Vineet Bhagwat at the University of Oregon and Timothy Burch at the University of Miami have found that US companies which are active on Twitter see their share prices behave differently after earnings news. Twitter-active firms see their shares rise more in the days after good earnings news than firms not on Twitter; they enjoy a strong post-earnings announcement drift. But they also see prices recover more strongly in the days after bad results. This suggests that being on Twitter attracts investors’ attention, causing trading which distorts prices.
Jennifer Itzkowitz of Seton Hall University in New Jersey has found something stranger. She shows that shares whose names begin with letters earlier in the alphabet trade more than shares beginning with later letters. This is consistent with earlier research which has found that there’s more trading in shares whose ticker symbols are actual English words than there is in shares whose symbols are just a jumble of letters.
There’s a reason for this. We have limited attention. When we’re looking for shares to buy or sell ones at the top of the alphabet sometimes grab our attention first - say because they appear at the top of portfolio statements. Similarly, ticker symbols that are proper words are easier to remember and so they loom larger in our minds.
For investors, the error can be costly. All this shows how easy it is to trade not upon hard information with predictive power, but simply because some things loom larger in our minds even though they are irrelevant.
However, it’s not just distorted attention that can generate noise trading. So too can correlation neglect - our failure to discount signals if they come from a common source and instead to treat them as if they were independent reports; this can cause us to put too much weight upon weak information.
Some experiments at the University of Bonn by Florian Zimmerman and Benjamin Enke have shown how common this error is. They got subjects to guess the number of balls in an urn they couldn’t see, based upon some computer-generated clues. Some were told that the clues were uncorrelated while others were told they were correlated. But the latter didn’t discount the signals accordingly. This meant they were overconfident about the reliability of the clues, and so guessed to high when the clues were high, and too low when they were low.
Professors Enke and Zimmerman then tweaked the experiment so that subjects traded an experimental asset whose pay-off depended upon the number of balls in the urn. They found that traders getting the correlated clues drove prices too high if the clues were high and too low if they were low. This shows that correlation neglect can generate excess volatility in share prices. This might be one explanation for a famous finding by Yale University’s Robert Shiller, that US share prices are many times more volatile than the underlying path of dividends. For example, since 1946 the annualised volatility of annual US dividend growth has been just 6.9 percentage points, but share price volatility has generally been far higher than this.
This matters because in the real world many signals might be correlated in quite subtle ways. It’s obvious that media reports about a company are correlated with each other, because they come from a common source. But if there are peer effects and herding, then the optimism or pessimism of our friends will be correlated - in which case it might be unwise to follow them. And the opinions of different experts can also be correlated, by virtue of their similar backgrounds and training.
It is, therefore, easy to become overconfident about perhaps weak signals.
There is, though, a simple antidote to this. Remember the background facts - that in investing the ratio of noise to signal is very high so very little information has predictive power, and that frequent trading is expensive. "Inaction delivers better financial outcomes than frequent buying" says Greg Davies, a behavioural economist at Barclays. "Do less than you’re inclined to."