Below are the annual logarithmic average and standard deviation of returns to some common stocks, and two primary ETFs, SPY and QQQ (percent returns look similar). I used as much data as every stock had, where IBM goes back to 1962, but QQQ starts in 1999. As you can see, the average returns for every stock except MSFT was significantly greater overnight than from open to close. In fact, on average, stocks had a negative return during trading hours, even though 2/3 of the 'risk' is intraday. This is yet another instance of how the theory that risk begets return is contradicted by reality.

Start Date | Ticker | Open-Close | Close-Open | stdev(O-C) | stdev(C-O) |

1999 | QQQ | -13% | 14% | 29% | 16% |

1993 | SPY | -3% | 10% | 17% | 10% |

1984 | AAPL | -13% | 30% | 41% | 29% |

2008 | TNA | 3% | 43% | 78% | 46% |

1986 | YRCW | -25% | -1% | 61% | 30% |

1986 | MSFT | 12% | 10% | 32% | 21% |

1962 | IBM | 2% | 6% | 22% | 12% |

| Avg. | -5% | 16% | 40% | 23% |

Tom Anichini has found this in several ETFs, see

here.

## 9 comments:

Isn't there a fairly simple "limits to arbitrage" story as to why this might happen -- namely that even 7bp of per-day transaction costs would erase the return difference between the two for SPDRs, for example?

I'm not saying there's an arbitrage. Indeed, if I thought there were, I wouldn't have posted it. It's just rather strange, and anomalous.

Interesting - I took a look myself at SPY cyclicality. The O-C return seems to dominate during speculative periods - 2003, 2009 rally - but otherwise C-O is fairly dominant.

I'm curious, what is the definition of risk? Is yours different from the mainstream?

Reading Benjamin Graham, his definition of risk was something to the extent of "the probability that I will lose all capital invested"?

Thats how I've always looked at it.

hello eric,

first of all congratulations for the blog. I ran into it by accident yesterday and only wish I found it earlier.

with regard to your post, I am no expert, but to me none of those numbers looks statistically significant. close to open and open to close could well be part of the same population (if we assume normal distribution with standard deviation as calculated by you). the difference between the two means is not that big either considering the expected distribution of the difference of two means.

I am probably missing somethig, and would be very grateful if somabody could point it out :)

once again thanks for the blog.

me again... actually your samples are huge in size. the difference between open to close return and close to open can't be random. sorry for that.

This gives me an idea for two new etf's.

Standard deviation is a poor measure of risk in this case. When holding stocks from close to open, you likely incur significant tail risk due to earnings reports etc. Skew and kurtosis are important.

Not sure this passes the ideological Turing test. I would doubt that Fama or Sharpe would say that the returns from Close to Open are due to increased risks. However, I suppose there are liquidity risks that aren't accounted for. The average person can't trade when the market is closed. Further, most economic/earnings data is released over this period, compounding some of that problem.

Post a Comment