Daily Speculations The Web Site of Victor Niederhoffer & Laurel Kenner Dedicated to the scientific method, free markets, deflating ballyhoo, creating value, and laughter;  a forum for us to use our meager abilities to make the world of specinvestments a better place.

Home

Write to us at: (address is not clickable)

12/14/2003
Calendar Effects by Charles Pennington  20:28:46)

Over the past year or two I think I've noticed increasing awareness of calendar effects, especially Yale's 6-month strategy of holding from around the first of November through the first of May, and then "Sell in May and go away." Naturally, the strategy didn't work this year--the Dow was flat from Nov. '02 to May '03, and then up 10% from May '03 to Nov. '03.

Recently the Chair urged us to re-read his chapter on Deception in "Education of a Speculator", which made
me wonder if calendar effects might be deceptive--performing well one year in order to lure folks in for the next year, which then underperforms.

To test, I took the calendar effect to be the percentage return of the DJIA for November 1 through May 1 of the next year, minus the percentage return from May 1 through the following November 1. This corresponds to the returns of the strategy of being long Nov-May and then short from May-Nov. (I take the annual cycle to end in November. Note also that Yale's strategy never has you going short. Nevertheless the number that I'm using is a good measure of how much the favored 6-month period outperforms the not-so-favored one.) Looking from 1930 through 2003 we've had 73 cycles, with an average calendar effect return of 3.3%. But to test for deception, we want to plot year (n+1)'s calendar effect vs year (n)'s calendar effect, and look for correlation.

The best fit is:

(Next Year's Calendar Effect in %)=3.84 - 0.185*(This Year's Calendar Effect)

R=-18.4%

The correlation, then, is negative and fairly large at 18%. With only 73 observations it's on the borderline of statistical significance, with t~1.6, which might not be impressive--yet it's larger than the t for the calendar effect itself! (The 3.3% average return of the calendar effect has a 19% standard of deviation. With 73 observations the standard error is 2.3%, giving a t of only 1.4 for the calendar effect itself.)

So there's another deception example. I expect a nice calendar effect for Nov 2003-Nov 2004 about 5.6% based on the regression above. (Of course more than a month of that period has already elapsed and gone in the right direction, so I have a head start.)