Scott Sumner  

Why the EMH is truer than supply and demand

Gary Johnson's Bold Attack on ... Krugman's Strange Post on Sola...

My previous post discussed the strangeness of the efficient markets hypothesis. Here I'll defend its utility.

In the field of economics, all models represent simplifications of reality. Thus when we consider whether the EMH is true, it makes no sense to compare it to something like Newtonian physics. Yes, even Newtonian physics is only approximately true, but the approximation is much better than almost anything we observe in economics. So let's compare apples with apples.

Supply and demand has become the archetype model of economics. It's a workhorse widely used to explain the behavior in all sorts of markets, from haircuts and dry clearers to automakers and PC producers. Of course if you read the fine print the model is, strictly speaking, only applicable to a very restricted set of markets, essential grain producers. But almost all economists use it in a much wider range of applications, with justification. It's really, really useful.

But is it true? One of the most important implications of S&D is that producers are price takers. This assumption is what underlies the existence of a supply curve. If firms are not price takers then no supply curve exists. You can't have a supply and demand model without the assumption of firms being price takers.

But are they price takers? Not really. The vast majority of firms, even in highly competitive industries such as laundromats, dry cleaners and pizza shops, could raise prices by 5% and still hold on to a substantial share of their customers. Exxon might not be able to do so, but most small businesses could. This means the supply and demand model is not literally "true."

Fortunately, S&D is incredibly useful, even if not strictly true.

I would argue the EMH is truer that the S&D model. And by EMH I am actually only talking about the assumption that asset price deviations from trend are essentially unforecastable. Specific versions such as the CAPM may be flawed in other ways. However I believe the random walk model is truer than S&D, and also quite useful. But how can we test the EMH?

Many academics look for "anomalies." This is asking both too much and too little. Contrary to widespread belief, the EMH does not claim that a search of 20 million statistical patterns would fail to identify 1 million anomalies that show non-random price movements at the 5% level, or 200,000 at the 1% level. On the other hand, I'd also argue that it's asking too much of academics to suggest they need to find get-rich-schemes that violate the EMH, it should be enough to prove that at least someone has done so (say Warren Buffett.)

As an analogy, an economist looking for signs that the most famous secret in alchemy--turning lead into gold--had been discovered should not have to identify the magic formula, but rather merely show that gold prices are behaving in a way consistent with the fact that someone had discovered this sort of chemical process.

Eugene Fama understood that the only meaningful test of the random walk was to look for evidence that others had found anomalies. The initial tests showed no evidence; mutual funds excess returns were serially uncorrelated, whereas they would be correlated if a subset of investors had found the magic formula. Later work with Kenneth French slightly modified that conclusion. There was some evidence of stock picking ability, but too small to overcome the expense ratios of mutual funds. So now the EMH is only approximately true. But since it's a part of economics, we should have known that all along. No economic model is precisely true.

OK, but is it useful? I see three uses:

1. For ordinary investors, it suggests you'd want to stick to indexed funds.

2. For academics, it suggests that asset prices contain the optimal forecasts, and hence you should use something like TIPS spreads rather than the output of VAR models when trying to identify the optimal forecast of inflation. And you should use the response of asset markets to monetary policy announcements to evaluate the effectiveness of programs like QE, not subsequent movements in the economy.

3. For policymakers, it suggests that central banks and/or bank regulators should not try to identify bubbles. And that central banks should create and subsidize NGDP future markets. And that SEC officials should actually pay attention when whistleblowers bring in evidence that hedge fund returns are inconsistent with the predictions of the EMH (i.e. the Madoff case.)

To conclude, the EMH is a very useful model. It's also more true than the S&D model, as the pricing power of firms in so-called "competitive industries" where S&D is widely applied is actually much greater than the excess returns identified by Fama and French.

From now on any commenter who tells me the EMH is not true, should also tell me whether they think S&D is true, and if so, why it's true.

Comments and Sharing

COMMENTS (19 to date)
David R. Henderson writes:

Excellent post, Scott. Particularly clearly written and reasoned.

Greg G writes:

EMH is a bit like the anthropic principle or the idea of natural rights. It means a lot of different things to different people and, in a weak enough form, it is almost certainly right.

That said, this post offers a well defined and ably defended version of the idea.

Joel Aaron Freeman writes:

I have a question about EMH that I've always wondered about. I've read the Fama paper, and I don't see how EMH precludes significant alpha. That is, the concept of EMH and the concept of beating the market are compatible.

EMH is about informational efficiency. However, information not only must be gathered, it must be interpreted. If I show the same scatter plot to two different people, and ask each of them to draw a regression line, I could get two completely different lines. Similarly, if I show two people the same information about a security, I may get two vastly different valuations, even if both people are "sophisticated."

Thus if a person had an edge in their ability to interpret data for a particular security or a particular industry, they could beat the market so long as they stayed within their domain of expertise. This similar to the adage "write what you know." You can do very well by "investing what you know."

Kevin Erdmann writes:

Excellent post.

Even for a speculator vocationally operating in an anti-EMH framework, inefficiencies are generally sparse and filled with uncertainty. And, profits generally depend on a tendency toward efficiency.

Also, I would say that broader notions based on inefficiency are highly dependent on the use of their own models and an overbearing hindsight bias.

Everyone knows now that home prices were too high in 2005, even though very few took that side of the position in 2005. The reason we know they were too high is because they subsequently collapsed. So, we apply a model that assumes collapse was inevitable. How do we know that is the right model? Because that's what happened.

A robust model of 2005 home prices clearly would have allowed for many possible outcomes. But, hindsight bias leads people to accept models that only allow for collapse as a possible outcome. This is at once a conceptually satisfying and a conceptually bankrupt position to take.

Andrew_FL writes:

I think, phrased the following way, EMH is almost trivially true:

There is no strategy to beat the market consistently and indefinitely.

Stated that way, suggesting that Warren Buffet is an example of a violation of the EMH is a misunderstanding of what the weak sense EMH actually means.

John Hall writes:

Scott, EMH doesn't say that price deviations from trend are unforecastable. Its more a statement that (log) price changes tomorrow can't be forecasted by today's (log) price changes.

Scott Sumner writes:

Thanks David and Greg.

Joel, I can't speak for Fama, but that's what I mean when I say no social science model is perfect. It is absolutely true that the EMH doesn't prove that someone like Buffett might have superior analytical ability--it just assumes it. So then you test the theory and start asking is it correct, roughly correct, or useless? I say roughly correct.

Kevin. Good points. Regarding housing, it will be interesting to see if the Australia/NZ/UK/Canada housing bubbles ever burst.

Andrew, Regarding the "trivially true" comment, I'll just reiterate the areas where I claimed it was useful. That's the bottom line for me.

John, Yes, that wording was sloppy on my part.

ChrisA writes:

Scott, nothing to add to David Henderson's comment, except that a lot of people feel threatened by the EMH hypothesis. By a lot I mean more than just day traders, there are entire financial industries employing hundreds of thousands of well paid people that are predicated on the basis that good research can predictably find arbitrage opportunities. And in Government and Academia we have huge empires of clever people working day and night on developing their own internal models of the economy which differ substantially from the "market" forecast (as you point out with TIPs and inflation). So I predict continual denial of the conclusions you draw from the EMH in terms of policies, even if people will say that they agree with hypothesis itself.

Andrew_FL writes:

@Scott Sumner-I should have been clear, that I do think a statement which is almost trivially true can still be useful. Especially because of the non trivial statements it's truth implies are false.

Chris H writes:

I have a quick question about EMH to make sure I understand it. It is fair to say about it that market are not the theoretical maximum of efficiency that could exist if market actors were maximally rational and informed, but that markets tend towards being as efficient as the most rational/informed actors within them allow? Given the high rewards for finding market inefficiencies, the actors in the market tend to be the most rational/informed people humanity has to offer. Thus improving on market outcomes from outside the market is not possible in the modern world. Is that a good summation of EMH?

Putting it like this rather than simply saying markets are efficient allows for the possibility of more rational agents than humans actually improving market outcomes. For one thing, it helps reconcile pro- and anti-EMH positions, as the anti-EMH notes that market actors are not perfect rational agents, but pro-EMH can point out no other humans will do any better. However if, for instance, a super-intelligent AI was developed it could likely find numerous inefficiencies and corner the market rapidly right? Of course at that point, the market would become as efficient as it's most rational agent (the AI) making EMH true again.

Kevin Erdmann writes:


I don't believe this is the case. Even with EMH, there is a tremendous amount of work to be done in finance. Each client has a given risk profile and time frame, and many decent folks earn an honest living by managing retail investors' portfolios within a basic EMH framework. I would say that most financial planning operates in this framework - and rightly so.

Go to a gathering of CFA's (certified financial analysts) and try to argue against EMH. You'll be met with skepticism.

Further, over the past few decades where finance has supposedly swallowed up the universe with rent-seeking, the main, overwhelming change in the industry has been the deep penetration of low-fee index and near-index funds. The data is overwhelming on this.

In fact, on net, I think you will find that money managers have clients that want them to magically beat the market, and money managers push clients more in the direction of indexing.

Yes, there are fees. But, most managers see those fees as a product of risk-matching, not as a product of pushing clients into unrealistic return expectations.

Joel Aaron Freeman writes:

Thanks for replying, Scott. I think there's a lot of confusion around EMH (I know there has been for me), as well as a lot of misinformation, and I'm glad you're dedicating time to it to shed some light on it.

David N writes:

Which EMH? One shouldn't argue for the weak form of EMH, then reason from the strong form.

Tom DeMeo writes:

I buy the assumption that asset price deviations from trend are essentially unforecastable. I don't buy the idea that this represents any sort of informational efficiency.

John writes:

Tom DeMeo, here is a thought experiment:

1. Assume that asset prices are unforecastable. (Your first sentence.) There is no "magic formula" to earn excess returns.
2. Assume that the current TIPS spread is "informationally inefficient." (Your second sentence.) If true, there exists some model that can predict future inflation better than the TIPS spread. This model may be a predictive algorithm, a very smart person's brain, or an advanced AI.
3. Can M make money by buying or selling TIPS? Yes.

But (3) contradicts (1). Therefore we must reject (1) or (2). Since the empirical evidence for (1) is overwhelming, we must accept that no currently existing model contains more information about expected inflation than the market. We might accept this, but still deny that the market is "informationally efficient"--after all, some undiscovered model might contain far more information than any existing model as well as the market itself. But if "informational efficiency" requires that undiscovered information be incorporated into market prices, it is an impossible bar to clear.

Tom DeMeo writes:

John- I question your conclusion on #2 (If true, there exists some model that can predict future inflation better than the TIPS spread.) Isn't it possible that sometimes the TIPS spread is an accident of circumstances, and doesn't always precisely reflect the market's expectations of inflation?

You can give the same market the same information, but the results will not always be the same.

Ari T writes:

Okay, I'm not an economist but how I think EMH goes a lot deeper than just economics, and it goes all the way to epistemology. And I think most articles that I have read about EMH, miss something very profound about nature of knowledge.

Saying EMH is right, is basically saying nobody could ever know anything. It's like Daniel Kahneman saying statistically you are almost likely to be right as wrong, so you can't really know anything. Tyler and Robin argued the value of economics, or any special knowledge regarding political outcomes. Robin said economists know more, Tyler argued that they don't adjusted for IQ and number of other measures.

That's like saying the only relevant question if a person should become an entrepreneur is how much risk he wants to take. Or rather, that there is no other knowledge (or model)!

EMH is a bit like Schrödinger's cat in physics. As soon as enough people observe the arbitrage, the arbitrage disappears.

I think some people want this permanent model which says when they have knowledge (arbitrage) and when not. But in reality there is usually not such model. You need to use your own brains to consider. A lot of people out there are selling you arbitrage which doesn't exist, but that doesn't mean arbitrage doesn't exist.

I also think the distinction between "public" and "private" knowledge is vague. A lot of knowledge is tacit and implicit. For example I had private knowledge that Apple was underpriced. This wasn't based on knowledge I acquired as any kind of insider, but reasoning I made about the product and company as whole. David Friedman said same thing.

So arguing whether EMH is right or not, is missing the point. You are basically asking where is my model that tells me if I can disagree with the majority or not, but I argue that such simple model doesn't exist. And usually the cost is the time spent acquiring that knowledge. Because of 2nd law of thermodynamics, complexity of the world increases, every situation is different. I think Lee Smolin (or some other physicists) argued also that even the laws of physics may not be constant (although they will be constant enough within our lifetime).

Steve Sailer writes:

The existence of the hedge fund industry, with its extravagant 2-and-20 pay system raises serious questions about EMH? Why hasn't efficient markets hammered down the cost of hedge funds to say 1-and-10 or 0.5-and-5?

EMH can be vindicated, however, by assuming that hedge fund profits stem largely from inside information.

I haven't seen too many economists wanting to make that suggestion, however.

Mark V Anderson writes:

And yet no one in the comments has questioned your comments on supply and demand. You say that if a laundromat raised its prices by 5% it would likely keep a substantial number of customers. Of course this is true, how does that contradict S&D? You move the price up slightly, then the quantity moves down slightly. That is exactly what the demand curve would tell you.

How macro theory over-simplifies the market is your discussion of suppliers being price takers and thus cannot change their prices at all. The theory seems to be that the S&D curves set the one price and that changes from this price will result in 100% loss of customers. That of course is nonsense, because suppliers change their prices all the time.

The problem isn't with the S&P curves but how economists aggregate these curves. In reality every business has a slightly different product and so they are not 100% price takers. Each firm has its own supply and demand curves. But in a competitive industry, the demand curve is very elastic because it is easy for consumers to switch to another company with a very similar product.

Comments for this entry have been closed
Return to top