An interesting fact of returns is that all of the stock returns since 1993 are from overnight returns. Here are the total returns using only Close-Open (overnight) vs. Open-Close (intraday). The intraday returns are basically flat over the past 20 years.
That's a curiosity because 2/3 of the risk of stocks is from their intraday returns--measured by beta or volatility--so if return is compensation for risk, it doesn't seem consistent with that theory. However, there is further nuance I discovered that I haven't seen anywhere else. If you take all the tickers, the top 1000 non-etfs over the past 2 years, and rank them by prior daily volatility, and then look at their overnight returns, you see that volatility is strongly positively correlated with subsequent overnight returns, which then reverse over the next day session.
So it appears that cross-sectionally, volatility receives a positive overnight risk premium, a negative intraday one.
I couldn't figure out a way to make money off this, obviously. Note that if the average price in this sample is $43, making 0.15% generates 6 cents. Sounds great, but actually the returns are more concentrated for the lower-priced stocks, generating a return very close to the spread, ticker-by-ticker.
While I think this pattern retains because it is too small to arbitrage, it is an interesting residual pattern. I think it is best explained by something like this: high vol stocks are targets of intraday trading. This demand is generally positive, and so what you have are returns being depressed at the end of day from day traders selling and closing their positions, returns at the beginning of the day pushed up by the day traders opening positions (generally buying).
That's a curiosity because 2/3 of the risk of stocks is from their intraday returns--measured by beta or volatility--so if return is compensation for risk, it doesn't seem consistent with that theory. However, there is further nuance I discovered that I haven't seen anywhere else. If you take all the tickers, the top 1000 non-etfs over the past 2 years, and rank them by prior daily volatility, and then look at their overnight returns, you see that volatility is strongly positively correlated with subsequent overnight returns, which then reverse over the next day session.
So it appears that cross-sectionally, volatility receives a positive overnight risk premium, a negative intraday one.
I couldn't figure out a way to make money off this, obviously. Note that if the average price in this sample is $43, making 0.15% generates 6 cents. Sounds great, but actually the returns are more concentrated for the lower-priced stocks, generating a return very close to the spread, ticker-by-ticker.
While I think this pattern retains because it is too small to arbitrage, it is an interesting residual pattern. I think it is best explained by something like this: high vol stocks are targets of intraday trading. This demand is generally positive, and so what you have are returns being depressed at the end of day from day traders selling and closing their positions, returns at the beginning of the day pushed up by the day traders opening positions (generally buying).
26 comments:
I noticed this effect a few years ago. Not only does all of the return come from overnight holds, but since 1994 something like 80% comes from overnight holds in the 24 hours preceding an FOMC statement. The really funny thing is how puzzled they are by it.
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1923197
Keith, did you read beyond the third sentence?
No, I dropped my last Twinkie and it closed the page before I could read any further. Glad I came back though, otherwise I would have missed your insightful question.
Why is the interday/intraday split so strange? (genuinely curious)
The stock market is only open 6 hrs/day. The 8th-grade-economics explanation of the stock market is that it's a machine that digests all information to produce an assessment of the future prospects of a company.
So shouldn't we expect bigger deltas when the 18 hours of accumulated information is allowed to catch up?
I just downloaded data for the S&P 500 from Yahoo, but got exactly the opposite result...
Just to confirm are you using (say):
1. Close on 19th Nov / open on 19th Nov -1 = Open-Close
2. Open on 20th Nov / Close on 19th Nov -1 = Close-Open
I can't see where I have gone wrong - see spreadsheet for formulas... Maybe I made a simple mistake but possibly you have...
https://docs.google.com/spreadsheet/ccc?key=0AkP4kYn6l7XQdFA4aXpNVXdQMTJfNkFqSUtVSnVEcEE
Have you tested this for European and Asian markets? Is it the same phenomenon or are the gains made during their market hours then?
I posted a question here - let's see what comes out:
http://quant.stackexchange.com/questions/4571/when-gains-are-made-overnight-or-during-trading-hours-what-is-the-connection-t
Aren't you aware that stocks trade almost 24 hours a day? It is natural that most of the return will be outside the normal trading session.
Did some research on this subject myself and came to the same conclusion. The strange thing is that it does seems to work for European stocks.
A similar anomaly has been documented here: http://www.livjm.ac.uk/Images_Everyone/2nd_revised.pdf
"Profitable Mean Reversion after Large Price Drops: A story of Day and Night in the S&P 500, 400 Mid Cap and 600 Small Cap Indices" by Dunis et al.
Damien's error is using the S&P500 index, which "opens" at ± the prior close value, then updates only as ticks start to print. This error should be obvious from the zero-ish vol of his series 'Sleeping.' As always, look at the data a bit before coming to conclusions.
Damien: Interesting! Note difference with the SPY, a tradable security. This highlights that indices are best for longer term trends, never high frequency issues.
Peifferc: the stock return is not a function of time, presumably, but risk. Risk is clearly generated primarily during trading hours. Of course, that's standard theory, which I think is wrong, why I note this datapoint in my book.
"...because 2/3 of the risk of stocks is from their intraday returns--measured by beta or volatility"
There are many measures of volatility. Volatility is usually measured by the diffrence in the logarithms of two consecutive closes.
I'm unaware of any volatility metric that would not show most volatility is Open-Close vs. Close-Open.
The high vol stocks are the stocks day-traders like.
By definition a day trader has to close out his position before the close. He/she never earns the overnight return.
=> Falkenstein's second chart implies it is not a good idea to day trade.
Could this be a symptom of computer driven trading and micro-arbitrage?
Typical overnight traders tend to hold stocks for longer, while during the day it has come down to algorithm vs algorithm with no intent of holding stocks at all. They can see everything you do and react to that pretty much instantly, while you just have to guess at what they might do.
Due to the way stock trades are settled on a per-day basis, rather than an intra-day basis, this makes some amount of sense from a capital perspective (albeit not from a risk perspective). If I have $1,000,000 in stock ABC and I want to sell it and buy $1,000,000 in DEF, there's no reason why I have to do both at the same time. In particular, if I buy DEF at the open and sell ABC at the close, I'll hold $2,000,000 of intra-day exposure but I won't have to pay any extra margin costs because both trades are settled on the same day. Since trades are settled on a per-day rather than an intra-day basis, you only need hard capital to hold positions overnight, so in a risk-neutral but capital constrained world, there should be no return to holding stocks intra-day but only from putting up the capital to hold them overnight.
This seems like proof of a positive return to risk. You have much more risk in a position that you can't really trade out of. So you get paid to hold on overnight. If earnings come out, a bomb goes off somewhere, etc, you are locked in until liquidity returns. Yeah, I know there's some volume outside of normal hours, but not enough.
Falken,did you read beyond the third sentence?
I take pleasure in, lead to I discovered exactly what I was having a look for.
You have ended my four day lengthy hunt!
God Bless you man. Have a nice day. Bye
Stop by my web site ... panic attack
Everything is very open with a really clear description
of the challenges. It was truly informative.
Your site is very helpful. Many thanks for sharing!
Feel free to visit my site :: get your ex back
EF: curious how you factored ex-div days into the calculations for the series. thnx
The effect may be coming from the bid-ask bounce noise in the close to open vs open to close returns. Check out the papers by Asparouhova et al (JFE and JF in 2010/2012) that discusses the related bias issues.
All I can say is keep it up. This blog is so necessary in a time when everyone just wants to talk about how many people want intraday trading tips. I mean, thanks for bringing intelligence back to the web, its been sorely missed. Great stuff. Please keep it coming!
Post a Comment