Short-term update: this article was written as Quote of the Day in Wall Street Journal network's MarketWatch.
The market has had a recent hiccup leaving the S&P in the low 1900's, and now there is a choir of professionals who state that we could -or could not- be in for the start of a larger correction (e.g., 10% or so). Of course if anyone knew this in advance for sure (even just once in their working adult life), then he or she would comfortably find themselves doing something else more interesting than constantly putting out correction counsel for anyone willing to listen.
In risk theory we see evidence of these want-to-be crystal gazers, as they leave in their wake a streak of Type 1 errors. How can you blame them either, as whenever the market falls 1%-2% in a day, then clearly some zoot should have known about it.
Let's say that a 10% -or greater- correction occurs maybe once or twice a year (and that's fairly generous.) Meanwhile a daily market drop of 1%-2% occurs about 20 times a year. If we allow all of these daily 1%-2% market drops to always cue us in on a broader 10% correction, which when it occurs only includes a handful of the 1%-2% daily drops, then more than 3/4 of these broader market correction guesses would be wrong. Those are horrible odds to bet with, and not worthy of thinking twice about. This is also an idea known as a false positive, in the sphere of biostatistics testing. Of course one can also often be right for the wrong reasons, though we save this for another note. Meanwhile, these daily drops of 1%-2% should just be considered oddities, and nothing more. They are typically not "signals" of something more significant, outside of normal market gyrations, within a longer-term up-trend.
But what about the 2% -or greater- daily drop, such as the one we saw last Thursday, the last day of July? This certainly occurs less frequently; normally about a half-dozen times a year. In our lower volatility regime we recently are not used to seeing them. But even so, they still occur frequently enough that we can not use it to falsely guide us on broader market corrections. About 3/5 of these types of guesses would be proven wrong. Better odds than using a 1%-2% daily market drop signal, but they are still bad odds. Even if the market correction guess turns out to come true (which after being wrong most of the time, a guess may correct some of the time), we must remember that it is generally made -to begin with- more than 2% away of the market peak. This timing gap is critical, as it severely erodes the performance alpha one expects from such speculation.
We can look at a more narrow market segment now, in cases where the market may drop about 3% in a span of six days. In doing this we are now data-mining with the recent past market history, trying to fraternize with an alarming market statistic. The issue with data-mining is that one will eventually find "something", though perhaps nothing of great value (else we'd all be creating data-mining software for fun and retaining any garnered secrets). Of course a 3%-or-so market drop, in six or so trading days, is not enjoyable. This still generally happens several times a year. In fact, we expect to see a 3% market drop in just a single day, a couple times a year (see us in this New York Times article). These relative frequencies of about week-length market drops suggest that half of these times it is wrong to plainly associate them with a broader market correction. This signal has better odds still, versus the other ones noted above, though these odds are still not good. And even if the guess proves correct, it only means that one had to have waited until after the market has dropped at least 3% to join the party. The broader market correction by that point could be at least a third over. If one wants to give counsel they need to do it earlier, though not too frivolously early that their guess is not connected to any proximate signal.
So we see in this note some of the problems of just waiting around to be right. It's a never ending game of full-throated data and conjectures, masquerading as shrewdness. By the time one knows they are correct for sure, so does everyone else in the world. One shouldn't then take more time, to irrationally try to to convince everyone else, that it is they who somehow best foresaw the risky market event a-priori and further reckon there even more of the correction is still to come. Because while this is occurring, the market has already violently reversed course and chased back up. Such a market rebound would surprisingly catch most want-to-be crystal gazers wrong-footed, yet again. And then a new choir begins, and so to does a new cycle of guessing passed off as intelligence.
The market has had a recent hiccup leaving the S&P in the low 1900's, and now there is a choir of professionals who state that we could -or could not- be in for the start of a larger correction (e.g., 10% or so). Of course if anyone knew this in advance for sure (even just once in their working adult life), then he or she would comfortably find themselves doing something else more interesting than constantly putting out correction counsel for anyone willing to listen.
In risk theory we see evidence of these want-to-be crystal gazers, as they leave in their wake a streak of Type 1 errors. How can you blame them either, as whenever the market falls 1%-2% in a day, then clearly some zoot should have known about it.
Let's say that a 10% -or greater- correction occurs maybe once or twice a year (and that's fairly generous.) Meanwhile a daily market drop of 1%-2% occurs about 20 times a year. If we allow all of these daily 1%-2% market drops to always cue us in on a broader 10% correction, which when it occurs only includes a handful of the 1%-2% daily drops, then more than 3/4 of these broader market correction guesses would be wrong. Those are horrible odds to bet with, and not worthy of thinking twice about. This is also an idea known as a false positive, in the sphere of biostatistics testing. Of course one can also often be right for the wrong reasons, though we save this for another note. Meanwhile, these daily drops of 1%-2% should just be considered oddities, and nothing more. They are typically not "signals" of something more significant, outside of normal market gyrations, within a longer-term up-trend.
But what about the 2% -or greater- daily drop, such as the one we saw last Thursday, the last day of July? This certainly occurs less frequently; normally about a half-dozen times a year. In our lower volatility regime we recently are not used to seeing them. But even so, they still occur frequently enough that we can not use it to falsely guide us on broader market corrections. About 3/5 of these types of guesses would be proven wrong. Better odds than using a 1%-2% daily market drop signal, but they are still bad odds. Even if the market correction guess turns out to come true (which after being wrong most of the time, a guess may correct some of the time), we must remember that it is generally made -to begin with- more than 2% away of the market peak. This timing gap is critical, as it severely erodes the performance alpha one expects from such speculation.
We can look at a more narrow market segment now, in cases where the market may drop about 3% in a span of six days. In doing this we are now data-mining with the recent past market history, trying to fraternize with an alarming market statistic. The issue with data-mining is that one will eventually find "something", though perhaps nothing of great value (else we'd all be creating data-mining software for fun and retaining any garnered secrets). Of course a 3%-or-so market drop, in six or so trading days, is not enjoyable. This still generally happens several times a year. In fact, we expect to see a 3% market drop in just a single day, a couple times a year (see us in this New York Times article). These relative frequencies of about week-length market drops suggest that half of these times it is wrong to plainly associate them with a broader market correction. This signal has better odds still, versus the other ones noted above, though these odds are still not good. And even if the guess proves correct, it only means that one had to have waited until after the market has dropped at least 3% to join the party. The broader market correction by that point could be at least a third over. If one wants to give counsel they need to do it earlier, though not too frivolously early that their guess is not connected to any proximate signal.
So we see in this note some of the problems of just waiting around to be right. It's a never ending game of full-throated data and conjectures, masquerading as shrewdness. By the time one knows they are correct for sure, so does everyone else in the world. One shouldn't then take more time, to irrationally try to to convince everyone else, that it is they who somehow best foresaw the risky market event a-priori and further reckon there even more of the correction is still to come. Because while this is occurring, the market has already violently reversed course and chased back up. Such a market rebound would surprisingly catch most want-to-be crystal gazers wrong-footed, yet again. And then a new choir begins, and so to does a new cycle of guessing passed off as intelligence.
Thanks much Market Map for your comments on this article. Unfortunately I can't address your concerns directly, but please continue to use this site for anything you need. And exchange your views through these comment sections via the social media icons here.
ReplyDeleteNote that comments left here will be removed, if they are only designed to draw people to an offline site.
Thanks much John Cavender for your offline comments on this and recent web log articles. Unfortunately I can't address your concerns directly, but please continue to use this site for anything you need. And exchange your views through these comment sections via the social media icons here. Note that this article has received wide attention from a number of media (http://statisticalideas.blogspot.com/2013/02/additional-resources.html), since it was once more prescient probability advice a couple weeks ago that industry prognosticators of 10%-corrections are generally wrong, but for the rare times they stumble upon luck.
ReplyDelete