Hi Gaurav,

We don't have the flow data in our internal dataset - at the moment we have just pricing data for the ETFs and other US equities. Is there a public datasource you use now?

thanks,
fawce

You can track the flows, but it's difficult to find causes in ETFs. You may want to research how ETFs work. It's been a few years since I was deeply in it, so the following my be off. Hopefully it will bring you closer to understanding. In mutual funds, if the public thinks nano-technology is hot, they'll funnel money into a mutual fund specializing in it. The next day, that fund will go out and buy stock in those companies (though it may just build it's cash position). So in that case, it seems that flows would indicate investor sentiment.

ETFs are not really investments in underlying companies. They are "baskets" of stock that large financial institutions own ("Authorized Participants"). These financial institutions basket the underlying stocks in an ETF stock and you buy that representation. In a mutual fund you can lay legal claim to your share of the underlying assets. In an ETF, you're really buying artificial stocks from large financial institutions that track the underlying value of their portfolios.

ETFs are also used for many reasons. From institutions looking for quick exposure, for hedging, etc.

Another thing about ETFs is it takes a few days for authorized participants to create/redeem shares (ETFs). They are always adjusting their own portfolio. Lots of mechanics that make it difficult to draw a connection.

Finally, they see data you'll never see. So if there is a profitable "relationship" they would benefit (and some say do). By the time new flows (ETF creation) is public the information is stale (except on a long-term basis).

The data is available (or at least was a few years ago). However, it only lasts for a day. So you have to get it every day (what I used to do). I forget the field.

Hope this helps.

Something I wanted to improve upon

http://www.trimtabs.com/global/pdfs/ETF_Flows_and_Market_Returns.pdf

TrimTabs has been around for a long time. Naturally, every data provider wants you to believe their data can be used to make money. Why would you need to improve on something that give you a 70% better return? Okay, so you're greedy ;)

Data is neither homogeneous or consistent. Or, "you can never step in the same river twice". First, the results of all empirical studies is that no one can prove they have beat the market from skill, rather than luck. Obviously, we'll skip that simple truth in the interest of fun.

How do you think you can improve on that model? I would need to know more, to give any bad advice.

Hehe.
The general idea is to use price indicators with flows.
I thought that they might be using mean reversion in some of their studies. Assuming flows are positively correlated to price. Negative correlation between flows and future price action could be due in part to negative correlation between price action and future price action.
But over time the negative correlation might not hold up. I wanted to isolate that indicator and then use longer term statistics of it to see whether completely removing that part is better or not.

First, I don't want to rain too much on your parade. If anyone succeeds in this, I believe they do so by VERY carefully isolating out specific data and coming up with a theory that explains WHY that data is predictive for a theorized point in time. I've only made one investment in my life. I invested in Federated a few years back because I knew, from all my data, that their stock price declines were going way beyond their actual fund outflows. However, we could have gone into a depression and then I would not have profited (so maybe I was just lucky). Federated is in a market I know a lot about and I had a lot of data. though I ran models, I didn't overdo it. My point is that you don't need much work to see a reversion. In any case, the data didn't drive my investment, the observation/theory did. (That the market over-panicked). Warren Buffet, for example, knows that people will continue to kill/harm themselves (Insurance) and eat bad food (Coke), so he buys when people think the world is going to be a better place. I could go on and on. The models only help with your timing. I don't believe they can make good choices on their own.

I would pick one class of ETFs. Maybe even just one ETF. You need a theory about why a certain reversion to the mean will happen. Then you can do your models. Math alone doesn't make money because the data is so messy/dirty. How can you run even multivariate analysis on data with a billion variations baked in?

Reversion to the mean isn't really a mathematical concept in investing. It's more a psychological one.

Hope this helps.

I think this is the page you want to scrape to get shares outstanding for an ETF. I used XLF (access VBA) when I was doing this.

Shares Outstanding
http://finance.yahoo.com/q?s=SPY-SO

NAV
http://finance.yahoo.com/q?s=SPY-NV

AGAIN! This data only stays up a day and then it's gone. [Symbol]-SO or [Symbol]-NV