Pertinent information about predictive pricing for PSA graded cards.

The sports card and memorabilia market is on fire.
The lockdown of 2020 and a $3 Trillion stimulus package led investors to seek safe havens for their cash. The perception of runaway money printing by the Fed has shone a major spotlight on collectibles.
Due to their inherent scarcity, collectibles have become the preferred investment of those looking to park money in something truly limited and secure. Time will tell if those decisions were prudent.

For the past several months, we’ve performed an intensive dive into predictive modeling for collectibles. Starting with graded basketball cards. Watching prices of a PSA 10 Michael Jordan Rookie card (1986 Fleer) explode from $35,000 to $100,000 in a matter of months had us exploring if these card movements can be predicted. Are there any variables that greatly affect the selling prices of basketball cards. If so, what are they, and how big of an effect do they have on the final sales price?

To what degree to those underlying conditions and variables explain the price movements?
Those are the questions we set out to answer.

We started by importing and analyzing just under 100,000 graded basketball cards. We used graded cards because each item is practically indistinguishable from another if graded the same (fungible) and the volume of transactions gave us a large enough sample to run our models. These ~100,000 cards were from a subset of 150 players that see significant sales transactions

These transactions were also across various auction houses ranging from Heritage Auctions, Robert Edward Auction, SCP Auctions, eBay, Mile High Auctions and more. One thing we explored was the impact certain auction houses and sellers have on final bid prices. Does seller feedback and/or transaction volume push prices one way or another? That’s something we’ll share in a later report.

The following is a very simplified summary of what we were able to create.

We are going to discuss just one model:
What macroeconomic factors push prices of sports collectibles? Can one predict future movements based on the economy (stock market, unemployment, money supply, volatility indices etc.)?

The best model we created used over 100 economic factors, a summary list is shown later below.

Once we attributed the correct ‘weight’ to each factor, we were able to predict the next sale of an identical sports card with a staggering accuracy. We found that up to 95% of the price of these cards could be directly attributed to the economic variables we gathered and adjusted. Put another way, 95% of a card’s price can be explained by further analysis of these underlying variables.

Our predictions for lower priced cards are spot on. After cards crossed the $10,000 threshold, predictions were harder to make. We know why. A hint on that at the end of the report.

What were these underlying variables? A sample of the 100 is shown below:

This image has an empty alt attribute; its file name is economicvariables2.png

Small sample of the economic variables we used.

From this data, we were able to predict the prices with a staggering accuracy. A very small sample is shown below. Our ‘next sale’ prices were extremely accurate over the entire 100,000 card database.

Actual sales prices are in the first column and then our price predictions for that sale are to the right.

Using these predictive models, we’re able to assist collectors with real-time pricing of their assets. What is my card worth today… right now… this second? If I sold my card today, what would it bring? Instead of searching past auction sales and estimating a price — we could provide this to you instantly.

Our very best model, and the variables we gathered, explained the next sales price with 98% accuracy. We will share this with you soon.

In a future report, we’ll explain other sets of variables and their impact on pricing. For example, what about player stats? We explored 200 different stats and ran models to determine if their career stats influenced pricing.

This image has an empty alt attribute; its file name is stat2.pngThis image has an empty alt attribute; its file name is stat1.pngThis image has an empty alt attribute; its file name is stat3.pngThis image has an empty alt attribute; its file name is stat4.png

We took each player and created a profile with stats – hundreds of line items… We’ll explore this in the future.

We’ve used dozens of variable categories to create a real-time snapshot of the market. This is the March Index… AND it goes beyond graded cards. It encompasses the entire sports memorabilia market – and well over 10,000,000 analyzed transactions. It’s constantly learning and updating, so check back often!

Our predictive models can give guidance on the future movements of our indices. An early model of a player index with predictions, is shown below. Our models can predict future movement of player collectibles — as this model from mid-2020 estimates Ken Griffey Jr. items increasing in value for 2 consecutive months:

In creating an index for each player, we assigned a value to each athlete and made an index with this information in order to track movement and interest in specific players.

We would love to hear from you.
Is this deep analysis beneficial to you?
Is our index beneficial to get a quick glance of the health of the collectibles market?

Reach out to us with your input!


Our market outlook for sports memorabilia is as follows: We found that many sports collectibles are negatively correlated to equities markets. If the stock market sees a sudden drop, expect more money to flow into scarce collectibles. An early version of our index, compared to the NASDAQ, is seen below.

See the two major NASDAQ dips — and how the March Index rose? We think that trend will continue.
Additionally: lower priced items are easy to predict. Frequency of transactions and the fact that these lower items are swapped like commodities makes them modelable. The higher end of the market, and the human emotion that accompanies winning an auction, takes our model off the trend line just a little bit, as you can see with our scatter plot image toward the top of this article.

Our index construction and methodology is proprietary. We aim to share as much information with you as possible. We have several algorithms that we created that run 24/7 to create the live index. We have a lot of work to do. What you see on is V1. More to come.

Thank you!

Please share via social media if you found this interesting!

Leave a Reply