Tuesday, May 5, 2009

May Effect In Stock Markets: True Or False?

The Malaysia-Finance blog the other day brought up an interesting seasonal effect that I never heard of before: to wit, when investing in stock markets sell in May and go away (until October). Essentially this divides up the year into a bull period (Nov to April) and a bear period (May to Oct). Ordinarily I dont't bother much with equity market analysis – it`s not really my field – but this sort of seasonal effect is amenable to statistical testing.

Rather than doing something complicated like a seasonal ARCH model or laborious like directly calculating compound returns for each subperiod, I hit on a fairly simple way to prove the May effect (anyone feel free to tell me this is wrong). In essence I cut the daily return* data I had through using two dummy variables (E1 for Nov to April, E2 for May to Oct) which took the value of 1 for each subperiod and 0 for the opposite period. This gave me two daily return series, each with its own sample mean and distribution. All that was then required to prove the May effect was to verify that the Nov to April series had a higher mean, and that this mean was statistically significantly different from the mean of the May to Oct series.

*returns are calculated as the log difference between two trading days.

It took longer to write about than to do.

Here are the results tabulated for the markets that I have coverage for. Mean difference is the net difference between the sample means of series E1 and E2; F-stat is the value of the ANOVA F-Statistic (t-tests yielded virtually identical results); and Prob. is the threshold on the test probability distribution where the null hypothesis is not rejected (1 minus prob. gives the confidence level). The results are ordered from strongest to weakest:

^STI
Mean difference: 0.0005
F-Stat: 8.94
Prob.: 0.003

^DJI
Mean difference: 0.0003
F-Stat: 8.24
Prob.: 0.004

^TWII
Mean difference: 0.0008
F-Stat: 6.97
Prob.: 0.008

^JKSE
Mean difference: 0.0009
F-Stat: 6.75
Prob.: 0.009

^GDAXI
Mean difference: 0.0005
F-Stat:F-Stat: 5.22
Prob.: 0.022

^FTSE
Mean difference: 0.0003
F-Stat: 5.15
Prob.: 0.023

^IXIC
Mean difference: 0.0003
F-Stat: 4.85
Prob.: 0.028

^N225
Mean difference: 0.0004
F-Stat: 4.45
Prob.: 0.035

^KLCI
Mean difference: 0.0005
F-Stat: 4.42
Prob.: 0.036

^GSPC
Mean difference: 0.0002
F-Stat: 4.23
Prob.: 0.040

^KS11
Mean difference: 0.0007
F-Stat: 3.51
Prob.: 0.061

^SSEC
Mean difference: 0.0007
F-Stat: 2.08
Prob.: 0.150

^AORDS
Mean difference: 0.0002
F-Stat: 1.99
Prob.: 0.159

^HSI
Mean difference: 0.0002
F-Stat: 0.81
Prob.: 0.370

^BSESN
Mean difference: 0.0001
F-Stat: 0.06
Prob.: 0.800

Quite a few surprises here, although I'd encourage anyone to read through the caveats before taking any of these conclusions at more than face value. First though is that the May effect appears to be pervasive – two thirds of the 15 markets covered here have better than a 95% chance that the May effect is true. The second surprise is that developing markets are as likely as developed markets to exhibit the May effect. The third surprise was the difference in probabilities for the Dow Jones Industrials and the broader S&P500, though I suspect the difference here is that less of the S&P500 counters are within the investment universe of investors.

Now for the caveats - the sample sizes for each index is different, with the SSEC the shortest at just four years. Most of the rest date back to the 1990's, although the developed country indexes go back further (US markets are from 1970). The impact of this is that for the short sample indexes, the degree of the May effect is less certain. That also makes the ordering that I've done here much less reliable than I would like. One other related problem here is that I've only covered a small sample of the available markets, and a larger sample might reach different conclusions.

In summary though, what's clear is that if you're thinking of trading on the basis of the May effect, you're likely to end up in the money – but you have to be careful and confirm that there's a historical basis for it in the market you're in. There's no substantive dividing line between developing and developed markets here. You should also be aware that the difference in yield between markets of the May effect can be strikingly different – it might not be worth the trouble, especially if you have a mandate that requires a certain level of constant exposure to equities.

2 comments:

  1. Hi Hishamh,

    Many of these anomalies observed on developed countries but investors are not sure whether it’s applicable to developing countries. It’s good that you did a study.. Thx.

    Haha, I think you can write many research papers on these anomalies and get them published.

    Wonder what could be the reason for this May effect.
    Could it be due to the period of Nov – April:
    1. contains more festives such as Christmas, New Year, Chinese New Year (for countries with Chinese), Hari Raya etc.. ?

    2. contains annual reporting season (management may try to “make” annual report performance better at the expense of 1Q report) ?

    ReplyDelete
  2. Lot's of these studies have already been conducted - you just have to know where to fund them. There's a couple more I do on a regular basis as a check, which I'll put up when I have the time.

    As for why this particular seasonal impact seems to work - I don't want to make generalisations, because every market is different, but your guess is as close as anyone elses.

    ReplyDelete