Excess Trading: How Activity Erodes Returns

Excess trading is activity beyond what the information justifies. The evidence is not that busy investors picked worse: across the five turnover groups the gross returns sit inside 0.3 points of each other. The difference is that they paid up to 4.1 points a year to act on picks that were no better, leaving net returns 4.0 points apart.

The whole finding rests on pulling apart two quantities that a single return figure quietly welds together. The pick is one thing. The cost of acting on the pick is a completely separate thing. A reported return has already subtracted the second from the first before it is ever seen, so a low number says nothing about which of the two produced it. Once the two are separated, the finding stops being a story about skill and becomes arithmetic anybody can check.

What is excess trading, and excess against what?

The word excess is doing all the work in the phrase, and on its own it does none. Excess means more than something, so until the something is named the phrase is just a way of calling somebody busy. So the first job is to fix the benchmark, and the benchmark that works is narrower and duller than the one most readers arrive with.

Three benchmarks are available and two of them are unusable. The first is other people: excess means trading more than the average person trades. Comparing against other people makes half of any group excessive by construction, and it would keep doing so if everybody halved their activity tomorrow. The second is a fixed number: excess means more than some count of trades a year. A fixed number is worse. The count would have to be made up, and a holding that genuinely needs rearranging twice would be charged with excess while a holding rearranged eleven times for no reason would pass. The third works. Excess tradingActivity beyond what the available information justifies, rather than activity above any particular count or above what other people do. is activity beyond what the available information justifies, and it works because it is the only one of the three that can be wrong.

Look at why the third one can be wrong. Being open to being wrong is what makes a benchmark worth having. If a busier investor really is acting on information the quieter one does not have, then the busier one's picks should turn out better before costs are taken off. The better picks are a prediction, and predictions can fail. If it holds, activity was justified and the word excess does not apply. If it fails, the extra activity bought nothing and was charged for. The test is not an opinion about how much anybody should be doing; it is a comparison between what the activity produced and what it cost.

The measuring stick throughout is turnoverThe share of a holding that changes hands over a year. At 100 per cent, the dealing across the year adds up to the size of the whole holding.. Turnover is how much of a holding changes hands in a year as a share of its size. Turnover of 100 per cent means an amount equal to the whole holding was bought and sold over the year. The measure is a blunt instrument on purpose. Turnover does not care what was dealt in, why, or whether the person felt confident, and that indifference is exactly what makes it comparable across sixty different people who were each doing their own thing. Terrance Odean set the question out in these terms in Do Investors Trade Too Much in the American Economic Review in 1999, and Brad Barber and Terrance Odean measured it against a large body of accounts in Trading Is Hazardous to Your Wealth in the Journal of Finance in 2000.

What did the five turnover groups actually pick?

The invented Palash decision log sorts its sixty investors into five groups of twelve by how much they dealt, and then reports what each group got. The five groups ran at annual turnover of 9, 34, 71, 128 and 210 per cent. The quiet end is somebody who moved less than a tenth of the holding in a year. The busy end is somebody who turned it over twice. A range of twenty three times is a wide spread of behaviour by any standard, and it is exactly the spread needed to find out whether activity buys anything.

Now the part that decides the question. Before any cost is taken off, what did each group pick? The gross returnThe return a holding produced before dealing charges, the spread and tax are taken out of it. is the answer to that question. Gross return is what the choices produced before anybody was charged for making them. Across the five groups the gross returns were 11.2, 11.0, 11.1, 10.9 and 11.0 per cent. Read those five numbers again in order. The five numbers are the finding, and they are easy to skim past. The busiest group, dealing at twenty three times the rate of the quietest, produced 11.0 against 11.2. The whole range of picking ability across a twenty three fold range of activity is 0.3 percentage points, and the ranking inside that range is scrambled rather than sloped.

The full table below carries all four columns in one place, and what follows is a series of comparisons between them. Costs cover dealing charges, the spread and tax together. The net column is the gross column with the cost column taken off it, and the subtraction can be checked on every row.

Costs are given in points rather than as a percentage of anything. A point taken off a return is what a reader can actually work with.

Why are the two spreads the whole finding?

Set the two ends of the table beside each other and the argument makes itself. On the gross side the five groups run from 10.9 to 11.2, a spread of 0.3 points. On the net side they run from 6.9 to 10.9, a spread of 4.0 points. Same sixty people, same eight quarters, same holdings available to them. The only thing that happened between the first spread and the second is that each group was charged for its own activity.

The shape has a name in practice, and recognising it transfers to any table built this way. When outcomes separate but inputs do not, the effect is a cost effect and not a skill effect. The test is always to find the number as it stood before the charge. If picking really differed across these groups, it would have to show up on the left of that chart, where picking is drawn with nothing taken off it. Picking does not show up. Whatever separated these five groups by 4.0 points a year had already finished its work before any question of judgement arose.

There is one trap in this arithmetic and it catches careful readers rather than careless ones. The busiest group paid 4.1 points and the quietest paid 0.3, so the difference in cost is 3.8 points. The difference in net return is 4.0 points. The two figures are not the same number, and neither is a mistake. The extra 0.2 comes from the gross side: the quietest group also happened to pick 11.2 against the busiest group's 11.0, so it starts 0.2 ahead before any charge. The 0.2 of picking added to the 3.8 of paying gives the 4.0 of keeping.

Where does the cost come from, item by item?

A cost of 4.1 points a year sounds like a fee somebody is charging, and it is not. The cost is three separate things that happen to arrive together in one column, and naming them separately matters because they respond to different things and one of them is not obvious at all. The net returnWhat is left of a return after dealing charges, the spread and tax have been taken out of it. is what survives all three.

The first item is dealing charges: what the intermediary takes for handling the transaction, plus whatever levies ride along with it. Dealing charges are the visible item. A charge appears on a contract note, it is denominated in money, and a person who deals twice as often pays it twice as often. The second item is the spread. The spread is the gap between the price at which something can be bought and the price at which the same thing can be sold at the same moment. Crossing that gap costs money even when no charge is shown anywhere. The spread is the item that catches people out. The spread is a cost that never appears as a line on any statement and is paid in full on every round trip regardless.

The third is tax. Whether a realised gain is taxed, at what rate, over what holding period and with what set-offs is decided by rules that sit entirely outside this subject area and change from time to time. Only the structural point matters for the argument. Realising a gain can bring forward a tax event that holding would not have brought forward, so activity can move a tax cost from later to now even where it changes nothing else. Anybody working out what that means for a real decision reads the position in force at the source.

There is an honest gap in that picture. The log records the three together and does not split them, so no split between them can be stated. The log does support the relationship between the total and the activity, and that relationship is close to a straight line.

Read that slope back into the net column and the whole table stops being surprising. If gross return is essentially a constant near 11 per cent and cost is essentially a straight line rising with turnover, then net return is a constant minus a straight line, and a constant minus a rising line falls. The net column falls in exactly that way: 10.9, 10.4, 9.6, 8.4, 6.9, in the same order as turnover every time, with no exceptions and no crossings. The gross column, by contrast, has the worst figure in the fourth group and identical figures in the second and fifth. A column of noise looks like that.

What is Tax-Loss Harvesting, and when is selling deliberate rather than excessive?

Here is the objection that arrives the moment somebody hears that activity costs 4.1 points a year. Plenty of selling is done on purpose, for a reason that has nothing to do with restlessness, and it would be absurd to file all of it under excess. The clearest case is tax-loss harvestingSelling something at a loss on purpose, so that the loss is realised and can be set against a gain when tax is worked out.. A holding is standing at a loss. Somewhere else in the same record a gain has been realised. Selling the loser turns a paper loss into a realised one, and a realised loss can be set against a realised gain when the tax is worked out. The sale is not a change of mind about the holding. The sale is administration with a date on it.

Notice what has just happened to the turnover figure. The harvesting sale adds to turnover exactly as any other sale does. The harvesting sale costs dealing charges and crosses the spread exactly as any other sale does. Nothing in the number marks it as deliberate. Deliberate selling and excessive selling are indistinguishable inside a turnover figure, and they separate on one question that lives outside it: was a reason written down before the sale rather than after it? The written reason is the whole test, and a test about records rather than about intentions is a test somebody can actually run.

A household handles a similar problem in a familiar way. A person clearing out a cupboard before a move has a reason, and the reason existed before the cupboard was opened. A person clearing out a cupboard because they are restless on a Sunday afternoon produces exactly the same pile of bags at the door. A photograph of the pile does not reveal which Sunday it came from. The list written before the cupboard was opened does. The Palash decision log carries a written reason on 84 of its 240 decisions, or 35.0 per cent, so on 156 decisions the test simply cannot be run in either direction.

Two cautions belong with this, and both matter more than the neat test does. The first is that the tax treatment of a realised loss, what it may be set against and over what period, is set by rules that sit outside this subject area and change without warning, and anybody acting on it reads the current position at the source. The second is subtler and is where the neat test earns its keep. A sale made for a tax reason is still a sale of something. The reason justifies realising the loss. The tax reason does not, on its own, justify the decision about what to hold afterwards, and a reason recorded in advance is precisely what lets those two decisions be read apart later.

What is Informed Trading, and how is it told apart from noise?

The second objection is stronger than the first. Some activity is informed tradingDealing on something not yet in other hands, so the price has not yet adjusted for it.. Informed trading means dealing on something other people do not hold yet, so the price has not moved to reflect it. Dealing on something nobody else holds is not excess by any definition. Informed trading is the mechanism by which a price comes to reflect anything at all, and a record with none of it in would be a record of a market that never learned anything. So the honest question is not whether informed trading exists. The honest question is how it would be recognised in a log.

A log would not recognise it. The answer is uncomfortable and worth sitting with. A decision log holds the date, the direction, the instrument and the amount. The log does not hold what the person knew at the moment of pressing the button, and it does not hold whether what they knew had already reached everybody else. Both of those live entirely outside the record. Informed trading and restless trading produce identical rows in a log. Knowledge separates them, and a log stores actions. The quintile result is therefore stated as a cost result and never as a claim that anybody was uninformed: the record was never in a position to tell.

There is one measurement in the log that gets close without ever getting there, and it is worth reading carefully because it is so easy to over-read. Of the 96 buys recorded across the eight quarters, 41 followed a media mention within three days, or 42.7 per cent. Over the same period, about 11.0 per cent of the eligible list was mentioned at all in a given week. Buying is therefore running about four times as heavy on the mentioned part of the list as the mentioned part's own share would suggest. Attention is doing work.

Now be careful. Here is the exact point where a reader is tempted to draw the conclusion the record cannot support. A mention is not information. A mention is the opposite of information in the sense that matters here. A mention is the part everybody saw at the same moment, and the part everybody saw is the part least likely to be sitting unreflected in a price. So the 42.7 per cent figure is evidence about where attention went, and it is not evidence that anybody was trading on something others did not hold. Nor is it evidence that they were not. The record simply has no field for it, and inventing one is how a careful finding turns into an accusation.

How is a holding reviewed for concentration context, without being told what to hold?

Turnover is one way a holding drifts away from what somebody meant. ConcentrationHow much of a holding sits in one place, expressed as that place's share of the whole. is the other, and it is quieter. Concentration can change without a single decision being taken. A position that rises faster than the rest becomes a larger share of the whole while its holder does nothing at all. A drift that arrives without a decision is worth a procedure rather than an impression, and the procedure below is deliberately silent on what any share should be. The procedure only asks what a share would have to mean for it to be deliberate.

How to Review a Portfolio for Concentration Context

The order matters more than it looks. Doing the arithmetic before asking the question stops the question being answered by whatever the arithmetic happened to produce, and writing the answer down is what turns this from a feeling into something that can be read back next time.

Run it on the invented case. At 30 September the four positions in Meera Sundaram's holding stood at Rs 3,36,000/- in the Vindhya index scheme, Rs 2,55,000/- in the Nilgiri mid-cap scheme, Rs 4,60,000/- in Suvarna Chemicals Limited and Rs 1,95,000/- in Kesari Logistics Limited. The four positions come to Rs 12,46,000/- in total against a cost of Rs 13,00,000/-, so the holding as a whole was down Rs 54,000/-, or 4.2 per cent. Step two turns those four amounts into four shares.

Step three has to be worded carefully. The question is not whether 36.9 per cent is too much. A judgement of that kind comes from a different subject area. The question is what would have to be true for 36.9 per cent to be deliberate. A holder who says the share is deliberate is committing to something checkable: that this position was chosen to be the largest, at roughly this size, for a reason that can be stated. A share is deliberate when a reason for its size exists in advance, and it is an accident when the only explanation available is that this one went up and the others did not.

Read the history of that particular share and the distinction stops being abstract. On 4 January the holding opened at Rs 12,00,000/- as four positions of Rs 3,00,000/- each, so Suvarna Chemicals was 25.0 per cent of it. On 19 February, after a television segment named it, Rs 1,00,000/- was added the same evening, taking its cost to Rs 4,00,000/- and its share of the Rs 13,00,000/- cost to 30.8 per cent. By 30 September its share of value was 36.9 per cent. Only the middle step was a decision. The other movement happened while nothing was being decided at all, and that is precisely the movement a review of this kind exists to surface.

Step four is the one people skip, and skipping it is what makes the next review start from nothing. Write the answer down with its date. Next time, the comparison is between a question and a recorded answer rather than between a question and a memory, and a memory has the inconvenient property of agreeing with whatever happened since. A household running on one salary is carrying a concentration of exactly this kind, and the useful version of that observation is not a rule about how many salaries there should be. The useful version is the sentence written down last year saying what the single salary was assumed to cover, read back against what it covered.

What does an adviser do with a finding that names nobody?

A measured result meets a person here, and the meeting is easy to get wrong in the direction of being useful. The adviser in the invented case, Devika Rao of Palash Advisory Services Private Limited, cannot walk into a review with the quintile table and tell Meera Sundaram she is trading too much. The table does not say that about Meera and it does not say it about anybody. The table supports a statement about a rate: dealing at this level of activity attracts a cost of about this size, whoever is doing it and whatever they pick.

Notice what the money translation does and does not license. The money figure makes the size of the cost feel real in a way points never do. The money figure does not turn the group result into a claim about Meera. The arithmetic is on the rate, not on her record. If she dealt at 210 per cent this year, then 4.1 points is what that rate costs and Rs 51,086/- is what it comes to on that holding. Whether her picks were good enough to be worth it is a completely separate question, and four logged decisions cannot answer it.

When is high activity the right thing?

Stated plainly: high turnover is sometimes exactly correct, and nothing measured here says otherwise. A holding that has to be rearranged because a goal date is arriving, a liability is falling due, a standing arrangement has drifted and is being put back, or a rule requires it, generates turnover for reasons that have nothing to do with restlessness. So does somebody whose occupation is dealing, for whom the cost is a cost of doing business that is counted before the year starts rather than discovered at the end of it.

The measurement says one thing and only one thing: in this invented cohort the extra activity was not paying for its own cost. The finding is about a price, not about a temperament, and a reader who converts it into a rule about how often anybody should deal has gone further than the record goes. Brad Barber and Terrance Odean connected turnover to overconfidence as a mechanism in Boys Will Be Boys in the Quarterly Journal of Economics in 2001. Overconfidence and optimism sets that mechanism out. None of it is needed to read this table, and the table works as arithmetic on its own.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited, the Palash decision log, the Palash 100 index, the Vindhya index scheme, the Nilgiri mid-cap scheme, Suvarna Chemicals Limited and Kesari Logistics Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.