The Follies Of ETF Flows: Part 1

August 13, 2013

Beware: ETF flows are not always a good gauge of investor sentiment.

We’re as guilty as anyone when it comes to drawing conclusions from data in excess of the facts. We publish flows information here at IndexUniverse every day, and for good reason—flows information tells you who’s winning and losing in the ETF space.

But does it really tell you much about investor sentiment? And should you be using that information to form trading strategies?

The answer to both questions is probably no.

I’ve noticed a disturbing trend with some of my friends at hedge funds and prop desks. They’re looking to daily flows data in ETFs as a source of signal. That is, they’re building algorithms to try and get some sort of information out of the daily ETF flow data published by us (and others) that can tell them where the prices of ETFs will head each morning.

There are numerous reasons why this is a bad idea. Let’s take a look at one of the biggest problems with flows data:

The First Problem: Bad Data

The first is simple: ETF flows numbers are buggy as all get-out. To know how much money went into or out of an ETF, you need to know three things: how many shares there were outstanding yesterday; how many shares outstanding there are today; and a price.

The price is generally the easy part—if we’re going to look at flows for yesterday, we use yesterday’s net asset value where possible, and only if that’s unavailable will actual closing price be used. The problem comes with the share count.

ETF share counts are reference data. They’re not like market prices, which live in a world of regulatory and competitive pressures. If you’re a real-time data vendor, your real-time market prices need to be good, or you’re not going to get much business.

Similarly, if you’re an exchange, your closing prices better be pretty good, or you’re going to lose business. Even NAVs, which we frequently find errors in, have a regulatory structure—funds are required by prospectus to publish NAVs, and actual business gets done based on that number.

But share counts? Share counts are just nice-to-have information, and they’re frequently just wrong. We currently look at share count numbers from four different data vendors on a regular basis. They rarely match on any given day.

 

 

For example, based on three different data vendors—BARL, the Morgan Stanly Crude Oil ETN—had either 30,000 shares outstanding, 400,000 shares outstanding or 72 million shares outstanding in July. (Our bet’s on the low number, this time around.)

Even if you’re getting consistent numbers, depending on where those numbers are sourced, the actual change in share counts (and thus, the basis for any flows analysis) can be lagged by a day, a few days or even longer.

Most data vendors get their share count numbers from the National Securities Clearing Corporation feeds, which, theoretically, should be accurate. However, they’re generally referencing creation and redemption activity from at least the previous day.

So if the NSCC share count this morning is 100,000, and was 50,000 yesterday, that likely doesn’t mean that 50,000 new shares were created last night in the settlement process. That 50,000 in flows likely resulted from creation activity the day before. It just took a day to work through the system and come out the other side as data.

Of course, it’s not that simple. Sometimes that data is a day or two later. And sometimes a data vendor is getting a direct feed from the issuer of an ETF or the issuer’s custodian; in which case, for those ETFs, the share count might actually be current and accurate.

And as a consumer of this data, you have essentially no way of knowing, when looking at all ETF data for a given day, which funds are lagged by how much, and which data might simply be bad. For these reasons, in our analysis we try to focus on longer-term trends—flows over a week, a month or a quarter—and take daily flows with an enormous grain of salt.

Further, we flag any big movements in shares outstanding and call the issuer to confirm before we publish anything outrageous. That happens a lot more often than you might imagine, and about half the time, it’s an error.

In part 2 of this series tomorrow, I’ll look at the second major issue with ETF flows data—interpretation.


At the time this article was written, the author held no positions in the securities mentioned. Contact [email protected].


 

Find your next ETF

Reset All