Overpriced/Underpriced Unusual Price Analysis (Hats & Taunts)

[ocu.TF ♥ TWP]

I've been thinking of publishing this for a long time now. I hope it helps the community. Enjoy.

 

 

Short version:

 

I made a fancy plot. Plot shows overpriced / underpriced unusuals. Plot iz gud. This iz plot:

 

 

This post is not for you if you're looking for a short version, you need to read and try to understand what's going on here before you actually use it.

 

The image above is a screen shot of the real thing. Please access the interactive version below (otherwise it's useless)

Interactive version: http://ocu.tf/apache/UnusualPrices.html

NOTE: Due to the complexity of the plot, it may take a while to open in your browser.

 

What is it for?

• You want to make a price suggestion and want to know what some of the most outdated / overpriced / underpriced hats are.
• You want to make a price suggestion, but there are multiple sales. Which ones are outliers?
• You’re going to buy an unusual but you’re not sure if its current price is right, especially if it’s outdated.
• You own a trading site and you don’t want to get spammed by overpriced unusuals (you know who).

I must admit it’s a bit hard to use this plot for trading purposes (because you have to find the hat that you’re interested in), but I hope it will still be somewhat helpful. I intend to make a price checking service later on, using ocu.TF .

 

Understanding the plot:

 

For the purpose of this post, an unusual hat-effect (or taunt-effect) combination (e.g. team captain – burning flames) is referred to as an item.

This plot shows whether an item is overpriced or underpriced based on nonlinear regression predictions.

• Y-axis: Low-end prices of every item in keys.
• X-axis: Predicted prices (see below for how this prediction is done) in keys.
• Pink line: Bp.tf price = Predicted price (this is where every item would be in a perfect world)
• Colors: Clustering based on a modified Absolute Percentage Error, calculated as (Actual-Predicted)/Predicted x 100 (yellow=good, grey=overpriced/underpriced).
• Sizes: How outdated the item is (bigger = more outdated)

Insight:

• Predictions are based on all items combined (exclusions apply, see below). They can be considered as the “correct†price of an item, given all (or most) of the variation in item prices are explained by the features explained below.
• Items (points) above the line are currently overpriced. Items below the line are currently underpriced according to this analysis. The farther a point is from the line, the more overpriced/underpriced that item is.
• If you are an unusual price suggestion maker, go for the big grey points. Once they’re done or you’re bored, go for the red ones, and so on.
• If you’re buying an unusual that is overpriced, try to not to pay more than the closest purple point directly below the item you’re interested in. Or pay even lower than that to be safe.
• Likewise, if you’re selling an underpriced unusual, you may want to ask for an increased (purple points above, or more).
• Predicted prices are not quicksell prices. You may need to apply further discounts to these.

Updates:

 

Each time an unusual is updated, the predictions will be effected (therefore this analysis can become obsolete).

• Always check back for the most up-to-date version. So if a hat is overpriced according to this plot right now, it may not be overpriced anymore 1 day later if “similar†hats have new price suggestions, even if that particular hat doesn’t get updated.
• For the same reason, please don't make posts about how inaccurate* this plot is. It will change again soon and your comment will confuse other people.

I will update this plot weekly, or whenever I get a chance. If it ever becomes too out of date (I would say a month or so), do not use this plot anymore.

 

You can manipulate the plot:

• Move your mouse over different points to see the item information. The numbers at the top in parenthesis represent (Predicted, Actual) prices.
• Draw a rectangle around a region of interest to zoom in. You need this to see the cheaper unusuals.
• Double click to reset zoom.
• Click on certain groups on the legend (top-right corner) to disable them. This will speed up your browser as well. You can disable yellow, blue and green ones for example.

 

How it’s made:

 

I am not going to get into math / statistics / computer science part of the result as this is not the place for that. However, here is some brief info:

 

Each item has a set of features that are related to it. These features include, but are not limited to (for proprietary reasons):

• Item’s name (e.g. team captain)
• Item’s effect name (e.g. burning flames)
• Item’s current low-end bp.tf price in keys (e.g. 8000, wow that’s expensive).
• Classes that can wear this item
• Whether the item is a cosmetic or taunt
• Whether the effect is Halloween or not
• And more

Based on these features, and existing prices, a nonlinear regression method is used to predict the items’ prices.

 

I purposefully did not use last update date as a feature because “wrong†prices could have been submitted anytime. Instead, it is indicated in the plot itself.

 

Exclusions: The data excludes all Unusuals with prices higher than 1000 keys. Only low-end prices are considered.

 

But TWP, why should I trust this?

 

This is a statistical result, it does not have to be 100% correct. It is meant to serve as a guide and should be taken with a pinch of salt. It does not capture themes, collector behavior (it also cannot cook food that you like, which is equally hard to accomplish). Use it at your own risk. That being said, I have a PhD in this field.

 

Disclaimer:

 

I own ocu.TF. The data and analysis shown here are not biased in any way, however, it is not guaranteed to be correct.

 

We will also be using similar approaches on our website as a base for our unusual pricing.

 

 

Changelog:

• 5/11/2016: Changed the plot to show Absolute Percentage Errors (formula above) so that error tolerance is smaller for cheaper hats, and larger for more expensive hats.

[JayTuut]

This is kinda cool, but all i see is like a graph showing the most expensive and least expensive, is that it?

[♠Derpeh♤]

This is pretty cool

 

Worth pointing out that most of the outliers are "themed", 1/1 unusuals or unusuals which havent seen the daylight for years xD

[E-Arkham]

Is there a way to search for a specific hat or combo to see where the chart places it?

 

It'd be great if you could also view all hats by effect.  For example, searching specifically for just Haunted Ghosts only.  Or even just being able to toggle effects by category (1st gen, 2nd gen, halloween, etc).

[ocu.TF ♥ TWP]

This is kinda cool, but all i see is like a graph showing the most expensive and least expensive, is that it?

 

Thanks. No, there's much more information in the graph, assuming you're looking at the correct one.

 

 

 

This is pretty cool

 

Worth pointing out that most of the outliers are "themed", 1/1 unusuals or unusuals which havent seen the daylight for years xD

 

Thanks. You can look at red, purple and blue ones as well.

 

 

 

Is there a way to search for a specific hat or combo to see where the chart places it?

 

It'd be great if you could also view all hats by effect.  For example, searching specifically for just Haunted Ghosts only.  Or even just being able to toggle effects by category (1st gen, 2nd gen, halloween, etc).

 

I cannot implement searching here, we will add that to ocu.tf soon. But 'll see if I can generate a few more plots that may show what you're looking for.

[λngelღмander]

This is great man. Thanks for being a part of this, it could be a useful tool for suggestors.

[ocu.TF ♥ TWP]

This is great man. Thanks for being a part of this, it could be a useful tool for suggestors.

 

Thank you. That's the goal.

[ocu.TF ♥ TWP]

Data & model updated 5/16/2016

[Jackk]

Tldr maybe?

[77777777777777777777777777]

Nice job

[E-Arkham]

Tldr maybe?

This post is not for you if you're looking for a short version, you need to read and try to understand what's going on here before you actually use it.

My guess is "no."

[ocu.TF ♥ TWP]

 

My guess is "no."

 

 

Your guess is correct

[Apeiron]

Good study.

But the interactive version looks like hard to search. 

[ocu.TF ♥ TWP]

Good study.

But the interactive version looks like hard to search. 

 

Thank you. I did not really make it to search a particular hat, it's the other way around. I made it to show the hats that are overpriced / underpriced.

 

However, you can still find the information you're looking for. If you know the low-end backpack.tf price of the hat you're searching for (you should), then zoom in to a horizontal cross-section of the plot near the backpack price range, and you can find the hat you're interested in.

 

---

 

Also, data & model updated 5/17/2016.

[ocu.TF ♥ TWP]

Data & model updated 5/26/2016.

[ocu.TF ♥ TWP]

Data & model updated 6/9/2016.

[A Delicious Cashew]

Is this really the most overpriced, or am I missing another one? Closest one I got is the stare from beyond cotton head at 133%, but that technically should be lower since not all of the effects have been updated since it went all class. Not surprising actually, considering burning is 350 and scorching is 200. 

 

Great, so I'm waiting for "the most overpriced hat + effect combo" to be unboxed, so I can buy it.. 

[ocu.TF ♥ TWP]

 

Is this really the most overpriced, or am I missing another one? Closest one I got is the stare from beyond cotton head at 133%, but that technically should be lower since not all of the effects have been updated since it went all class. Not surprising actually, considering burning is 350 and scorching is 200. 

 

Great, so I'm waiting for "the most overpriced hat + effect combo" to be unboxed, so I can buy it.. 

 

 

Based on the data at the time of this post and according to my calculations, yes the Circling Heart Head Warmer is the most (by percentage and absolute error) overpriced hat. The estimated value (despite it being an outlier) is 320.513 keys. But remember, this analysis does not include hats priced more than 1000 keys.

 

Also, I did not update some of the recently updated (all-class enabled) hats in the calculations (eg cotton head), so prices on those may be less accurate. I could use some help tracking them actually.

[=TNW= Maximus™]

How did you create the predicted values plot? (what is it based off of)

[ocu.TF ♥ TWP]

How did you create the predicted values plot? (what is it based off of)

 

I thought I explained it in the first post. Do you have a more specific question?