On another thread, Walt sets up a 4 rule system for avoiding Chinese fraud. His four rules produce a data set that is exactly one item big: XIN. Walt then tests that one item, XIN, against whether it's an unmasked fraud, and, finding (shocker) that it's not, he goes on to conclude that these rules offer a 100% protection from fraud.
On the surface that might seem valid. I mean, if we apply the data set's current 100% pass rate to future admissions, the prediction it makes is that 100% of all of those future companies will also pass.
Except there's a problem. All models are only as strong as the data that supports them, and the data supporting Walt's is a grand total of one item deep. That (very considerably) weakens the expectation that the current pass rate, 100%, will remain constant through the next 10 or whatever companies, although to support Walt's claim that these rules protect an investor from fraud, it would have to maintain its 100% rate.
Don't get me wrong, the model's not hopeless. Whenever possible, I like to quantify things. But the model won't mean anything until its conclusions are based on more than just one piece of data.
If I'm remembering right, the article where Professor Gillis' Big 4 Chinese fraud rates are cited produced an average of, what, 16% fraud rate, was it? I can't remember now, but I think that's close. Therefore, if Walt's data set had in it, say, 10 items, and all 10 passed, we might be able to start saying there's some significance to the rules, that 10 out 10 passes truly does seem to be diverting from the 16% overall fraud rate.
But one item tells you nothing. Even just choosing a China stock completely at random gives you an 84% chance of passing, which, if it passed, Walt would consider validly proving the conclusion that the rules that got it there, random selection, provided the same 100% fraud protection (1 pass out of 1 company), but for a data set we know for sure would ultimately be 84%.
As it stands now, we get the exact same result in fraud protection from Walt's rules as we get from using the following rule:
"No China stocks whose ticker symbols start with the letter "X" are fraudulent because the letter X would never stand for that."
My data set: XIN, so the evidence supporting my rule and Walt's rules are exactly the same: 1 item, 1 pass. Heck, it's even the same item.
A complete lack of evidence (tiny data set) is not the same thing as having a well proven conclusion (these four rules protect from fraud).
There is one thing we need to know from you, Hmmm. How do you define fraud in Chinese companies. While XIN, in my opinion, may well likely exaggerate its revenue and earnings, it may not deserve the word - fraud. Accounting manipulations are extremely common among Chinese public companies to serve different purposes, but majority of them are still valid profitable business.
The reason why XIN is different from those infamous fraud in China may be XIN is never meant to operate in a scheme to get numerous cash from the stock market. XIN may just overstate its earnings in order to boost its stock price, which helps its employees. That's why XIN gave out dividend while some other frauds did not. Remember, dividend is given to all shareholders, which include the chairman and XIN's employees.
The current dividend payout ratio of XIN is below many XIN's peers while the dividend yield is substantially higher than many its peers. XIN should pay the same level of dividend payout ratio as its peers' if it really earned that much profits. However, it didn't. The message the market got is that it could not.
1. Good post. First, thanks for using the term "Dividend Payout Ratio". I didn't know the term, but I've discussed that concept several times before when the discussion turned to the size of the dividend XIN can afford to pay and people were always pointing at the yield based on share price, rather than the Dividend Payout Ratio, based on a percentage of the company's net income. The concept made sense to me, I just didn't know it had a name.
2. If XIN were exaggerating its earnings, how would that benefit its employees? Are their bonuses tied to net income?
3. The different types of fraud question deserves its own thread. I'll write one up, but I think we're in agreement in principle: the kind of "fraud" I was talking about was "fraud" that was big enough that if it were discovered, the company's stock price would crash. So, Longtop would qualify, whereas the less material violations that FMCN committed would not. Maybe call that "material fraud" or something? I dunno; that's just off the top of my head. I'm open to suggestions.
Basic logic seems to be a weakness of yours. Xin is one example that passes my fraud test. Of course there are others. Petro China for one. The point is that a dividend and share buy backs demonstrate that a company has MONEY. An ipo requires strict scrutiny. And there are fewer frauds on the NYSE than any other exchange. These are rational criteria for determining fraud. Of course they don't guarantee that Xin is not a fraud. They do indicate that the chances are small. Given that Xin remains an excellent value play.
If they haven't, then your data set remains 1 item big.
If they have, that's progress, you're up to 2.
But still, a 2 passes out of 2 is the most likely outcome gained if we just grabbed two China stocks randomly, Using Professor Gillis's 16% fraud rate, there's still a 70.6% chance of two randomly chosen China stocks producing two passes: .84 times .84