Weekday Patterns

Discussion in 'Technical Analysis' started by travis, Sep 13, 2006.

  1. travis


    List any weekday patterns that you know of.

    I did some study on past daily data of the Dow Jones, and I think we could say that there's been a strong (and constant) tendency in the past 20 years, to go up in the first three days of the week (especially on mondays), and go down, or nowhere on average on the last two days. In other words, if you sum up all gains and losses of the past 20 years, and divide them by week day, you get this:

    Monday 105%
    Tuesday 58%
    Wednesday 39%
    Thursday -14%
    Friday 7%

    In the past 20 years the Dow Jones daily monday to wednesday gains add up to 200%, which is not just 200% because they are "compounded" daily (it goes from 2000 to 11000). Yet on thursday and friday, they add up to -7%. I think it is a pattern. Any comments? You know of any others?
  2. m4a1


    are you including gaps? could it be that mondays show such a big difference because there are 2 days in between, so that gap is larger?
  3. travis


    How would I measure that mondays or any other day goes up or down, if not by subtracting yesterday's close from today's close? Gaps are not relevant to this process. I am not talking about volatility where you distinguish between Range (not counting gaps) and True Range (counting gaps).
  4. m4a1


    what does monday 105% mean?
  5. travis


    It's what you get if you sum up all gains and losses of all mondays in the past 20 years.
  6. just to clarify you are taking the close on Friday and subtracting the close on Monday to come up with the %'s for Monday?

  7. going back to Octber 1980 here are the results i get: (these are % returns)

    Monday 75%
    Tuesday 97%
    Wed 79%
    Thurs 10%
    Friday 24%

    While mine aren't exactly the same (I did use different starting date) they are somewhat consistent with your results.

    Also seeing how the 1987 crash happened on a Monday and Tuesday the results are that much more impressive.
  8. how about breaking the whole period into 4 different periods and testing each period separately..Larry Williams at some point was a fan of using these dayofweek filters in combination with other indicators
  9. travis


    This is to clarify the very simple calculation I did.

    To come up with the figure for each day, all I do is add up the specific day's gain/loss for the years of the test period, which on my daily closes data for the Dow Jones means the following - monday's minus (previous) friday's close, tuesday's close minus (previous) monday's close...very very simple.

    Take the 52 mondays in a year, and add all their % gains/losses, and do the same for every other day.

    Well, if you add up all mondays in the past 18 years, you get what I told you, and I think it's not a random even, but to the contrary very significant.

    Regarding giggollo's idea of breaking the 75 years into four periods, I tried all that as well - with Excel is quite simple, as there is an immediate average function on the lower right, on the status bar. No matter what periods you divide the data into, as long as it's large enough (at least 10 years), for years 1930 to 1980, you get just the opposite, with monday being a constant down day. I don't know why it happens, however for my system I will use the findings of the last 18 years.

    Any ideas on why this could have happened? It doesn't seem to be related to the overall up or down trend.
  10. m4a1


    did it happen abruptly in 1981 or was it a gradual change?
    #10     Sep 14, 2006