Penny Auctions and Applied Statistics

Disproving the idea of a constant "strike price"

2013-01-31T15:23:00.000-08:00

I had an interesting discussion today with one of my professors about volatility and how it fits into the penny auction world.

The preface, his background is in Risk Analytics and econometrics. One of his main topics of discussion is around volatility in the stock market. Financial theory suggests that in stock prices, higher volatility is correlated with higher risk. Likewise, lower volatility is correlated with lower risk. This counters the idea that "strike price" is a fixed value for all possible auctions.

At a macro level view, the auctions for gift cards appear to have varying levels of clustered volatility (The reference line in both charts is the mean):

With clustered volatility, a GARCH model can be useful in predicting the expected variance, or volatility, around the mean. For users looking to "strike a deal", bidding in one of the bands of low volatility would likely be a better use of their money, as the volatility and expected costs are lower. Using the entire data set from June 2-5, we can use an GARCH model (1,(1 15)) with an exponential distribution to see some of the change in expected volatility:

Exponential GARCH Estimates
SSE	7535.92932	Observations	671
MSE	11.23089	Uncond Var	.
Log Likelihood	-1743.004	Total R-Square	0.0050
SBC	3544.58696	AIC	3504.00804
MAE	2.62665134	AICC	3504.28036
MAPE	3234.0269	HQC	3519.72475
		Normality Test	459.8457
		Pr > ChiSq	<.0001

Parameter Estimates
Variable	DF	Estimate	Standard Error	t Value	Approx Pr > \|t\|
Intercept	1	-4809	2677	-1.80	0.0725
auctionendtime	1	2.9085E-6	0	Infty	<.0001
AR1	1	-0.0204	0.0372	-0.55	0.5839
AR2	1	0.0745	0.0363	2.05	0.0403
EARCH0	1	0.6534	0.0613	10.66	<.0001
EARCH1	1	-0.1449	0.0341	-4.25	<.0001
EGARCH1	1	0.8578	0.0221	38.76	<.0001
EGARCH15	1	-0.1340	0.0157	-8.56	<.0001
THETA	1	0.4318	0.1766	2.45	0.0145

The output here suggests that lag 15 is significant (seems kind of odd). We probably need more data to support this. In layman's terms, this suggests that an auction end price 15 periods prior has an effect on the next one.

This chart shows the predicted standard deviation in black. As you can see, there are significant differences of what the standard deviation is at certain points in time. So for those of you who use the idea of a "strike price" at 2 or 3 standard deviations away, you are ignoring the fact that the volatility isn't constant across time.

Extracting a business proposition from data

2013-01-11T10:28:00.002-08:00

It has been a few months since my last post. Turns out a 10 month masters degree takes most of your free time!

One of the last things I was working on was turning some of my insights into value. My original idea was to use statistics to win auctions, sell items, and try to optimize a rate of return. Going back to the drawing board, I came up with an idea that flipped the previous one on its head:

What if I provided a service that would give a user the statistics they needed to be successful (for a fee), while minimizing overhead costs and risk on my end?

Using a business strategy like this would allow me to tap the vast user base that come to websites like Quibids.com without risking much capital on my end. In return, I could provide the data to users with advanced statistics for a monthly fee along with other types of value items (such as tutorials). There are many things you can do with a subscription service that would let scale up much easier than if you were buying the items outright. Plus it would give me a chance to do statistics and teach people how to use them (Which is what I wanted to do anyways).

Feel free to contact me or leave a comment with what you think.

More survival analysis stuff

2012-10-08T15:49:00.000-07:00

One of the things I am trying to understand are the patterns at the end of an auction. If there is a way to visually understand who has entered and exited an auction along with who remains, you would have a significant advantage over other users. In this case, basic survival charts can show a play-by-play analysis of what occurred in an auction. The chart below is from a 50 bid Voucher auction:

A visual inspection suggests that many of the bidders have a fixed rate of bidding. Once they exhaust their bids, they abruptly stop bidding. These people would be easy to model into a given auction (if we have enough historical data for the bidder). Even without it, we could create a function to predict the expected quantity of bids based on the rate of bidding. Some confidence interval could also be used to suggest entry points in an auction based on their expected exit from the auction.

The real problem is predicting these values. In a live auction, we can only plot the survival of a user once they leave an auction. One idea would be to plot the "expected" survival plot of each user as they enter the auction. Once in, we could provide a pseudo-survival rate based on historical data or a function as described previously.

I am also looking into using proportional/nonproportional hazard models to try to predict an expected exit time of each bidder. More to come.

Survival Analysis and Hazard Functions

2012-10-02T08:05:00.000-07:00

If we want to predict the ending point of an individual auction, it is important to understand the hazards that occur at certain points in the auction. By understanding these patterns in an auction, we can understand specific points of attrition for individuals. It is important to note that the hazard function is not related to winning an auction. Just because you last the longest in an auction does not necessarily mean you won an auction. In many cases, others will swoop in near the end an win with relatively few bids.

For all of the charts, they are from auctions for iPads. They are a random sample and should give a general idea of the increasing hazard rates (NOT hazard probabilities) during each auction. If you are not familiar with hazard rates, they are proportional to hazard probabilities. In essence, a higher value suggests that the hazard of leaving (or likelihood of leaving the auction) is higher. The spikes suggest that there is a high likelihood of quitting at low time duration, medium time duration, and high time duration.

Auction 1 for iPad

Auction 2 for iPad

Auction 3 for iPad

Auction 4 for iPad

In general, there appears to be a W-Shape or U-Shape to the hazards. There may be some relationship between hazard pattern and the maximum time in the auction. The maximum time in the auction is highly correlated with the ending auction price (for someone to be in the auction for a long period of time, they have to bid for a long period of time OR start and stop over a long period of time).

Later on I will model expected end points for auction using this phenomena and other variables.

Insights into bidding behavior

2012-08-13T05:58:00.001-07:00

Until this point, all I had to work with were total bid quantities per bidder. This was useful for basic modelling, but not much more than that. With the new data set provided by http://www.quibidsinsider.com, I have been able to drill down much deeper into bidding behavior. This data has bid times tracked by the second as well as their bid number.

This information reveals insights into each bidder. For reference, this auction was for a $50 Walmart Gift Card and ended with 1720 bids ($17.20). Each circle represents a bid. The line is just a general trend line.

Some bidders will bid continuously until they decide that they don't want to play anymore. These are generally the strict BidOMatic bidders who set strict bidding ranges.

Most bidders do not bid linearly. In fact, there are a variety of different strategies employed by each user. Some bid hard early, wait, and try to bid again later (as seen below).

Others will bid lightly and sporadically during each auction without any discernible pattern.

In reality, these patterns aren't particularly useful when trying to understand how auction winners win. To determine that, we need to examine what an auction winner does.

In this case, 5starebayer won the auction. He/she employed a concentrated burst attack over about a 20 minute time period. Perhaps this was when the user detected the exit of an aggressive bidder or when a BidOMatic bidder ran out of bids. In any case, the strategy worked (in this particular auction) while others did not.

It remains to be seen if each user has similar bidding patterns in each auction. This is something I will be investigating further.

A new shift in direction

2012-08-05T14:10:00.001-07:00

Several things have occurred since my last post. I would like to summarize them as good and bad news. Let's start with the bad, just to get it out of the way:

The Bad:

First, I started my graduate program. It has been extremely time intensive and has put me into a position where I have little time to maintain or fix the existing script. I am announcing that I will not be fixing or updating the script in any way. However, the script is still a great jumping off point for anyone looking to try to do something similar.

The Good:

I have teamed up with a fellow North Carolina State researcher who operates a fully automated Quibids scraper (http://quibidsinsider.com/). Since he is actually a programmer, the capabilities of the website are significantly greater than the script I wrote. The best part is that I will now have access to more data. And with more data, I can do more analysis (You know, the original purpose of this project).

So from now on, the blog will be solely based on the analysis of that data. Ultimately, I would like to create some kind of predictive modeling to determine probabilities of winning auctions based on lots of factors (User behavior, time of date, etc.) It may even spin off into a pay service similar to www.allpennyauctions.com. Who knows. Should be an interesting adventure.

Quibids website changes

2012-06-28T07:12:00.001-07:00

Well it was bound to happen. Quibids changed the design of their website. By doing so, they crippled my data mining script. I will have to investigate and update it to get data mining again.

Pennystats Chart of the Day - 6/19/12

2012-06-19T10:54:00.001-07:00

Its been a few days since my last post (just got back from vacation), so I tried to come up with something that will really help bidders understand what they are up against in a competitive auction.

I decided to look at $25 gift cards again to quantify how much people will typically bid in a given auction. To my surprise, they were fairly consistent with what I uncovered in the $10 gift card auctions. Most of the aggressive bidding occurs between 2am and 8am. This further compels me to believe that bidders are more aggressive when there are fewer auctions available.

Bidding aggressiveness nearly plateaus between 12pm-8pm, then steadily ramps up for the evening. In general, the weekends appeared to be the time for least aggressive bidding. I would suggest that the best time to aggressively bid on $25 gift cards would be on the weekends between 12pm-4pm, when there are plentiful auctions available.

Pennystats Chart of the Day - 6/15/12

2012-06-15T18:03:00.000-07:00

Profiting from winning auctions is another challenge. Since retail prices on Quibids is often an exaggeration or even outrageous, winning auctions that can easily be flipped is important in generating a steady revenue stream.

The Plastic Jungle is a company that offers to resell your gift cards for a percentage of the value. Like the varying auction prices on gift cards, Plastic Jungle offers varying redemption rates. They range from nearly 97% to about 84% of retail price.

An important side note is that the minimum gift card size they accept is $25. This limits the auctions you can target, but the majority of the auctions are $25 and up.

Pennystats Chart of the Day - 6/14/12

2012-06-14T19:46:00.000-07:00

All equal value gift cards shall not be treated as such!

Looking at $25 gift cards (my favorite range since they are the majority of the data), not all gift cards sell for the same price. In fact, some of them are down right neglected. The majority of the bass pro shop giftcards sell, on average, for less than 40 total bids. Compare that to the Walmart gift cards who average nearly 80 every day of the week. By far, the best value for gift cards are the Sunoco gas cards on the weekend.

Surprisingly, it appeared that the best time to attempt to win these gift cards is on the weekend. Since there are an increase in auctions on the weekend, more of these cards are listed. One theory suggests that competitive bidders are focused in other categories during the weekend while $25 gift cards are neglected. I saw a similar trend while looking at $10 gift cards.

Protip while using XLHost

2012-06-14T10:53:00.002-07:00

Here's a little tip for anyone who decides to use XLHost for dedicated server hosting....

TURN OFF WINDOWS UPDATE!

If you don't, Windows will restart and it will screw up your data collection...

Like it just did for me....

More data for analysis and insights

2012-06-13T09:59:00.001-07:00

This week I am collecting data from all the Electronics and Voucher auctions. From a quick glance of the data, it appears that auctions behave differently than gift cards. This is an interesting and important indicator. It proves that in order to win auctions for a low cost, bidders should understand the types of auctions they are dealing with.

Instead of blindly bidding on an item based on vague "averages" that do not consider time and day, bidders need to consider the charts that have been posted to know if they truly got a bargain. Most importantly, it is clear that winning low cost auctions requires discipline and calculating risk before you place bids.

For instance, if it was Saturday afternoon and you were attempting to win a $15 gas card, your first instinct should be figure out what the average selling price is for that day and time. To get a low cost card, you may way to try your hand at multiple auctions. If you notice that one auction has more action than the other, you should focus your interest on the least active auctions. You should also have an exact number that you are willing to spend in order to get that auction. Should you surpass it, it's time to regroup and identify another potential auction. Run away bidding and Buy-It-Now auctions are a waste of time and money.

Methodical bidding is key to making money in these auctions. If you have a good strategy and stick to your bidding system, you will make money.

Pennystats Chart of the Day - 6/13/12

2012-06-13T07:15:00.000-07:00

Following the theme of the $10 gift card data from yesterday, I took a look at $15 gift cards selling averages. In general, the trends were similar to $10 gift cards. Averages and variation was significantly higher in the early morning and were minimized in the mid-afternoon.

One interesting trend that occured was the minimization of average bid totals on the weekend. Saturday between 2-10pm were the best times to land a $15 gift card for the lowest average price. The data is very different from what we showed in the $25/$50 gift card range.

Pennystats Chart of the Day - 6/12/12

2012-06-12T12:02:00.000-07:00

Reviewing $10 Gift Card auction data revealed another interesting aspect of Penny Auction psychology. Today's chart shows the average total bids of each auction based on the day of the week and the end time of the auction. Unlike the chart from yesterday, the average auction prices increase during off-peak hours.

Stranger still, the lowest, most consistent averages with the least variance occured during peak times between 2-8pm. One theory I have is that during off peak hours, there are fewer auctions listed. Because of this, competition grows on low cost items (Such as $10 Gift Cards).

Pennystats Chart of the Day - 6/11/12

2012-06-11T10:24:00.000-07:00

After a week of collecting some interesting gift card data, I started poking around to find potential auction value. More specifically, I wanted to know what types of auctions sell cheap (in terms of total bids) and when they happen. Not surprisingly, the best deals on gift cards occurred early in the morning ( between 2am and 10am). The data was fairly consistent on any given day.

So this is the chart of the day -- Percentage of auctions for $25 and $50 gift cards that sold for less than 20 total bids. The average of those sub-20 bid auctions were typically between 3-6 bids with a standard deviation between 3-6.

AuctionScrape for QuiBids launch!

2012-06-06T14:28:00.001-07:00

Well the day has finally come. This is the initial release of AuctionScrape (for Windows) for mining auction data on QuiBids.

Requirements:
Selenium Server
Java Developer Kit
Python 2.7
-Python Dependencies Selenium and Base Distribution
A fast server with lots of RAM (Tested on 64 bit Windows 2008 Standard Edition configuration. Script does not currently work with Windows XP. Untested on Windows 2003 Server.)
Firefox (on that server)

How to setup AuctionScrape
1. Adjust the data/time on your computer clock to match the real time. AuctionScrape sets up auctions to launch based on the clock in Windows. If it is incorrect, you will get bad data.

2. Download/purchase/setup all of the requirements above.

3. When completed, run the 'setup.py' script in the attached package.

4. Move all of the files in the downloaded ZIP to the AuctionScrape directory.

5. Install the Python dependencies. To get Selenium and the required package urllib (included in the base distribution), you can launch a new command window and type:

pip install -U selenium

pip install -U urllib3

6. Decide what section of the website you want to scrape. It can be specific like $10 gift cards, or it can be general, like all active auctions. By default, the scraper is setup to scrape all gift card auctions. To change this, open auctiongrabber.py and look for this text (there are 2 locations that you will need to change):

#### CHANGE THIS LINE TO ADJUST SECTION YOU WANT TO SCRAPE

sel.open('/categories/Gift-Cards/')

#### CHANGE THIS LINE TO ADJUST SECTION YOU WANT TO SCRAPE

url='http://www.quibids.com/categories/Gift-Cards/' + str(z)+ ''

selurl = '/categories/Gift-Cards/' + str(z)+ ''

In both cases, you will need to change (and match) the section you want to scrape. For instance, if I wanted to scrape $10 gift cards, I would go to quibids.com, select live auctions and click on "$10 Gift Cards". Once loaded, the URL bar would look like this:

http://www.quibids.com/categories/Gift-Cards/10-Gift-Card/

To scrape this section, I would adjust the code in auctiongrabber.py to the following:

#### CHANGE THIS LINE TO ADJUST SECTION YOU WANT TO SCRAPE

sel.open(' /categories/Gift-Cards/10-Gift-Card/')

#### CHANGE THIS LINE TO ADJUST SECTION YOU WANT TO SCRAPE

url='http://www.quibids.com/categories /categories/Gift-Cards/10-Gift-Card/' + str(z)+ ''

selurl = ' /categories/Gift-Cards/10-Gift-Card/' + str(z)+ ''

Be careful not to add any spaces as it will probably crash auctiongrabber.

How to run AuctionScrape:
1. Start selenium server. To do this, you will need to open a command line window and type in the following:

java -jar selenium-server-standalone-2.22.0.jar -singleWindow

If it doesn't work, check if you are in the correct directory with selenium in it and that the name of the application is correct.

2. Launch auctiongrabber. You can do this by command line or in the Python Shell included with Python 2.7. It will setup all the auctions in the task manager.

How to stop AuctionScrape:
1. Close the Selenium Server command window.

2. Run auctionstop.py

3. Close all Firefox and Python windows.

Final notes and comments:
1. Be aware that AuctionScrape cleans itself and backs up files every day at 11:55pm. The files are moved into the "backup" directory. During the daily scraping, the current list of auctions scraped is in "AuctionScrape/Total-Auction-Data.csv". DO NOT CHANGE THE FILE OR OPEN THE FILE WHEN THERE ARE AUCTIONS RUNNING. YOU WILL LOSE DATA.

2. Be aware that when AuctionScrape cleans up the scheduled auctions, it DELETES ALL SCHEDULED TASKS. That means if you have other tasks scheduled to run everyday, they will be PERMANENTLY REMOVED. The reason for this is a glitch in the schtasks command with Windows. I didn't want it to work this way, but it was the only way I could get it to do what I wanted.

3. DO NOT USE WITH WINDOWS XP. It is a known issue that the task scheduler cannot be automated with Windows XP. Unless you like submitting a password for every scheduled auction, do not try to run this on Windows XP.

4. DO NOT SET WORKING DIRECTORY WITH SPACES IN IT. This is another known issue. It will be fixed in a future update. For now, no spaces.

3. Auctions that failed to get setup in the task manager are dumped into "AuctionScrape\FailedAuctionList.csv". If you want to monitor these auctions, you will need to setup the launch time in the task manager by hand.

DOWNLOAD and DONATE
Please note that I spend nearly 150 hours developing this script so that everyone would have equal access to record and analyze data as they please. I didn't put any stipulations on how much/what you can record (like other websites). Use the software as you please. Modify it as you please.

If you appreciate the work that I did and want me to continue developing the software and analyzing the data, please donate. I quit my job to go back to school so I don't have any income coming in for the next year. Any amount will help. Thank you.

Some more auction data

2012-06-06T10:35:00.000-07:00

I realized that the data I provided wasn't actually a complete day.

Here is all recorded auction data for all gift cards for 6/5/12.

Download 6/5/12 Auction Data

The actual script will be uploaded shortly.

Auction Scraper is up and running!

2012-06-03T18:11:00.000-07:00

Today was a milestone.

After working out server host problems (I leased a quad core Xeon at XLHost.com) and several programming problems, I have FINALLY gotten the software up and running.

To test how robust my software is, I tried collecting all 4000 active auctions at once. After about an hour, the server began to seize up with 95 Python windows and 95 Firefox windows. The data recorded from the auctions wasn't good and lacked various pieces of information. In general, it was unreliable.

The next test I ran was for gift cards only. There are still a significant amount of auctions (nearly 2000 listed), but it seems to be running comfortably at about 20 Python and Firefox windows at a time. Nearly all of the auction data is complete.

Not surprisingly, the software uses a lot of bandwidth. I was shocked to see it generate over 20 GigaBYTES of transfer in only a matter of hours. I may need to purchase an additional bandwidth allotment!

You can download a preview output of data you can expect with the auction software here.
(Sorry for the crappy upload host. It was the only one I could use where I didn't have to signup for anything.)

Script finished, crawling imminent!

2012-05-30T13:19:00.001-07:00

With the scraping script complete, I have decided to begin the data scraping process. Due to the nature of how the program scrapes auctions, I need some pretty beefy hardware and connection speed to make it happen.

After a little browsing, I settled on a Windows server from iWeb.com. For $135 a month I get:
-10,000GB of Bandwidth per month
-Core i3 3.0Ghz HT processor
-6GB RAM
-320GB Hard Drive
-Windows Server 2008 Standard Edition

The only thing I have left to do for a script release is write a setup file. Hopefully that will be completed by this weekend.

For those who are curious how it works, I am using Selenium RC to parse the individual auctions and then read specific information (Bidders, Bid Type, and Price) during the course of the auction. I used this instead of BeautifulSoup and other modules because it was the only one that could recover the information after the server side AJAX requests. Once it is completed, it sorts and organizes the auction information into a cleaned up auction and bidder file. The last step adds the final information to an overall summary file. All the files are CSV so they can be easily exported to your favorite data analysis package (Excel, SAS, etc.)

Warning to those who are programmers that look at the code -- please note that I am not a trained programmer. Some of the loops I wrote could easily be condensed, but I don't have the understanding, motivation, or interest in doing so since I am on a tightening deadline.

UPDATE:
iWeb may not have been the best choice of web server. They are located in Quebec so they reach the Canadian version of Quibids.com. I am trying to develop a work around, but Selenium does not want to play nicely with authentication for a proxy. More to come...

Beta Testers

2012-05-16T15:49:00.001-07:00

I am looking for a few people to beta test/improve my script. Ideally, I would like to work with people who have previous python experience, are willing to help document bugs and have the capability to cleanup the code. If you have all of those credentials (or even if you don't but are still interested), please reply to this topic with a way for me to contact you. I would like to start testing this within the next week.

Auction grabber nearing completion

2012-05-14T12:07:00.001-07:00

The auction grabbing script I am using is nearly complete. It grabs each auction that is currently available on the penny auction website and dumps the auction number, auction type, and end time.

From this information, I want to setup a scheduled script launcher for the last 5 minutes of the auction (since typically no one bids until around that time). That would be easy to do on Linux, but it may take a little more effort on Windows.

Here is the typical output that the script generates:

Finalizing python script, data output and personal thoughts

2012-05-11T12:09:00.001-07:00

With the script nearly completed, I still have a few minor decisions to make regarding how data should be written for analysis.

Currently, I have three files for individual auction output. First is the raw, individual auction information. Next is an organized and cleaned up auction list (With Bid Cost, Bidder, and Bid Type). Lastly, I have an output file with the bidders and their respective bid quantity for each auction.

To run an effective statistical analysis, I need to create a summary data file that would have the most important data in it from the auction. This would be information like Auction Number, Auction Item, Total Bid Count, Winner Bid Count, etc. I could create some new metrics like "After Cost Profit", taking into account the bids spend, item cost and shipping cost.

Another item I would like to investigate is the quantity of bidders and their bids. For instance, if a bidder has spent nearly $50 in bids on a $50 item, the likelihood that they will give-up increases with every bid (in theory). Running some analysis could determine the point where most bidders will begin to give up in a bid war.

The most important situation I want to recognize is where low auction selling price occurs. The only buying strategies I have come across on the internet are those where people try to steal an auction at the end of it when the price is already inflated and other bidders are exhausted. Since nearly half of all auctions for low priced items end with low bidding prices (see the charts from my first post), low cost situations is what smart bidders should be looking for.

The last minor aspect I would like to investigate is the relationship between auction selling price and the number of similar auctions ending at the same time. From anecdotal evidence, it appears that when an item is aggressively sought after, bidders will ignore other auctions ending at the same time, even if it is the same item. Recognizing this situation would contribute to the understanding of when lost auction selling price occurs.

Example output of data

2012-05-04T11:29:00.001-07:00

If you are curious, here is the data output you can expect to see with my script. I am still in the development stage, so I may add or remove things that add useful information.

Currently, I cannot dump the bid time of each individual bid (Not sure that it matters, but it may be insightful into bidding psychology). I have considered adding a feature to check if the auction is hidden upon completion of the auction. That categorical data will probably be helpful in regression analysis. I am also considering adding an output feature that would list the bidders with how many bids they have placed over the course of the auction in real-time (as opposed to after the auction ends).

Here is the typical auction output data for the script:

I am also considering adding output that would dump general auction information to a comprehensive auction file (i.e. Auction Type, Total Bids, Bids by Winner, Winning Bid Type, Auction Duration (past anticipated end time), and Total Bidders). That stuff might be useful!

Scraping auctions with urllib2

2012-05-03T20:44:00.000-07:00

After completing the individual auction scraper, I needed to create a script that would scrape all available auctions. To my dismay, there were nearly 200 live auctions in each section. The computing resources required to run each auction simultaneously would be prohibitive.

To mitigate this problem, I came up with a clever method of checking the auction time left with what I decided to be an appropriate starting point. After all, there's no sense recording an auction with 30 hours left when you could get the same data if you started with only 5 minutes left.

Even so, I am concerned about the computing resources needed to record all of the auctions. Right now I am looking at hosting options to run the script and perform the excessive back end resource requirements.

Selenium for scraping in python

2012-05-01T11:09:00.000-07:00

After several days of searching, I found a viable solution for large scale data scraping!

By using a combination of Selenium RC (remote console) and the client version on my local machine, I have put together a working script that will pull data values from the page while dumping them into a CSV file.

The biggest struggle behind the coding (other than having no experience with python) was from the AJAX code on the auction website server. Every time I parsed the page with Beautifulsoup , Scrapy or any other scraping script, it would return null values for the containers I wanted.

By using Selenium, I can automate the process of grabbing info from the page on each refresh. The code itself is fairly simple. The next step would be to write a scraping function (probably using Beautifulsoup since it is easy for static html) to grab all the auctions, then dumping them into a list to be monitored by Selenium.

I have a final exam this week so I don't expect to get much done. But I will be on vacation the next week, so hopefully I can complete the code and start harvesting some rich data sets!