Pricing Improvements

It's very important to us that our prices be accurate.  Thousands of visitors each day depend on the prices being correct.  They use the prices to keep their family business profitable or save a little cash on their favorite hobby.  Of course, we also eat our own dog food, so inaccurate prices annoy us as much as they annoy you.

One of the trickiest parts of running our site is trying to calculate accurate market prices for illiquid markets.  It's easy to price Call of Duty: Black Ops which sells more than 50 copies each day and has 200 listings between Half and Amazon.  It's quite a bit harder to price items like Cannibal Women in The Avocado Jungle of Death which only sells a couple times each month and has a couple dozen listings.  Many items have even less liquidity.

For the last two summers, we've added features allowing visitors to report bad auctions and incorrect prices.  We've analyzed this data and adjusted our pricing techniques to improve on the problems you discovered.

Removing Bad Sales

About 65% of inaccurate prices were caused by incorrectly listed eBay auctions.  Common examples are sellers who list Famicom games as if they were SNES games or list their collection of Xbox games as if it were a single game.  Of course, a brand new Chrono Trigger magnet is not the same as a sealed copy of the classic RPG itself.  Our pricing bots now recognize these kinds of problems and automatically remove 100-200 junk auctions each day.

You can still report bad auctions with the icon.  That helps us further refine our system and keep prices accurate.

More Price Signals

A quarter of our inaccurate prices came from placing too little emphasis on high quality auction sales.  We previously weighed all eBay sales the same.  Over the last couple years, we've noticed that fixed price sales on eBay are less accurate than hotly contested auctions.  With a high-bid auction, at least three people agreed on the final price (seller, buyer, second-place bidder) and possibly more.  We now give more weight to healthy auction sales.

Ridiculous listing prices caused about 10% of the inaccurate prices.  Just because someone lists his favorite copy of Shaq Fu for $100 on Amazon doesn't mean it's actually worth that much.  For thinly traded games, this is a real problem.  By analyzing listing price variations over time, we can determine which prices are reasonable and which are wishful sellers fishing for easy money.  In our analysis of thinly traded games, where this new signal is used, prices improved 66% of the time and only rarely got worse.

Conclusion

We continue to improve our pricing algorithms, as we've done in the past, to make sure we provide the most accurate movie and game prices possible.  If you have any suggestions for further enhancements or questions about these changes, please let us know in the comments below.