sp500 statistics since 1984

Discussion in 'Strategy Development' started by man, Apr 7, 2003.

  1. man

    man

    I want to discuss some properties of daily data on the sp future. I use the prompt contract. data source is bloomberg. To explain the following: avg daily gain is the return of the day, calculated as ln(Cto/Oto). To those who do not use log returns: within these small numbers it does not make any significant difference whether one uses percentage or log.

    I gonna put in some conditions to see if the properties remain constant. Environment 1 uses the previous days range to see if this has an impact on todays gain.

    Environment 2 calculates yesterday's range as a percentage of the previous 5 day's range and calculates today's average gain.

    Environment 3 is the same as 2 plus yesterday must be an inside bar in reference to the day before.
     
  2. man

    man

    data
    SP500 prompt future
    start 04.01.1984
    end 04.04.2003
    observations 4848

    avg daily range 1.39%
    avg daily gain abs 0.72%

    ENVIRONMENT 1
    yesterdays range
    observations 3058
    upper 1.39%
    lower 0.00%
    avg daily gain abs 0.61%

    yesterdays range
    observations
    upper 10.00%
    lower 1.39%
    avg daily gain abs 0.90%

    yesterdays range
    observations 381
    upper 10.00%
    lower 2.50%
    avg daily gain abs 1.35%

    ENVIRONMENT 2
    yesterdays range in % of previous 5 days average range
    observations 4848
    upper 10000.00%
    lower 0.00%
    avg daily gain abs 0.72%

    yesterdays range in % of previous 5 days average range
    observations 2752
    upper 100.00%
    lower 0.00%
    avg daily gain abs 0.72%

    yesterdays range in % of previous 5 days average range
    observations 2096
    upper 10000.00%
    lower 100.00%
    avg daily gain abs 0.70%

    ENVIRONMENT 3
    yesterday was an inside bar
    yesterdays range in % of previous 5 days average range
    observations 557
    upper 1000.00%
    lower 0.00%
    avg daily gain abs 0.70%

    yesterday was an inside bar
    yesterdays range in % of previous 5 days average range
    observations 517
    upper 100.00%
    lower 0.00%
    avg daily gain abs 0.70%

    yesterday was an inside bar
    yesterdays range in % of previous 5 days average range
    observations 391
    upper 75.00%
    lower 0.00%
    avg daily gain abs 0.67%

    yesterday was an inside bar
    yesterdays range in % of previous 5 days average range
    observations 123
    upper 50.00%
    lower 0.00%
    avg daily gain abs 0.62%
     
  3. man

    man

    It seems that narrow range inside bars do not give a shift in probability of the expected change for today.

    Yesterday's daily range compared to the previous seems also irrelevant.

    But it makes a difference if yesterday had a range above or below the average. Wide range days seem to lead to days with more gain.


    Nevertheless this last conclusion is biased since the average day was know before the analysis took place. So I will rerun it without that bias.
     
  4. man

    man

    running the same test as in environment 1, but making the average dynamic (thus at any point of time the average is calculated only with data prior to that day), changes numbers slightly but leaves the picture in place.


    Summing up this would mean that narrow range inside days are not followed by days that trend more than an average day.


    peace
     
  5. man

    man

    in addition to the sp's reaction to itself I want to look for dependencies with exogene data.

    the general target is to define what influences the sp, but not so much in terms of directional forecasting (at least not in the frist place), but in other trading related forecasting like expected range or expected trendiness.
     
  6. man

    man

    some statistics on month and weekday effects since 1984. first column is the number of observations. second is selfexplaining. third is median of todays gain, fourth is mean of today's gain.

    419 Jan 0.05% 0.02%
    384 Feb 0.00% 0.01%
    435 Mar 0.03% 0.04%
    396 Apr 0.06% -0.01%
    404 Mai 0.04% 0.04%
    402 Jun 0.00% 0.01%
    403 Jul 0.09% 0.01%
    421 Aug 0.04% -0.04%
    380 Sep 0.00% -0.03%
    423 Okt 0.02% 0.04%
    388 Nov 0.08% 0.04%
    399 Dez 0.02% 0.07%

    926 Mo 0.09% 0.06%
    994 Di -0.02% 0.00%
    991 Mi 0.05% 0.05%
    977 Do 0.02% 0.03%
    966 Fr 0.03% -0.03%
     
  7. man

    man

    same analysis since march 2000

    63 Jan -0.02% -0.03%
    57 Feb -0.23% -0.17%
    63 Mar 0.10% 0.07%
    65 Apr -0.08% -0.09%
    66 Mai -0.21% -0.04%
    63 Jun -0.11% -0.02%
    63 Jul 0.00% -0.13%
    68 Aug 0.21% 0.05%
    55 Sep -0.12% -0.18%
    68 Okt 0.27% 0.18%
    62 Nov 0.07% 0.11%
    61 Dez -0.11% -0.05%

    143 Mo 0.14% 0.04%
    153 Di -0.09% -0.12%
    153 Mi -0.13% 0.00%
    152 Do -0.01% 0.02%
    153 Fr -0.02% -0.03%
     
  8. man

    man

    since 2.1.2002, Monday and Friday effect have reverted.

    41 Jan -0.06% -0.12%
    38 Feb -0.17% -0.06%
    41 Mar 0.10% 0.22%
    26 Apr -0.33% -0.25%
    22 Mai -0.23% -0.17%
    20 Jun 0.51% 0.00%
    22 Jul -0.73% -0.24%
    22 Aug 0.48% 0.17%
    20 Sep -0.05% -0.36%
    23 Okt 0.45% 0.37%
    20 Nov 0.33% 0.27%
    21 Dez -0.17% -0.07%

    59 Mo -0.12% -0.10%
    65 Di -0.09% -0.24%
    63 Mi 0.52% 0.27%
    64 Do -0.11% -0.11%
    65 Fr 0.11% 0.09%
     
  9. man

    man

    my personal conclusion from the weekly analysis to keep hands off since these statistics are way too volatile to enhance trading.
     
  10. what was your hypothesis here?

    what were you looking for ?
     
    #10     Apr 7, 2003