Efficient Market Hypothesis vs Adaptive Markets Hypothesis

Both hypotheses say prices respond to information. The two disagree about one thing: whether the degree of that response is fixed or varies with conditions. Efficiency treats it as a standard a market meets or misses. Adaptive markets treats it as a quantity that moves. Nothing observable separates them, so no verdict is available, and the reason none is available is structural rather than a shortage of data.

A pair of rival ideas is only worth comparing if evidence could pick between them. The evidence test comes before any comparison, and this pair fails it. The reason was settled when the three forms of efficiency were set out: any check on whether a price is right is also a check on the model that said what right would have been, so a failed check never shows which of the two halves broke. Because no measurement reaches past that problem, the honest comparison is about what each idea is useful for rather than which one is true. Usefulness is a smaller question than truth, and it is the only one of the two that can actually be answered.

A comparison needs three things before it can be settled at all. The pair here meets the first of the three and misses the other two.

What does each hypothesis actually claim?

A comparison in which one side is set up to lose teaches nothing. Take each at its strongest. The efficient market hypothesisThe claim that prices already reflect a named set of information, so nothing in that set is left over to act on. says that a price already reflects a stated set of information. Paul Samuelson, in Industrial Management Review in 1965, showed the mechanical version of this: if a price already carries everybody's best expectation of what comes next, then what comes next has to arrive as a surprise, and a series of surprises looks random. Eugene Fama, in the Journal of Finance in 1970, turned that into something testable by naming which information set is meant. The claim is not that everybody is clever. The claim is that competition between people trying to be first has already pushed the price to where the named information puts it.

The adaptive markets hypothesisThe view that how completely prices reflect information is not fixed, but rises and falls with conditions and with how many people are competing., set out by Andrew Lo in the Journal of Portfolio Management in 2004, does not deny any of that. Lo says the degree to which efficiency holds is not a constant. Competition varies, the number of people chasing the same information varies, and conditions change what people can bear to do. So efficiency, on this view, is high in some stretches and low in others, and asking whether a market is efficient is like asking whether a road is busy: the honest answer names a time.

Here is the difference outside finance first. A street has four vegetable sellers. On a Saturday morning with sixty buyers walking the row and comparing, the price of a kilo of tomatoes at the four stalls sits within a rupee of each other. Anyone pricing high sells nothing. On a wet Tuesday afternoon with four buyers, the four prices drift apart and stay apart for hours. The efficiency reading is that each price reflected what was known at that stall at that moment; the adaptive reading is that the tightness of the four prices moved with how many people were comparing them. Neither sentence is wrong. The two sentences describe one street.

What do the two agree on, and how large is that agreement?

Before the dispute, count the zone of agreementThe claims both hypotheses make in the same words, which turns out to be nearly all of them.. The dispute reads very differently once the small share of the ground it covers is clear. Both say prices respond to information as it arrives. Both say prices are moved by competition between people trying to use that information first. Both say repeatable free lunches are rare. Rare is a much weaker statement than impossible. Both say costs and taxes come out of any effect anybody measures. Both accept that a check on a price is also a check on a pricing model. And neither of them tells any reader what to hold.

Six claims shared against one disputed. The two hypotheses are not rival descriptions of a market; they are one description with a single knob set differently. Count the shared claims yourself rather than taking the count from a summary. Almost every popular account of this argument spends its space on the one disputed claim and leaves the impression that the six shared ones are also in play. The six shared claims are not in play. Sanford Grossman and Joseph Stiglitz, in the American Economic Review in 1980, showed why even the strict position cannot mean perfect reflection: if prices already revealed everything, nobody would be paid for gathering information, and if nobody gathered it, prices could not reveal it. Some inefficiency has to survive to pay the people whose work does the reflecting.

Where do they genuinely disagree?

The single dispute is about the status of the degree. For the strict position, reflection of a named information set is a property a market either has or does not have, and the whole point of writing it that way was to make it possible to be wrong. For the adaptive position, that degree is a quantity that rises and falls, so it can be high in one stretch and low in the next without anything being contradicted.

Now look at what the invented Palash 100 record offers as the closest thing to a measurement of that. Turnover in the record ran 3.1 times its eight quarter median in the quarter the index peaked at 131.0, and 0.4 times its median in the quarter it bottomed at 104.0. Activity was heaviest near the top and nearly absent near the bottom. The adaptive account expects exactly that pattern, given its claim that the number of people competing over the same information changes. And the strict account expects exactly the same pattern. A quarter in which a great deal of information arrives is a quarter in which a great deal gets traded. The measurement is consistent with both, and consistency with both is not evidence for either.

What evidence would settle this, and does any of it exist?

The settling question, asked properly, has a definite answer, and that answer is the finding. Which measurement would come out one way if the strict position were right and the other way if the adaptive position were right? To say that a price reflected information less completely in one quarter than another requires knowing what the price should have been in each quarter. The price that should have been is an unobservableA quantity an argument depends on that no measurement reaches, so both sides can assume their own version of it.. Every candidate for it comes out of a pricing model, and the moment one is used, a disagreement between the model and the price can be blamed on either side.

So there is no settling evidenceData that would come out one way under one hypothesis and the other way under the rival, rather than fitting both. here, and the shortage is not a shortage of data. Another twenty quarters of the invented index would give both accounts twenty more observations to fit comfortably. The disagreement is structural rather than empirical. No better study is on the way, and no study could settle it. A verdict would have to be smuggled in from somewhere other than the evidence, so the absence of one is worth more than a verdict would be.

What does each one say about the same index path?

Put both against one object and watch how little separates them. The invented Palash 100 opens at 100.0 and records eight quarter ends: 118.0, 131.0, 112.0, 104.0, 116.0, 124.0, 121.0 and 127.0. The peak is Q2 at 131.0, a level 31.0 per cent above the open. The low is Q4 at 104.0, down 20.6 per cent from the peak. From that low the path recovers 23.0 points to 127.0 by Q8, a rise of 22.1 per cent. Nine numbers in total, and every reading below has to fit all nine.

Read as eight separate moves rather than as one path, the record turns over from rise to fall and back again four times in the seven chances it gives, and neither account finds that surprising.

The strict reading of those nine numbers goes like this. At each point the level reflected the public information available when it was struck. The fall from 131.0 to 104.0 does not mean the 131.0 was a mistake; it means information arrived between the two dates that was not available at the first. Nothing in the nine numbers contradicts a word of that.

The adaptive reading goes like this. The degree to which those levels reflected information varied across the eight quarters. Competition was intense while the index climbed and thin while it fell. The turnover multiples of 3.1 and 0.4 are consistent with that. Nothing in the nine numbers contradicts a word of that either. Both readings fit all nine levels, neither is contradicted by any of them, and no tenth quarter would separate them. Separating them would require knowing what the level should have been. The absence of a separating measurement is the finding.

Which one is more useful for which subject?

Since neither can be shown to be right, the useful question is which one to hold in mind for each of the neighbouring subjects, and the answer changes by subject. For a market anomalyA documented departure from what a stated efficiency claim implies, which needs that stated claim to exist before it can be called a departure at all., and for the limits to arbitrage that let mispricing survive, the strict position supplies the benchmarkA fixed standard a market either meets or does not, which is what makes a departure from it measurable. without which the word anomaly means nothing. Narasimhan Jegadeesh and Sheridan Titman, in the Journal of Finance in 1993, and Werner De Bondt and Richard Thaler, in the Journal of Finance in 1985, could only report departures because there was a precise statement to depart from.

For speculative bubbles, for the reflexive loop in which belief changes the thing believed about, and for investor sentiment measured at the level of a whole market, the adaptive account supplies the vocabulary. All three are stories about conditions changing rather than about a fixed state being met or missed. The strict position is the measuring stick and the adaptive position is the description of the weather, and a reader who insists on one of them for every subject will read half of what follows badly.

Does either hypothesis license a course of action?

No, and the reasons differ enough that they are worth stating one at a time. The strict position is a claim about what a price reflects rather than a claim about what anybody should hold, so it does not license sitting still. Saying that the degree of reflection varies is not saying when it is low, by how much, or in a direction anybody could reach, so the adaptive position does not license acting.

An account of why a price moved is an explanation and never a trading signal. Three separate reasons keep that honest, and any one of them would be enough on its own. First, an effect is almost always measured before costs. In the invented Palash record the most active fifth of investors gave up 4.1 points a year to dealing charges, spread and tax. The least active fifth gave up 0.3, and gross returns across all five groups sat inside 0.3 points of each other at 11.2 down to 10.9 per cent. Net returns ran from 10.9 down to 6.9 per cent, a spread of 4.0 points, and that spread is wider than most documented effects are before costs. Second, a published effect has been read by everybody else who read the paper, so what it did before publication is not what it does after. Third, the same obstacles that let a mispricing survive are the obstacles that stop a reader capturing it, a point Andrei Shleifer and Robert Vishny set out in the Journal of Finance in 1997. The explanation and the obstacle are one fact seen twice.

What does the whole comparison look like in one table?

Set out side by side, with the shared assumptions written once on a shared row rather than twice, the argument shrinks to its last two rows. Everything above the double rule is agreed, and everything the two of them fight about sits underneath it.

What is to be done with a disagreement nobody can settle?

A practitioner uses the pair as two instruments rather than as one contest. Devika Rao, the adviser at the invented Palash Advisory Services Private Limited, reaches for the strict position when the question in front of her is whether an explanation for a move is even needed: if the ordinary account already covers it, an exotic one is not required. She reaches for the adaptive framing when a client asks why a stretch of months felt so different from the stretch before. A description in which conditions change is honest about something a fixed state cannot express.

Meera Sundaram, who invests on her own account with no committee behind her, gets the same use from the pair without any of the machinery. Her holding cost Rs 13,00,000/- and stood at Rs 12,46,000/- when the eight quarter valuation was struck, a fall of 4.2 per cent, and no reading of either hypothesis changes that arithmetic or tells her what to do next. The pair changes what she can honestly claim to know, and that is a real change even though it moves no money. She can say the market got noisier, and she cannot say she can tell when. A lender reading a borrower's holding, an analyst writing up a stretch of prices and a household reviewing its own decisions all land in the same place: the language for describing what happened improves, and the ability to say what happens next does not.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited, the Palash 100 index and the Palash investor record are invented.
Educational material. Not advice on any investment, tax, budget or market position.