The most recent winning Powerball numbers were 9,30,31,50,54,39. But a month ago, nobody would have placed any significant probability on those exact numbers coming up. What better illustration that questions about the future cannot be answered, even in the probabilistic sense?
If that made you scratch your head, your scalp will probably be rubbed raw before you’re finished reading Anatole Koletsky‘s Wall Street Journal essay, excerpted from his book Capitalism 4.0. (Caveat: I have not read the book, so I’m not sure how much danger the rest of it poses to your scalp, or to your sanity.) Mr. Koletsky’s “proof” that some questions “cannot be answered, even in a probabilistic sense” is this:
In 1980, nobody would have put any significant probability on computer sales exceeding car sales by a factor of 10 to 1.”
But that’s not all! There’s also this:
What is the probability that someone in the next hundred years will invent a soft drink more popular than Coca-Cola? This probability must surely rate at almost 100%, yet that would also have been true in 1910. There is no rational way of making such an assessment(!!!!!!!!!!!!!!!!). [Snarky emphasis added.]
Look. This isn’t rocket science. Over the course of a hundred years, a whole lot of things happen. When a whole lot of things happen, some of those things (like a particular set of Powerball numbers or the computer revolution, or the continued success of Coca-Cola) will have been extremely unlikely. That’s how probability works. If, over the course of the past century, no very unlikely things had happened, then we’d know that our probabilistic models weren’t working. So far they seem to be working just fine.
Of course the same models predict that more likely events will occur more often. For example, I could easily have predicted a year ago—or even a decade ago—that with extremely high probability, a lot of nonsense would be published in the year 2010. Thanks to Mr. Kaletsky for doing his bit to confirm that prediction.
Good point.
It reminds me of a popular argument in the middle-brow press that the recent turmoil in equity markets or the fall of mortgage backed securities disproves all forms of the efficient capital markets hypothesis (ECMH) and shows up all those gullible fools who believe that “the market is always right.”
In one, extremely limited sense, that is true. If you believed in a super-strong version of the ECMH claiming that market prices reflect all perfect knowledge of all events current, past, and future, then indeed you were a gullible fool and volatile markets prove it.
Thing is this super-strong ECMH is proven false not only in market turmoil, but every single day that not *all* capital market prices move in perfect lock-step. Absent extraordinary events (such as dividend or interest distributions) and regardless of your assumptions about discount rates and the like, if the price ratio between any two securities changed from day to day, at least one the market prices involved must have been wrong sub specie æternitatis (sorry about that).
So much for the super-strong ECMH which avoids being discredited only by the circumstance that–as far as I can tell–nobody ever believed in it.
If you want to discredit markets try disproving that prices fully reflect all publicly available information (that is, the semi-strong ECMH). Proving that the semi-strong ECMH is significantly violated really would offer support for the proposition that more decisions should be made through the coercive, non-market, technocratic mechanisms enjoying such a vogue among the middle-brow these days.
But no fair pointing to past prices which were wrong. Almost all of them are practically guaranteed to be, regardless of whether the semi-strong ECMH is true. And as almost any way the semi-strong ECMH would be significantly wrong implies a *working* get-rich-quick scheme, you better have an astronomical net worth or have a very, very good explanation for why you haven’t.
If you have neither, you have no serious argument against the accuracy of prices in capital markets.
Seems to me this guy is trying to write something similar to Taleb’s Black Swan, but I’m not sure how he manages to morph unpredictability into:
“Governments or regulators must have the power and the self-confidence to second-guess and override market signals. They must accept responsibility for managing economic activity and employment. And they must stand ready to support the financial system if regulation fails.”
Governments must manage economic employment? That’s the problem with government, not enough self-confidence to regulate. Have strength, oh meek and humble government legislators and regulators, thine holy mission is to save the market from itself!
To sum up:
1) Black Swan events occur that are wholly unpredictable.
2) ……
3) Socialism!
He sets up this:
“Modern economists sometimes pretend to overcome this problem by assuming that financiers make decisions by calculating future probabilities in the same way that normal businesses, operating in the present, count current profits and losses”
In order to knock it down by saying we can’t calculate the pobability of 9/11. I am pretty sure that no-one claimed to predict these kind of events.
As æternitatis says above, prices reflect publically available information. All Koletsky seems to be saying is there is a lot of information that is not publically known, not known at all, or not knowable. Therefore prices do not (and can not)reflect “reality”. Dressing this up as probability and predictions seems an unnecessary distraction.
I think he has a point that many people (possibly not economists) thought that their models did in fact reflect reality.
The difficult bit to understand is this:
Probability is not a property of the universe.
Probability is a property of our KNOWLEDGE of the universe.
(Modulo quantum wavefunction collapse at least)
To illustrate. I’ve tossed a fair coin. What is the probability it is heads? Before I look, it is 1/2. After I look, it is either 1 or 0. The coin has not changed, but my knowledge has.
I think what he means by ‘there is no rational way of making such an assessment’ is: “it’s sometimes quite hard to make assessments like this, and people sometimes get them wrong”.
It’s quite possible that we would have been *wrong* in 1910 to say that Coca-Cola would be superseded with high probability (Overconfidence Bias is a well-documented phenomenon – experts aren’t always well-calibrated). It’s also quite possible that such a statement would have been right, and unlucky. You can’t possibly decide either way on questions by citing specific examples.
And it’s just silly to say that because people sometimes say things which are wrong, it’s impossible to be right.
Reminds me of Littlewood’s Law – “miracles” happen about once a month.
So do you just not but into the “risk vs. uncertainty” distinction at all?
I have to disagree with you on this one. The probability of a particular combination of lottery numbers being picked can be calculated, because the process by which they are chosen is fixed, known, and has certain mathematical properties. The probability of something currently unknown being invented in the future, or of major shifts in the markets for various goods, cannot be modeled in this same way.
Koletsky goes a little too far in saying there’s no rational way to judge, at least assuming you buy into (as I do) the Misesian definition of rationality which rests on deliberate action. But just because people act in a way that economists believe says something about their estimates of probabilities does not mean that we can actually calculate those probabilities.
Yogi said it best “Prediction is hard, especially about the future.”
@Aeternitatis,
It’s interesting to note that all those people who say “it’s so obvious that the efficient market hypothesis is wrong” always just seem to point past examples of disruptive market movements.
It’s interesting how none of these people actually give you a strategy or prediction that earns excess returns for the future. If you’re going to call people who believe in EMH fools and say that the market is so obviously distorted in so many ways, then my question is why haven’t you quit your day job to go start a hedge fund?
I don’t think that Koletsky is right about computers and 1980. Moore’s Law had been underway for 30 years already, and people were predicting that computers would eventually be in everyday appliances like bread toasters.
He also recites a phony Watson quote about computers. Watson bet the company on computers, and he would not have done so if he thought that the market was limited to five machines.
Sean, what this guy is arguing is the *opposite* of Taleb.
Taleb: we (1) underestimate certain probabilities, and (2) overestimate the probability of random things that have already happened having happened (that is, when a .0001% chance event happens, we assume that it must have been much more likely to happen than we thought, when in fact it need not be).
Koletsky: I am going to write an essay in which I repeatedly make mistake (2)!!! Yay!!!!!!
Isn’t Koletsky just talking about Knightian uncertainty? But since I never clearly understood that idea, I may be wrong.
@Harold “All Koletsky seems to be saying is there is a lot of information that is not publically known, not known at all, or not knowable. Therefore prices do not (and can not)reflect “reality”. Dressing this up as probability and predictions seems an unnecessary distraction.”
Prices don’t reflect reality omnisciently. But, in open and liquid markets, they do represent our best collective estimate of reality and the future. That estimate can be, and usually is to some degree, wrong. But ordinarily it is the best guide to the future that we have. That both real and pretty neat.
@Sean Just so. Market prices can be and often are wrong. But why in the world would we assume that even a clever, educated, well-meaning government technocrat would be more likely to be right? And why would such a hypothetical technocrat work for a modest government salary and forgo the available opportunity to become a hedge fund billionaire? The only reasons for such a strange belief are hindsight and anti-market bias.
@Noah Yetter “The probability of a particular combination of lottery numbers being picked can be calculated, because the process by which they are chosen is fixed, known, and has certain mathematical properties. The probability of something currently unknown being invented in the future, or of major shifts in the markets for various goods, cannot be modeled in this same way.”
I’m not sure I see the distinction. Setting aside certain question regarding quantum mechanics equally applicable to both examples, the difference seems to be merely one of degree, rather than kind.
If I knew exactly in what state the lottery machine was at the beginning of the draw and I had sufficient computational resources, I could predict with absolute certainty what numbers would result. If I knew exactly in what state the world is today and I had sufficient computational resources, I could predict with absolute certainty all future inventions, market shifts, etc.
But as I have neither perfect knowledge nor sufficient computational resources for either examples, I am reduced to merely making more or less valid probabilistic statements about either. As @Ben noted, when speak of probability, we speak as much about our ignorance as we do about the universe.
Or perhaps, I misunderstood you, Noah. Are you arguing that for the lottery machine there exists a simplified computable model (i.e., each number is equally likely to come up at each draw) which accurately describes our limited knowledge of the machine, while we do not even have such a simplified model for the world? I think that is right, but I am not sure what important consequence follows from that.
@Roger Schlafly In fairness, I think that Moore’s law was only 15 years old in 1980 and, even if you understood as some clearly did, that computers would become as ubiquitous as toasters, it is not clear that they understood that the total size of the computer industry would dwarf that of the auto industry. The toaster industry never did.
What exactly does it mean for a price to be right or wrong? Is a price right when marginal value equals marginal cost?
Economic values are subjective. It seems bizarre to use objective terminology to describe something that’s subjective.
æternitatis: “Prices don’t reflect reality omnisciently. But, in open and liquid markets, they do represent our best collective estimate of reality and the future. That estimate can be, and usually is to some degree, wrong. But ordinarily it is the best guide to the future that we have. That both real and pretty neat.”
I think the point is that the “best” estimate, whilst good for a time, sometimes goes very badly astray, with bad consequences for everyone.
Part of the problem is making sure the full costs and benefits are felt. The bankers can take one-sided bets, knowing that the worst that can happen is they lose their jobs, money from the wins already salted away. This makes the market much more prone to “exuberance”. Cautious bankers see their short term profits fall, and get taken over. The prophets of doom do not attract investors.
There seem to be several ways to combat this.
1) Bring back debtors prisons for bankers who lose money. This will keep their incentives aligned.
2) Set in place rules that will limit the growth of the bubbles. This will also limit “proper” growth to some extent.
3) Hope you can spot the bubble forming, and borrow loads of money to bet the other way. If you are wrong, you may end up in debtors prison.
4) Authorise reulators to step in if they think things are getting out of hand. But if they think it is, then why don’t they take optoion 3 and make themselves a mint? Presumably because they are not that confident that things are overheating.
Overall, I think a mixture of 1 and 2 is the best option. But fortunately, no-one is asking me.
I think the point Anatole Kaletsky is trying to make (albeit very badly) is thus:
There is a probability that I will invent a super-new drink that will displace coca-cola. There has never been a drink like it in the past, so we have no empirical evidence to work with, and it is impossible to estimate logically what the probability of such an outcome might be.
Therefore, the probability that I will invent a successor to coca-cola lies between 0 and 1 (inclusive), but we have no idea where exactly it is. This, as I understand it, is Knightian uncertainty.
NB When you consider that the unknown probability is also likely to be conditional upon other events, which may or may not have known probabilities, and that the probability that I may invent “new coke” changes over time (more likely to happen when I’m 30 than when I’m 3!), we take the uncertainty to the nth degree.
@JLA
Ultimately you are right: what how and we value are subjective, so one can’t generally say whether any particular price difference reflects differences in taste or factual error.
In the case of investments, though, the issues narrow considerably.
For investors (with the possible exception of the handful who hope to gain secondary benefits not directly related to price, such as corporate control), the sole value of an investment is the stream of dollars–interest, dividends, and ultimately resale price–that investment generates.
While rational people might have different subjective view of the relative values of a dollar today and a dollar tomorrow, there seems to be no psychologically (or introspectively) plausible difference in tastes that values two dollars from an investment A tomorrow more highly than three dollars from investment B tomorrow.
If somebody chooses investment A over B, they either have truly bizarre preferences or they were mistaken about the value of the investments. If the today’s market price for investment A is higher than that for investment B, either the entire market has bizarre subjective preferences or today’s market prices for investments A or B (or both) are wrong in an objective sense.
Now you might respond that nobody knows for sure what investments will be worth tomorrow and the market prices for investments A and B may accurately reflect all available public information. That may be true–in fact I think it usually is–but that merely offers a good explanation for why the prices are wrong–universal, unavoidable ignorance of tomorrow’s events–but does not show that they are right from the point of view of a perfectly informed observer.
I think Joe Bingham has got it right. Taleb says that people often think that something with a very low probability cannot happen, but also overestimate the likelihood of something with a high probability happening. The USA has never won the World Cup. What’s the probability that they will win it in the next 100 years? It’s certainly not 0%, as it is certainly possible that they can; and it’s not 100%, as it’s not guaranteed. And if they do win, that doesn’t change the probability. I’d bet that we could collectively come up with a pretty good estimate.
What’s the probability that the sun won’t come up tomorrow. Pretty low, but not 0%. Maybe the LHC will generate a black hole and the earth will collapse. Very unlikely, but not impossible. Of course, if that happens there won’t be anyone around to complain about black swans.
Well in (very mild) defence of “not knowing things in a probabalistic sense”, the recent wimbeldon match prompted this post by terry tao.
http://www.google.com/buzz/114134834346472219368/DgaPpRfB5fH/Currently-at-Wimbledon-one-of-the-sets-between
The sense of which is that the naive model of assigning a value p to the probability of both players holding there serve is at odds with the 70-68 score (considering that the previouse record was 24-22).
i.e. the number of games won seems to not be distribued in this psedo-geometric way.
Of course this could be an example of littlewood’s rule but it’s also possible that the naive model is wrong, which is more likely depends on your prior belief about how likely the model is to be wrong appori. I’m leaning towards the naive model being wrong.
Of course we could just construct a better model but I’m not sure we’d ever be prefectly sure about it (more precisely we wouldn’t be sure about how unlikely a 100-98 game is until we get one and a 1000-998 game until we see one etc).
I guess the soundbyte I’m looking for is “all models are wrong but some are useful”, and if our models are wrong then we don’t KNOW anything even in the probabalistic sense. What we do have is a good enough idea of probabilities to be going on with.
Jonathan