Fun fact: overriding regulatory data with #alternativedata is also what Dennis, Gerardi & Schenone do to support the EFA-best-paper-award-winning claim that #airlinepaper’s results aren't robust to reasonable alternative empirical choices.
(The difference to Rock & Rubinfeld is that they don’t call it “reformat” when they come up with #alternativedata. They call it “alternative mapping”, among other things.)
Wanna hear more about how they continued furthering a narrative of a false AST paper, even after they realized their EFA-best-paper-winning claims had no basis in fact?
Come join me for a fun ride through the land of #alternativedata!
Come join me for a fun ride through the land of #alternativedata!
(Hopefully) humorous - and on twitter - because other avenues proved ineffective in my quest to #makefactsfunandmatteragain media.giphy.com/media/VGG8UY1n…
Disclaimer: if you’ve ever heard me tell a joke, you know it’ll take some time for the punchline. But sometimes, it’s worth it.
Plus, the journey is the goal etc. Here in particular.
Plus, the journey is the goal etc. Here in particular.
Let’s start with the highlighted claim from the abstract of the first version of their paper: 
Sounds like AST have a big problem with their very particular choice of voting designations, doesn’t it?
Let’s find out if that is true.
Let’s find out if that is true.
First, what’s DGS V1’s basis for proposing an alternative to AST’s empirical treatment of voting rights?
Sounds potentially reasonable. AST only show what happens if you treat “shared” rights as *having* voting rights.
Seems sensible to ask what happens if you instead assign “shared” rights to “no” / *not* having voting rights instead.
Seems sensible to ask what happens if you instead assign “shared” rights to “no” / *not* having voting rights instead.
So what’s the effect of the alternative empirical choice?
If you ask DGS V1, the answer is this:
If you ask DGS V1, the answer is this:
But, remember, although they falsely claimed otherwise, they hadn’t *actually* replicated AST at the time.
(Among others?) because they chose to skip the step of aggregating 13Fs to the level at which the shares are voted.
(Among others?) because they chose to skip the step of aggregating 13Fs to the level at which the shares are voted.
So their “robustness tests”in V1 don’t actually apply to *AST*’s results. They just apply to … well … DGS-V1’s-incomplete-replication-of-AST’s-baseline results.
(DGS V1 didn’t even manage replicate my name – another commonality with Rock & Rubinfeld. See if you can spot the error! #economistspellingbee)
Come on, people! It’s really not *that* hard, is it? Be a little more considerate of my citation count!
media.giphy.com/media/3orieOj4…
media.giphy.com/media/3orieOj4…
Btw, don’t you look up what that last name of mine means.
DON’T!
Urrh, ok. Now you know that, too. Welcome back.
So what *does* happen when you try this potentially reasonable alternative empirical choice with the *actual AST* results?
So what *does* happen when you try this potentially reasonable alternative empirical choice with the *actual AST* results?
I haven’t checked. But the latest version of DGS claims they did, using the AST replication kit. Here it the result, at the firm-market- and at the market-level:
Huh?!?! That looks *remarkably* robust to me?! Robust TO THE THIRD DECIMAL nonetheless??? In both firm-market and market-level regressions!
Looks like a pretty substantial correction to the EFA-prize-winning first version!
Looks like a pretty substantial correction to the EFA-prize-winning first version!
… you might think further:
“Huh, that’s kind of bad, actually! The entire field was misled to believe AST wasn’t robust to this choice all this time?! And regulators, worldwide, were misled! Did DGS ever apologize to AST...
“Huh, that’s kind of bad, actually! The entire field was misled to believe AST wasn’t robust to this choice all this time?! And regulators, worldwide, were misled! Did DGS ever apologize to AST...
...for bringing their paper in disrepute with factually incorrect claims? And to the public for misleading a policy debate of substantial social importance? And to the EFA prize committee that believed their claims? And to the profession for undermining its credibility?” Etc.
“Why did I never hear about DGS having retracted their claim then? That must have been a huge scandal!” #fomo
(Actually *all* factual claims DGS made with respect to robustness of AST’s results have since been shown to have been incorrect.)
(Actually *all* factual claims DGS made with respect to robustness of AST’s results have since been shown to have been incorrect.)
But relax. You didn’t miss out. That mea culpa never happened.
(Or I wouldn’t be sitting and tweeting here.)
In fact, the abstract of the latest version still claims:
(Or I wouldn’t be sitting and tweeting here.)
In fact, the abstract of the latest version still claims:
“How can that be?? Didn’t you just say that they show, in just that same very version of the paper, that AST’s choice of what you do with `shared’ voting rights doesn’t matter *at all*, and that it was all a false alarm back in the days of the EFA?” you might ask.
Well, yes, I did say that. And, yes, they do show that.
But: they don’t particularly emphasize that message in the writing. Get ready for some mindboggling stuff.
But: they don’t particularly emphasize that message in the writing. Get ready for some mindboggling stuff.
Instead, of acknowledging the error …
OH! LOOK! A SQUIRREL PLAYING BANJO!!
media.giphy.com/media/WIBreSdA…
OH! LOOK! A SQUIRREL PLAYING BANJO!!
media.giphy.com/media/WIBreSdA…
… all our attention gets absorbed by a *new* result that appeared alongside the (implicit) retraction of the old attention-grabbing claim.
As per column 4 (out of 3!) in this table -- of which I earlier, nefariously, only showed the first two columns -- *something* in AST is found to be not robust to *something* related to voting rights!
Ha! Here we go again!
media.giphy.com/media/QnyX01Tv…
media.giphy.com/media/QnyX01Tv…
Well, AST’s result is actually robust at the market level, although a bit attenuated. It’s just not significant at the market-carrier level, in this new “Owned” treatment.
But robustness at the market-level wouldn’t deter DGS from making a generalized abstract-level claim that “THE previously documented resultS ARE fragile to alternative reasonable mappingS from equity votes to control rights”. (caps added)
Anyways, so, what is this new “Owned” treatment all about?
Lest you think I exaggerate, read for yourself:
Lest you think I exaggerate, read for yourself:
“So…” you are thinking. “They just assume that … `no’ means `yes’?!”
-Correct. Because what could possibly go wrong!
-Correct. Because what could possibly go wrong!
“No matter whether the investors *actually* had control over the shares, as opposed to having loaned them out?”
-That’s right. No matter what.
-That’s right. No matter what.
Even though “asset managers who loan their shares give up their right to vote attached to those shares”? reuters.com/article/us-fun…
-Yes.
-Yes.
“But then the right-hand-side variable makes no economic sense and misses a bunch of its variation!? How is this a robustness test, rather than a placebo test?!”
Also: “What could possibly justify altering the data like that, in blatant contradiction of the regulatory data and in apparent contrast to practical reality?”
If you’re desperate to look for a defense of this choice, you might wonder:
“Perhaps this is kind of wrong, but innocuous? I mean, these big asset managers, the big common owners, do they even lend out their shares all that much?”
“Perhaps this is kind of wrong, but innocuous? I mean, these big asset managers, the big common owners, do they even lend out their shares all that much?”
Look at one of these Vanguard fund reports you receive in the mail, for once, before throwing them out. Look where the revenues come from.
Hint: in a zero-fee index fund, the revenues don’t come from the fees.
Hint: in a zero-fee index fund, the revenues don’t come from the fees.
They cover the costs with securities lending. (And you call yourself a finance professor?!? 😉 )
“And that’s supposed to matter for corporate control?” You might ask.
Where do I even start.
Where do I even start.
Here’s a recent case reported by @dawnlim: BLK-Vanguard-Fidelity only controlled 5% of ballots instead of the 40% they nominally owned, in Whatever Company, due to having lent out the shares.
wsj.com/articles/how-i…
wsj.com/articles/how-i…
What would DGS do? Forget 5% control!
Just assume 40%!
Just assume 40%!
“But surely, that WSJ one is an exceptional case, whereas the big asset managers would *generally* recall the votes before an election, and therefore in fact have full control most of the time. Even when they tell the regulator they don’t have any control perhaps.” You might ask.
“`We believe that, *generally*, the likely economic value of casting most votes is less than the securities lending income,’ BlackRock said in an Oct. 26 fund disclosure.” (* added)
media.giphy.com/media/GDnomdqp…
media.giphy.com/media/GDnomdqp…
But surely BlackRock is an exception?!?
Wellington writes that “Efforts to recall loaned securities are not always effective, but, *in rare circumstances*, Wellington Management may recommend that a client *attempt* to have its custodian recall the security to permit voting of related proxies.” (* added)
In sum, once you "ignore" what happens in the real world and regulatory filings that supposedly reflect the facts, and override the data with your – THEN you are ready to construct a #commonownership variable that doesn’t correlate with U.S. airline ticket prices. Hence:
QED.
Wait, don’t leave! That wasn’t the punchline yet! I’m still going! We’re almost there!
While you were breathing -- were you just thinking:
“Wait -- my mind is spinning -- didn’t they say earlier that:
`it is clear what institutions record under the designation of shares owned with no voting rights’
or something to that effect?”
“Wait -- my mind is spinning -- didn’t they say earlier that:
`it is clear what institutions record under the designation of shares owned with no voting rights’
or something to that effect?”
Don’t scroll up, this is a full-service tweetstorm. Here it is again:
Yes, you’re right. They did say exactly that.
“But then, how can they now override exactly that thing that was “clear” to them, with their novel `no’ means `yes’ approach?”
“But then, how can they now override exactly that thing that was “clear” to them, with their novel `no’ means `yes’ approach?”
They key word in my question above (“Didn’t they say earlier that…”) was: EARLIER!
They said it was "clear" in the highly publicized *first* version of the paper! You know, the one that won the EFA prize!
They said it was "clear" in the highly publicized *first* version of the paper! You know, the one that won the EFA prize!
The one that falsely claimed it was AST’s empirical choice regarding “shared” voting rights that AST’s results crucially depend on. That choice they now show doesn’t matter AT ALL, in either firm-market or market-level regressions. *That* earlier version!
So what has changed from that first to the latest DGS version?
media.giphy.com/media/3oEhn98n…
media.giphy.com/media/3oEhn98n…
What changed is the authors’ (reported) perception of the SEC instructions.
The corresponding paragraph in the latest DGS version reads:
The corresponding paragraph in the latest DGS version reads:
The change from V1 to the latest version, illustrated: media.giphy.com/media/q3ieN0SA…
Apparently, it is no longer clear - to DGS, that is - what institutions record under the designation of shares owned with “no” voting rights.
“No”, “sole”, “shared” – ALL of them are considered “vaguely defined” now.
“Arbitrary” even.
“Inconsistent” also.
Many such words.
“No”, “sole”, “shared” – ALL of them are considered “vaguely defined” now.
“Arbitrary” even.
“Inconsistent” also.
Many such words.
Not just the “shared” designation is the problem! The SEC data can’t be trusted on *any* of the voting designations. Or so DGS ask you to believe.
“Wait!? What?! Why?? How can I forget what I used to know a minute ago?”
media.giphy.com/media/26tjZFqE…
media.giphy.com/media/26tjZFqE…
I guess, you just have to … trust the authors. They’ve “assumed” a superior data set for you. Or, if not superior, at least they “assumed” data that constitutes a “reasonable alternative mapping,” or so the abstract promises.
But who would be so cynical to think that DGS removed from the text that it’s “clear” that “no” means “no”, just because it no longer fits the (time-invariant) narrative and title of the paper? Let’s give the benefit of the doubt!
Look, maybe they learned something along the way that made them realize how overconfident they were back in the day when they thought they knew that “no” means “no”.
And they somehow forgot to share that newfound knowledge with the reader.
Perhaps the reader is cynical, rather than the authors! (You really *are* a finance professor, after all. 😉)
Perhaps the reader is cynical, rather than the authors! (You really *are* a finance professor, after all. 😉)
I’m sure that after this tweetstorm, there will be a thorough explanation in the next draft.
… or maybe even some *evidence* to support the assumption that the voting designations in the SEC data have *zero* informational content.
… perhaps even some evidence that assuming ownership = control is more economically meaningful than what the funds disclose to the regulator?
… and that all these worries in the newspapers about big asset managers loaning their shares around elections are completely unfounded.
… and an explanation why using this SEC-provided variable leads to a false positive in AST, rather than attenuation bias from measurement error.
If all that happens, it’ll turn out I’ve been pulling a @billackman/Herbalife with this tweetstorm, and made the world a better place.
In which case I get to be grateful for having had the opportunity to help improve the paper.
You’re welcome!
media.giphy.com/media/Q7y3K35Q…
You’re welcome!
media.giphy.com/media/Q7y3K35Q…
Now, to conclude, whether overriding the regulatory data counts as “reasonable alternative mappings of equity votes into control rights”, or whether changing the data … is a somewhat less condoned practice in empirical research …? You be the judge.
Does it resemble this?
media.giphy.com/media/q3ieN0SA…
media.giphy.com/media/q3ieN0SA…
… or rather this?
gph.is/g/E0nbzXY
gph.is/g/E0nbzXY
Just perhaps do give it some thought, while you ponder that question, what would happen to *your* research results if someone replaced the underlying data used to construct your explanatory variable with all 1’s, for example.
Because that certain someone thought the variation in your RHS variable was “implausible”, without further explanation or reasoning.
media.giphy.com/media/xT5LMrDX…
media.giphy.com/media/xT5LMrDX…
And that was just *one* of the claims that shape public perception about the #airlinepaper’s credibility. From *one* of the attack papers.
ARE YOU NOT ENTERTAINED?
media.giphy.com/media/d7mMzaGD…
ARE YOU NOT ENTERTAINED?
media.giphy.com/media/d7mMzaGD…
Which one do you want me to do next?
Perhaps DGS’s claim regarding the effect of “unusual” filters on AST’s results? (It just requires one tweet.)
Perhaps DGS’s claim regarding the effect of “unusual” filters on AST’s results? (It just requires one tweet.)
Here it is:
