I am trying to create the algorithm that will tell me the good price to buy stuff and a correct re-selling price based on eBay item history.
The main variables would be the auction type (buy now or bidding), price, if the item was sold and some sort of margin that has to be made on each transaction (in percents or dollar value)
This is the spreadsheet with example data: Excel sheet. The first few rows appear below.
AUCTION_TYPE PRICE DATE_SOLD SOLD_Y_N BUY-NOW $300.00 May-10 22:26 Y AUCTION $300.00 May-13 09:11 Y AUCTION $300.00 May-13 09:12 Y BUY-NOW $474.89 May-14 12:30 N
It is a table of Auction_Type (a binary covariate equal to "Auction" or "Buy now"), Date_Sold (to the minute), Price (consistently in American dollars)--which is an outcome for auctions but is predetermined for "Buy now" sales--and Sold, a binary outcome indicating whether the item was sold.
Herein is one of the principal complications: Price, which is something I would like to estimate, functions sometimes as an "independent" variable and sometimes as a "dependent" variable, depending on the type of auction. This situation therefore does not seem to fit within a standard regression setting. A good solution may require some innovative statistical thinking.
Well, I think that I we need to establish some sort of MINIMUM amount of sold/unsold items per week. For example, it is hard to predinct anything if there are 5 records exist. But if we would work with 50 records per week and with the period of coupld of weeks, we could narrow the probability.
In addition, autioned items have the higher probability to be sold for lesser value, than the buy-it now. But, on another hand, some unsold auctions can be a warning sign that indicated that this item might not be that valuable.
What I am trying to figure out- is there any resonable solution can be created that would allow to buy things based on the item selling/unselling history, and then to be able to resell the same item with the profit and with very high probability of success.