Not signed in (Sign In)

Auctioneer::Technical: Statistics and Auctioneer

Bottom of Page

1 to 10 of 10

  1.  
  1.  
    First of all hallo to everybody, this is my first registred post here. I've been playing WoW for ½ year, and used Auctioneer for three months or so. It is truely a nice addon.

    Now to the question. Am I correct in my understanding that Auctioneer calculates the BO prices based on the 35 items closest to the median price? I understand the need to avoid using a "clean" mean number because of items priced far to high or low (outliers), but there are most sensitive ways of removing these. In statistics one usually classifies a outlier as having a value that falls more than 1,5 IQR (Inter Quartile Range) above the upper quartile or more than 1,5 IQR below the lower quartile. IQR is identified by sorting the sample into four groups according to size with an equal amounts of cases in each. Then one notes the lower quartile, which is the highest value in the group with the lowest values. And on notes the upper quartile, which is the lowest value in the group with the highest values. Deduct the lower quartile from the higher and you have the IQR. An example:

    Sample:

    2, 3, 1, 7, 3, 7, 9, 2, 7, 1, 7, 2

    Sorted sample:

    1, 1, 2, 2, 2, 3, 3, 7, 7, 7, 7, 9

    Lower quartile is 2, upper quartile is 7 => IQR is 5

    Hence any number great than 14,5 or lower than -5,5 would be a outlier. In this case there are none.

    Therefor a mean could somewhat safely be calculated. Had there been outliers these should be ignored when calculating the mean.
  2.  
    Posted by: Sutekh on Jan 18 2006, 08:29 AM
    First of all hallo to everybody, this is my first registred post here. I've been playing WoW for ½ year, and used Auctioneer for three months or so. It is truely a nice addon.

    Now to the question. Am I correct in my understanding that Auctioneer calculates the BO prices based on the 35 items closest to the median price? I understand the need to avoid using a "clean" mean number because of items priced far to high or low (outliers), but there are most sensitive ways of removing these. In statistics one usually classifies a outlier as having a value that falls more than 1,5 IQR (Inter Quartile Range) above the upper quartile or more than 1,5 IQR below the lower quartile. IQR is identified by sorting the sample into four groups according to size with an equal amounts of cases in each. Then one notes the lower quartile, which is the highest value in the group with the lowest values. And on notes the upper quartile, which is the lowest value in the group with the highest values. Deduct the lower quartile from the higher and you have the IQR. An example:

    Sample:

    2, 3, 1, 7, 3, 7, 9, 2, 7, 1, 7, 2

    Sorted sample:

    1, 1, 2, 2, 2, 3, 3, 7, 7, 7, 7, 9

    Lower quartile is 2, upper quartile is 7 => IQR is 5

    Hence any number great than 14,5 or lower than -5,5 would be a outlier. In this case there are none.

    Therefor a mean could somewhat safely be calculated. Had there been outliers these should be ignored when calculating the mean.

    You're very welcome to code this into Auctioneer yourself . An algorithm like this has been suggested lots of times before, however none of us (that I know of) are anywhere near the field of statistics so we can't implement this ourselves.
  3.  
    Posted by: MentalPower on Jan 18 2006, 09:17 PM
    You're very welcome to code this into Auctioneer yourself . An algorithm like this has been suggested lots of times before, however none of us (that I know of) are anywhere near the field of statistics so we can't implement this ourselves.

    I'm taking a university course in statistics (I'm studying Political Science). I don't know nothing about coding though. So while I could find the matematical formulars I wouldn't have any idea of how to implement them.
  4.  
    test
  5.  
    Where is the median price calculated?
    I've taken a few stat classes and am a programmer by trade... if you point out where I should be looking for the median calculation I could make the modification Sutekh suggested and post the diff here
  6.  
    Posted by: Guest on Feb 23 2006, 02:54 AM
    Where is the median price calculated?
    I've taken a few stat classes and am a programmer by trade... if you point out where I should be looking for the median calculation I could make the modification Sutekh suggested and post the diff here

    I have a better idea, why don't you register and we can get you dev access to the repository.
  7.  
    Hey,

    If possible I would like to take a look at the algorithm as well. The main problem I've seen with auctioneer (and it is a great project) is that it does not identify the market clearing price, only the price the item is listed at even if it doesn't sell. Thus there is an upwards bias in the estimate. Discounting outliers is useful in this regard but still doesn't solve the problem, nor would increasing the number of observations.

    Anyway, I think there are ways to more clearly identify the market clearing price, which should be the goal. Does anyone feel like working on this? I have some knowledge of programming but my strength would be in economics and price theory.
  8.  
    On the subject of outliers, it would be awesome if there was a function to flag the low outliers. That way you could pick up steals on items placed well below their approximate value.

    Posted by: ummd on Mar 7 2006, 05:20 AM
    Hey,

    If possible I would like to take a look at the algorithm as well. The main problem I've seen with auctioneer (and it is a great project) is that it does not identify the market clearing price, only the price the item is listed at even if it doesn't sell. Thus there is an upwards bias in the estimate. Discounting outliers is useful in this regard but still doesn't solve the problem, nor would increasing the number of observations.

    Anyway, I think there are ways to more clearly identify the market clearing price, which should be the goal. Does anyone feel like working on this? I have some knowledge of programming but my strength would be in economics and price theory.
  9.  
    Posted by: ummd on Mar 6 2006, 11:20 PM Hey,

    If possible I would like to take a look at the algorithm as well. The main problem I've seen with auctioneer (and it is a great project) is that it does not identify the market clearing price, only the price the item is listed at even if it doesn't sell. Thus there is an upwards bias in the estimate. Discounting outliers is useful in this regard but still doesn't solve the problem, nor would increasing the number of observations.

    Anyway, I think there are ways to more clearly identify the market clearing price, which should be the goal. Does anyone feel like working on this? I have some knowledge of programming but my strength would be in economics and price theory.


    I have rewritten the algorithm in Auctioneer to use IQR for the calculations. It is available in builds 734 and up. The new calculations are as follows:

    Determine IQ1 (25%), IQ3 (75%) IQR ((IQ3-IQ1)*1.5) then exclude anything below IQ1-IQR and anything above IQ3+IQR. Median is then calculated on the resulting dataset. Important to note that the outliers are NOT removed from the dataset to allow for market trends up/down, they just are not used to determine current median.

    I'm not a statistics genius by any stretch, but this seems to have helped more accurately show central tendency than before the change.
  10.  
    Posted by: Karavirs on Mar 7 2006, 04:09 PM
    I have rewritten the algorithm in Auctioneer to use IQR for the calculations. It is available in builds 734 and up. The new calculations are as follows:

    Determine IQ1 (25%), IQ3 (75%) IQR ((IQ3-IQ1)*1.5) then exclude anything below IQ1-IQR and anything above IQ3+IQR. Median is then calculated on the resulting dataset. Important to note that the outliers are NOT removed from the dataset to allow for market trends up/down, they just are not used to determine current median.

    I'm not a statistics genius by any stretch, but this seems to have helped more accurately show central tendency than before the change.


    That's a good first step... Does auctioneer discriminate between auctions that have a bid and auctions that don't, or does it just use all listings to construct this median?

1 to 10 of 10

  1.  
World of Warcraft™ and Blizzard Entertainment™ are trademarks or registered trademarks of Blizzard Entertainment, Inc. in the U.S. and/or other countries.