above. The record locks two points and no others: the portfolio started the period at its policy weights, and its worst peak to trough fall inside the period was 9.7 per cent. The path between those two points is not in the record and is not drawn. The count of zero firings rests on the harshest single reading the record permits, not on a reconstructed path.

Step eight: what does the policy cost, and where does that appear?

The last step is the one nobody enjoys writing. Every rebalance trades, and every trade costs something. The cost is real, the case record contains no number for it, and a policy written without acknowledging it will look free in every review it ever appears in.

Where the cost does and does not appear is worth marking. The portfolio return for the stated twelve months was 14.2 per cent and the composite benchmark returned 12.6 per cent, both figures invented and both belonging to that one period. Neither of them says what the trading cost. A committee reading the return line alone sees a rebalancing policy that appears to cost nothing at all. Costing nothing is the one thing it is certain not to be.

So step eight writes two sentences. The first names the cost as real and says where it will be reported from. The second states plainly that the policy carries no estimate of it. The absence is then on the record as an absence rather than passing for a zero.

What happens when the whole sequence is run on one portfolio?

Here is every figure the eight steps produced for the Anantara Multi-Asset Portfolio, on one table, from a policy weight of 60.0 per cent and a range of 50 to 70 per cent and nothing else.

The last two rows read together carry the whole argument. The trades are the same tenth of the portfolio and the moves that produced them are Rs 100 crore and Rs 166.67 crore, two thirds apart. A committee that signed the band without step three would have had the bottom row and none of the rows above it.

Notice three things the table does not carry and the eight steps never once produced. Whether 60.0 per cent was the right weight is settled in the investment policy statement, what any of these trades would cost to place is settled at execution, and whether following the rule helped is settled under performance evaluation.