Analysis of the determinants of selling price for Vancouver NHL ticket auctions

Introduction

I’ve been scraping a lot lately. If data is the new oil, then it’s time I started pumping it. I came across this old project of mine from a statistics course in my masters program, dated December 15, 2006. Stubhub today should be a goldmine of data and I hope appropriate data and/or analysis is being passed to the leagues to help them set prices.

Analysis of the determinants of selling price for Vancouver NHL ticket auctions

Data was collected for a one month period of auctions completed on eBay, a popular internet auction site, for tickets to see the Vancouver Canucks play at home.  The data was analyzed to determine a good regression model for the selling price per ticket. The most significant indicator was face value, but auctions by sellers with higher feedback scores and transactions that are completed longer before a game also finished with a higher price per ticket. Noteworthy insignificant factors were number of tickets in a lot, the feedback percentage rating for the seller, and the length of the auction.

 

Findings

My most interesting finding would provide guidance to price setting.

Of most interest is the regular game Upper Bowl IV and Upper Bowl V price categories. The eBay market price difference from Ticketmaster is significant and in the case of Upper Bowl V seats, sizable. This indicates that these tickets are priced well below market demand. It would seem that from the consumer’s perspective Upper Bowl IV and V are indistinguishable and the Canucks should seriously consider charging more for these seats.

I also found that:

  • Tickets sold further ahead of game day sold for more. There is a fundamental tension in the market for tickets, a time-valued good (after game day a ticket is worth zero). Some shoppers are will pass on an auction hoping to find a better deal later. Some shoppers are willing to pay more for the certainty of having a ticket now. Some sellers want to unload their tickets now before they become worthless. Some sellers will patiently wait out the buyers in order to fetch a higher price. In my data there was a correlation of 0.3 between minimum bid and time until game day. This is evidence  of patient sellers setting and getting higher prices further ahead of game day.
  • Feedback rating was a more important factor than feedback percentage. In the marketplace that I studied where most operators were small-scale with 100% positive feedback, their tickets sold at a higher price if they gathered more positive feedback. On the flip-side, big-time operators with only a little less than 100% feedback did not suffer greatly.
  • The number of tickets in a lot did not affect selling price per ticket. You would expect four tickets together to be worth more than two separate pairs, and surely they are, but there was not enough evidence in the data, so it is not a strong impact.
  • Length of the auction did not have an effect. Longer auctions presumably get seen by more people and therefore sell for more, unless the market is active enough such that a critical mass of buyers will see auctions of short length.

How I Did It

  1. A scraper was written in C# to collect html pages from eBay of completed auctions for Vancouver Canucks tickets
  2. Extensive regular expression matching was done on the auction’s free text in order to determine what game the tickets were for, how many tickets there were, and other factors like where the seats were in the stadium
  3. Regression models were tried in order to find a good fit
  4. Regression coefficients were interpreted to draw conclusions

Average shot locations in the paint and at midrange

Yet another post on visualising and analysing NBA shot location data using location averaging methods.

Previously I have shown averages by team for all shots taken. What about shots taken by zone? Consider the following two charts:

avg_inthepaint_teams

avg_midrange_teams

Now we can see a level of detail that we couldn’t in the average of all shots.

Previously:

for_and_against_640w

Observations:

  • Previously we saw that the GSW were the longest shooters in the league and indeed they were also long shooters in the paint and at midrange. It is not simply that the GSW take a lot of 3s.
  • Other teams like the NYK take shots from close up in both the paint and at midrange. They were generally longer shooters in the previous analysis, suggesting that they balance those close 2-pts with many 3-pt attempts
  • Previously we saw DEN as the closest shooters overall, and indeed they appear to be close shooters in the paint and moderately close shooters at midrange.

NBA players shoot from further as the game progresses

On average, players in the NBA take shots 6.5% further from the basket in the 4th period than in the 1st. This is a subtle, but consistent trend across all periods. 1.2% further in the 2nd than the 1st, 2.4% further in the 3rd than the 2nd, and 2.7% further in the 4th than the 3rd.

Most teams show the same trend, consider the graphic Average Distance of Shots by Period below:

avg_distance_period_allteams

 

Nearly all teams show an increasing distance by period, but there are some notable exceptions. A few teams like NYK, UTA, ATL, and ORL show an opposite trend.

It’s not clear at this point what is leading this. Greater defense in later periods forces further shots? A greater need for 3-pts pushes shots away from the basket? A greater need for points in less time forces less ideal shots from further out? What about the exceptional teams where they get closer? What are these teams doing in the 4th quarter? Leaning more heavily on particular strengths or players?  Tired players not driving to the net?

What is true at the global level and largely true at the team level is again reflected in the players. Here are the average distance for the top 100 scorers:

avg_distance_period_100players

More noise, but the trend is still visible!