Fantastic work! I found the article (and the one that preceded it) to be exceptionally informative. I understand that these kinds of articles may not be for everyone, but they certainly appeal to readers such as myself.
It's especially interesting to me that we don't know the number of cards in circulation at this point, considering that one of the most influential determinants of card prices is the size and number of print runs. I'll be sure to e-mail Wizards. If that doesn't work, as you pointed out, it may be possible to infer that kind of information.
For the card types analysis (2.2), I assume you ran it with Kamigawa block included. The legendary theme of that block may be an outlier. What is the relative value of the legendary type looking only at other blocks, i.e. those without a legendary theme?
Private Mod Note
():
Rollback Post to RevisionRollBack
- My Powered Cube (draft it here)
- My Full Mirrodin Cube (draft it here)
- My One-Drop Cube (draft it here)
MCC Winner Nov ‘14 & Nov ‘15
I know this is a completely pointless and unhelpful nitpick, but my head almost exploded after reading an article actually written with unabbreviated Latin notes, umlauts, and the æ ligature only to see a your/you're error in the very last sentence.
This is not a useful article without reference to your data and methodology. Where did you get the price data from? What software (and commands) did you use to analyze the data?
I know this is a completely pointless and unhelpful nitpick, but my head almost exploded after reading an article actually written with unabbreviated Latin notes, umlauts, and the æ ligature only to see a your/you're error in the very last sentence.
I am probably every color-combination it is possible to be, though it's really hard to figure out what it would mean to be 4-colored....it doesn't seem logical to be 4-colored without being 5-colored.
Someone posted a link to the Google docs version on another website and I commented over there:
It's definitely interesting as a descriptive thing. Statstalk: I'd squibble about analyzing the entire range of printed cards in one go; I'd argue there are structural breaks as to how pricing works for cards from 1995 and cards from 2012. It would also make sense to treat extremely cheap cards separately, since his data is coming from online stores who must overvalue commons for transaction cost reasons. Plus, the fact that he's using cross-sectional data means he's claiming way more than he can actually say about the effect of reprints or print runs (since print runs are correlated with time). He also runs a few questionable regressions. For example, there are a lot of fishy things going on in Table 8 (if it's a regression, there's no excluded category so he's relying on gold cards to avoid collinearity, which will not be good for his estimates. If it's means or a regression that omits the constant, which would make sense given the coefficients are all positive the significance terms aren't meaningful, and you shouldn't be taking means anyway for such a skewed distribution, which is maybe why those standard errors are so huge for a regression with 18k observations), and he's not correctly interpreting his coefficients in table 9, with the closely related CMC and CCMC.
It's still good to see this kind of thing, at least as a primary step, but what's an applied stats article without a wet blanket?
e: and on the time regression:
he doesn't show his equation, which makes interpreting his regression very difficult to do (the best i can guess is a regression of log price on card age?), and i can't think of a reasonable specification that gets him a 7% annual decrease from those coefficients.
he also talks about inflation which doesn't make sense because he collected the prices all at the same time. he's confusing the issue of card price growth over time with the correlation of prices and a card's age. he also doesn't take into account the standard legality bump he finds elsewhere.
plus his results are insignificant once you drop ABU, so :I
oh he gets the 7% by taking 4% and adding inflation. i also can't see how he's getting 4%, unless he rescaled card age to a 0-1 scale (why?)
also the fact that the coefficient drops when you leave out ABU is suspicious. surely ABU cards are above-average in price? that would suggest that the line would slope down more steeply if you leave them out. did he run the regression backwards?
e2: also, someone else pointed out that your imputed print run for Mirrodin Beseiged is smaller than the actual print run for Ice Age.
I very much enjoyed reading this. Despite the above mention, uh...criticisms(?, not to imply that they are destructive in nature or negate the value of the data itself), I find that most of what I read didn't really shock me. Most of what was found lays fairly well in line with my own view and the view of others that I am aware of (though again, not implying it is a comprehensive view of the community).
On the note of colors, you mention that there does not appear to be a color bias. My question is, by that do you mean:
1) The data does not indicate a large population of players who generally "like" a certain color more.
2) The data does not indicate that any one color appears to be more competitively viable.
3) The data does not indicate that WotC favors the production of more "good" cards in a specific color.
It seems like there is a bias but I'll reserve that to myself as I'm not quite sure how to word it without sounding...bad for lack of a better term.
Also, it would seem to me that another variable should be considered in regards to color: the density of cards with a specific function per color.
(i.e.- Removal vs. Creatures vs Utility Cards vs Planeswalkers)
That would seem to me to help explain why black is valued so high on the list. It boasts some of the most versatile removal which is always valued by players. Its been stated many times on the mothership that their feedback states that players value interaction and black, could be argued, to supply the most of that (depending on definition of interaction etc). I think that this kind of variable could also have an impact on the price by color. If you assume that Planeswalkers are heralded as the most powerful type of card, and thus among the most expensive, and blue has typically had the highest number of successful Planeswalkers it also helps explain why it ranks the way it does in the top 1-3% of expensive cards and yet still not rank as high in the rest of the pack.
Private Mod Note
():
Rollback Post to RevisionRollBack
The following link is an invitation to join Pucatrade (card trading service though similar to TCGPLayer). If you follow the link then it awards me with tokens to exchange for actual cards. Thanks! https://pucatrade.com/invite/gift/86097
Correct me if I'm wrong, but would the high price of legend cards be due to the fact that they are pretty much always rare? Not sure if the 'prices based on card types' was done over all of magic, or just the rares. Just curious!
Correct me if I'm wrong, but would the high price of legend cards be due to the fact that they are pretty much always rare? Not sure if the 'prices based on card types' was done over all of magic, or just the rares. Just curious!
Great duo of articles though! Amazing work.
His commentary on Table 9 suggests he's doing it for all card types, so you're right, that's by far the better inference.
I think it's actually impossible to perform a truly satisfactory analysis of the price premium for legends because there's no control. That's because being a legend is generally a drawback as far as card mechanics go, which should result in the card being more aggressively costed, which should result in an overall relatively more powerful single card than would otherwise be the case. In other words, there are no or incredibly few functional copies of legends, so there's nothing with which to compare them. That doesn't even get to the aesthetics.
That being said, comparing them to all rares would probably be an improvement over all cards. Even then, potential print run variances between sets with different concentrations of legends or power level variances within sets would seem to make anything somewhat questionable.
As far as legends are concerned, the existence of Commander can probably be said to create categorically different price behavior -- certainly for those that have been printed since the rise in popularity of EDH/Commander, in the last... five years or so? Commander is a format where a playset means 1 copy, not 4, which would certainly encourage different buying/selling behavior. The fact that EDH doesn't work the same way formats with GPs and Pro Tours do should create different price trajectories somehow.
There's a lot of theoretical work still to be done here, but this paper could be the start of something brilliant. Please don't sacrifice rigor; some of us really appreciate it.
(There's no need to put accents on "debacle", though; the word has been Anglicized without them.)
I like the rigor of this article series, and how clear you are with your research assumptions and data analysis. It's refreshing to see somebody taking such a data-based approach to price speculation!
One of my concerns about your paper is your metric for card quality. The "cube power ranking" is for a very specific format and covers a minimal amount of cards in existence. It is effectively meaningless here. Card price is definitely driven by card quality, but by card quality in the context of particular constructed metagame; for example, many current commons/uncommons in Eggs and Storm were unplayable before a particular deck allowed them to shine, and Helm of Obedience experienced a huge price jump when Rest In Peace entered its format. I think your analysis is fascinating, but including commons and uncommons in your data set without regards to their playability they see in constructed formats -- which appears to me to be the #1 factor in determining these cards' prices -- seems like an oversight to me, I think.
Also, T2 prices really should be the subject of their own paper, if you want to analyze them further. They require a longitudinal analysis due to their high volatility and the effects of limited play and MTGO redemption.
Private Mod Note
():
Rollback Post to RevisionRollBack
To post a comment, please login or register a new account.
Author of "Flogging the Data"' econometric article series.
It's especially interesting to me that we don't know the number of cards in circulation at this point, considering that one of the most influential determinants of card prices is the size and number of print runs. I'll be sure to e-mail Wizards. If that doesn't work, as you pointed out, it may be possible to infer that kind of information.
Anyway, keep up the good work!
My LinkedIn profile... thing (I have one of those now!).
My research team's webpage.
The mtg-rnn repo and the mtg-encode repo.
- My Full Mirrodin Cube (draft it here)
- My One-Drop Cube (draft it here)
MCC Winner Nov ‘14 & Nov ‘15
I'm Mike, from The Mana Pool.
Check out my Tapped Out profile and comment on my decks!
He's got a point there :/
People need to read this fanfiction, though:
www.hpmor.com
I am probably every color-combination it is possible to be, though it's really hard to figure out what it would mean to be 4-colored....it doesn't seem logical to be 4-colored without being 5-colored.
It's still good to see this kind of thing, at least as a primary step, but what's an applied stats article without a wet blanket?
e: and on the time regression:
e2: also, someone else pointed out that your imputed print run for Mirrodin Beseiged is smaller than the actual print run for Ice Age.
On the note of colors, you mention that there does not appear to be a color bias. My question is, by that do you mean:
1) The data does not indicate a large population of players who generally "like" a certain color more.
2) The data does not indicate that any one color appears to be more competitively viable.
3) The data does not indicate that WotC favors the production of more "good" cards in a specific color.
It seems like there is a bias but I'll reserve that to myself as I'm not quite sure how to word it without sounding...bad for lack of a better term.
Also, it would seem to me that another variable should be considered in regards to color: the density of cards with a specific function per color.
(i.e.- Removal vs. Creatures vs Utility Cards vs Planeswalkers)
That would seem to me to help explain why black is valued so high on the list. It boasts some of the most versatile removal which is always valued by players. Its been stated many times on the mothership that their feedback states that players value interaction and black, could be argued, to supply the most of that (depending on definition of interaction etc). I think that this kind of variable could also have an impact on the price by color. If you assume that Planeswalkers are heralded as the most powerful type of card, and thus among the most expensive, and blue has typically had the highest number of successful Planeswalkers it also helps explain why it ranks the way it does in the top 1-3% of expensive cards and yet still not rank as high in the rest of the pack.
https://pucatrade.com/invite/gift/86097
I'm glad to hear there is at least some interest.
I reran the regression as requested:
With Kawigama block: Legendary = +37% price
Without Kawigama block: Legendary = +55% price
Interpretation: All the crappy legends from Kawigama dilute the value of legendary as a game design tool to moderate the power level of a card.
I like my ligatures and Latin. Like you I also appreciate non-phonetic spelling.
As part 1 states: www.TCGPlayer.com
R
It's pretty messy, but you asked:
Author of "Flogging the Data"' econometric article series.
Great duo of articles though! Amazing work.
His commentary on Table 9 suggests he's doing it for all card types, so you're right, that's by far the better inference.
That being said, comparing them to all rares would probably be an improvement over all cards. Even then, potential print run variances between sets with different concentrations of legends or power level variances within sets would seem to make anything somewhat questionable.
There's a lot of theoretical work still to be done here, but this paper could be the start of something brilliant. Please don't sacrifice rigor; some of us really appreciate it.
(There's no need to put accents on "debacle", though; the word has been Anglicized without them.)
One of my concerns about your paper is your metric for card quality. The "cube power ranking" is for a very specific format and covers a minimal amount of cards in existence. It is effectively meaningless here. Card price is definitely driven by card quality, but by card quality in the context of particular constructed metagame; for example, many current commons/uncommons in Eggs and Storm were unplayable before a particular deck allowed them to shine, and Helm of Obedience experienced a huge price jump when Rest In Peace entered its format. I think your analysis is fascinating, but including commons and uncommons in your data set without regards to their playability they see in constructed formats -- which appears to me to be the #1 factor in determining these cards' prices -- seems like an oversight to me, I think.
Also, T2 prices really should be the subject of their own paper, if you want to analyze them further. They require a longitudinal analysis due to their high volatility and the effects of limited play and MTGO redemption.