Thanks for sharing, Andrew. I enjoy getting the insight into the underlying model, and I like what you've done here.
I do wonder if the regression is struggling with fit on this pretty high variance dataset. I only had time to look at the age 0 GKs, but what I see is below. These are counts of how many auction buys fall in how each of our value bins. S50+ indicates the counts from only the past 13 seasons:

So 0/x GK might struggle in the model as an edge case, but it shows under-pricing bias that gets worse when we look at only more recent pricing (as I think money has gone up in SESL, leading to a climb in auction prices). 95% of auction wins are outside of the middle pricing bands all-time, with all of them landing as either excellent or very expensive when looking only at recent seasons.
Given the amount of data we have, I wonder if a percentile-based approach could help with prediction accuracy. We are likely relying on too sparse of data if we try to segment completely on position/age/SL. But I do wonder if grabbing weighted percentiles within the position/age combination and then weighting by both recency and SL distance would give fairly accurate bands. I think we would struggle with UT data density somewhat with that approach, but I think it would work with the rest of the position/age combinations.

Bournemouth 2 - 2 Liverpool

Brentford 2 - 1 Chelsea

Brighton 1 - 2 Arsenal

Everton 1 - 0 Ipswich Town

Fulham 1 - 2 Man. United

Leeds United 2 - 0 Crystal Palace

Man. City 3 - 0 Sunderland

Newcastle 1 - 2 Hull City

Nottingham Forest 3 - 0 Coventry City

Tottenham 0 - 0 Aston Villa

First of all, very cool app, Andrew! I love it.

One question:
- I assume the guide price is a median/50th percentile price? And the value ranges are other percentiles? Is that correct, or are they calculated some other way?

Comments:

- If that is accurate, I wonder if it would make sense expand the percentiles somewhat. A range of 330k-480k for an 0/8 GK seems too tight for me

- In the same vein, I wonder if recency weighting would be helpful here. These prices haven't been seen in some time.

- On the topic of accuracy...I see 16 0/8 GKs having been sold at auction. Of those 16, 4 are in the range given, and the other 12 GKs that have been sold would all be catigorized as 'Very Expensive'. If I set percentiles of P10, P25, P50, P75, and P90 then I calcualte the range for an 0/8 GK as approximately:
240k

371k

807k

912k

1110k
 

So, given the input data of player auction prices sold, I find it hard to see how we got the prices of 330k to 400k to 480k

Cameron - you can go to 'Features' -> 'Player Search' -> 'Player Value Guide' and you will be able to search by age, SL, and position to see historic auction pricing. Good for seeing the wide historic range, but also you can key in on only recent seasons to see the spread.

Last youth auction can be pretty swingy. Some players might go for very high fees when multiple teams are desperate for that position. And then others might end up at 1k. Hard to predict.

Several very good deals won in the auction

...MOT not part of any of them. Instead paying probably a record fee for an 0/8. Paying UT premium for a GK. Only for the exact same player to show up the next auction. Whoops!

Aston Villa 2 - 1 Nottingham Forest

Bournemouth 1 - 2 Brentford

Chelsea 4 - 1 Hull City

Coventry City 1 - 3 Brighton

Crystal Palace 2 - 1 Ipswich Town

Leeds United 1 - 1 Newcastle

Liverpool 3 - 2 Fulham

Man. United 1 - 3 Man. City

Sunderland 0 - 3 Arsenal

Tottenham 1 - 1 Everton

Could I please change my captain to Gary Clark Jr (5/19 DM)? Thanks