I have recently downloaded daily market data for salvaged materials at the market hubs, including the median, min, max and standard deviation values. Occasionally, there are large discrepancies between the max and stddev. For instance, the median ‘buy’ price for Contaminated Nanite Compound in Amarr on 4/4/21 is 7060 ISK. The stddev is 5154.2. However, the min /max range is 5580 to 8460. Clearly, given an stddev of over 5000, the max value of 8460 is wrong. It should be somewhere in the vicinity of 19,000+ ISK. (I’ve made actual purchases at this level in Amarr in early April.)
Does anyone know what the definition of min/max is? Or some explanation of these large discrepancies.
There’s several possibilities…
You are right if the price distribution is a Gaussian - a feature of a random distribution. I’m fairly confident that that doesn’t apply to market orders where the value of a new order is related to the existing orders - and not just random in the range.
It may be that there is a lot of the prices out near the maximum/minimum mark and it’s fairly thin in the mid_range of prices. Sort of two peaks - and in this case they are fairly even so the median is falling between them. That will skew SD.
I would need to dig into the details - pull down the orders and have a look. It is quite possible that “max/min” is actually based on the transactions (filled offers) rather than the orders (offers) on the market.
If that’s the case, I suspect that the Standard Deviation could be the distribution of offers/orders.
Where I’ve pulled down pricing information from ESI then I’ve not worked with any of the aggregated statistics, but the actual orders on the market at the time of the query (and whatever latency there is in that record set).
Basically the usual interpretations of some of the metrics (SD) isn’t going to apply because the assumptions underlying those interpretations (random distributions) aren’t valid in the market.
Thank you.
I agree you’re notion of a median falling between double peaks is a possible explanation for the SD. That leaves the question of max/min. Are these calculated values or actual? I can confirm that the actual max values (based on my purchases) are low by a factor of 2 to 3 (for some salvaged materials) compared to the reported max.
So, it would appear that, along with SD, the max/min values are calculated (inferred from the SD?), rather than actual transactions.
May I impose on you with further questions?
Background:
I’m an alpha; in Eve less than a year. I’ve accumulated enough ISK to go Omega. Would like to generate enough ISK to maintain Omega status. I’m involved in farming and ratting. Recently, ratting wrecks no longer release cargo containers; instead only ‘salvaged materials’ are found at higher frequencies and values than the past. For most of these materials, there are large deltas in buy/sell values between trading hubs. I’m interested in whatever reliable stats I can find to accurately identify those deltas.
If max/min’s don’t reflect actual transactions, and SD doesn’t help because the distributions aren’t normal, is there any available metric that could describe the distribution? Or, might you have another suggestion?
Your feedback would be much appreciated.
I believe price outliers can be omitted from the statistics? Would this account for it?
1 Like
I don’t think so. I haven’t seen any price outliers. Do you have an example?
At this point, the problem is that the reported ‘buy’ max values can be 2 -3 times lower than the actual ‘buy’ prices. This is consistent with the large reported stddev values. So, either the stddev is wrong or the max values are wrong, OR the max value is defined to be something different than the actual transaction max (perhaps that value is an ‘offer’ rather than a transaction?).
Does anyone know where the stats are defined? (Specifically, what is the definition of ‘percentile’?)
Is there a way to download actual transaction data to compare against the aggregate data?
It’s more than that, CCP delete outliers.
The reasons
A) It crushes in game bar graph/line graphs.
B) If reinforced it messes up average prices which disables regional price warnings.
The above is not automated in anyway I can tell. Some item markets have crushed graphs for months others get “fixed” 24hrs later.
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.